A calibration program that reports 2 % probability of false accept sounds reassuring, until you realize that number is an average across every instrument in the population. It says nothing about the measurement result sitting on your bench right now.

This is the distinction between global risk and specific risk, and confusing the two is one of the most consequential mistakes in measurement decision-making. Global risk tells you how a process performs over time. Specific risk tells you whether this particular result, at this particular point relative to the tolerance limit, is trustworthy. They answer fundamentally different questions, and treating one as a substitute for the other leads directly to conformity statements that overstate confidence.

The difference is easier to grasp than most people expect. Think of a field goal kicker in American football.

Figure 1 Specific Versus Global Risk

Global (Unconditional) Risk

Global PFA asks: “Across all the instruments we calibrate, what percentage of truly bad ones do we accept?”

This is an average across your entire population of instruments and calibration events. It depends on two things: your historical reliability data (how often do instruments actually drift out of spec?) and the uncertainty of your calibration process. You must know how your instruments behave over time to properly calculate global risk.

In programs that contractually invoke ANSI/NCSL Z540.3 [1], a common target is controlling consumer’s risk (PFA) for acceptance decisions to ≤ 2 % under the specified decision rule. This is typically framed as an unconditional, population-level requirement. Achieving that target generally requires not only a measurement uncertainty model but also a defensible model for the population of items being calibrated (i.e., historical records showing how often instruments of that type actually drift out of tolerance between calibrations), meaning prior information supported by reliability and interval data such as end-of-period reliability (EOPR) or similar historical evidence.

Where such data are limited, some Z540.3 [1] handbook approaches use managed guard banding to control reported risk under stated assumptions, at the cost of increased false rejects, without claiming full population characterization.

The catch: an instrument that passes a 2 % global PFA requirement can still present much higher specific risk on individual measurements near the tolerance boundary. The global number is an average; it masks what’s happening at the edges. ILAC-G8:09/2019 [2] makes clear that an instrument passing a global risk criterion of 2 % PFA may have a specific risk for false acceptance that can be as high as close to 50 % as shown in the previous section.

Specific (Conditional) Risk

Specific PFA asks: “Given this particular measurement result, and given my uncertainty, what’s the probability this specific instrument is actually out of spec?”

This is a conditional probability; it depends on where the observed value landed relative to the tolerance limits. A result right in the middle of the tolerance range? Very low specific PFA. A result right at the edge of the acceptance limit? The specific PFA can be substantial, even if the global number looks fine.

Specific risk is the right framework when you don’t have prior population data, what JCGM 106:2012 [3] (clause 7.2.2) calls ‘meagre prior information,’ or when a more cautious approach to false accepts is required, regardless of available data.

Figure 2 Field Goal Example

Specific (Conditional) Vs Global (Unconditional) Risk, A Simple Example: The Field Goal Kicker

In American football, a field goal kicker either kicks the ball between the goal posts or does not.

Specific risk is the single kick in front of you right now. The ball either traveled between the goal posts or it didn’t. Pass and points or fail and no points for your team (sorry, Ravens fans). When we make a measurement and record the observed value, it is either between the acceptance limits or it is not. One kick, one result, one decision.

Global (average) risk steps back and looks at all the kicks. Say the Steelers kicker makes 97 % of his field goals from the 35-yard line. This kick is from the 35-yard line, yet it didn’t go in. Based on population data alone, you would say it is likely that it did go in. And the Steelers won the game, hooray!

That is exactly the tension between the two approaches. Global risk uses historical population data to assess the probability of an incorrect decision, such as a false accept or false reject, across a population. Specific risk evaluates the single result in front of you. One tells you how reliable the process is over time. The other tells you what happened right now.

Now shift your focus from the kick to the goal posts themselves. The distance between them represents your tolerance. A kicker with a tight, repeatable trajectory effectively has more room to succeed because their variability is low. In contrast, a kicker with inconsistent performance needs nearly perfect alignment every time. The same principle applies to measurement: as uncertainty decreases, the “usable” portion of the tolerance increases, and confidence in the result improves.

With that perspective in mind, decision rules become the method for determining whether a result truly falls within those goal posts. Most practical decision rules stem from three foundational approaches, each offering a different way to balance risk and confidence.

If you found this breakdown useful, share it with your colleagues. The gap between global risk and specific risk is one of the most misunderstood concepts in measurement, and the consequences show up every time someone treats a 2 % population average as proof that an individual result is trustworthy. The more people in your organization who understand the difference, the better your conformity decisions become.

Share this article with anyone who writes decision rules, reviews calibration certificates, or signs off on pass/fail statements. These are the decisions that determine whether your measurements actually mean what you think they mean.

[1] ANSI/NCSL Z540.3-2006, Requirements for the Calibration of Measuring and Test Equipment. American National Standards Institute / National Conference of Standards Laboratories International.

[2] ILAC-G8:09/2019, Guidelines on Decision Rules and Statements of Conformity. International Laboratory Accreditation Cooperation.

[3] JCGM 106:2012, Evaluation of measurement data — The role of measurement uncertainty in conformity assessment. Joint Committee for Guides in Metrology.

I’ve worked on this topic for many years and continue to teach, refine, and learn. This article reflects that ongoing process and is intended to clarify what measurement risk is, how it shows up in practice, and why it matters.

Measurement risk is often misunderstood because we rarely discuss it as part of the full measurement system. In conformity assessment, calibration, and testing, the conversation can narrow to the numbers—uncertainty values, probabilities, and whether a particular statistical method was used. Those details matter, but when they become the whole discussion, the real drivers of risk get pushed aside.

In practice, measurement risk rests on three equally important foundations: the requirements being applied, the equipment used to make the measurement, and the process used to perform it. Scott Mimbs described these as a three-legged stool. The system is only stable when all three legs are sound. If any one leg is weak—unclear requirements, incapable equipment, or a fragile process—the decision becomes unstable, no matter how sophisticated the math looks.

ISO/IEC 17025 is consistent with that reality. The standard does not require laboratories to calculate probabilities, apply guard bands, or adopt a particular risk model. Instead, it expects laboratories to understand and manage risk the way risk is commonly understood: the likelihood of a wrong decision and the consequences if that wrong decision occurs. The standard leaves the “how” flexible on purpose, because no single prescribed method can fit every application.

At its core, measurement risk is not a statistical obligation; it is a decision-making responsibility. Organizations must decide what level of uncertainty is acceptable, what happens when a decision is wrong, and who ultimately bears the risk. Those choices are shaped not only by technical capability, but by safety, cost, reputation, and business objectives.

Why Measurements Exist—and Where Risk Enters

Measurements are not made for their own sake. They exist to support decisions, such as:

· Accepting or rejecting a product

· Releasing or stopping a process

· Declaring something safe—or unsafe

Every measurement-based decision follows the same fundamental path:

Measurement → Decision → Consequence

Most of the time, the measurement itself is not what creates harm. The harm shows up after the decision, when consequences materialize—sometimes immediately, sometimes much later, and sometimes undetected for years.

When a measurement cannot reliably support the decision it is meant to inform, the risk doesn’t disappear. It moves downstream. Downstream risk is usually larger, more expensive, and harder to correct than addressing the issue at the point where the measurement is made.

Why Measurements Are Never Exact

Every measurement carries uncertainty. Even with capable equipment and skilled people, there is always some doubt about how close a result is to the true value. That isn’t a defect in measurement; it’s a fundamental reality of measurement.

The problem usually isn’t uncertainty itself. The problem is making decisions without understanding what the uncertainty implies.

When results fall well inside specification limits, uncertainty may have little impact. Yet when results fall near a limit, uncertainty dominates the decision. This is precisely why decision rules exist—not to eliminate uncertainty, but to define how much uncertainty is acceptable given the consequences of being wrong.

Understanding Risk: Likelihood and Consequences

Across engineering, safety, finance, and project management, risk is consistently defined as a combination of two factors:

· Likelihood – How likely is a wrong decision?

· Consequence – What happens if that wrong decision occurs?

ISO/IEC 17025 follows this same, well-established definition. When the standard requires decision rules to consider the level of risk, it is referring to both elements.

One point that often helps: risk is not the same thing as probability. Probability is one way to express likelihood, but it is not always necessary or even useful, and ISO/IEC 17025 does not mandate that likelihood be quantified as a numerical probability.

That flexibility matters because likelihood can be managed in practical ways without ever calculating a probability—reducing uncertainty, improving equipment capability, applying conservative acceptance criteria, or using an agreed decision rule that fits the application. These approaches still manage risk, even when no probability is computed.

The Two Decision Errors That Drive Risk

Every measurement-based conformity decision can be wrong in two ways. When quantified, these are often expressed as the probability of false acceptance (PFA) and the probability of false rejection (PFR). The tradeoff exists even when you never calculate probabilities.

False acceptance risk

Accepting something that is actually nonconforming. This is typically the higher-consequence error and can result in:

· Safety incidents

· Product failures

· Recalls

· Legal liability

· Loss of trust

False rejection risk

Rejecting something that is actually conforming. False rejects usually do not endanger safety, but they can cause:

· Scrap and rework

· Delays

· Higher operational costs

· Customer frustration

Every decision rule represents a tradeoff between these two risks. Reducing one almost always increases the other. There is no “zero-risk” option—only informed choices.

Note: false rejection risk only applies when you issue a nonconforming (“fail”) statement. If you are not making that kind of statement, false rejection is not the risk you are managing at that moment.

What ISO/IEC 17025 Actually Requires

ISO/IEC 17025 requires laboratories to define and document a decision rule when making statements of conformity. That decision rule must:

· Take measurement uncertainty into account

· Consider the level of risk associated with the decision

· Be communicated to and agreed upon with the customer (when not inherent in the specification)

That is the requirement.

Just as important is what the standard does not require. ISO/IEC 17025 does not mandate guard bands, probability calculations, PFA/PFR calculations, or any specific mathematical method for risk evaluation. In fact, the word “probability” does not appear anywhere in ISO/IEC 17025, underscoring the standard’s intent to allow multiple valid approaches to managing risk.

Probability-Based Methods Are Optional Tools

Some organizations choose to calculate probability-based metrics such as PFA and PFR. These methods can be valuable, particularly when consequences are severe, regulatory frameworks demand quantified risk, or statistically meaningful data exists.

Value does not equal requirement. Some standards or customer specifications may set explicit probability targets (for example, limiting false acceptance to a defined threshold), but that is not a default expectation under ISO/IEC 17025.

Many organizations manage measurement decision risk effectively without probability calculations by controlling measurement uncertainty, applying minimum capability ratios such as TUR, and aligning acceptance criteria with real-world consequences.

Decision Rule Guidance and Risk Ownership

One point emphasized in the Decision Rule Guidance (Zumbrun, Cenker, Shah) is that measurement risk isn’t something the lab can “solve” on its own. The lab can quantify uncertainty and apply a decision rule correctly, but it can’t determine how much risk is acceptable in the real-world use case.

Only the organization relying on the measurement can make that call.

ISO/IEC 17025 recognizes this explicitly. When a statement of conformity is required, the laboratory must apply a documented decision rule and take uncertainty and risk into account. Yet unless the decision rule is inherent in a referenced specification, the customer must define—or explicitly agree to—the decision rule being used. This is not a paperwork exercise; it is a risk ownership decision.

Organizations that fail to understand their own requirements often default to whatever decision rule is offered, without realizing they may be accepting far more risk than intended.

The Hidden Risk of Simple Acceptance

The Decision Rule Guidance document calls out a common and potentially costly misconception: that simple acceptance (declaring a pass whenever the measured value falls within the tolerance) is a neutral, low-risk choice. It might not be.

With simple acceptance, the decision boundary is the specification limit itself. When measurement uncertainty is large relative to tolerance, results near that limit can carry a specific risk of false acceptance that becomes surprisingly high. In common scenarios, the guidance indicates that this specific risk can approach 50%, meaning a result that appears to “pass” may be no better than a coin flip in determining whether the item is truly conforming.

In practical terms, this means:

· A result can look like a clear “pass” while still having a substantial chance of being out of tolerance.

· Customers may unknowingly accept a much higher level of risk than intended because the decision rule was never made explicit.

· Downstream consequences—failures, rework, recalls, disputes, or safety concerns—become more likely.

The takeaway is straightforward: accepting risk by default is still accepting risk. When a decision rule or risk requirement isn’t defined, risk isn’t eliminated; it is implicitly shifted to whoever ultimately bears the consequences.

Note: the above figure shows only one probability density function, indicating that the end-user is concerned with an instantaneous liability and does not consider the population. It is important to note that when using global risk models, the 50% risk could be much lower. In some cases, it might even fall below a PFA of 2 %.

With that in mind, the Decision Rule Guidance document expands the discussion beyond specific (single result) false acceptance to a broader view of how risk is created, interpreted, and allocated across an operation, whether you’re making one decision at the bench or managing a stream of decisions across an entire product population.

Risk Exists at Multiple Levels

Another critical insight from the Decision Rule Guidance document is that measurement risk is not confined to a single reported result. Risk shows up at multiple levels, and organizations need to understand each one to manage it intentionally rather than accidentally.

Specific Risk (also called bench-level risk) is based on a specific measurement result.

• It triggers a response based on measurement data gathered at the time of the test.

• It may be characterized by one or two probability distributions, depending on the method.

• Any representation with only one probability distribution is always a specific risk method.

Global Risk (also called process-level risk) is based on a future measurement result.

• It is used to ensure the acceptability of a documented measurement process.

• It is based on expected or historical information and is usually characterized by two probability distributions.

Focusing only on a single pass/fail decision while ignoring global and lifecycle effects is how organizations end up with measurements that appear technically compliant yet still produce expensive—or even catastrophic—outcomes in the real world.

When Measurements Go Wrong: Real-World Consequences

History provides painful reminders of what happens when measurement risk is misunderstood or ignored:

· Hyatt Regency Walkway Collapse — A design change doubled the load on a critical connection without proper verification. The structure met only about 60 % of the required capacity. 114 people died.

· Aircraft Sensor Miscalibration — Sensors were calibrated without accounting for moisture contamination. The calibration was technically performed, but under the wrong assumptions, resulting in a multi-billion-dollar accident.

· Medical Radiation Overdosing — A miscalibrated radiation system overdosed 152 cancer patients, some by as much as 70 %, due to inadequate processes and a lack of verification.

· Hubble Space Telescope — Test data revealed a mirror error far outside specification. The data was dismissed. Fixing the problem required a $1B+ servicing mission.

In every case, the measurements themselves were only part of the story. The real failures were decisions made without fully understanding risk and consequences. There are several more examples, which we cover in more detail in a separate article.

Requirements Drive Risk—Not the Other Way Around

A recurring theme in Decision Rule Guidance is that requirements must be understood before equipment is purchased, processes are designed, or decision rules are selected. Too often, organizations do the opposite: buy equipment first, pick a vendor second, and ask about risk last. That sequence almost guarantees suboptimal outcomes.

Understanding requirements means more than knowing the tolerance. It includes where the tolerance came from, what nonconformance means in the real world, how you want to balance false acceptance versus false rejection, and what costs actually matter (rework versus failure), including safety, liability, and reputational impact. Only then can you decide what uncertainty is acceptable and what decision rule makes sense for the application.

Decision Rules Are Tools—Not Answers

The guidance document intentionally presents multiple decision rule options. This is not because one is universally “better” than another; it is because different risks demand different tools.

Some applications benefit from:

· Simple acceptance with strict capability requirements (for example, 4.6:1 TUR/C_m)

· Guard banding to reduce false acceptance risk

· Explicit probability limits when consequences are severe

· Reporting results with uncertainty only, leaving the decision to the customer

What matters is not the sophistication of the method, but whether the method aligns with the organization’s actual risk tolerance.

Using an advanced probability model without understanding the consequences is just as dangerous as ignoring uncertainty altogether.

Why Organizations Must Be Honest About Risk

One of the strongest and most uncomfortable messages in the guidance is that many organizations do not truly want to know how much risk they are accepting.

Risk analysis can reveal equipment that is not capable, unstable processes, unrealistic requirements, or decisions driven by cost rather than safety or quality.

Ignoring these realities does not make them go away. It simply delays the appearance of the consequences and increases their severity.

As the guidance bluntly notes, choosing not to decide is still a decision.

Measurement Risk as a Strategic Asset

When organizations understand measurement risk properly, it stops being a compliance burden and becomes a strategic tool.

Effective decision rule selection can reduce unnecessary rework and scrap, prevent costly failures and recalls, improve long-term reliability, align quality decisions with business objectives, and save millions—or billions—over the product lifecycle.

This is why ISO/IEC 17025 intentionally avoids prescribing a single “correct” approach. The standard assumes that competent organizations will evaluate their risks and make informed decisions.

Final Perspective

Measurement risk does not belong to the laboratory. It belongs to the organization making the decision.

Laboratories provide the data, uncertainty, and technical framework. Standards provide the structure. But only the organization can determine what level of risk is acceptable for its products, processes, and customers.

Different measurements involve different risks. A cosmetic dimension does not carry the same consequences as a safety-critical measurement. Rejecting a conforming item may be trivial in one application and extremely costly in another. Accepting a nonconforming item may be manageable in one case and catastrophic in another.

Every decision rule answers the same fundamental questions: How much uncertainty are we willing to accept? What happens if we are wrong? Who bears the risk—producer, consumer, or both? These questions cannot be answered by mathematics alone. They require judgment, context, and an understanding of safety, cost, liability, reputation, and customer trust.

Understanding that responsibility—and acting on it deliberately—is the difference between merely complying with ISO/IEC 17025 and actually using measurement to protect people, products, and long-term value.

Measurement risk is not about doing more math. It is about making better decisions.

Risk is about likelihood and consequences. Probability is optional, not mandatory. Guard bands are a tool, not a requirement. Decision rules are chosen—not imposed.

When organizations understand this, they move beyond box-checking and begin using conformity assessment as it was intended: to protect people, products, and long-term value.

For additional information on measurement risk and decision rules, please download our guidance document at https://mhforce.com/wp-content/uploads/2024/04/Decision-Rule-Guidance-1st-Edition-V1.1.pdf.

Additional Information

Appendix A — Decision Rules and Risk Management Compliance (ISO/IEC 17025:2017)

The laboratory’s approach to decision rules complies fully with ISO/IEC 17025:2017.

ISO/IEC 17025 defines a decision rule as the method by which measurement uncertainty is accounted for when making statements of conformity (Clause 3.7). The standard does not prescribe a specific decision rule methodology, nor does it mandate the use of guard bands or probability-based calculations.

In accordance with Clause 7.1.3, when a statement of conformity is requested, the applicable decision rule is defined and, unless inherent in the referenced specification, is communicated to and agreed upon with the customer.

Measurement uncertainty is evaluated in compliance with Clause 7.6, ensuring that all significant contributors are identified and appropriately considered.

In compliance with Clause 7.8.6.1, the laboratory applies the documented decision rule and considers the associated risks, including the risks of false acceptance and false rejection.

The applied decision rule is clearly identified in the report, as required by Clause 7.8.6.2.

ISO/IEC 17025 permits multiple decision rule methodologies. The absence of guard bands or probability calculations does not constitute a nonconformance provided that measurement uncertainty, risk consideration, and customer agreement requirements are met.

Appendix B — Decision Rules & Measurement Risk Compliance Checklist (ISO/IEC 17025:2017)

This checklist demonstrates the laboratory’s conformance with ISO/IEC 17025 requirements related to decision rules, measurement uncertainty, and risk management when issuing statements of conformity.

1. Decision Rule Definition (Clause 3.7)

☐ A decision rule is defined as the method by which measurement uncertainty is accounted for when making statements of conformity.

☐ The laboratory has defined applicable decision rule(s) for conformity assessment activities.

☐ The decision rule(s) selected are appropriate for the measurement application and associated risks.

☐ The laboratory recognizes that ISO/IEC 17025 does not prescribe a specific decision rule methodology.

2. Statement of Conformity Determination (Clause 7.1.3)

☐ The laboratory determines whether a statement of conformity is required for each applicable request.

☐ When a statement of conformity is requested, the applicable decision rule is identified.

☐ If the decision rule is inherent in a referenced specification or regulation, it is applied as written.

☐ If the decision rule is not inherent, the decision rule is communicated to and agreed upon with the customer prior to issuing results.

☐ Customer agreement is documented in accordance with laboratory procedures.

3. Measurement Uncertainty Evaluation (Clause 7.6)

☐ Measurement uncertainty is evaluated for applicable measurement activities.

☐ All significant contributors to measurement uncertainty are identified and considered.

☐ Appropriate methods are used to estimate measurement uncertainty.

☐ Measurement uncertainty is reviewed to ensure it is suitable for the intended conformity decision.

4. Risk Consideration in Decision Rules (Clause 7.8.6.1)

☐ The laboratory applies the documented decision rule consistently when issuing statements of conformity.

☐ The level of risk associated with the conformity decision is considered.

☐ Risks of false acceptance and false rejection are recognized and addressed.

☐ The decision rule selected aligns with the consequences associated with incorrect decisions.

☐ The laboratory acknowledges that risk may be managed through multiple valid approaches, including but not limited to uncertainty control, capability requirements, or decision rule selection.

5. Reporting of Decision Rules (Clause 7.8.6.2)

☐ The applied decision rule is clearly identified in reports containing statements of conformity.

☐ Reporting clearly indicates how conformity was determined.

☐ Reports are sufficiently clear to allow the customer to understand the basis of the conformity decision.

6. Guard Bands and Probability Calculations

☐ The laboratory understands that ISO/IEC 17025 does not require the use of guard bands.

☐ The laboratory understands that ISO/IEC 17025 does not require probability-based calculations (e.g., PFA or PFR).

☐ Guard bands or probability calculations are used only when appropriate to the application or when required by specification, regulation, or customer agreement.

☐ The absence of guard bands or probability calculations does not constitute a nonconformance when measurement uncertainty, risk consideration, and customer agreement requirements are met.

7. Risk Ownership and Customer Responsibility

☐ The laboratory recognizes that the customer is responsible for determining acceptable risk for their application.

☐ The laboratory supports the customer by providing measurement results, uncertainty information, and decision rule options.

☐ The laboratory does not assume or impose risk acceptance beyond the agreed decision rule.

8. Application-Specific Decision Rules

☐ Decision rules are selected based on the specific measurement application.

☐ The laboratory recognizes that different measurements may require different decision rules.

☐ Decision rule selection considers safety, regulatory, cost, and business impacts where applicable.

9. Continual Review and Improvement

☐ Decision rules and uncertainty evaluations are periodically reviewed for continued suitability.

☐ Changes in requirements, equipment capability, or process are evaluated for their impact on risk.

☐ Lessons learned from nonconformities, customer feedback, or measurement failures are considered.

10. Overall Conformance Statement

☐ The laboratory’s decision rule practices conform to ISO/IEC 17025:2017.

☐ The laboratory’s approach supports informed risk management and transparent conformity assessment.

☐ The laboratory’s practices align with the intent of ISO/IEC 17025 to allow flexibility while ensuring technically valid and defensible results.

Share

Think of a load cell like a really precise digital scale built into a machine. When you push or pull on it, it measures that force and sends a number to a display. Engineers use load cells in everything from testing how strong a bridge bolt is to making sure an aircraft part won’t break.

But here’s the thing — if your load cell isn’t calibrated correctly, the numbers it gives you could be wrong. And wrong numbers in engineering mean unsafe products, failed tests, and wasted money.

That’s where ASTM E74 and Morehouse Instrument Company come in.

What Is ASTM E74 and Why Does It Matter?

ASTM E74 is the official standard for calibrating a load cell in North America. Think of it like the rules of a science fair; everyone follows the same process, so the results can be trusted and compared.

Other important standards that depend on ASTM E74 include:

• ASTM E4 – used to verify that testing machines (like the ones that crush concrete or stretch metal) are accurate

• ASTM C39 – testing concrete strength

• ASTM E18 – Rockwell hardness testing of metals

In short, if you want trustworthy test results in almost any industry, someone upstream had to follow ASTM E74 to calibrate the load cell used in that process.

The Calibration Pyramid: Who Calibrates What?

Imagine a pyramid of accuracy:

Primary Standards are typically quite large and expensive, precisely known deadweights (actual physical weights). Morehouse uses a fully automated deadweight machine capable of uncertainty as low as 0.0008 %. That’s incredibly accurate, like measuring the length of a football field and being off by less than the width of a fingernail.

• Secondary Standards (Class AA) are load cells calibrated by those deadweights.

• Working Standards (Class A) are calibrated by Secondary Standards and are used to calibrate the machines in your lab.

• Testing Machines are what your engineers actually use day-to-day.

Important: Only labs that have deadweights (Primary Standards) can assign a Class AA range. If a lab only uses another load cell to calibrate yours, they cannot legally assign a Class AA rating. Watch out for this!

What Is the ASTM LLF and Why Should You Care?

The Lower Load Limit Factor (LLF) —a statistical estimate of the error in forces computed from the calibration equation of a force–measuring instrument when the instrument is calibrated in accordance with these practices.

Or simply, it is the number that determines the lowest force your load cell can reliably measure after calibration.

• Class A: usable range starts at 400 × LLF or Instrument Resolution, whichever is higher

• Class AA: usable range starts at 2,000 × LLF or Instrument Resolution, whichever is higher

Example: If your LLF = 0.05 lbf, then:

• Class A range starts at: 400 × 0.05 = 20 lbf

• Class AA range starts at: 2,000 × 0.05 = 100 lbf

The better (lower) the LLF, the wider the useful range of your load cell. A shear web load cell from Morehouse will typically have a much lower LLF than an S-beam cell, meaning it’s more useful over a wider range of forces.

Before Calibration: Setting Up for Success

Getting the best calibration requires preparation. Here’s what Morehouse recommends:

1. Let Everything Stabilize

• Allow 24 hours for the load cell and room to reach a stable temperature

• Warm up your electronics for 15–30 minutes before taking readings

• The ideal temperature is 23°C (73°F). If you’re not at that temp, corrections need to be applied

2. Exercise the Load Cell First

• Before calibration, load the cell to its full capacity a few times (we find 3-5 times is sufficient, even though the standard only requires 2).

• Why? A load cell that hasn’t been “warmed up” with force can show errors up to 0.01 % or more just from sitting idle

• Skipping this step is like sprinting without stretching; you likely won’t perform your best, and if you’re older like some of us, you may hurt yourself!

3. Use Enough Calibration Points

• ASTM E74 requires a minimum of 30 data points

• More points = a better mathematical fit = more trustworthy results

Ascending vs. Descending: Don’t Skip the Down Direction

This is one of the most commonly overlooked mistakes in load cell calibration.

When you load a load cell going up (adding force), the reading can be slightly different than when you come back down (removing force). This is called hysteresis.

• If your application ever measures force while the load is decreasing, you must calibrate with a descending curve too

• A single ascending-only calibration can produce an error of up to 0.042% if you’re using the cell in both directions

• Separate ascending and descending curves give you lower uncertainty; a combined curve gives you a higher (worse) standard deviation

Note: The difference in output on an ascending curve versus a descending curve can be significant. A particularly good 100K load cell had an output of -2.03040 on the ascending curve and -2.03126 on the descending curve. Using the ascending-only curve would result in an additional error of 0.042 %.

Load cells from five different manufacturers were sampled, and the results are recorded above. The numbers varied from 0.007 % (shear web type cell) to 0.120 %. On average, the difference was approximately 0.06 %.

Six of the seven tests were performed using deadweight primary standards, which are accurate within 0.0016 % of the applied force.

The conclusion from these tests is clear: If a load cell calibrates both ascending and descending forces, it must be calibrated in both modes.

Tell Morehouse how you plan to use the load cell. We will calibrate using the adapters and modes (ascending, descending, compression, tension, or any combination) that match your real-world application.

Alignment: Keep the Force Straight

Imagine trying to weigh something, but you’re pushing the scale sideways instead of straight down. You’d get a wrong reading.

The same thing happens with load cells. Misalignment is a major source of error.

• For compression (pushing): use a flat load pad and make sure the surface is smooth and flat

• For tension (pulling): use adapters with a spherical seat to eliminate bending forces

• Morehouse found that a manually aligned setup had a standard deviation of 0.1 lbf with max deviation of 0.2 lbf, while a properly aligned setup using adapters had 0 deviation

Misalignment in an S-beam load cell can cause 0.75 % error. The same misalignment in a Morehouse Shear Web cell? Just 0.0022 %. That’s roughly 340 times better.

Rotate the Load Cell During Calibration

Don’t just push or pull straight down in one direction. Rotate the load cell to at least 3 positions (0°, 120°, 240°) during calibration. This:

• Randomizes the loading conditions, such as any uneven spots in the machine or load cell mounting

• Catches errors you’d miss with a single-direction test

• Is required by ASTM E74 for a proper calibration

Note: A load cell performs best when it is installed in a load path that is truly aligned and mechanically stable, meaning it is plumb, level, square, and rigid with minimal twisting. In that condition, the force is applied mainly along the intended axis, so the load cell output is more repeatable and accurate. If any of these conditions are compromised, the setup can introduce side loading, bending, or torque, which can increase variation and apparent error beyond what the load cell can achieve on its own.

Use the Right Load Cell for the Job

Not all load cells are created equal. Here’s a quick cheat sheet:

For ASTM E74 calibration work, Morehouse recommends shear web load cells for 100 lbf through 100,000 lbf applications. They’re more accurate, more stable, and less sensitive to side forces.

Keep Your Indicator (Display) Consistent

Your load cell talks to an indicator that displays the force reading. This is often overlooked, but it matters a lot.

• The easiest path is to always use the same indicator that was used during calibration

• If you swap indicators, you introduce additional measurement uncertainty

• The replacement indicator’s uncertainty must be less than 1/3 of the total system uncertainty

Think of it like this: If you calibrate a thermometer using one voltmeter, but then measure with a different voltmeter, your readings might shift. Same principle here.

✅ The “Never Do” List

To avoid ruining your calibration, never:

• ❌ Assign Class AA ranges unless you used deadweight Primary Standards

• ❌ Start the verified range below your first non-zero calibration point

• ❌ Skip descending calibration if you measure forces in the decreasing direction

• ❌ Use only part of the calibrated range regularly without exercising to full capacity

• ❌ Re-zero a load cell after an overload and assume it’s fine (overloading causes permanent damage)

• ❌ Swap out cables without verifying them with a transducer simulator

• ❌ Use mass weights (like gym weights) to calibrate a force device without converting to force units — the error can be up to 0.53%

What to Tell Morehouse Before Your Calibration

To get the best possible calibration from Morehouse, tell us:

1. What forces you plan to measure (minimum and maximum)

2. Whether you’ll be loading up, down, or both (ascending only, or also descending)

3. The indicator you’ll use in the field (so they can calibrate the whole system together)

4. How you load the cell, adapters matter!

5. Whether you need Class A or Class AA (depends on your application)

Summary: Your “A+ Calibration” Checklist

Learn More

Morehouse Instrument Company — “We create a safer world by helping companies improve their force and torque measurements.”

📎 Website: www.mhforce.com

Whether you’re calibrating a 50 lbf force gauge or a 5,000,000 lbf universal calibrating machine, following ASTM E74 correctly and working with a qualified lab like Morehouse is the difference between a measurement you can trust and one that only looks right.

This guide walks you through the key considerations so you can make an informed decision for your specific application.

First, Understand What You Are Working With

The Proving Ring

A proving ring is a precision elastic ring that deflects under load. That deflection, measured with a dial micrometer or a displacement sensor, is converted into a force value using the ring’s calibration equation, which is typically a polynomial. The concept is elegantly simple and mechanically robust. In fact, Morehouse has proving rings still in service after more than 60 years.

Analog proving rings with dial micrometer assemblies. The larger ring on the right shows the classic design still widely used as a check standard.

A Morehouse digital proving ring with Mitutoyo EH-10P readout. The digital displacement sensor eliminates dial reading error and significantly reduces operator influence.

The digital version reduces reading error significantly compared to the old dial-and-reed designs. If you are committed to proving rings, always choose digital over analog.

The Load Cell

A load cell is a force transducer that converts mechanical force into an electrical signal using bonded strain gauges arranged in a Wheatstone bridge circuit. As force is applied, tiny elastic deformations change the bridge output, which a digital indicator converts into engineering units such as pounds-force or Newtons. Modern indicators can apply multi-point linearization or polynomial coefficients, so the display reads direct force without any manual computation on your part.

The Four Factors That Should Drive Your Decision

1. Tare Management

This is where the two technologies diverge most sharply in everyday use.

Proving Ring: Any preload from fixtures or adapters must be mathematically corrected using the ring’s calibration coefficients. Every time a significant tare is present, you are running calculations before you even take a measurement. This is not difficult if you are disciplined, but it adds steps, introduces transcription risk, and slows your workflow considerably.

Load Cell: Press TARE on the indicator. Done. The instrument zeros the fixture mass electronically, and you proceed in engineering units. For small tares less than one percent of capacity, the error is typically under 10 ppm. Even at five percent of capacity, the error generally stays below 100 ppm, which is acceptable for most calibration work.

Verdict: Load cells win on tare simplicity, especially in high-throughput labs.

2. Temperature Effects

Temperature is a silent source of error in force calibration, and the two technologies handle it very differently.

Proving Ring: The calibration is valid at the ring’s stated calibration temperature. If the temperature of the ring differs from that reference condition, which it will after multiple loading cycles or in a shop environment, you must either apply a temperature correction or carry that effect explicitly in your measurement uncertainty budget. Analog rings are particularly vulnerable because multiple loading cycles generate heat in the ring body itself.

Load Cell: Modern reference load cells include internal temperature compensation networks built into the strain gauge bridge. The residual temperature effects are small, well-documented, and handled in the uncertainty budget through published specifications rather than hand-applied correction equations. You still need to allow thermal equilibration and control your environment, but you are not manually correcting for temperature on every run.

Verdict: Load cells offer a more practical thermal management path for routine calibration work.

3. Operator Influence

This factor is often underestimated when comparing the two technologies, but it has a real impact on measurement uncertainty.

Proving Ring (Analog): Reading an analog dial or micrometer introduces parallax error, interpolation uncertainty, and the well-known challenge of hearing when the vibrating reed first contacts the ball. Electric read vibrators add even more variability. The Type A uncertainty component for analog proving rings is typically about 30% higher than for digital rings. That is a meaningful difference when you are trying to achieve low uncertainty.

Proving Ring (Digital): A significant improvement over the analog version. Digital displacement sensors like the Mitutoyo LGF series eliminate dial reading error. If you are committed to proving rings, make it a digital one.

Load Cell: The indicator displays force directly in engineering units. There is no dial to read, no polynomial to enter manually, and no interpolation. The operator’s primary job is to apply the load correctly and record the number. This dramatically reduces the human contribution to measurement uncertainty.

Verdict: Load cells and digital proving rings significantly reduce operator influence compared to analog rings.

4. Long-Term Stability and Lifespan

Here is where the proving ring earns its reputation.

A well-maintained proving ring can last 25 years, 40 years, or more. Morehouse has rings that have been in continuous service for over six decades. That kind of mechanical stability is genuinely impressive and makes proving rings excellent check standards, as their drift year-over-year is often negligible.

Load cells have a shorter lifespan, typically around half that of a proving ring. However, a good shear web reference load cell paired with a high-quality indicator will deliver year-to-year stability well better than 0.05%, which is more than sufficient for the vast majority of calibration force applications. And when it is time to replace the load cell, the cost is substantially lower than a proving ring. Proving rings typically cost more than double the price of a comparable load cell, so even accounting for the shorter lifespan, load cells are generally the more cost-effective choice for most laboratories. The lower upfront cost, combined with reduced calibration time and operator training requirements, means the total cost of ownership strongly favors load cells in most modern lab environments.

Verdict: Proving rings win on lifespan. Load cells win on cost-effectiveness over the instrument lifecycle.

Which Load Cell Type Should You Choose?

If you have decided a load cell is the right path, the next question is which type. Not all load cells perform equally as calibration reference standards.

Shear Web Load Cells are the preferred choice for reference and field calibration standards from 100 lbf through approximately 112,000 lbf (500 kN). They exhibit very low creep, excellent resistance to off-axis loading, and outstanding year-to-year stability. This is the type Morehouse recommends for most force calibration force applications.

Column Load Cells can handle very high forces but are more sensitive to off-center loading and exhibit greater non-linearity. They are workhorses in large-capacity force applications but require more careful attention to alignment.

S-Beam Load Cells are useful for tension force applications and small forces but are susceptible to misalignment and generally not the right choice for force applications requiring calibration to ASTM E74, ISO 376, or similar standards.

Button and Pancake Load Cells are compact but highly sensitive to misalignment. Errors of 1 to 10% of rated output are not uncommon without proper adapters. Specialty adapters can dramatically improve performance, and Morehouse has achieved over 500% improvement in repeatability with custom button cell adapters.

Do Not Overlook the Full System

Choosing the transducer is only part of the decision. The indicator and adapters matter just as much.

The Indicator: For highest accuracy, choose an indicator capable of applying polynomial coefficients such as the Morehouse 4215 Plus or C705P, which mathematically correct for the load cell’s non-linear response and significantly reduce measurement bias. Resolution, stability, and environmental suitability all factor in as well.

Adapters: Correct alignment is critical. Spherical alignment adapters in compression force applications ensure force is applied along the correct axis. ISO 376 explicitly highlights the role of adapters in measurement reproducibility. Using a load cell without proper adapters can introduce errors of 0.1 to 2%, negating much of the accuracy advantage you chose the load cell for.

Calibration Provider: Your reference standard is only as good as its calibration. Look for a laboratory with Calibration and Measurement Capability (CMC) as low as 0.002% of applied force for critical work, a Test Uncertainty Ratio (TUR) of 4.6:1 or greater, and accreditation to ISO/IEC 17025 with compliance to ASTM E74 or ISO 376 where applicable.

Quick Decision Guide

Bottom Line

For most modern calibration laboratories, a master load cell with a high-quality digital indicator delivers the best combination of accuracy, speed, operator independence, and workflow efficiency. The built-in temperature compensation, simple tare handling, and direct digital readout remove several sources of error that analog proving rings carry by design. Add to that the significantly lower cost, and the case for load cells in routine calibration work is compelling.

Proving rings remain highly capable and respected instruments, particularly in legacy force applications, as check standards, or where exceptional mechanical longevity is the priority. If you go the proving ring route, choose digital over analog every time.

Whatever path you choose, pay equal attention to your indicator, your adapters, and your calibration provider. The best transducer in the world will underperform if any of those three elements are not matched to it properly.

Have Questions About Your Force Calibration System?

Morehouse Instrument Company has been designing, manufacturing, and calibrating force measurement equipment for over a century. Our accredited force calibration laboratory in York, PA operates to the lowest measurement uncertainties available in North America.

Visit our website at mhforce.com, or call 717-843-0081.

Scope. This article summarizes what our calibration certificates show over many years when the same load cell is calibrated in a low range (LO) and a high range (HI). All findings below are derived from the calibration data provided (ASTM LLF history, standard deviation, stability between calibration, and deviations on the calibration reports.)

Example: of a 10,000 lbf load cell calibrated at 10,000 lbf (HI) and 1,000 lbf (LO).

1. What HI and LO Range Mean (in plain terms)

A single load cell can be calibrated over different force spans. The HI range is the upper span (higher applied forces). The LO range is a lower span (lower applied forces). These are not different sensors, just different verified operating ranges of the same sensor.

Calibrating a load cell with both a HI and LO range or a Dual range might make sense, depending on the application.

Dual-range calibration certificates can be useful because they allow a single load cell to be used over an extended operating range. However, they do not guarantee identical performance in both spans (HI and LO).

To better understand how performance compares between ranges, we sampled over 30 load cells across four different manufacturers and multiple load cell designs. These included a variety of capacities and calibration histories.

While it is possible that exceptions exist, the overall behavior was remarkably consistent across the full sample.

For clarity and readability, we are highlighting 6 representative load cells in the detailed analysis below. These 6 were selected because they clearly demonstrate the same patterns observed across the entire 30-load-cell population.

Across the broader sample, the same general conclusions held: performance in the low range behaves differently than performance in the high range, and the differences are repeatable across manufacturers and designs.

2. Data Sets and Ranges Reviewed

The analysis includes all uploaded calibration certificates for the following load-cell groups and calibrated ranges:

Note: For comparisons, each calibration’s LLF is normalized to the maximum applied force used in that specific range. One of the load cells had a LO range which varied depending on the customer’s request.

3. What Metric We Compare (and why)

The primary number we decided to compare was the ASTM E74 Lower Limit Factor (LLF), reported in force units (lbf). However, LLF is reported on the force scale of the range being calibrated. This means LO-range LLF will often look “smaller” simply because the forces are smaller.

To compare ranges fairly, we reduced the data using these metrics:
• LLF%Max = LLF / (maximum applied force in that range)
• StdDev%Max = standard deviation of residuals / (maximum applied force in that range)

These normalize the results to how each range is actually used.

Comparing the HI and LO range certificates on the 10,000/1,000 lbf load cell above. The LLF for the LO range is about 2.67 times lower than for the HI range. However, when comparing to % of maximum force, the HI Range is 0.0015 % and the LO Range 0.0056 %. The LLF for the LO range percentage-wise was 3.73 times higher.

4. Results: What the Multi-Year Data Shows

4.1 LLF normalized to range maximum (all load cells)

Figure 1 shows LLF expressed as a percent of the maximum applied force for each calibrated range, plotted over time for every load cell and for both HI and LO spans.

This removes the misleading effect of comparing LLF in absolute force units (lbf) across drastically different force spans. A smaller LLF in pounds does not necessarily mean better performance if it represents a larger fraction of the usable range.

For example, a 2 lbf LLF in a 100 lbf range represents 2 % of that range, while a 10 lbf LLF in a 10,000 lbf range represents only 0.1 % of that range. Comparing 2 lbf to 10 lbf directly would suggest the 2 lbf case is “better,” when it consumes far more of the usable range.

Thus, a load cell with a HI range of 10 % of the range will almost always have a LO range that is greater than 10 % or that of the high range.

By normalizing LLF to the maximum applied force of each range, we evaluate performance relative to how the range is actually used. This allows a fair comparison between LO and HI ranges, regardless of their vastly different capacities.

Figure 1. LLF normalized to the maximum applied force for each calibrated range (all load cells, all years).

Across the full dataset, LO-range points consistently sit above HI-range points once normalized.

By “usable margin,” we mean the portion of the calibrated range between zero load and the ASTM E74 verified lower force limit. Because the lower force limit is calculated as 400 × LLF (for Class A), a higher LLF directly raises the minimum verified usable force.

When LLF%Max is higher, the verified lower limit moves upward as a percentage of capacity. That reduces how much of the low end of the range can be used while still meeting the required class. In practical terms, less of the calibrated span remains available for verified measurement.

4.2 Scatter about the fitted curve (standard deviation)

ASTM E74’s LLF is driven by scatter about the fitted curve. Figure 2 plots the standard deviation of the residuals (Residual = Applied Force − Calculated Force (from the fitted curve) which are normalized to the HI Range Max lbf. Because these residuals are what the LLF is built from, this plot is a direct view of repeatability/consistency behavior in each range.

Figure 2. Scatter about the fitted curve (standard deviation), normalized to the maximum applied force for each range.

The same pattern is visible here: LO range generally exhibits higher normalized scatter than HI range across the different load cells and years.

4.3 Overall stability comparison (LO vs HI)

To summarize stability across all instruments, Figure 3 compares the distribution of normalized LLF values for LO vs HI ranges. A narrower distribution indicates more stable, repeatable behavior across calibrations.

Figure 3. Overall stability comparison: distribution of normalized LLF for LO vs HI ranges.

The HI-range distribution is tighter and lower than the LO-range distribution in the aggregate. This matches what we observe in the time-series plots: HI ranges are typically more stable over time on a normalized basis.

5. What This Means Under ASTM E74 Stability Expectations

ASTM E74 emphasizes that devices should be calibrated at defined intervals to evaluate stability. In practical terms, stability is assessed by how repeatable the residual behavior is from one calibration to the next, and therefore how consistent LLF remains over time for a given range.

Using the complete dataset, the practical ASTM-style takeaway is:
• HI ranges tend to maintain tighter normalized LLF bands across years.
• LO ranges tend to show larger normalized LLF and larger variation, which translates to a higher verified lower-force limit for a given class.

This does not necessarily mean a load cell is ‘bad’—it means the low-force operating window is more constrained and may require derating sooner(either more frequent calibrations to maintain the desired Class A verified range of forces, or raising the Class A verified range of forces to accommodate degraded performance).

6. Challenges of Low-Range Calibrations (what the data repeatedly shows)

Across all reviewed load cells (spanning max forces of 600 lbf up to 600,000 lbf), the LO range tends to show:
• Higher LLF%Max directly reduces the usable verified range. For example, a HI range with an LLF equal to 0.01% of full scale results in a Class A lower force limit of approximately 4% of capacity (400 × LLF). If the LO range has an LLF greater than 0.02% of full scale, the Class A lower force limit becomes greater than 8% of capacity. In practical terms, doubling the normalized LLF doubles the minimum verified usable force.
• Higher StdDev%Max (more scatter about the curve)
• Larger year-to-year movement

In certificate terms, this shows up as a higher verified lower force limit for a target class and less stable low-force performance over time.

7. Dual-Range (HI/LO) vs Two Load Cells: What Users Gain and What They Sacrifice

7.1 Why users choose dual-range HI/LO calibration (one load cell)

Users choose dual-range calibration because it simplifies real work:
• One sensor covers more of the force envelope

• Fewer changeovers and less fixturing (this may reduce timing errors from having to switch the reference standard).

• Faster setups and fewer opportunities to introduce setup variability
• Reduced calibration inventory and logistics

7.2 What you sacrifice (and what the certificates show you’re trading away)

Compared with using two dedicated load cells (one optimized for low forces and one for high forces), the trade-offs seen in the data are:
• Reduced low-force stability (LO normalized LLF is higher and more variable)
• Higher low-force scatter about the fitted curve (StdDev%Max trends higher)
• Earlier need to derate or restrict the LO range to meet a given ASTM class

In simple terms: you’re trading low-force performance margin for convenience and coverage.

7.3 Why you might NOT choose dual-range (choose two sensors instead)

If low-force performance is critical—tight acceptance bands, small-force threshold detection, or long-term stability at the bottom end—two dedicated sensors often make sense:
• A smaller-capacity sensor used only at low force typically shows better normalized performance in that regime.
• The high-capacity sensor remains available for large-force work without compromising the low-force requirement.

The cost is additional equipment, additional calibration cost, and more setup complexity.

8. Practical Decision Guide

Dual-range HI/LO calibration is usually a good choice when:
• The application tolerates a higher low-force verified limit (Some ASTM E4 calibrations)
• Setup speed and simplicity matter
• The low-force region is used for trending or screening rather than tight-acceptance measurements

Two sensors are often the better choice when:
• Low-force uncertainty/stability is a primary requirement
• The test relies on small-force changes

• The indicator doesn’t have enough resolution to gain any benefit from a LO range
• You need long-term stability at the bottom end with minimal derating risk

Bottom line: dual-range calibration is often the right operational choice when reduced setups are a priority. If the low end is critical (tight acceptance limits, small-force changes, or long-term stability demands), the data strongly supports using a dedicated low-capacity sensor for that portion of the work.

So should you use one dual-range load cell or two load cells?

Across the sampled load cells and capacities, the certificates repeatedly show:
• LO range LLF is usually smaller in lbf, but higher once normalized to how the range is used. (For example, On a 10,000 lbf load cell the high range LLF could be 1 lbf, Class A = 400 lbf, and the low range 0.2 lbf, Class A = 80 lbf. Thus the HI range would be 0.01 % of FS, and the LO range 0.02 % of FS)
• LO range scatter about the fitted curve (normalized) is typically higher and varies more year-to-year.
• HI range tends to always have tighter normalized LLF bands and better repeatability.

Note: Changing polynomial order or recalibration can refine the curve fit, but it does not remove low-range scatter when the system’s repeatability is the limiter.

Highlights from the sampled load cells (what repeats across many years)

Figure 4 translates LLF into the practical question users care about: “How far down can I go and still claim a verified Class A range?” Across the sampled load cells, LO ranges generally imply a higher minimum usable force (as a percent of range max) than HI ranges. In other words, the low range tends to lose verified usable margin sooner.

Figure 4. ASTM E74 Class A implied lowest usable force as a percent of each range’s maximum applied force (computed as 400 × LLF%Max).

ASTM E74’s LLF is a statistical summary of how tightly the calibration points repeat about the fitted curve. For Class A verification, ASTM E74 uses a simple rule: the lower force limit is 400 × LLF (in force units). That means any increase in LLF—especially in the low range—pushes the verified minimum usable force upward.

What ASTM E74 implies at the low end (in plain language)

1) Resolution and noise floor. At low forces, a fixed resolution or small random variation consumes a larger fraction of the applied force. That shows up directly as larger normalized scatter and a higher normalized LLF.
2) Stability over the calibration interval. ASTM E74 ties usability to whether the system holds calibration values over time. Low-range performance typically changes sooner and more noticeably than high-range performance.
3) The verified lowest usable force point. Both ISO 376 and ASTM E74 recommend avoiding extremely low percentages of capacity; ASTM E74 also provides a direct way to compute the verified lower limit for a class from LLF.

Top 3 considerations when deciding “how low” to go

This short update is based on sampling several dual-range (HI/LO) load cells across multiple capacities and years. The goal is simple: when a customer asks to use one load cell “for everything,” what do the calibration certificates say we can realistically expect at the low end?

It’s common for organizations to push equipment toward the lowest possible operating point to reduce inventory and setup time. Fewer sensors, fewer setups, and fewer calibrations is an attractive goal. The data in this paper shows that the trade is real: convenience usually comes with reduced low-force performance margin and reduced stability.

1. Resolution and Noise Floor

At low forces, small effects become large as a percentage of the applied load. Resolution of the readout, electrical noise, and mechanical seating effects consume a greater fraction of the signal.

Even when absolute deviations are small, they represent a larger percentage of applied force at 2% than at 20%. This shows up directly in higher normalized LLF and higher normalized scatter.

Low-force performance is often limited not by curve fit, but by repeatability.

2. Verified Lower Force Limit (ASTM E74 Impact)

ASTM E74 ties usability to repeatability through the Lower Limit Factor (LLF). For Class A verification:

Lower Force Limit = 400 × LLF (in force units)

This means that as LLF increases, the verified minimum usable force increases quickly.

If LLF is 0.01 % of full scale, the Class A minimum is about 4 % of capacity.
If LLF is 0.02 % of full scale, the minimum becomes about 8 % of capacity.

The multi-load-cell sample consistently shows that LO ranges tend to have higher normalized LLF than HI ranges, meaning the verified usable low end is often higher than users expect.

3. Stability Over the Calibration Interval

Low-range performance is typically less stable over time.

Across multiple load cells and years, the data shows:

• LO ranges vary more year-to-year.

• Drift appears in the LO range before it appears in the HI range.

• Small shifts can push the load cell outside ASTM E74 stability criteria, shortening the calibration interval.

Even if a load cell performs acceptably at 2 % today, maintaining that performance over a two-year interval may be significantly more difficult.

Bottom line for one load cell

• If your low end is for calibrations with higher tolerances, such as 1 % of applied force, a single dual-range load cell can be a smart operational choice (fewer setups, fewer sensors).

• If your low end is critical (tight acceptance limits, small-force changes, long calibration intervals), the sample supports using two sensors: a dedicated low-capacity load cell for the bottom end, and a higher-capacity one for the top end.

It isn’t.

The hard part is building an organization that can execute the strategy every day, even when priorities collide, mistakes happen, and people get busy.

I was fortunate to attend a presentation on Strategy Implementation by Gideon Maherbe and took so many notes that, as a reflection of this presentation, I figured I would review my notes and try to create a post that others might find beneficial. Looking back at them later, a few themes jumped out immediately:

• clarity beats complexity (I admittedly struggle here as everything is always clear to me)
• culture drives execution
• priorities matter more than plans
• the right people make the system work

The ideas weren’t revolutionary by any means.

Yet they were important reminders of how leadership really works.

Looking at these notes later, I realized something:

Most organizations already know what they should do.

They just struggle to consistently execute it.

And execution almost always comes down to people, culture, and clarity.

The leadership equation

Here’s the simplest way I can describe what I took away from the discussion:

Strategy × Culture × People × Systems = Execution - It’s this simple, right? :)

If strategy is weak, people work hard but end up moving in different directions. When culture is weak, problems tend to be hidden instead of solved. Weak people capability means the organization simply lacks the skills needed to succeed, and weak systems make execution inconsistent.

In calibration and metrology, this becomes obvious quickly. Precision work exposes weak systems! And, oh boy does it ever. This is where I’d recommend going back to a previous article on pick two of the three.

Culture vs Strategy

One of the ideas that kept appearing in my notes was the tension between strategy and culture.

Strategy = What we want to achieve

Culture = How we behave while trying to achieve it

Many organizations assume strategy drives performance.

In reality:

Daily behavior drives performance.

Culture shows up in small moments:

• whether someone questions a suspicious measurement, or an obvious outlier. (Are your people engaged enough to question measurements ?)
• whether mistakes are discussed openly
• whether teams help each other solve problems, or managers are constantly asking “How may I help?”

If the culture discourages transparency, strategy doesn’t matter.

Pride in the work

In metrology, there’s one trait that separates strong teams from average ones.

Pride in the work.

Pride shows up in small details:

• what adapters/fixtures are used (does calibration replicate field use)
• how data is reviewed (is it reviewed, or is a box checked)
• how teams investigate anomalies (when a shift occurs, is it noticed, or is all of the data analyzed for trends?)

You can’t fake pride. - Many companies and people might try or lead you to believe they are great! They might even tell you they are world-class :). That statement alone means you have to believe them, right?

Pride grows when people understand that their work matters. And our team members know that it absolutely does; in calibration, they see it more clearly than in other areas. That might be because the consequences are known, and many mistakes will be questioned by the end-user. Though not all end users scrutinize their data enough, that is another topic altogether.

The clarity principle

One of my favorite notes from the session said simply:

“I don’t want more complexity. I want clarity.”

Many organizations confuse complexity with sophistication.

More dashboards.
More endless meetings with no objective.

Personal Note: I’m certainly guilty of extending meetings and having silly and sometimes meaningful discussions. Though that is definitely not on our agenda.
More reports.

Note: There are really only 5 or 6 Key Performance Indicators that are important to drive your business. No need to argue it, what can you not live without? What are you looking at to make decisions?

The fact is Complexity slows execution. Clarity accelerates it. And if you have dealt with many in operations, or quality, SOP’s (Standard Operating Procedures) can be anything but clear and simple. There are people in the world who want every single detail in these. We all know some of them.

My personal opinion here is to go read the Checklist Manifesto!

Clarity makes priorities obvious, helps people understand their roles, enables faster decisions, and ultimately improves execution. Without clarity, teams spread their effort across too many directions.

The right people in the right seats

Another concept that kept appearing in my notes and a concept in one of my favorite books, Good to Great by Jim Collins:

Right people on the bus. Right people in the right seats.

It sounds obvious. Yet many organizations struggle with it.

Too often, companies try to fix execution problems with new software, more meetings, or more reporting dashboards. But the real issue is often much simpler: alignment.

People cannot execute well if they are unclear about what matters, where they fit, and what success actually looks like.

In practice, people need clarity about three things.

1. What are we trying to achieve?

What really matters?

This sounds simple, but many organizations unintentionally create dozens of competing priorities. When everything is important, nothing is.

Clarity around priorities helps teams focus their energy on the work that actually moves the organization forward. Without that clarity, people stay busy but progress slows.

2. What is my role?

Who owns it, and is it me?

Ownership matters. When responsibilities are vague, work falls into the gaps between roles. People assume someone else is handling the problem.

Strong organizations remove this ambiguity. People know what they own, what they support, and when they need to raise their hand.

At Morehouse, we use EOS, and are not perfect, though we do have defined roles; sometimes, more people are copied on emails that need to be, though we are always working to be better and fix things.

3. What does success look like?

Are we solving the right problems?

Activity is not the same as progress. Teams can work incredibly hard while still focusing on the wrong problems.

Success needs to be defined clearly enough that people know when they are moving in the right direction.

When those three things are clear, execution improves dramatically. It is why we spend two days and a lot more time on developing and refining strategy every year and then meet quarterly to see if we are on track or if anything has changed.

Priorities Matter More Than Plans

Another insight from the discussion was the difference between operations and strategy.

Operations focuses on questions like:

• Was the work completed?

• Did we meet the requirements?

• Did we hit the numbers?

These are important questions. Operations keeps the organization running.

But strategy asks different questions:

• Where should the organization go next?

• What capabilities do we need to build?

• How will we remain competitive over time?

Both perspectives are necessary. The challenge is that in most organizations, operations naturally dominate.

Urgent problems appear.
Emails arrive.
Customers need answers.
Deadlines approach.

Daily firefighting almost always wins.

Without intentional leadership discipline, strategy slowly gets pushed aside by operational noise.

Core Values and Culture are so Important

This is where core values and culture become incredibly important.

Culture shapes execution in ways that leaders often underestimate.

A healthy culture reinforces several key behaviors:

• Priorities must be protected.
  Teams cannot succeed if leadership constantly shifts focus.

• Leadership behavior sets the tone.
  People watch what leaders do far more than what they say.

• People must feel safe raising problems.
  The earlier a problem surfaces, the easier it is to solve.

None of these ideas are overly complicated; they just require consistency and people who will not overly complicate them.

Culture Is a System Outcome

One of the most important leadership lessons I’ve learned though certainly have not mastered, is that:

Culture is the outcome of the systems leaders create.

Culture reflects the environment people work in every day.

It grows out of:

• incentives

• expectations

• routines

• accountability

• communication patterns

If a system rewards speed over accuracy, accuracy will suffer (missing the little things on a calibration certificate, like a checkbox).

If a system rewards transparency and learning, improvement accelerates.

This is why culture cannot be built through slogans or posters on the wall, though many try. I’ll just buy that cat “Hang in there Poster,” and everything will be grand :).

Culture is not what leaders say.

Culture is what the system encourages people to do.

Leadership Is System Design

Leadership is often described as inspiration or vision.

Those things matter. But in practice, leadership is also something more practical:

Leadership is system design.

Leaders design the environment in which people work.

That environment includes:

• priorities

• decision frameworks

• feedback loops

• communication norms

When these elements align, strategy becomes easier to execute.

When they don’t, even the best strategy struggles.

What This Means on Monday Morning in a Calibration Lab

Leadership ideas often sound good in a meeting room.

The real test is what happens on Monday morning.

A technician walks into the lab and begins setting up a calibration. Something about the setup doesn’t look right. Maybe the data trend is unusual. Maybe the fixture alignment feels slightly off.

What happens next depends entirely on culture.

In a weak culture, the technician might think:

“I’ll just keep going. It’s probably fine.”

In a strong culture, the technician thinks:

“Something isn’t right. I’m going to stop and figure it out.”

Where are you going, and when will your team stop firefighting to figure out your destination?

Thanks for reading, if you feel that there is value in any of this post, please share it.

There’s a saying that goes something like pick and two because there’s no one that can do all three. That makes a lot of sense when you think about it as cheap, fast, and quality is something that labs can never balance. When you have a reputation for fast, cheap , and quality, eventually everyone comes to you, and something will break, or multiple things will break. Even in this AI age, we can have reviews, and speed the process, though a race to the bottom, is not something any quality company is likely going to pursue.

Every time we start to think about boosting some area or strategically having a better goal, I cannot help myself. I start think of two things.

1. How much better than 98 % on time can we really be?

2. The Meatloaf Song, “Two out of three ain’t bad”

Yes, we could shoot for 99 % or higher on time, though would that really move our needle? Do not get me wrong, as we discussed our three year plan, we do plan on improving the 98 % number, though in our case, it’s just not strategically that beneficial. So, we will move it, though that’s not going to be our strategic initiative. Our strategic initiative for the next three years is going to be to double down on creating real value for our customer base, listening to them and implementing something everyone seems to want. Some might be wandering what that is, and I’m going to hold this one a little tighter until we are closer to fully implementing it. Though, I can say it will likely make the value of the what we provide to you, even better. It is also something that not many are offering now, which is crazy to me.

So, with that in mind, let’s get to the fabulous Meatloaf song, I write this today, not being a lyricist, or claiming to be. I write this as a fun little ditty, in reimagining the song to suit what Morehouse does.

Here goes at my attempt at this.

And all I can do is keep on telling you
We move it fast (we move it fast)
We do it right (we do it right)
There ain’t no way we’re ever gonna compromise quality

Now don’t be mad (don’t be mad, ’cause)
Two out of three ain’t bad

You want it cheap, you want it fast, you want it right
Everybody does, but you can’t win that fight
At Morehouse, we choose speed and strength
Back to work for you in record length

So don’t be sad (don’t be sad, ’cause)
Two out of three ain’t bad

Your equipment back, back on the floor
Back to work, back making more
Lost time costs more than price ever could
Fast and right is always good

Two out of three really is the reality in our world, and at Morehouse, we have made our choice. We are going to keep prioritizing quality and lead time, because compromising accuracy is not an option, and waiting too long to get critical equipment back in service costs more than a small difference in calibration price ever will. When your tools are down, production slows, schedules slip, and the cheaper option quickly turns into the expensive one.

That does not mean we ignore price. It means we focus on real value. Over the next few years, our goal is to keep listening closely to what customers actually need and deliver improvements that make your experience smoother, faster, and more dependable. If you are looking for a partner that will not race to the bottom, but will get you back up and running fast and right, that is exactly where we are staying locked in.

Making Informed Decisions

The Real Cost of a Poor Calibration

Price matters, but it should never be the only deciding factor when choosing a calibration provider. A low-cost calibration can look like a win upfront, but if the work is inconsistent or inaccurate, the downstream costs can increase quickly.

Here are a few things worth weighing before you choose:

• Long-term performance: A bargain calibration may save money today, but it can create inaccurate measurements, unexpected downtime, and lost revenue later.

• Turnaround time: If you need equipment back quickly, you may pay more for speed. The better question is what downtime costs you if your equipment is delayed or fails because the calibration was not done right.

• Brand reputation: Your customers count on you for reliable results. One incorrect measurement can lead to bad decisions, product issues, and damage to your credibility.

When you evaluate a calibration provider, consider the full cost of the decision, not just the invoice. Cutting corners on calibration is one of those “savings” that can become very expensive.

Why “Cheap” Can Cost More

Some providers may offer a lower price, but that discount often comes with tradeoffs you will feel later. At Morehouse, we focus on getting your equipment back to you quickly and accurately so you can stay productive and confident in your results. We offer standard 7–10 business day turnaround on working equipment, expedited options when you need them, and very low measurement uncertainties to reduce risk and minimize downtime.

Think about what happens when an instrument comes back out of tolerance, or worse, the calibration misses something and the problem shows up later. If equipment fails and you cannot produce to spec, the impact can include scrap, rework, missed deliveries, lost revenue, and even safety exposure. That is why choosing the lowest bidder can become the most costly choice.

Quality, Lead Time, and Price Conclusion

Choosing a calibration provider is a critical decision, and the cheapest option is rarely the best value. At Morehouse, we deliver high-quality calibration with fast lead times, backed by more than a century of experience. When you choose Morehouse, you are not just paying for a service. You are investing in confidence that your equipment will be calibrated accurately and returned on time.

Force and torque calibration is a key part of any quality program, but not all calibrations are created equal. We believe the true cost of calibration includes quality, lead time, and the risk you take on with each option. That is why we work to educate customers beyond the “just calibrate it” mindset and help you avoid costly mistakes.

If you are looking for a calibration partner that prioritizes what matters most, contact Morehouse and get the quality and service you deserve.

Most measurement failures don’t start with a dramatic mistake. They start with something ordinary: a unit that isn’t clearly defined, a calibration performed under the wrong conditions, a process step skipped because “it’s always been fine,” or a requirement that no one can clearly explain. Then a decision gets made, accept, reject, proceed, launch, and the consequences show up downstream.

That “downstream” point is the key. Measurements exist to support decisions, and every decision follows the same path: Measurement → Decision → Consequence. The measurement rarely causes damage by itself. The damage happens when a bad measurement (or a misunderstood one) becomes a confident decision.

A practical way to think about measurement reliability is as a three-legged stool:

• Requirements (what you need to know, and why)

• Equipment (what you use to measure)

• Process (how you perform and control the measurement)

If any leg is weak, the “decision” you’re sitting on becomes unstable—no matter how good the math looks. Your presentation frames this as: know the right requirements, choose the right equipment, and have the right processes [A].

What follows are real-world examples—each one a reminder that measurement risk is not theoretical. It’s operational, financial, and sometimes fatal.

1) The Vasa — When “an inch” isn’t an inch

In 1628, the Swedish warship Vasa sank less than a mile into its maiden voyage. One contributor discussed in modern reporting is that shipbuilders used rulers based on different measurement systems, creating inconsistent construction and instability [1].

What failed: requirements (common measurement system) and process (standardization/verification).
Lesson: If definitions aren’t shared and enforced, the measurement system can’t save the decision.

2) Lake Peigneur — A positioning/reference failure that cascaded

In 1980 at Lake Peigneur (Louisiana), a drilling operation punctured a salt mine beneath the lake, triggering flooding and a dramatic sinkhole/whirlpool event. Accounts emphasize an incorrect location/reference understanding that led to the drill intersecting the mine [2].

What failed: requirements (correct location definition) and process (independent verification).
Lesson: Some “measurement” failures are really reference-frame failures—coordinates, maps, assumptions, and verification steps.

Video on youtube.

3) Hyatt Regency Walkways — Verification missed, consequences were immediate

On July 17, 1981, the Hyatt Regency skywalks collapsed in Kansas City, killing 114 and injuring hundreds. A key factor documented in major investigations is that a design/change detail dramatically altered load paths, and the connection failed catastrophically [3][4].

What failed: process (engineering change control and verification) and requirements (ensuring design meets load/code intent).
Lesson: High-consequence systems require high-discipline verification—especially when changes alter the risk landscape.

4) B-2 crash (Guam, 2008) — “Calibration done right” under the wrong conditions

A B-2 accident investigation found that moisture in air data components during air data calibration distorted sensor output, contributing to erroneous flight control inputs and loss of the aircraft during takeoff [5].

What failed: equipment/environment controls and process controls (detecting/mitigating moisture effects during calibration).
Lesson: Calibration isn’t a magic reset button. A calibration performed without controlling critical conditions can create false confidence.

5) CoxHealth/BrainLAB — Wrong tool, right intent, years of undetected harm

In Springfield, Missouri, a stereotactic radiation therapy system delivered incorrect doses for years. Reporting describes systematic error linked to calibration/measurement method problems and weak independent verification [6][7].

What failed: equipment selection (appropriate measurement method/device) and process (independent verification and detection controls).
Lesson: Repeatability is not correctness. A process can be stable and consistently wrong if the setup is wrong.

6) Hubble Space Telescope — Measurement warnings were good enough… and still ignored

The Hubble launched with a mirror figure error that severely degraded optical performance. Investigations trace the issue to test equipment and process/decision failures that allowed contradictory evidence to be dismissed before launch [8][9][10].

What failed: process (cross-checking, criteria, escalation) and decision discipline when results contradict expectations.
Lesson: Sometimes the failure isn’t measurement capability—it’s organizational decision risk: what happens when measurement evidence conflicts with schedule or assumptions?

7) BP Texas City — When signals don’t drive decisions

The 2005 BP Texas City refinery explosion and fire killed 15 and injured 180+. The U.S. Chemical Safety Board documented major process safety management breakdowns, including weaknesses in instrumentation awareness, safeguards, and decision-making under abnormal conditions [11][12].

What failed: process and requirements—what triggers action, how warnings are handled, and how risk is governed.
Lesson: In high-consequence environments, measurement only protects you if the organization has reliable decision pathways that respond to it.

Pulling it together: measurement risk is likelihood and consequence

These cases all show the same pattern: the catastrophe is rarely “the measurement.” It’s the decision made using a weak system—unclear requirements, incapable/compromised equipment, or poor processes—and the downstream consequences that follow.

Risk management is not about obsessing over one metric. It’s about making deliberate choices:

• How likely is a wrong decision?

• What happens if we’re wrong?

• Who bears the consequence?

The practical takeaway from every example above: reduce risk by strengthening the whole system—requirements, equipment, and process—not by relying on a single technique or assuming “the number” is enough.

Engineering failure is measured in two ways:

1. Human death toll

2. Materials lost.

Even when the technical concepts are clear, the real vulnerability often isn’t math—it’s mindset. When processes seem to work, it’s easy to treat the current approach as “safe enough,” especially when no major failures have surfaced. But decision rules that default to simple acceptance can quietly erode safety margins over time, normalizing higher risk without anyone explicitly choosing it. That’s how uncertainty becomes invisible: not through bad intent, but through routine.

“Failures appear to be inevitable in the wake of prolonged success, which encourages lower margins of safety. Engineers and the companies who employ them tend to get complacent when things are good; they worry less and may not take the right preventative actions.” - Heny Petroski

Petroski’s claim about complacency might merely describe human nature, or it might point to the old but well-known motto, “If it ain’t broke, don’t fix it.” After presenting the information above, the question for you is what are you going to do in your organization to help make the world a safer place.

Please visit mhforce.com for more information on force and torque measurements.

References

[A] Zumbrun, H. “Measurements Gone Wrong” (webinar, presentation), Morehouse Instrument Company, June 6, 2024.

[1] GBH (WGBH). “New Clues Emerge in Centuries-Old Swedish Shipwreck (Vasa).”

https://www.wgbh.org/news/national/2011-03-01/new-clues-emerge-in-centuries-old-swedish-shipwreck

[2] 64 Parishes (Louisiana Endowment for the Humanities). “Lake Peigneur Drilling Accident.”

https://64parishes.org/entry/lake-peigneur-drilling-accident

[3] NIST (NBSIR 82-2465A). “Investigation of the Kansas City Hyatt Regency Walkways Collapse.”

https://nvlpubs.nist.gov/nistpubs/Legacy/IR/nbsir82-2465A.pdf

[4] NIST Publication Landing Page (Hyatt Regency Walkways Collapse).

https://www.nist.gov/publications/investigation-kansas-city-hyatt-regency-walkways-collapse-nbs-bss-143

[5] U.S. Air Force. “B-2 accident report released” (moisture in PTUs during air data calibration).

https://www.af.mil/News/Article-Display/Article/123360/b-2-accident-report-released/

[6] Radiology Business. “76 patients over-radiated at Missouri hospital” (CoxHealth/BrainLAB reporting).

https://radiologybusiness.com/topics/medical-imaging/oncology-imaging/76-patients-overradiated-missouri-hospital

[7] HealthLeaders Media. “Radiation Errors Reported at Missouri [Hospital]” (additional coverage).

https://www.healthleadersmedia.com/strategy/radiation-errors-reported-missouri

[8] NASA Science. “Hubble’s Mirror Flaw.”

https://science.nasa.gov/mission/hubble/observatory/design/optics/hubbles-mirror-flaw/

[9] ESA/Hubble. “The Aberration Problem.”

https://esahubble.org/about/history/aberration_problem/

[10] NASA Technical Reports Server. “Hubble Space Telescope Optical Systems Board of Investigation Report.”

https://ntrs.nasa.gov/api/citations/19920008654/downloads/19920008654.pdf?attachment=true

[11] U.S. Chemical Safety Board (CSB). “BP America Refinery Explosion” (case page).

https://www.csb.gov/bp-america-refinery-explosion/

[12] CSB. “Final Investigation Report PDF” (BP Texas City Refinery Explosion and Fire).

https://www.csb.gov/file.aspx?DocumentId=274

Share

Let’s break down what’s really happening.

The Two Errors Nobody Talks About Enough

Every pass/fail decision in calibration carries two potential errors:

Probability of False Accept (PFA) — also called Consumer’s Risk is the chance that an out-of-specification instrument gets measured as in spec and passes. You accept something you should have rejected. That instrument goes back into service, makes measurements, and introduces errors downstream. Nobody knows. The consequences can range from minor product variation to safety failures, depending on your application.

Probability of False Reject (PFR) — also called Producer’s Risk is the opposite: a perfectly good instrument gets measured as out of spec and fails. You reject something you should have accepted. This one hits you in the wallet. The instrument gets pulled from service, sent for adjustment or repair, or replaced, none of which was actually necessary.

Both errors cost you. PFA costs you quality and risk. PFR costs you time, money, and operational disruption. The frustrating thing is that they’re linked if you try to reduce one without doing anything about uncertainty, you tend to increase the other.

Figure 1 Courtesy of Dilip Shah

And what it all actually means

Figure 2 A Cartoon of What Happens in These Quadrants.

The two charts below show the same instrument calibrated at two different labs. The only difference is measurement uncertainty. Notice how the wide bell curve on the left bleeds past the spec limits in both directions, creating meaningful PFA and PFR risk. The narrow curve on the right stays well within the limits. Same spec. Same instrument. Completely different risk picture.

Figure 3: Example of PFA and PFR in Practice

Where Does Uncertainty Come into This?

Figure 4. What Uncertainty Does to Conformance.

This graphic illustrates a fundamental problem in metrology: what you specify on paper and what you can reliably verify in practice are not the same thing.

The Design / Specification Phase (top band)

When an engineer designs a product or process, they define a tolerance, typically a Lower Specification Limit (LSL) and an Upper Specification Limit (USL). This is known as a symmetrical tolerance. They can also design with an asymmetrical tolerance. At this stage, the boundary is assumed to be a sharp, clean line. Something is either in spec or it isn’t. There is no ambiguity. The world is simple.

The Verification Phase (bottom band)

The moment you pick up a measurement instrument to check whether something actually meets that specification, reality gets complicated. Your measurement tool has uncertainty; it doesn’t tell you the exact true value, it gives you an estimate with a range of possible error around it. That uncertainty is represented by the triangles: a single sharp spec limit at the top fans out into a fuzzy zone at the bottom.

The Uncertainty Range (amber zones)

This is the critical insight the graphic is communicating. Near each spec limit, there is a band of measurement values where you genuinely cannot tell based on the measurement alone whether the item is truly in spec or truly out of spec. If your measurement falls within this zone, the true value could lie on either side of the limit. The wider your measurement uncertainty, the wider these amber zones become, and the more of your usable specification range gets consumed by ambiguity.

What this means practically

You face three distinct regions in the verification phase, not two:

• Non-conformance zone — far enough outside the limit that even accounting for uncertainty, the item is almost certainly bad. Reject with confidence.

• Uncertainty range — close enough to the limit that you cannot reliably determine conformance. This is where false accepts (PFA) and false rejects (PFR) happen.

• Conformance zone — far enough inside the spec that even accounting for uncertainty, the item is almost certainly good. Accept with confidence.

The practical consequence

If your measurement uncertainty is large relative to your tolerance, the amber zones eat up a significant portion of your specification range. Your “safe” conformance zone shrinks. You’re forced to either accept more risk (by ignoring the uncertainty) or tighten your effective acceptance limits, which means rejecting more borderline-good parts and increasing cost.

In most practical calibration programs, the most direct way to recover usable specification range without increasing decision risk is to reduce measurement uncertainty at the source. That is precisely what a lower uncertainty calibration lab like Morehouse delivers, keeping those amber zones narrow so your full specification range remains usable, and your pass/fail decisions actually mean something.

Figure 5 A Normal Distribution (Bell-Curve) with PFA and PFR.

Picture the measurement result as a probability distribution, a bell curve centered on the observed value. The width of that bell curve is your measurement uncertainty. It represents your ignorance about where the true value actually lies.

When that bell curve is wide (high uncertainty), the tails of the distribution extend well past the specification limits. That means:

• A measurement result that looks in spec might have a true value that’s actually outside the limit that’s PFA.

• A measurement result that looks out of spec might have a true value that’s comfortably inside the limit that’s PFR.

When the bell curve is narrow (low uncertainty), the tails barely reach the spec limits. The measurement result is much more likely to mean what it says. PFA drops. PFR drops provided guard bands are recalculated to reflect the new, lower uncertainty.

Your pass/fail decisions start meaning what they say. Your pass/fail decisions start meaning what they say. Yet not all risk is the same. There are two distinct ways to think about it, depending on whether you are evaluating a population of instruments or a single result in front of you right now.

Figure 6. Example of boundary-case conditional risk under simple acceptance (illustrative ~50% at the limit under symmetric assumptions).

This is not a subtle effect. Under a simple acceptance (zero guard band) decision rule, if you assume a symmetric error model (often approximated as normal) and meagre prior information, a measurement result reported exactly at the tolerance limit implies an approximately 50% conditional probability that the true value lies on either side of that limit. In other words, at the boundary the decision can be no better than a coin flip. Reducing uncertainty (or applying an appropriate guard band/decision rule) sharply reduces this conditional risk for results near the tolerance boundary.

Global Risk vs. Specific Risk: Two Different Questions

This is where calibration gets nuanced, and where a lot of quality systems fall short.

Figure 7: Global Versus Specific Risk Defined.

Global (Unconditional) Risk

Global PFA asks: “Across all the instruments we calibrate, what percentage of truly bad ones do we accept?”

This is an average across your whole population of instruments and calibration events. It depends on your historical reliability data. How often do instruments drift out of spec? and on the uncertainty of your calibration process. You have to know how your instruments behave over time to calculate them properly.

In programs that contractually invoke ANSI/NCSL Z540.3, a common target is controlling consumer’s risk (PFA) for acceptance decisions to ≤2 % under the specified decision rule (often framed as an unconditional/population-level requirement). Achieving an unconditional risk target generally requires not only a measurement uncertainty model but also a defensible model for the population of items being calibrated (i.e., prior information), which is often supported by reliability/interval data such as end-of-period condition (EOPR) or similar historical evidence. Where such data are limited, some Z540.3 handbook approaches use managed guard banding to control reported risk under stated assumptions, at the cost of increased false rejects, without claiming full population characterization.

The catch: an instrument that passes a 2 % global PFA requirement can still present much higher specific risk on individual measurements near the tolerance boundary. The global number is an average it masks what’s happening at the edges.

Specific (Conditional) Risk

Specific PFA asks: “Given this particular measurement result, and given my uncertainty, what’s the probability this specific instrument is actually out of spec?”

This is a conditional probability it depends on where the observed value landed relative to the tolerance limits. A result right in the middle of the tolerance range? Very low specific PFA. A result right at the edge of the acceptance limit? The specific PFA can be substantial, even if the global number looks fine.

Specific risk is the right framework when you don’t have prior population data, what JCGM 106 calls ‘meagre prior information,’ or when a more cautious approach to false accepts is required, regardless of available data.

Guard Bands: The Balancing Mechanism

So how do labs manage the PFA/PFR trade-off in practice? With guard bands.

Figure 8: Image from ILAC G8 Showing Acceptence Zones with a Guard Band and Measurement Uncertainty.

A guard band is a buffer zone inside the specification limit. Instead of accepting any result that falls within the full spec range (LSL to USL), you only accept results that fall within a tighter acceptance limit (AL). The gap between the AL and the spec limit is the guard band.

Guard bands deliberately trade some PFR (you’ll reject more borderline-good instruments) in order to reduce PFA (you’ll accept fewer borderline-bad ones). The width of the guard band is set based on the measurement uncertainty and the desired maximum risk level.

Figure 9: A Specific Risk Example of 2.275 % PFA.

One common conditional-risk guard band approach is to choose the acceptance limit so that, for a stated probability model (often normal) and stated uncertainty (u or U), the conditional probability of nonconformance at the acceptance boundary equals a chosen maximum consumer’s risk, α (one-sided). A convenient form is:

Acceptance limit = Spec limit − g·U, where U = k·u and g = Φ⁻¹(1−α)/k for a normal model (Φ⁻¹ is the inverse standard normal cumulative distribution function, CDF).
A simpler, more conservative rule (often associated with ILAC guidance) is to set Acceptance limit = Spec limit − U, which does not explicitly target α but generally reduces consumer’s risk at the expense of more false rejects.

Here’s the key insight: the narrower your uncertainty, the narrower your guard band needs to be. With lower uncertainty, you can set tight guard bands without sacrificing much of the specification range, keeping PFR low while still controlling PFA. With large uncertainty, you either accept high PFA risk or apply wide guard bands that eat into your usable specification range and drive up false rejects.

An Important Clarification on what ISO/IEC 17025 Actually Requires

Note: It is worth being clear about what the standard does and does not mandate. ISO/IEC 17025:2017 requires that when a statement of conformity is made, the laboratory define and apply a decision rule that accounts for measurement uncertainty and considers the associated risk. It does not require guard bands, probability calculations, or any specific mathematical risk model.

Simple acceptance (a shared-risk decision rule with zero guard band), where a result is passed or failed purely on whether the reported value falls within the tolerance limits, may be used under ISO/IEC 17025, provided the decision rule is defined and (where applicable) agreed with the customer. The practical implication is that conditional consumer’s risk (PFA) for results close to a specification limit can be high. In the boundary case, depending on the assumed probability model and prior information, a result reported exactly at the limit can correspond to roughly 50% conditional probability of being on the nonconforming side. ISO/IEC 17025 does not mandate a specific guard band, but it does require that the decision rule and associated risk treatment be defined and applied.

Guard bands, probability-based decision rules, TUR requirements, and specific or global risk models are tools used to manage decision risk. They exist because the conditional risk near specification limits under shared-risk decision rules is often unacceptable in practice, especially when uncertainty is large relative to the tolerance.

If your calibration certificate shows a pass or fail with no mention of a decision rule or how uncertainty was accounted for, it is worth asking your lab which option they applied, because the answer has direct implications for how much you can trust that result.

What This Means in Practice

Figure 10: Why You Might Consider Using Morehouse for Force and Torque Measurements!

Let’s make this concrete. Imagine you’re calibrating a force measurement instrument with a ± 0.1 % of full-scale specification.

• Lab A (Not Morehouse) has measurement uncertainty of ±0.05 % (half the spec range). Their uncertainty distribution is wide relative to the tolerance. Any measurement near the spec limit carries meaningful PFA risk. They need wide guard bands, which means they’re effectively only accepting instruments that fall within perhaps ±0.04 % of nominal, a much tighter window than the stated spec. Their false reject rate is high.

• Lab B (Morehouse) has measurement uncertainty of ±0.002 % (25x lower). Their uncertainty distribution is narrow. The result at the spec limit has a tiny tail past the boundary. Guard bands can be narrow. The full specification range is practically usable. False accepts are near zero. False rejects are minimal.

• Same instrument. Same specification. Completely different risk profiles driven entirely by the calibration lab’s uncertainty.

Why This Matters for Your Quality System

If your calibration lab has high uncertainty, you face a choice that nobody usually makes explicit:

1. Accept the PFA risk — pass/fail decisions near tolerance limits are unreliable, and some fraction of out-of-spec instruments are in service right now.

2. Apply wide guard bands — increase your false reject rate and the associated costs.

3. Reduce calibration intervals — catch drift more often. This can help, but interval adjustments should be grounded in reliability evidence (EOPR data where available, or other defensible historical/engineering information). Shortening intervals arbitrarily does not guarantee an unconditional (population-level) risk target is met and can add cost without a quantified benefit.

None of these is a great option. The better solution is to reduce the measurement uncertainty at the source by using a calibration lab with lower uncertainty.

For force and torque calibration specifically, the difference between a general-purpose lab and a specialist with primary deadweight standards can be 10–50× in uncertainty. That’s not a marginal improvement. That’s the difference between a quality system that’s actually working and one that’s generating paperwork while hoping for the best.

The Bottom Line

Measurement uncertainty isn’t just a number on a certificate. It’s a direct driver of how many bad instruments you accept into service and how many good ones you unnecessarily reject. Both have real costs. One shows up in quality and risk. The other shows up in your budget.

Understanding PFA and PFR, the difference between global and specific risk, and how guard bands work gives you the tools to have an intelligent conversation with your calibration provider about whether their uncertainty is actually fit for your purpose.

The next time you get a calibration certificate, don’t just check the pass/fail result. Look at the uncertainty. Then ask yourself: given where that measurement landed relative to my tolerance, what’s the actual probability it’s telling me the truth?

Morehouse is an ISO/IEC 17025 accredited force and torque calibration laboratory (NVLAP Lab Code 600259, A2LA Cert. 1398.01), founded in 1920 and based in York, PA. Morehouse specializes in ultra-low uncertainty force calibration using primary deadweight standards known to better than 0.0025 % of applied force up to 120,000 lbf, and other calibrations better than 0.01 % of applied force up to 2,250,000 lbf (10MN).

mhforce.com · (717) 843-0081

This article focuses only on load cells, with load cell specific examples and the setup errors that most often cause measurement problems.

What a Bad Load Cell Setup is at a high level

Load cells are designed and calibrated to measure force along a preferred axis, usually axial tension or compression. A setup becomes a Bad Load Cell Setup when the applied force no longer travels through that intended axis consistently.

Most bad load cell setups fall into these buckets.

Off-axis loading
The force is not centered or aligned with the load cell’s intended axis, which creates bending moments and cosine error.

Side loading and torsion
The fixture applies force with a lateral component or twists the load cell stack.

Uncontrolled threaded interfaces
Thread engagement depth is inconsistent, bottoming out occurs, or torque varies. Any of these can change the internal load path.

Unstable contact surfaces
Soft or uneven surfaces indent, slip, or seat during loading, changing the contact area and altering the output.

The key idea is simple. If the mechanical interface is not repeatable, the measurement will not be repeatable.

Some common measurement errors from bad load cell setups

Bias shifts
A small alignment change can cause a large shift in indicated force because the stress state inside the load cell changes.

Note: Misalignment typically occurs between the measuring and applied force vectors.

Nonlinearity and hysteresis changes
Off-axis forces and unstable contact conditions can change the apparent nonlinearity and hysteresis because the cell is not loaded the way it was designed or calibrated. Off-axis loading can also alter the slope and span of the load cell.

Poor reproducibility after rotation

A setup may look fine until you rotate a component or swap an adapter. Then readings move because contact points and force lines move.

Drift and return to zero problems
If the load cell is pressed into a softer surface or a surface that seats or creeps, you can see drift during loading and poor return to zero after unloading. Though a poor zero return could also be due to a poor load cell design.

Note: the figure shows an upward drift, though some load cells may have negative creep.

Specific load cell setup examples that create bad results

Note: Non-center load should theoretically run lower, not higher, on a good load cell.

Example 1: S-beam load cell with misaligned rigging that creates a side pull (off-axis load)

What it looks like
An S beam load cell is installed with rigging hardware such as clevises, shackles, and pins that does not keep the line of force centered. Clearance in the hardware lets the connection shift under load. The load cell sees a side pull or twisting component.

Why this becomes a Bad Load Cell Setup
S beam load cells are commonly used in tension applications, but they are sensitive to off-axis loading. If rigging allows the force line to wander even slightly, the load cell experiences bending and side load instead of pure axial tension.

Typical measurement problems
Even slight misalignment less than 1/8 inch has been documented to produce a large errors (our example shows 0.752 % error on a 10,000 lbf S-beam), this error caused the expanded uncertainty to increase approximately 9-fold (from roughly 10 lbf to 90 lbf). This is sufficient to cause a technician to adjust a machine that is actually in-tolerance, with potential downstream recall consequences. Repeatability is poor from run to run.

What to watch for
Use hardware that keeps the force line centered and stable. Minimize clearance and uncontrolled rotation at pins. Avoid geometry that allows side pull.

Example 2: S-beam load cell in compression loaded through threads versus loaded flat

What it looks like

The same S-beam load cell is used in compression, but the compression load can be introduced in different ways depending on the adapter and setup. In one configuration, the force is applied through a threaded interface (for example, through the top and bottom threads), which is generally preferred when symmetry error is a concern. Symmetry error is the difference in output between the maximum force in compression and the maximum force in tension.

In other configurations, the load cell may be tightened against the threads or loaded flat-on-flat against a base or platen. The figure above shows the output of the S-beam load cell under these different adapters and loading conditions. In practice, the adapter often changes between calibration and use, even when the load cell itself stays the same.

Why this is a bad setup

Some load cells, especially S-beam designs, can produce different outputs depending on how compression enters the body. Threaded versus flat loading changes the stress distribution through the element and can change the indicated force.

When loaded through the threads in both modes, the S-beam load cell’s symmetry is often extremely good. However, flat-on-flat loading is not recommended because the output can vary significantly depending on exactly where and how the force transfers through the contacting materials. Even changes in the interface area at the top or bottom can alter deflection and, therefore, the reading.

Typical measurement problems

The most common issue is that the load cell calibrates one way but reads differently in production when the loading interface changes. This typically shows up as a bias shift and can also appear as a change in nonlinearity.

The worst-case documented in testing is comparing full thread-through loading to flat-on-flat loading, which produced a maximum output difference of 0.369 %. A smaller but still meaningful error, almost 0.03 %, was observed simply from the difference between loading through the threads versus loading flat on the bottom while still threading on top. These are not edge cases; they represent transitions that commonly occur when a load cell is calibrated one way and shipped to a customer who uses it another way.

What to watch for

Calibrate using the same compression interface you will use in the field. Keep adapters consistent and treat them as part of the measurement system. If symmetry error matters, loading through both top and bottom threads is often preferred, but whatever configuration you choose, the key is to avoid switching interfaces between calibration and application.

Example 3: Button load cell with off-center loading and a soft contact surface

What it looks like (Bad setup on the left side)
A button load cell is placed between platens for a compression test. The contact point is slightly off center, the platens are not perfectly parallel, or the button is centered by eye instead of being mechanically aligned. At the same time, the button is pressing into a surface that is softer than the load cell or not truly flat, so the contact spot shifts and grows as the force increases.

Why this becomes a Bad Load Cell Setup
Button load cells have a small contact area and are highly sensitive to how force enters the sensor. A 0.1 % misalignment can produce a large cosine error, and cells calibrated without proper adapters have been documented to show errors from 1 % to 10 % of rated output. In Morehouse testing, using proper button load cell adapters improved reproducibility by 525 % compared to manually aligned setups.

When the load is even slightly off center, the button sees bending and side loading instead of clean axial compression. If the mating surface is soft or seats under load, it can indent and reform the contact patch during the test. That means the load path is changing while you are trying to measure it, which is exactly what drives unstable force readings.

Typical measurement problems
The readings often shift as the setup changes, even if you are holding a constant force. Drift becomes more noticeable during force holds because the surface continues deforming. Hysteresis and return to zero get worse because loading and unloading do not follow the same contact condition. Repeatability suffers when you reposition the button, rotate it, or rebuild the stack, and the scatter usually grows at higher loads, where small alignment errors create bigger bending effects.

What to watch for
Use proper button load cell adapters and alignment features so the force stays centered and consistent. Keep platens flat, parallel, and hard enough that they do not indent under load. Avoid hard on soft contact by using hardened loading blocks or stable interfaces. Treat adapters and contact surfaces as part of the measurement system and keep them consistent between calibration and use.

Example 4: Top block hardness and flatness can change a load cell’s output

What it looks like
A hard load cell is loaded against a softer platen or material. Under load, the surface deforms and the contact patch changes during the test.

Why this is a bad setup
If the mating surface indents or creeps, the load path changes during the measurement. With button and miniature designs, small contact areas make this especially damaging because the setup depends on stable contact conditions.

Typical measurement problems
Our example shows the output can shift by up to approximately 0.3 % when the top pad material or hardness changes. Flatness is a separate and compounding contributor: a non-flat adapter or base has been documented to triple repeatability errors, and a non-flat base can increase rotational error by approximately 5×. Both hardness and flatness must be controlled they are not the same problem. Expanded uncertainty can jump because the adapter becomes the dominant contributor. Repeatability also worsens noticeably when the top block or load cell base is not flat on the ground, and results may vary more between rebuilds or different block configurations.

What to watch for
Use hardened, flat, stable loading blocks or platens. Replace or recondition contact surfaces when they wear or dish. Avoid hard on soft contact when accuracy matters.

Example 5 Shear web or shear beam load cell with inconsistent thread engagement

What it looks like
A shear web load cell is installed using threaded adapters. Thread engagement depth varies between setups. The adapter bottoms out differently or is tightened differently. Sometimes the adapter is omitted when it should be used.

Why this becomes a Bad Load Cell Setup
Even if a load cell is robust to some off axis effects, the threaded interface can dominate the error. Thread engagement changes can shift how the load transfers into the sensing element, changing output. One specific, documented example: if an alignment plug is not seated flush to the base, leaving even one or two threads exposed below the load cell, the load transfers through the internal threads rather than the base, adding approximately 0.012 % calibration error on shear web cells and often damaging the alignment plug. The correct standard is to thread the plug flush to the bottom of the load cell, then one additional turn in, verifying no threads are exposed. This should be inspected visually every time before loading.

Typical measurement problems
Bias shifts when the adapter is installed differently. Differences appear between calibration and use. Results change when swapping adapters even if the load cell is unchanged.

What to watch for
Standardize thread engagement depth and tightening method. Prevent bottoming out. Treat the adapter as part of the system and keep it consistent from calibration to use.

What should your load cell look like

Having the adapter installed is better because it keeps the load cell’s force path controlled and repeatable, which is the whole goal of accurate force measurement.

When the correct adapter is installed, the load is introduced into the load cell the way it was designed to be loaded. That means the contact surface, alignment features, and thread engagement are consistent, so the strain pattern inside the sensing element stays predictable. This reduces off-axis loading, bending moments, and “mystery” stress paths that can change the output even when the applied force is the same.

It also protects you from common setup errors. Without the adapter, it is easier to bottom out threads, vary engagement depth, or load through unintended surfaces. Those small assembly differences can shift the load path and create bias shifts, worse repeatability, and changes in apparent nonlinearity or hysteresis. With the adapter installed, you remove a lot of that variability because the interface is standardized.

Finally, keeping the adapter installed makes calibration more meaningful. If the load cell is calibrated with a specific adapter and then used without it, you are no longer matching the calibration conditions. Using the same adapter in calibration and in the field makes your calibration results transfer more reliably to real measurements.

What else to look out for

In this picture, it is apparent that someone tried to remove the top adapter. If we calibrate the load cell and observe a large shift exceeding expectations, we have good reason to believe the evidence before us indicates that someone removed the adapter. The simple solution is DO NOT DO OR TRY THIS!

Example 6: Shear web load cell loaded through the base versus flat

What it looks like
An alignment plug or threaded alignment feature is installed incorrectly. Threads remain exposed where the plug should be seated.

Why this is a bad setup
When an alignment plug is not seated correctly, the load can be introduced in a way that changes the internal load path or loads through threads in an unintended way. This can create measurable error and can damage components.

Typical measurement problems
Small but real bias errors that are hard to diagnose. Setup to setup differences because assembly is not repeatable.

What to watch for
Use a clear assembly standard for seating depth and inspection. Verify installation visually every time before loading.

Example 7: Torque and Bolting a Load Cell

What it looks like
A load cell that is normally bolted to a base in a test machine is unbolted for calibration or moved between machines, then reinstalled later. The load cell may be bolted down by a different technician, with different bolts, a different base, a different tightening method, or a different torque value. Sometimes the bolts are not tightened in a consistent star or crisscross pattern, and sometimes one or more bolts are not fully torqued. In other cases, the load cell is shipped in without being bolted to its base at all, then mounted at the calibration lab in a way that does not match how it is installed in the field.

Why this becomes a Bad Load Cell Setup
Bolting is part of the force path. Changing the bolting procedure changes how the load transfers through the load cell and into the base, which changes strain distribution and output. Differences in thread class, bolt properties, base material, surface finish, hardness, stiffness, alignment, and flatness all become additional error sources when the load cell is removed and reinstalled. Even if the same torque value is targeted, variability between technicians and tightening methods can be large and hard to quantify. If the calibration is performed with a different mounting condition than the real application, the calibration data may not represent how the load cell actually behaves when it is used.

Typical measurement problems
The most common symptom is a measurable shift from previous calibration results after the load cell is reinstalled. Output can change simply from tightening to a different torque value. In a real calibration example on a 60,000 lbf load cell, changing the bolting torque from 25 lbf ft to 45 lbf ft produced about a 0.7 percent output difference, and the deviations from the fitted curve were much lower when the load cell was bolted properly. Reproducibility can also collapse when bolting is inconsistent. Missing one or two bolts or failing to torque evenly can make performance significantly worse, and errors can be several times higher than expected. These changes often show up as poor repeatability between rebuilds, larger scatter, unexpected nonlinearity, and results that depend on who installed the load cell and how they tightened it.

What to watch for
Bolt the load cell to its base before sending it for calibration, and use the same bolting method you use in the field. Send the actual bolts used to attach it, and send a base that matches the material and flatness of the one used in service whenever possible. Use a consistent tightening sequence, hand-tighten in a crisscross or star pattern, then final torque with a calibrated torque wrench in the same pattern. Make sure every bolt is present and torqued correctly. Missing or unevenly torqued bolts have been documented to produce errors three times the expected performance.

Example 8: Thread Reducer and Thread Depth

What it looks like
A threaded reducer, bushing, or thread adapter is installed in the load cell’s main threaded port so a smaller stud, eye bolt, or fixture can be used. The reducer may sit proud, sit too deep, bottom out early, or have inconsistent engagement from one setup to the next. In some cases it is installed with whatever tool is available, without a clear standard for torque, thread locker, or engagement depth. The reducer can also introduce extra interfaces, meaning more places for looseness, rocking, or misalignment to start.

Why this becomes a Bad Load Cell Setup
A thread reducer changes the way force enters the load cell. Just like varying thread depth, changing engagement depth or bottoming conditions changes the internal load path and can shift the load cell’s output. A reducer also adds a second threaded joint, which increases variability and increases the chance of unintended bending or side load if the load line is not perfectly centered. The bigger issue is safety. A reducer can reduce the effective thread engagement, change the shear area, and create stress concentrations at the interface. The best practice is to avoid thread reducers entirely by using correctly-sized threaded hardware or purpose-built adapters for force measurement. When a reducer cannot be avoided, it must be treated as a controlled measurement component, not a convenience part, and rated for the full load, material, and duty cycle.

Typical measurement problems
You will often see output shifts between setups that look similar on the outside because the reducer seats differently each time. This can show up as poor reproducibility after reinstalling, swapping fixtures, or rebuilding the stack. The load cell may also appear to “change calibration” because the strain pattern changes with engagement depth and seating condition. In addition to measurement error, reducers introduce safety-related failure modes such as backing out under cyclic loading, thread galling or stripping, cracked reducers from over torque, and sudden load release if the reducer or mating stud fails. Even when it does not fail, slight loosening can add movement that increases scatter and makes the system feel unstable.

What to watch for
Use a reducer only when it is rated for the full load, material, and duty cycle, and only when it provides adequate thread engagement for both sides of the joint. Standardize the installation, including engagement depth, torque, and whether thread locking is allowed for that application. Make sure the reducer does not bottom out in a way that changes seating from one build to the next. Inspect threads for wear, galling, and deformation, especially after overloads or cyclic testing. Keep the load line centered and avoid any setup that adds bending or side loading through the reducer. If safety is critical or the loading is dynamic, treat the reducer as a controlled component, not a convenience part, and consider eliminating it by using the correct threaded hardware or a purpose-built adapter designed for force measurement.

Example 9: Cable Length, Gauge, and Configuration Errors

What it looks like

A load cell is calibrated with a specific cable of a fixed length, and the gauge is then returned to service with a different cable. The replacement may be longer, shorter, a different wire gauge, or a different configuration entirely. In other cases, the original cable is damaged and spliced or swapped without recalibration. Sometimes, a 4-wire cable that shipped with the cell is replaced with whatever cable is available in the shop.

Why this becomes a Bad Load Cell Setup

Cable errors fall into two distinct categories that are easy to conflate but have different causes and fixes.

The first is a mechanical error: on smaller load cells in particular, a stiff cable can provide a parallel load path, meaning some of the applied force travels through the cable rather than through the sensing element. This is a setup problem, not an electrical one. The mitigation is to orient the transducer so the live end faces toward where the force is being generated, and to ensure the cable does not create a lateral pull or introduce bending at the connection point.

The second is an electrical error caused by resistance change in 4-wire systems. In a 4-wire cable, the excitation and signal lines share the same conductors. Cable resistance is a function of temperature, and any change in cable length, gauge, or operating temperature causes a voltage drop that directly affects span output. A 4-wire system has no way to compensate for variations in lead resistance.

The sensitivity error from changing a 4-wire cable is quantifiable and significant:

• Approximately 0.37 % per 10 feet of 28-gauge cable

• Approximately 0.09 % per 10 feet of 22-gauge cable

In documented testing, interchanging cables in a 4-wire system produced an error of 0.106 %, and adding that error to the combined uncertainty roughly doubled the uncertainty at mid-range forces compared to the same cell with a 6-wire cable.

Typical measurement problems

Span shifts after cable replacement or repair, without any change to the load cell itself. Temperature-dependent drift in readings that is mistakenly attributed to the load cell or the indicator. Results that differ between the calibration lab (with one cable) and the field (with a different cable). Recalibration appears to fix the problem, but the error returns when the cable is changed again.

What to watch for

For 4-wire systems: treat the cable as part of the calibrated measurement system. Mark each cable with a serial number and keep it paired to the system it was calibrated with. Do not substitute cables of a different length or gauge without recalibrating the complete system. If a cable is damaged and must be replaced, recalibrate immediately after.

The better long-term solution is to move to a 6-wire (sense lead) system. In a 6-wire setup, dedicated sense lines allow the indicator to measure and compensate for the actual voltage at the load cell terminals, eliminating the effect of lead resistance variation. A 6-wire system produces consistent readings regardless of cable length or gauge, and is particularly important for long cable runs or environments where temperature changes significantly.

If you are currently using a 4-wire system in a fixed installation where the cable never changes, the error is systematic and will be absorbed into calibration. The risk is when the cable is changed, and the system is not recalibrated.

Load cell types and set up risks with what to watch for

S-beam load cells
High risk includes side pull, twisting, and changing rigging geometry. Watch for a centered force line, stable pin and clevis connections, and consistent compression interface.

Button and miniature load cells
High risk includes off-center contact, dependence on surface condition, and rotation sensitivity. Watch for alignment features, hardened flat parallel platens, and a stable contact patch. Errors of 1 % to 10 % of rated output without proper adapters; proper adapters can improve reproducibility by 525 %

Shear web
High risk includes threaded interface errors, adapter omission, and inconsistent engagement length. Watch for controlled engagement depth, no bottoming out, and consistent adapters. Misalignment error is much lower (Our examples document the error at 0.0022 % versus 0.75 % for S-beams under comparable conditions). Loading through the bottom threads versus on the base is a consistent error of 0.01 – 0.012 %, while varying the thread depth can be 10x that or greater.

Column load cells
High risk includes off-center loading, overhung moment, and inconsistent top adapters. Watch for centering, a rigid short stack, and consistent adapter material and geometry.

Bad Load Cell Setup checklist before you test

Stop and fix the mechanics first if any of these are true.

• The force line is not clearly centered and axial through the load cell

• There is clearance or geometry that allows the load to walk under force

• You are using different adapters than the ones used during calibration

• Thread engagement depth or tightening method is not controlled

• The load cell contacts a softer surface that can indent or seat during loading

• The stack is tall, flexible, or overhung and likely introduces bending or torsion

• The adapter or contact surface is not flat, worn, or dished

• The bolts are missing, unevenly torqued, or tightened in a different sequence than during calibration

• The load cell requires bolting and was calibrated unbolted, or bolted to a different torque value than used in the field

• The torque wrench used to mount the load cell has not been verified for accuracy

• An alignment plug is installed but not seated correctly — any threads visible below the base of the load cell mean the load is transferring through the internal threads, not the base

• The loading interface changed between calibration and use — for example, thread-loaded during calibration but flat-on-flat in the field, or vice versa

• A thread reducer or thread adapter is in use without a controlled standard for engagement depth and torque

• The cable has been replaced, extended, or changed to a different gauge since the last calibration and the system uses a 4-wire configuration

• The cable is stiff enough on a smaller load cell that it may be providing a parallel load path and carrying part of the applied force

Practical takeaway

A Bad Load Cell Setup is a mechanical problem before it is a measurement problem. The best accuracy improvement is usually a repeatable axial load path with controlled interfaces and stable contact surfaces.

Reference basis: Zumbrun, Henry A., Force Calibration for Technicians and Quality Managers, 4th edition (2026).

Key topics used throughout: stress/strain fundamentals; compression vs tension setups; loading-condition impacts;

off-center and side-load sensitivity; adapter hardness/flatness; thread engagement; measurement uncertainty and risk.

Example 1 - Small pin in a larger hole; cracked clevis with an improvised C-clamp fix (Joking on the fix 😊)

What happened: A small pin was installed in a larger clevis hole. The resulting stress concentration and bending cracked the clevis. A C-clamp was added as an improvised “fix.”

Why this is bad practice:

· A pin-to-hole mismatch shifts load transfer to a small contact area and produces localized high stress (stress increases as effective area decreases).

· The load path is no longer axial; eccentric contact introduces bending and side load components.

Measurement consequences:

· Introduces bending strain that the sensor may interpret as force, causing bias and nonlinearity.

· Creates uncontrolled Type B uncertainty contributors that are not represented in the calibration certificate.

Safety consequences:

· Crack propagation and sudden fracture risk under load.

· Clamp slip or fracture can create projectile or pinch hazards.

Bottom line: If the mechanical joint is not designed for the load path, calibration traceability is irrelevant - the setup itself dominates the error and risk.

Example 2 - Welded clevis not perpendicular; undersized pin

What happened: A clevis was welded out of square (not perpendicular) and assembled with a pin much smaller than the hole.

Why this is bad practice:

· Non-perpendicular geometry forces the load line off-axis, introducing side loading and torsion.

· Weld distortion and heat-affected zones can change material properties and stiffness.

· Undersized pins repeat the stress concentration and bending problems from Example 1.

Measurement consequences:

· Side load sensitivity can become a dominant error term; readings can shift with small alignment changes.

· Performance metrics such as Non-Linearity and hysteresis can change due to the different stress states. Zero shifts may also occur.

· Results are not reproducible unless the exact geometry and contact conditions are replicated.

Safety consequences:

· Fatigue and fracture risk at the weld and clevis corners.

· Potential for pin shear or clevis tear-out under shock loading.

Bottom line: Out-of-square fixtures convert an axial force measurement into a combined bending/torsion problem, increasing error and hazard.

Example 3 - Small pin and large hole (repeat failure mode)

What happened: Another setup where a small pin is used in a larger hole, producing clearance, impact loading, and uneven contact. In addition to the large clearance for the pin, this setup is also wider than it needs to be.

Why this is bad practice:

· Clearance allows rotation and shifting contact points, so the line of force changes during loading.

· Changing contact area changes stress distribution and strain output, violating repeatability conditions.

· Causes extra bending stresses in the pin. The loading capacity of the pin depends on both the diameter and the length between the clevis.

· If loaded off center in a Morehouse UCM or Deadweight Machine, it can create higher stresses on one side of yoke. This can create bending or even breakage. If the pin was strong enough, spacers could be used to center the shackle.

Measurement consequences:

· Expect large run-to-run variability and poor reproducibility, especially if the setup is rebuilt.

· Potential for step changes in output as contact shifts during loading.

Safety consequences:

· Impact loading and sudden shifts can overload fixtures or sensors unexpectedly.

Bottom line: If the force line moves during the test, the measurement is not controlled - and the uncertainty can grow dramatically.

Example 4 - Using a compression device in a tension application (unsafe orientation)

What happened: A compression digital ring force gauge was used in tension. The application appears unstable and unsafe.

Why this is bad practice:

· Compression and tension setups require different adapters and loading paths; many devices are not designed to be pulled in the same way they are compressed.

· Changing loading direction changes the stress state and can invalidate calibration assumptions.

Measurement consequences:

· These devices do not have great symmetry and thus the compression calibration will have different deflection from that of the tension calibration.

Safety consequences:

· Dropped load risk if the device or adapters are not rated for tension.

· Sudden failure can create severe injury risk in line-of-force tension setups.

Bottom line: Do not change the intended loading direction without proper fixtures, ratings, and a calibration that matches the use condition.

Example 5 - Uneven loading against a surface (UCM loaded flat without proper contact control)

What happened: Several Devices Under Test were loaded against a surface in a way that creates uneven loading and contact.

Why this is bad practice:

· Adapter flatness, hardness, and controlled contact geometry are required to keep loading uniform.

· Non-flat or uneven contact causes stress gradients and off-axis components.

Measurement consequences:

· Increases off-center loading error and can shift output between runs.

Safety consequences:

· Contact damage (brinelling) and sudden slip risk as surfaces seat.

Bottom line: If you cannot control contact surfaces, you cannot control force data - the setup will dominate the result.

Example 6 - Load cell harder than the material it is loaded against (needs softer sacrificial block)

What happened: A hard load cell was loaded against a softer material, risking damage to the mating surface and changing contact conditions.

Why this is bad practice:

· Soft mating materials can indent under load, changing contact area and load distribution during the test.

· This violates the requirement that loading conditions be stable and repeatable.

Measurement consequences:

· Output can drift as indentation progresses; hysteresis and return-to-zero error increase.

· Repeatability degrades because each loading cycle changes the contact surface.

Safety consequences:

· Surface damage can cause unexpected seating or slip, leading to sudden load shifts.

Bottom line: Use properly designed sacrificial blocks or adapters so contact hardness and flatness remain controlled and repeatable. Typically, the blocks used in calibrating machines will be softer than the load cell and the loading surfaces of the machine. These blocks will likely need to be ground flat prior to using them again. Best practice is to have matching top and bottom blocks for any load cell that is sent in with it for calibration.

Example 7 - Uncontrolled thread engagement, adapter geometry, and contact surfaces

What happened: Thread engagement, adapter geometry, andontact surfaces are not controlled or repeatable.

Why this is bad practice:

· Thread engagement and bottoming-out can change the load path and strain distribution.

· Different adapters (or the same adapter installed to a different depth/torque) can produce materially different outputs.

Measurement consequences:

· Expect bias shifts and nonlinearity changes when adapters are swapped or installed differently.

· Significant errors (often 0.1 % to 2 % depending on the design and engagement) are possible if calibration conditions are not replicated.

Safety consequences:

· Thread stripping or sudden disengagement risk if engagement is insufficient or mismatched.

Bottom line: If the mechanical interface is not repeatable, your force measurement is not repeatable - and your calibration data will not transfer.

Example 8 - Applied force outside the intended force line (off-axis loading shown)

What happened: Force is applied well away from the intended line of force through the center of the sensing element.

Why this is bad practice:

· Off-axis loading introduces bending moments and cosine error.

· Many sensors have specified off-center and side-load sensitivities that become dominant when misaligned.

Measurement consequences:

· Large indication errors are expected; small misalignments can create outsized measurement bias.

· Rotation or repositioning often yields different results, indicating poor reproducibility.

Safety consequences:

· Bending stresses can overload fixtures and cause sudden failure.

Bottom line: Correct the load path first; do not attempt to “calibrate out” a bad mechanical setup.

Example 9 - Overhung moment and torsion (errors likely > 0.3 %)

What happened: The setup introduces a clear overhung moment and torsion rather than pure axial load.

Why this is bad practice:

· Force measurement devices are typically calibrated for axial loading. Adding torsion and bending changes the stress state and the sensor output.

· The machine and fixtures must be rigid, plumb, and aligned to minimize extraneous loads.

Measurement consequences:

· Errors greater than 0.3% are plausible; hysteresis and nonlinearity increase.

· Repeatability may look acceptable in one orientation but fail when rebuilt or rotated.

Safety consequences:

· Overhung loads can cause fixture rotation, slip, or sudden release.

Bottom line: Eliminate torsion and bending moments - otherwise the measurement becomes an uncontrolled multi-axis problem.

Example 10 - Reference standard placed below the jack (easy to distort the axial line)

What happened: The machine was designed for the reference to be above the jack, but the reference is placed below. The axial load line is easy to distort.

Why this is bad practice:

· Machine configuration affects alignment and rigidity; changing load stack order can introduce angular misalignment and torsion.

· Calibration results are only portable when the use conditions replicate the calibration conditions.

Measurement consequences:

· Increased bias and instability due to shifting load line.

· Greater uncertainty and higher probability of failing comparisons (e.g., PT/ILC) due to non-reproducible loading conditions.

Safety consequences:

· Unstable stack height and side loads increase the chance of buckling or sudden slip.

Bottom line: Use the machine as designed, or treat the alternate configuration as a different method that requires validation and matching calibration conditions.

Example 11 - Failed PT test due to misalignment (not surprising)

What happened: The setup is visibly misaligned throughout, resulting in poor agreement and a failed proficiency test.

Why this is bad practice:

· Measurement accuracy is a combination of trueness (bias) and precision; poor alignment creates systematic bias.

· Higher uncertainty tightens decision limits (guard banding), increasing false failures or false accepts if not handled correctly.

Measurement consequences:

· High between-lab variability and inconsistent results across setups.

· Elevated measurement decision risk when uncertainty is underestimated or ignored.

Safety consequences:

· Misaligned mechanical stacks are more likely to slip, buckle, or fracture under load.

Bottom line: PT/ILC failures are often mechanical, not mathematical - fix alignment, adapters, and loading conditions before blaming the reference standard.

Final technical summary

· Large force measurement errors almost always originate in mechanics: alignment, adapters, contact surfaces, thread engagement, and load path.

· Calibration results are only portable when the use setup replicates the calibration conditions (fixtures, adapters, rigidity, and loading direction).

· Off-axis loading introduces bending/torsion that can dominate the signal and create large Type B uncertainty contributors.

· When loading conditions are not controlled, repeatability can appear acceptable while reproducibility collapses (especially after disassembly/rebuild or rotation).

· Unsafe fixtures increase the likelihood of sudden slip, fracture, or dropped loads - treat safety as part of the measurement system.

Training note (recommended): Review these pictures during onboarding and require technicians to identify:

1) The intended line of force; 2) Where bending/torsion is introduced; 3) What must be controlled to make results repeatable.

If the team cannot describe the load path, do not run the test.

However, intermixing any load cell and meter combination makes little sense. If someone needs two load cells and a meter to do a specific job, they can check out the two load cells and a meter, along with adapters, to perform the calibration or test. The unused load cells would then be available for someone else. If done correctly, there is quite a bit of benefit. However, there are several disadvantages and obstacles to overcome.

Let’s start by addressing the requirements of ISO 376 and ASTM E74. These are standards required to calibrate force-proving instruments, most commonly known as load cells, to calibrate other force-measuring instruments, force machines, hardness machines, and testing machines, using ASTM E74, ASTM E4, ASTM E-10, ASTM E-18, ISO 376, ISO 7500, and so on.

ISO 376 and ASTM E74 requirements for meter calibration

ISO 376 in section C.2.11 Effect of a replacement indicator states,

“The deviation between the two indicators should be determined (there are several methods, e.g., calibration of both indicators, use of a common bridge simulator), and the uncertainty of this deviation should be estimated (including factors such as the calibration uncertainty of the indicators and the stability of the common bridge simulator). If corrections are made, the uncertainty of the deviation should be taken into account. If no corrections are made, the deviation and its uncertainty should be considered.”

Thus, if the calibration equation is not used and instead various span points are programmed into an indicator, enabling input of points from the calibration curve for display in force or torque units with linear interpolation between them, it is crucial to examine the impact of this approximation on the curve. An assessment of uncertainty contribution should be incorporated.

ISO 376 section 3.1 defines a force-proving instrument as a “whole assembly from the force transducer through to, and including, the indicator.”

One might be thinking, I do not calibrate following ISO 376. One could only use the ASTM E74 standard or a commercial calibration.

ASTM E74 is a bit more prescriptive in the requirements for substitution. Section 12 is explicitly titled Substitution of Electronic Indicating Instruments Used with Force-Measuring Systems.

The standard acknowledges that it might be desirable to treat the indicator and force-measuring instrument separately.

A huge benefit is that if you purchase the same indicators, one could be used as a backup if the primary unit fails. Hence, there is a possibility of circumventing the costly calibration of the entire system.

Then, the standard goes on to list conditions that shall be satisfied to substitute a metrologically significant element of the electronic indicating instrument.

ASTM E74 Section 12.1.1 specifically states, “The electronic-indicating instrument used in the initial calibration and the instrument to be substituted shall each have been calibrated and their measurement uncertainties determined. The electronic indicating instrument to be substituted shall be calibrated with traceability to the SI over the full range of its intended use, including both positive and negative values if the system is used in tension and compression. The calibrated range shall include a point less than or equal to the output of the force transducer at the lower force limit, and a point equal to or greater than the output of the force transducer at the maximum applied force. A minimum of five points shall be taken within this range. The measurement uncertainty of each electronic indicating instrument shall be less than or equal to one-third of the uncertainty for the force-measuring system over the range from the lower force limit to the maximum force.”

To summarize, you must calibrate a simulator to comply with the standard. The simulator needs to support both positive and negative values if the load cells are used in both directions.

The simulator must have at least one point less than or equal to the lowest force point value in the range and one for the highest point.

Below is a picture of a Morehouse simulator. This simulator likely cannot meet these requirements.

Morehouse Budget mV/V Load Cell Simulator

The first point is 0.5 mV/V, and the last is 4 mV/V. If someone had a 4 mV/V (10,000 Force Units) load cell and the range of verified force was 500 through 10,000 Force Units, the simulator at 0.5 mV/V is 1,250 Force Units. If the verified range of forces started at 200, a 0.08 mV/V first step would be required.

Note: The best high-end simulators typically have the first step of 0.04 mV/V or lower, as 0.04 mV/V on a 2 mV/V load cell equates to a 2 % LLF. A simulator that starts at 0.1 mV/V would equate to a 5 % LLF on a 2 mV/V load cell.

Using this simulator, 2 mV/V is 5,000 Force Units. The end-user would need to raise their Class A verified range of forces to 1,250 FU. This situation does not work for many as they want to capture force values from the first non-zero calibrated point, typically below 5 % of the load cell’s capacity.

The ASTM E 74 standard gives further guidance by stating the measurement uncertainty of the indicator shall be determined by one of the methods in Appendix X2. It recommends the simulator has a series of mV/V steps of the measurement range with similar impedance characteristics and then states this requirement in section 12.1.2.

“The measurement uncertainty of the transducer simulator shall be less than or equal to one-tenth of the uncertainty for the force-measuring instrument.” ASTM E74 further states, “Excitation voltage amplitude, frequency, and waveform shall be maintained in the substitution within limits to ensure that the affect on the calibration is negligible. It is the user’s responsibility to determine limits on these parameters through measurement uncertainty analysis and appropriate tests to ensure that this requirement is met. Substitution of an interconnect cable can have a significant effect on calibration. If an interconnect cable is to be substituted, see Note 15.”

This is interesting as the interconnect cable for the simulator does not always share the same connection as the load cell. If the system is not 6-wire, meticulous care will be needed to ensure the same gauge and length of wire are used for the simulator-to-meter connection as for the load cell.

Note 15 goes into more detail, and Morehouse covers much of this in our section about four and six-wire indicators.

Appendix X2 details all the steps necessary to determine the uncertainty. Morehouse fully supports ASTM E74 and feels the membership is incredible. For around $ 100.00, one can join and access a catalog of standards. This author believes this is one of the best deals in the industry. Signing up is simply at astm.org.

Summary of Top Requirements for Load Cell Simulator Calibration Needed for Meter Substitution

The summation of what is needed is as follows:

1. At least five readings for each polarity over the range must be taken.

2. The points need to be less than or equal to the first point in the Class A or AA verified range of forces, and the capacity needs to have a point equal to or greater than the maximum output observed during calibration. So, if loading a 10,000 Force Unit load cell to 11,000 Force Units, which might read 4.4 mV/V, a 4.0 mV/V simulator is not good enough.

3. The simulator shall provide at least one point for every 20 % interval throughout the range. (Interesting tidbit here, as the standard says five points, though the simulator likely needs to have the low force point and an additional 5 points to cover up to capacity or higher for a total of six points throughout the range)

4. Section 8. Calculation and Analysis of Data of the ASTM E74 standard provides guidance to determine the standard deviation Type A uncertainty component for calibration of the simulator.

Okay, so the benefits might still outweigh the additional headache of using a simulator and being able to separate one’s load cells from the indicator or decouple the system.

However, there are many more error sources one needs to be aware of.

These include Calibration Uncertainty (Gain Error), Zero Offset, Temperature Effect on Sensitivity, Quantization Error, Normal Mode Voltage, Power Line Voltage Variation, Non-Linearity, Temperature Effect on Zero, Gain and Zero Stability, Common-Mode Voltage, Noise, Electrical Loading, Error Signals due to thermal EMF, Difference in cabling if not an actual 6-wire system. Each of these error sources needs to undergo a thorough evaluation.

Typical Error Sources for Meter Substitution

When calculating Measurement Uncertainty for a meter to be used for substitution, the following are typical error sources:

Simulator Uncertainty includes the resolution of the meter, calibration of the simulator and the associated reference standard uncertainties, stability of the simulator, and the ratio uncertainty. At Morehouse, we achieve about ± 0.00005 mV/V uncertainty on our high-end simulator using different cables for positive and negative output as the polarity switch introduces additional uncertainty.

On the meter side, non-linearity, stability, environmental, ref uncertainty from the simulator, additional cable Uncertainty, noise or resolution, repeatability, and reproducibility are all additional measurement uncertainty components.

In our experience, most who use meter substitution add about 0.02 % - 0.04 % uncertainty to their systems. This is too much uncertainty for ASTM Class AA calibrations that are expected to be better than 0.05 %, too much for ISO 376 Class 00, 0.5, and likely too much for Class 1 & 2. The requirement for an ASTM Class A is to be better than 0.25 %.

The contribution to uncertainty is often significant, though somewhat manageable.

Morehouse does offer calibration of load cell simulators to comply with either standard. Below is a page from our calibration report for one of our reference standard simulators.

These simulators have high-quality aged resistors and steps from 0.04 – 4.4 mV/V. The standard deviation is less than the resolution, hence the importance of having the resolution as part of the overall measurement uncertainty.

Morehouse Page from Simulator Calibration Report in mV/V

Morehouse Budget Load Cell Simulator

Back to the simulator, that likely is not enough to calibrate the meters for substitution. Why would anyone want this? The best answer is cost.

The Morehouse budget simulator costs around $600.00, compared with a higher-end model that costs closer to 8-10 times that pictured below, plus calibration. So, what? It does not allow me to calibrate my meter. That is technically correct, though it is a potent tool.

Our simulator allows the end-user to do the following:

1. Perform cross-checks on equipment.

2. Help control stability/drift.

3. Verify that the coefficients are correctly entered in our 4215 Plus and C705P meters; both use the actual coefficients from the calibration report. Verify coefficients for other programs, such as Morehouse calibration software.

4. Check for linearity issues in any meter.

5. Use as a diagnostic tool to rule out the load cell meter, leaving the load cell, cables, or adapters as the issue.

6. It can be used to calibrate A/D offset and gain settings.

7. It can be used to set up a new indicator before system calibration.

Morehouse Simulator that would allow for meter substitution at lower points, notice the first point is 0.02 mV/V.

Meter Substitution Conclusion

Morehouse is not the calibration police; we are here to serve our customers’ requirements best. I feel it is much better to calibrate everything as a system. I always strive to do what yields the lowest measurement uncertainty to limit the overall risk.

There is a risk/reward scenario for separately calibrating the indicator and load cells. Much additional work is required to comply with either ISO 376 or the ASTM E74 standard. It might be worth it if that extra work saves time and money. Plus, the overall uncertainty increases by an additional 0.02 – 0.04 %, which will be absorbed by everyone else down the metrological traceability pyramid.

Though unsuitable for meter calibration following ISO 376 or ASTM E74, our budget simulator can save much time when troubleshooting equipment and verifying everything was keyed incorrectly via coefficients from a calibration report without breaking the bank. We can provide higher-end simulators for indicator substitution, though the cost is likely over $5000.00, depending on the exact steps and requirements.

The next email will start the deeper dive into calculating Measurement Uncertainty for Force.

4215 Plus uses coefficients to solve for Force.

Comparing the load cell meter filter differences on the 4215 Standard

Applying the minimum filter produced a result different from the maximum level of filtering.

However, the standard deviation of these two filters only varied on the 4215 Plus by 0.00001 mV/V.

The Average difference between the two options was about 0.00005 mV/V or about 0.0025 %.

The maximum deviation of the 22,000 plus readings was between 0.00003 mV/V to 0.00010 mV/V.

Comparing the Filter Differences on the 4215 Standard Vs the 4215 Plus

Conclusion Comparing Load Cell Meter Filter Settings on Morehouse 4215 Load Cell Meters.

• The filter settings can make a difference in the indicator readings.

• The data speaks for itself on both meters.

• The maximum difference in the average readings was about the same, from 0.0025 % to 0.003 %.

• In addition, the 4215 Plus is a way more accurate meter than its predecessor.

• The data speaks for itself on the improved performance, equating to about a 3 x improvement.

Use the Proper Ball Diameter

When applying force through a steel ball, always select a hardened chrome alloy steel ball in the manufacturer-recommended diameter. Using the wrong size or material can compromise safety and measurement accuracy.
Never use a carbide ball. Carbide is brittle and can shatter under load, sending debris flying at high velocity.

Avoid Loading Between Unstable Spherical Surfaces

Never apply force between two opposing spherical surfaces without specially designed adapters. This configuration is highly unstable, and under load the instrument can be expelled from the machine with dangerous force.
Morehouse has developed adapters specifically for this type of application. Without proper containment, the risk of the instrument being launched from the setup is significant.

Unsafe Horizontal Loading—What Not to Do (Figure 79)

Avoid loading between two steel balls unless the assembly is mechanically restrained to eliminate lateral movement. Even slight misalignment under load can cause catastrophic slippage.
Safe use requires:

• Properly sized and aligned ball seats

• Components aligned axially in both directions

• Ball retainer clips to prevent ejection

Correct and Incorrect Loading of Spherical Surfaces

When using tension-member assemblies with mating spherical surfaces, ensure they’re installed exactly as designed. Improper installation can introduce eccentric loading, leading to inaccurate results or unsafe conditions.

Morehouse Quick-Change Tension Members are engineered to minimize off-axis loading and significantly improve safety and accuracy.

Use Parallel Surfaces for Proper Loading

Loading through flat, parallel surfaces helps maintain alignment and minimizes the risk of side loads or bending forces. Always verify that mating surfaces are clean, parallel, and properly seated.

Critical Warning—Adapter Safety

MOST IMPORTANT:
Any adapter or accessory you design, modify, or purchase for use with a force-measuring instrument must be properly engineered and made from steel with adequate strength for the forces involved.
Adapters must be tested in controlled conditions before use, and equipment must never be loaded beyond its maximum rated capacity.
Improperly designed components can fail violently, resulting in severe injury or death.

Always read and follow the instructions and safety precautions for every instrument and machine involved.

Proper Mounting in a Morehouse UCM

The above figures illustrate the correct mounting practices for reference load cells and adapters in a Morehouse Universal Calibration Machine (UCM). Proper mounting prevents alignment errors, binding, slipping, and unsafe load paths.

Additional Load Cell Safety Guidelines

Safe, reliable load-cell performance depends on proper installation, handling, and operation. Follow these best practices to avoid damage and ensure accurate results:

Prevent Mechanical Damage

• Avoid overloading. Exceeding capacity can permanently damage the load cell.

• Prevent drops or impacts. Many precision load cells are highly sensitive to shock. As an example, dropping a 300-lbf shear-web cell from a desk will likely destroy it.

• Control side loads and moments. Use appropriate adapters—spherical seats, clevises, tension members—to maintain an axial load path.

• Use correct hardware. Follow thread-engagement and torque specifications; never pry on the sensor.

• Protect the cable. Never lift the load cell by its cable. Use strain relief, avoid tight bends, and secure routing to prevent snagging.

Environmental and Electrical Protection

• Environmental considerations: Temperature extremes, humidity, and corrosive environments can all degrade performance.

• Prevent power surges: Use stable power supplies and ensure nearby metal structures are properly grounded.

• Isolate from welding currents: Disconnect or shield the load cell before welding near the setup.

• Reduce stray electrical currents: Secure wiring, avoid unshielded lighting near the setup, and ensure proper grounding.

• Shield from magnetic/electrical fields: Use protective enclosures to limit electromagnetic interference.

Operational Best Practices

• Avoid sudden impact loads. Dropping weights onto a load cell platform can cause immediate failure.

• Perform regular calibration. Schedule periodic calibration to maintain accuracy and meet documented standards.

• Ensure proper mounting and installation—plan for adequate space, alignment, and future maintenance.

• Select the right load cell. Choose a model with the correct capacity, sensitivity, response time, and environmental rating for your application.

Want a .pdf to share regarding force safety? One can be downloaded at https://mhforce.com/wp-content/uploads/2021/04/Force-Measurement-Equipment-Safety-Guide-PG-9000.pdf

Share

Figure 1Henry Zumbrun uses a Morehouse Cable Tensiometer Machine (PCM-2MD-T1).

Introduction

Cable tensiometers are precision instruments used to measure cable tension in aerospace, industrial, and structural applications. Accurate calibration of these devices is essential to ensure reliable force readings—especially in mission-critical environments such as aircraft flight control systems.

Tensiometer calibration, however, is highly sensitive to several variables, including cable length, cable type, and calibration methodology. If these factors are not properly controlled, measurement deviations of up to ±20 % can occur, compromising both safety and performance.

This document identifies the most common sources of calibration error and establishes standardized best practices derived from current NAVAIR specifications, updated Navy METCAL methodologies, and relevant industry experience. It provides guidance for technicians and calibration laboratories responsible for certifying cable tensiometers used in aircraft maintenance and structural testing.

Common Error Sources

Figure 2 Comparing Different Length Cables.

1. Cable Length Effects

Tensiometers work by applying a small offset to the cable, effectively increasing tension during measurement. This additional force—referred to as induced tension—is inversely proportional to the cable length. For instance, the same tensiometer that induces 8 lbf of tension on a 3-ft cable may induce only 2 lbf on a 12-ft cable.

Shorter cables are stiffer and more sensitive to displacement, resulting in larger relative changes in tension.

Note: Tests conducted on 3/16” SSAC cables under 50–500 lbf tension indicated induced-tension bias of 12–22% for 3-ft cables versus <5% for 12-ft cables.

Standardization Recommendation:

· Use cables ≥3 ft in length for calibration.

· Ensure calibration cable length approximates actual in-use conditions to minimize induced error.

2. Calibration Methodology

Figure 3 Morehouse Cable Tensiometer Calibration Using Two Fixed Points

There are generally two primary approaches to tensiometer calibration:

- Clamp-and-read method (legacy): Calibrate by clamping the tensiometer onto a tensioned cable. This introduces additional tension but does not correct for it. (Example 500 lbf of cable tension may become 540 lbf when the tensiometer is clamped.)

Note: The reason this method would not be preferred is that it would eliminate calibrating cable tensiometers using deadweight.

- Clamp-then-adjust method (preferred): Clamp the tensiometer, observe the increase in tension, and adjust the applied force back to nominal. This simulates a deadweight calibration condition.

Standardization Recommendation:

· Adopt clamp-then-adjust as the preferred method to replicate deadweight equivalence and minimize offset bias. Align with revised NAVAIR 17-20MF-46 (post-2000) procedures that explicitly correct for clamping-induced tension.

3. Cable Type

Cable construction impacts tension response. For example:
- GAC (Galvanized Aircraft Cable) and SSAC (Stainless Steel Aircraft Cable) differ in modulus of elasticity and response to strain.

Standardization Recommendation:

· Always calibrate using the same cable type (GAC vs SSAC) that the instrument will be used with.

· If end-use cable types vary, produce separate calibration curves for each.

Cable Tensiometer Best Practices Checklist

The following checklist summarizes the standard calibration sequence for cable tensiometers:

1. Inspection: Ensure the tensiometer is clean, free from damage, and set to zero with no load applied.
2. Cable Setup: Use the correct cable diameter and ensure it is free from defects (no kinks or broken strands).
3. Fixture Alignment: Ensure calibrator frame is level (the best rule of thumb is to use equipment that is plumb, level, square, rigid, and has low torsion).

4. Clamp and Adjust: Apply known force, clamp the tensiometer, and adjust load to nominal if using dual fixed-point machines.
5. Repeatability: Conduct at least three repeat readings at each calibration point, ideally covering the low, mid, and high ends of the scale.

6. Maintain ambient temperature within ±2 °F of calibration standard conditions.

7. Data Correction: Convert dial readings to force using calibration charts if using analog devices; compare with direct readings on digital models.
8. Documentation: Record calibration results, identify the method used, and affix calibration labels, noting cable type, cable length, and method.

Summary: This post recommends standardized calibration procedures to minimize induced-tension error, enhance repeatability, and ensure measurement traceability across varying cable lengths and types. Adopting these best practices promotes greater consistency and reliability in cable tensiometer performance across laboratories and field applications.

More information on Morehouse Cable Tensiometers can be found here.

Share

“The quote earlier was for Ultra-Precision load cells. Do we need to go for that grade to calibrate devices known to within 0.1 %, or is there an alternate model that could still give us acceptable TUR?”

That’s the question most labs face when budgeting for force calibration equipment. The answer depends not on brand or preference, but on what happens in the denominator of the TUR equation.

We ran another article called Start with the End in Mind, in which knowing the tolerance of what you need to calibrate is the first step. Then, you can work through determining your calibration and measurement capability, selecting the equipment (such as various Morehouse load cells), and ensuring the calibration provider has the appropriate capability to achieve the lowest possible measurement uncertainty for that equipment.

Essentially, you must account for the measurement uncertainty provided by your calibration provider; this serves as a good starting point. If you need a measurement uncertainty of 0.02 %, your calibration provider likely has to be at least five times better.

This article focuses on the Measurement Uncertainty for your Calibration and Measurement Capability (CMC), which includes the calibration provider you use, the load cells you purchase, and influences from the machines you use and the devices that are submitted to you for calibration.

1. What TUR Really Is—and Why It Matters

The Test Uncertainty Ratio (TUR) defines how capable your measurement process is compared to the tolerance of the unit you’re calibrating. According to ANSI/NCSLI Z540.3, TUR is:

TUR is commonly used as a simplified approach of evaluating global risk. When we know the tolerance we are working to, we have a sufficiently large sample size to determine the shape and distribution of the calibration results and our end-of-period reliability. We can calculate the appropriate uncertainty that corresponds to the maximum amount of false accept risk we are willing to accept.

2. The Denominator: What’s Actually Inside

The denominator—sometimes referred to as the Calibration Process Uncertainty (CPU)—is not just your reference’s stated CMC. It includes every relevant contributor that affects the calibration result

Per the ANSI/NCSL Z540.3 Handbook, “For the denominator, the 95 % expanded uncertainty of the measurement process used for calibration following the calibration procedure is to be used to calculate TUR. The value of this uncertainty estimate should reflect the results that are reasonably expected from the use of the approved procedure to calibrate the M&TE. Therefore, the estimate includes all components of error that influence the calibration measurement results, which would also include the influences of the item being calibrated except for the bias of the M&TE. The calibration process error, therefore, includes temporary and non-correctable influences incurred during the calibration such as repeatability, resolution, error in the measurement source, operator error, error in correction factors, environmental influences, etc.” [1]

The ANSI/NCSLI Z540.3 Handbook specifies that the denominator must include:

• The CMC component of the reference measurement system,

• The resolution of the UUT,

• Repeatability (Type A variation),

• Environmental factors (temperature, alignment, stability),

• Operator and procedural effects, and

• Any temporary or non-correctable errors observed during calibration.

ILAC P14:09/2020 clarifies this directly:

“Contributions to the uncertainty stated on the calibration certificate shall include relevant short-term contributions during calibration and contributions that can reasonably be attributed to the customer’s device”

This means the denominator—the “U” in your TUR—represents everything that can influence the calibration result at that point in time. Ignoring contributors like resolution or repeatability can create a false sense of capability and shift risk downstream.

3. Why CMC Alone Isn’t Enough

Some calibration providers simplify the denominator to only the reference standard’s CMC. That’s not valid per the definitions in ANSI/NCSLI Z540.3 or ILAC G8:09/2019.

Here’s why:
The CMC is the best-case uncertainty your lab can achieve using its best measurement system and the best possible device.
But your customer’s unit isn’t that best device. It has its own resolution, repeatability, and environmental sensitivity that increase uncertainty during calibration.

When those contributors are ignored, the denominator becomes artificially small, inflating the TUR. That leads to incorrect conformity statements and higher Probability of False Accept (PFA)

4. How Different Load Cells Change the Denominator

The CMC portion of the denominator depends heavily on the load cell’s grade, design, and calibration method.
Here’s how the three main classes differ:

Note: These numbers are estimates; they can be calculated if you know the overall contributions of what you are testing, as well as calibration frequency, the number of standard changes, or transducers used. For Example, a UPC cell can maintain 0.02 % CMC from 20 % of capacity through full scale on a Morehouse Automated UCM with the right meter, though most might choose to run the cell down to 10 % of its capacity, and MU might become higher.

Below is an example of a 30 k load cell and a 100 k load cell.

Applied

Expanded Uncertainty

Expanded Uncertainty %

3000

0.38162

0.01272 %

6000

0.46095

0.00768 %

9000

0.54281

0.00603 %

12000

0.69859

0.00582 %

15000

0.89587

0.00597 %

18000

1.02646

0.00570 %

21000

1.31034

0.00624 %

24000

1.45856

0.00608 %

27000

1.64071

0.00608 %

30000

2.02653

0.00676 %

Applied

Expanded Uncertainty

Expanded Uncertainty %

10000

2.30327

0.02303 %

20000

3.10506

0.01553 %

30000

4.17180

0.01391 %

40000

5.30092

0.01325 %

50000

6.47068

0.01294 %

60000

7.68192

0.01280 %

70000

8.88338

0.01269 %

80000

10.07856

0.01260 %

90000

11.30984

0.01257 %

100000

12.53016

0.01253 %

These budgets are both for UPC cells, and the Expanded Uncertainties do vary a bit. The 30k has much better stability than the 100k load cell. All of this plays into that CMC portion of the denominator.

You can think of the denominator like a stack of uncertainty components. The CMC term is one layer, but so are resolution and repeatability. When you use a better load cell (lower CMC), you can improve one component of several. If the UUT’s own uncertainty is large, that improvement can get buried under other contributors.

5. Example: The Effect of Load Cell Class on TUR

Let’s take a 10,000 kg capacity system with ±0.10 % tolerance.

Now add a “poor UUT” scenario—coarse resolution and inconsistent readings. The denominator swells due to resolution and repeatability, while the CMC contribution becomes proportionally small. Suddenly:

In this case, using a higher-grade load cell changes little—because the UUT dominates the denominator.

That’s why we say:

You can buy a better cell to lower your CMC, but it won’t help if your process or UUT drives uncertainty.

6. What You Can Control—and What You Can’t

When optimizing TUR, you control the measurement process, not the UUT’s inherent characteristics. Focus on reducing contributors in the denominator that belong to the system:

You can control:

• Reference CMC (use better cells, better indicators, or more load cells)

• Environmental stability (temperature, alignment, vibration)

• Repeatability of setup

• Operator technique

• Proper adapters and loading alignment

You can’t control:

• UUT resolution (it’s fixed)

• UUT internal hysteresis or design limits

• End-user tolerance definition

If you’ve minimized your controllable components, then moving from a PC to a UPC makes sense—because it’s the most effective way to reduce the CMC part of the denominator.

7. Using Multiple Load Cells to Improve the Denominator

You can also improve the denominator by spreading your range across multiple cells:

• Better range coverage keeps each calibration near its optimal operating point.

• Interpolation errors shrink when you avoid working too low in the range.

For example, use a 1,000 lbf cell for low forces and a 10,000 lbf cell for high forces, rather than a single 10,000 lbf cell for everything. That reduces CMC and process variability simultaneously, tightening the denominator and improving TUR.

8. When a Better Cell Really Matters

UPC load cells pay off when:

• The UUT has fine resolution and repeatability.

• TUR targets are high (≥4:1).

• Process uncertainty dominates the denominator.

• Decision rules like Method 6 (Z540.3) require PFA < 2 %.

PC load cells are best for:

• Calibrating analog or moderate-tolerance devices.

• When environmental and repeatability effects already exceed 0.02 %.

Calibration-grade cells are best for:

• Routine production verification or field checks (TUR ≈ 2:1–3:1).

• Scenarios where UUT behavior dominates uncertainty, making lower CMC irrelevant.

9. Why It’s About Risk, Not Just Ratio

The purpose of TUR isn’t to chase numbers to control measurement risk. A TUR of 4:1 means you can make a confident pass/fail decision with a low risk of false acceptance, but only if the denominator is calculated correctly.

A correct TUR denominator ensures that global consumer risk—the risk that a nonconforming product is accepted—stays within acceptable limits across the traceability chain.

10. Bottom Line

• The TUR numerator (tolerance) is easy.

• The denominator (expanded uncertainty of the entire calibration process) is what separates good metrology from risky shortcuts.

• You can reduce the denominator by using better load cells (lower CMC), multiple ranges, and controlled processes—but if the UUT dominates, upgrading the reference won’t help.

• The right cell depends on what drives your denominator, not just the ratio on paper.

In the end, the goal isn’t to meet an arbitrary TUR ratio—it’s to ensure that your measurement results are trustworthy and that your conformity decisions truly reflect the risk.

Morehouse has a spreadsheet for anyone wanting to use different scenarios. This sheet will tell you how good your reference CMC needs to be to hit a specific TUR ratio.

Appendix Worked Example Using the Morehouse Spreadsheet that can be downloaded here.

In this example, let us suppose we are calibrating a 10,000 lbf device with a tolerance of ± 0.05 % of full scale. The device has a resolution of 0.5 lbf, and we have run repeatability studies showing that the standard deviation of repeat measurements runs around 0.138 lbf.

Scenario
You are calibrating a 10,000 lbf device with a tolerance of ±0.05 % of full scale.

• Full scale (FS): 10,000 lbf

• Tolerance = 0.0005 × 10,000 = ±5.0 lbf

• Device resolution: 0.5 lbf

• Repeatability: standard deviation of repeats = 0.138 lbf (already a standard uncertainty)

• Compare two reference options for the calibration system’s CMC (k≈2):

  • Calibration Grade Morehouse Load Cell©: 0.02 % of reading (k=2)

  • Precision Grade Morehouse Load Cell (PC): 0.01 % of reading (k=2)

  • Ultra-Precision Grade Morehouse Load Cell (UPC): 0.005 % of reading (k=2)

We’ll compute the denominator of TUR as the expanded uncertainty of the calibration process (k≈2), including: reference CMC, resolution, and repeatability. (You can add other contributors—environment, alignment, adapters—if needed.)

Step 1 — Convert contributors to standard uncertainties

Reference CMC → standard (u_ref):

• Calibration Grade (C): 0.02 % of 10,000 lbf = 2.0 lbf (k=2) →

• Precision Grade (PC): 0.01 % of 10,000 lbf = 1.0 lbf (k=2) →

• Ultra-Precision (UPC): 0.005 % of 10,000 lbf = 0.5 lbf (k=2) →

Resolution (digital):
Resolution step = 0.5 lbf, assume uniform distribution:

Repeatability (already standard uncertainty):

Step 2 — Combine (RSS), then expand (k≈2)

For each reference option:

A) Calibration Grade (CMC = 0.02 % of reading, k=2)

B) Precision Grade (CMC = 0.01 % of reading, k=2)

C) Ultra-Precision Grade (CMC = 0.005 % of reading, k=2)

Note: We are using the specification sheet values; the actual CMCs will likely be higher as additional Measurement Uncertainty Contributors will be included that are not included here, such as reference standard stability, repeatability tests, and more. These values are shown as a best-case scenario to explain the concept.

Step 3 — Compute TUR = Tolerance / U

Interpretation

• Reducing CMC from 0.02 % → 0.005 % produces a substantial improvement: 2.45: 1 → 7.87: 1.

• The Ultra-Precision load cell moves the process well beyond the 4:1 TUR benchmark, offering significant margin and risk reduction.

• At this fine resolution (0.5 lbf), the reference CMC is the limiting factor, so upgrading the reference directly improves TUR.

• For coarser-resolution devices, the benefit would taper off as resolution becomes the dominant contributor.

Now, if we changed the resolution of the UUT to 5 lbf and ran the same calculations the difference between a UPC and PC load cell would not be nearly as dramatic

Comparing Three Morehouse Load Cells with various resolutions of the UUT.

Calibration (0.02 % → 2 lbf @ k≈2 →Uref = 1.000 lbf)

Precision Calibration (0.01 % → 1 lbf @ k≈2 →Uref = 0.500 lbf)

Ultra Precision Calibration (0.005 % → 0.5 lbf @ k≈2 →Uref = 0.25 lbf)

Takeaways

• Resolution is the lever. Going from 5 → 2 → 1 lbf resolution markedly lifts TUR for all references.

• Higher-grade references help most when resolution is not a dominant MU contributor. At 1 lbf, moving C (0.02 %)→ PC(0.01 %) → UPC(0.005 %) improves TUR from 2.38 → 4.21 → 6.16.

• At coarser resolutions (e.g., 5 lbf), the UUT dominates the denominator, so lowering the reference CMC yields only modest gains.

• The same is true for UUT repeatability; if the device is not repeatable, it can dominate the denominator.

The standards stood ready, all polished and bright,
Each one reflecting the lab’s steady light.
The techs sipped their coffee, their patience in gear,
Hoping traceability soon would appear.

The 90% platinum, 10% iridium kilogram grinned ear to ear,
Said, “No drift this month—I’m holding steady, my dear!”
The metre stood tall, proud and precise,
While the second kept ticking, annoyingly nice.

The kelvin stayed chill (as per thermal tradition),
The ampere hummed softly with calm repetition.
The candela was glowing with luminous flair—
And said, “Precision and sparkle? I’ve got both to spare!”

The Christmas tree sparkled with nerdy affection,
Each branch labeled Traceability Chain Connection.
From NMIs above to the field work below,
Every ornament whispered, “Keep uncertainty low!”

With presents of Resolution and Repeatability,
And bows marked Stability and Reproducibility.
The torque wrenches dozed, the load cells were dreaming,
Of calibration stickers, all fresh and gleaming.

Then up from the bench came a loud Christmas cheer,
As Santa appeared in a lab coat this year!
He winked and declared, “I’ve got certs in my sleigh—
All ISO/IEC 17025 compliant, by the way.”

So as readings grow steady and deadlines draw near,
We raise our coffee—and maybe a beer.
For metrologists to know, when the world hits “reset,”
It’s the measured among us who’ll keep it all set.

Merry Christmas from Morehouse—where accuracy’s bright,
May your torque be true, and your force be right!

Measurement traceability is the foundation of accurate and reliable measurements.
It means that any measurement result can be related to a reference standard—typically the SI (International System of Units)—through an unbroken chain of calibrations, each contributing its own measurement uncertainty.

Think of it like a family tree of precision:

• At the top is the National Metrology Institute (NMI), maintaining the primary realizations of SI units.

• Below that, Primary Reference Laboratories calibrate instruments against NMIs.

• Then come Accredited Calibration Service Providers, who calibrate customer instruments.

• Finally, Working Standards and Field Measurements—the tools we use daily—are linked back through this chain.

Each link ensures that when you say “1 newton,” everyone in the world agrees on what that means.

The Seven SI Base Units

1. Metre (m) – The unit of length, defined by the distance light travels in vacuum in 1/299,792,458 of a second.

2. Kilogram (kg) – The unit of mass, defined using the Planck constant (h = 6.62607015×10⁻³⁴ J·s).

3. Second (s) – The unit of time, defined by 9,192,631,770 transitions of the cesium-133 atom.

4. Ampere (A) – The unit of electric current, defined via the elementary charge (e = 1.602176634×10⁻¹⁹ C).

5. Kelvin (K) – The unit of thermodynamic temperature, defined by the Boltzmann constant (k = 1.380649×10⁻²³ J/K).

6. Mole (mol) – The unit of amount of substance, containing exactly 6.02214076×10²³ elementary entities (Avogadro’s number).

7. Candela (cd) – The unit of luminous intensity, defined using a fixed luminous efficacy for monochromatic radiation of 540×10¹² Hz.

Together, these seven base units form the foundation of all measurements—the SI system—ensuring that science, industry, and calibration stay aligned across the globe.

May your torque be true, your force be strong, and your uncertainty stay low!
Wishing you a precise and joyful holiday season—from all of us at Morehouse.

Share

This document explains how each method realizes force, the respective traceability paths, advantages, and limitations, helping you determine which approach best suits your laboratory’s accuracy requirements, workload, and facility constraints.

When you calibrate force, you are essentially deciding how to realize force and how to transfer that realization to the device under test. In practice, most laboratories use either a deadweight machine or a transfer standard system that employs load cells.

• Deadweight machines that realize force directly from SI-traceable mass (force = m·g, corrected for local gravity, air buoyancy, and material density).

• Hydraulic machines that generate or multiply force using fluid pressure—either as hydraulic amplification standards (piston-area ratios, less common) or the more common hydraulic comparators (e.g., Morehouse Automated Universal Calibrating Machines, UCMs) that apply force with a hydraulic actuator while comparing the device to a calibrated reference transducer (load cell).

Choosing the right calibration method affects uncertainty, productivity, and compliance with ISO 376/ASTM E74/Conformity Assessment Ability.

Deadweight vs hydraulic force calibration: How does each method generate force

Deadweight (primary standard)

Deadweight machines apply known masses directly to a loading frame. To be a primary standard, the masses are determined by comparison to SI-traceable references and adjusted for local gravity, air buoyancy, and material density; the realized force is direct and does not pass through levers or hydraulic multipliers. In well-designed deadweight machines, typical expanded uncertainties are on the order of 0.001–0.005 % of applied force, enabling the tightest classifications (e.g., ISO 376 Class 00, ASTM E74 Class AA).

Why this matters: ISO 376’s best classification (Class 00) expects very low reference uncertainty—deadweight is the preferred path because it “usually” achieves better than 0.0025 % of applied force throughout the range.

ASTM E74 requires deadweight machines to be used to generate a Class AA verified range of forces for a reference transducer. That transducer is then used in a machine such as the Morehouse Automated UCM as the reference standard.

Hydraulic

Hydraulic approaches come in two flavors:

1. Hydraulic amplification standards. A known deadweight load is applied to a small piston; fluid pressure transfers to a larger piston. The force scales with the ratio of effective piston areas (Pascal’s law). This extends force ranges to multi-meganewton levels (national labs routinely use direct deadweight to a few MN and hydraulic multiplication above that—e.g., PTB up to ~16.5 MN). Typical expanded uncertainties are roughly 0.01–0.05 %; fine, low-end control can be harder, and operation is often slower than comparator machines.

2. Hydraulic comparator machines (UCM). These use a hydraulic actuator to apply force and compare the device under test to a reference transducer previously calibrated (ideally by deadweight using ISO 376 or ASTM E74). Properly designed systems routinely achieve ~0.02–0.05 % expanded uncertainty (even better with high-end references & automation), with excellent throughput for daily work.

Note: Morehouse also utilizes machines such as the BCM and PCM, which operate at lower forces driven by a screw jack.

Deadweight vs Hydraulic Force Calibration Traceability & standards (ISO 376 / ASTM E74)

• Deadweight is a direct SI realization through mass, with the shortest, simplest traceability chain. That’s why it’s the default recommendation for ISO 376 Class 00 reference transducers and for generating ASTM E74 Class AA verified ranges of forces.

• Hydraulic amplification is still ultimately traceable (to mass and piston areas), but the chain includes geometric area measurements, alignment of pistons, temperature/viscosity effects, and seal friction. It can support ISO 376 Class 00 at suitable ranges, but will usually carry larger uncertainty at the low end and operate more slowly. Most of the time, multiple transducers are used to perform cross-checks and SPC.

• Hydraulic comparators (UCMs) depend on the reference standard’s calibration (ideally deadweight to ISO 376 or ASTM E74). Achievable uncertainty in the comparator is therefore dominated by the reference’s uncertainty plus the machine’s repeatability, environment, alignment, and operator effects; with good design and control, ≤0.02 % is realistic. Automating the machine often improves throughput and lower measurement uncertainty by improving repeatability.

In general, Deadweight machines are the most accurate path to ISO 376 Class 00/ASTM E74 Class AA, while UCMs are described as hydraulic force calibration systems are often optimized for versatility and productivity.

Alignment, adapters, and the machine itself

Regardless of method, you must keep the loading line pure—plumb, level, square, rigid, and free from torsion—and use proper adapters (e.g., tension adapters for tension work, compression platens of the right hardness/flatness). Poor fixturing and off-axis loading can dominate the error budget (it’s common to see errors above 0.05 % simply from the wrong top block hardness in compression).

Hydraulic comparators vary in design; the better ones mount reference transducers to replicate how they were calibrated (e.g., in compression with alignment balls/sphericals) and integrate controls that avoid overshooting and hold points long enough to satisfy ISO 376 timing.

What this means for uncertainty (and your audit)

Every calibration ends in an uncertainty budget. In comparator systems mainly, major contributors include reference standard uncertainty, repeatability, environmental stability, operator variability, adapter alignment, and machine design. With disciplined control of these terms, a modern UCM can credibly document ≤0.02 % expanded uncertainty; deadweight can go lower when you need it.

Tip: If you’re targeting ISO 376 Class 00, plan on deadweight for the reference calibration and ensure your machine stores raw data and meets ISO timing (no readings within 30 s of force changes; uniform loading intervals).

Deadweight vs Hydraulic Force Calibration Selecting the right path: quick decision guide

1. What uncertainty do you actually need?

   • ≤0.005 % required for ASTM E74 Class AA or ISO 376 Class 00 across a broad range → Deadweight.

   • ~0.02–0.05 % adequate with high throughput and automation → Hydraulic comparator (UCM) with deadweight‑calibrated references.

2. What’s your workload?

   • Many devices per day, frequent setup changes → UCM/PCM (automation boosts productivity and consistency).

   • Fewer, highest‑stakes references → Deadweight sessions with meticulous adapter control.

3. What can your facility support?

   • Floor loading/ceiling height/gravity survey okay? Deadweight gives the best TURs.

   • Limited space/budget? A UCM provides excellent uncertainty for most industrial needs.

For product‑level context from Morehouse: deadweight systems are positioned for ISO 376 Class 00/ASTM E74 Class AA work, while UCMs are hydraulic comparators built for long life and high throughput. Morehouse Instrument Company, Inc.

Deadweight vs Hydraulic Force Calibration Common myths & gotchas (and how to avoid them)

• “Hydraulic machines can’t meet Class 00.”
  Not quite. Hydraulic amplification systems at national labs can support Class 00 classification, especially at higher ranges; however, they typically carry a higher CMC than direct deadweight and may be slower to operate. Verify the uncertainty at your points of interest.

• “Comparator = deadweight accuracy.”
  A UCM inherits the reference’s uncertainty and adds its own contributors (repeatability, environment, alignment, control). With good practice it’s excellent, but deadweight still sets the floor. Build—and show—your uncertainty budget.

• “Adapters don’t matter.”
  They do. Use ISO‑recommended tension/compression fittings; control hardness, flatness, and geometry. Misfixturing can add ≥0.05 % error—more than the difference between methods. Replicate field use during calibration.

• “Just span the meter; you’ll be fine.”
  When your reference is calibrated to ISO 376/ASTM E74, use the calibration coefficients (polynomial fits) rather than multi-point “span only”—it lowers interpolation error and protects your uncertainty.

Field‑tested best practices (apply to both)

• Keep the machine plumb, level, square, rigid, and low‑torsion; design and adapters that maintain a pure line of force reduce bending and side loads.

• Warm up electronics; exercise to max force; respect ISO timing and, if needed, run creep/reversibility checks per ISO 376 Case C/D.

• Control the environment (temperature stability, humidity) and log it—auditors will ask, and your uncertainty depends on it.

• Train for reproducibility. R&R studies (technician‑to‑technician) often reveal hidden variability you can eliminate, or automate the UCM for better R & R.

Deadweight vs Hydraulic Force Calibration Bottom line

• Choose deadweight when you need the lowest possible uncertainty and the simplest SI traceability—especially for ISO 376 Class 00 and ASTM E74 Class AA references.

• Choose hydraulic amplification to reach very high forces while maintaining strong traceability—expect excellent but not deadweight‑level CMC, and slower operation.

• Choose a hydraulic comparator (UCM) for day‑to‑day productivity and ~0.02–0.05 % uncertainty—particularly effective when references are calibrated by deadweight and you control adapters, alignment, and environment.

Deadweight vs hydraulic force calibration References used in this article

• ISO 376 guidance on reference classification, machine design (alignment/fixtures), and the role of deadweight for Class 00; plus coefficient use and common error sources.

• Setting up a Force Calibration Laboratory—typical uncertainties by machine type, hydraulic amplification at NMIs, and facility considerations.

• Force Calibration for Technicians & Quality Managers—why calibration ≠ verification, traceability, and practical impacts of adapters and loading conditions.

• Uncertainty budget for a UCM—contributors, methods, and realistic targets (≤0.02 %).

For more information on deadweight machines

For more information on Morehouse UCM Machines.

About Morehouse

We believe in changing how people think about Force and Torque calibration in everything we do, including discussions on Deadweight vs hydraulic force calibration machines.

This includes setting expectations and challenging the “just calibrate it” mentality by educating our customers on what matters and what may cause significant errors.

We focus on reducing these errors and making our products simple and user-friendly.

This means your instruments will pass calibration more often and produce more precise measurements, giving you the confidence to focus on your business.

Companies around the globe rely on Morehouse for accuracy and speed.

Our measurement uncertainties are 10-50 times lower than the competition, providing you with more accuracy and precision in force measurement.

We turn around your equipment in 7-10 business days so you can return to work quickly and save money.

When you choose Morehouse, you’re not just paying for a calibration service or a load cell.

You’re investing in peace of mind, knowing your equipment is calibrated accurately and on time.

Through Great People, Great Leaders, and Great Equipment, we empower organizations to make Better Measurements that enhance quality, reduce risk, and drive innovation.

With over a century of experience, we’re committed to raising industry standards, fostering collaboration, helping with understanding risk, and delivering exceptional calibration solutions that build a safer, more accurate future.

Contact Morehouse at info@mhforce.com to learn more about our calibration services and load cell products.

This post compares Bayesian and Frequentist reasoning using the Monty Hall problem. Frequentist statistics treat probability as a long-run frequency and express uncertainty through procedures such as confidence intervals. Bayesian statistics, by contrast, treat probability as a degree of belief, updated through Bayes’ rule — where the posterior is proportional to the prior times the likelihood.

The inspiration for writing this came from a memorable discussion with Paul Reese during an NCSLI workshop in Florida. Several of us—Greg Cenker, Dilip Shah, and others—listened as Paul explained these ideas through simple but powerful analogies, which helped the concepts “click.” Later, while reading Steven Pinker’s When Everyone Knows That Everyone Knows, I encountered additional clear Bayesian examples. Combining the NCSLI conversation with Pinker’s framing convinced me that these ideas can be made intuitive for anyone curious about Bayesian vs. Frequentist thinking—even if, like me, they do not claim to be an expert.

This note aims to illustrate how Bayesian and Frequentist reasoning differ in practice. It proceeds using the Monty Hall problem. The goal is not to take sides, but to show how both approaches can be helpful in metrology and decision-making.

Bayesian Reasoning via the Monty Hall Problem

Paul had such a good discussion with drawings on this one that I hope to replicate somewhat here.

Setup: Three doors, one hiding a Morehouse Deadweight Machine and two hiding less-than-desirable machines.

You pick Door 3. The host knows where the prize is, never opens your chosen door, and always opens one of the others to reveal a non-prize. Suppose he opens Door 2 - this observed action is our evidence E.

Using Bayes’ rule, switching raises your chance of winning from 1/3 to 2/3 by conditioning on how the evidence (the host’s action) is generated.

Figure 1: Any door picked has a 1/3 or 33.3 % chance

Figure 2: The host shows one of the doors that is not correct

Most people say the chances are now 50/50, because “two doors remain, and one must have the Morehouse Deadweight Machine.” Two doors remain, but they’re not symmetric because Monty’s choice was constrained; that constraint is precisely what the likelihoods capture.

Figure 3: Switching (Door 3 → Door 1) doubles your odds to 2/3.

Bayesian reasoning allows us to condition on how the information was generated. Monty’s action is not random—he is forced to reveal a non-winning door. When we account for that rule, the updated probability of winning the prize is:

· 1/3 chance if you stay

· 2/3 chance if you switch

In other words, switching doubles your odds. The host’s behavior gives you information, and Bayes’ rule updates the probability once that information is revealed.

Intuition:

When you first pick a door, there is a 1/3 chance you chose the deadweight machine and a 2/3 chance you chose the wrong door. Monty’s reveal doesn’t change the initial odds—it only removes one losing door from the 2/3 side. That entire 2/3 probability effectively transfers to the remaining unopened door. So switching is simply betting on the original 2/3.

Figure 4: The pick is switched, and the contestant wins :).

Normalize prior x likelihood

Sum = 1/3 + 0 + 1/6 = ½

Posteriors:

· P(Door 1∣E) = (1/3)/(1/2) = 2/3

· P(Door 3∣E) = (1/6)/(1/2) = 1/3

The product step is the heart of Bayes’ theorem:
it combines your prior belief and new evidence to compute how plausible each hypothesis is given what you observed.

· Multiply → Combine prior knowledge with evidence.

· Sum → Normalize to ensure total probability = 1.

· Divide → Get the updated, conditional probability.

That’s what “Bayesian updating” really means in practice.

There’s a distinction here.

To see the same problem through a Frequentist lens, let’s reinterpret Monty Hall in terms of repeated experiments rather than updated beliefs.

Frequentist Reasoning via the Monty Hall Problem

The Frequentist perspective doesn’t assign probabilities to beliefs or to individual events, but rather to long-run relative frequencies in repeated experiments. Instead of updating beliefs based on evidence, a Frequentist asks: “If this game were repeated many times under identical conditions, what proportion of times would switching win?”

Imagine running the Monty Hall game 300 times under the same rules:

• You always pick Door 3 first.

• Monty always opens another door that doesn’t have the prize.

• You always choose to stay or always choose to switch.

If you always stay, you’ll win roughly 1/3 of the time—because only one of the three initial choices is correct.
If you always switch, you’ll win roughly 2/3 of the time—because in the 2/3 of trials where you started wrong, switching corrects it.

From the Frequentist point of view, the 2/3 figure isn’t a belief update; it’s a limiting proportion observed over repeated trials. The interpretation is operational:

“If you play this game infinitely many times under identical conditions, switching will win about two-thirds of the time.”

The Frequentist framework doesn’t emphasize conditioning on Monty’s specific action as evidence about this particular instance—the probability statement refers instead to the long-run performance of the strategy.

Key Distinction

• Bayesian: “Given Monty opened Door 2, the probability that the prize is behind Door 1 is 2/3.”

• Frequentist: “In the long run, the strategy ‘always switch’ wins about 2/3 of the time.”

Both describe the same numeric ratio, but they differ philosophically:

• The Bayesian statement is about degrees of belief in this instance (conditional probability given evidence).

• The Frequentist statement is about the performance of a rule across repeated, identical trials.

Frequentist vs. Bayesian Reasoning — Easy Comparison

Both views agree on the numbers but differ in meaning. The Bayesian approach quantifies uncertainty in the state of the world, while the Frequentist approach quantifies uncertainty in the procedure. In metrology, both perspectives are valuable: Bayesian methods naturally incorporate prior calibration knowledge, while Frequentist tools provide standardized repeatable intervals for traceability.

Frequentist reasoning counts what happens over many trials.
Bayesian reasoning updates your belief in real time as new information arrives.

The Role of the En Ratio in Force Calibration Confidence

In the world of calibration, proving that your measurements are both accurate and defensible goes beyond simply watching control charts. While control charts are powerful for spotting drift or unusual trends, they don’t answer a critical question: are our stated measurement capabilities truly valid?

To demonstrate this, laboratories need to look outward. External benchmarks such as Proficiency Tests (PTs) or Inter-Laboratory Comparisons (ILCs) provide that reality check by comparing your results against a trusted reference. Among the tools used to make these comparisons meaningful, the En ratio stands out.

The En ratio doesn’t just compare two numbers—it accounts for the uncertainties behind those numbers. By considering both your laboratory’s claimed uncertainty and the reference laboratory’s uncertainty, the En ratio offers a clear, objective measure of agreement. In short, it bridges the gap between internal quality control and external validation, making it indispensable for demonstrating compliance and building confidence in calibration results.

Interpretation:

· |En| ≤ 1 → The result agrees with the reference, within stated uncertainties.

· |En| > 1 → A significant difference exists, suggesting underestimated uncertainty or bias.

One useful tool is to calculate the En ratio as stated above.

Figure 1: En Ratio Graphical Plot.

Application at Morehouse

At Morehouse, En ratio calculations are used whenever possible to compare the performance of in-house reference standards against NIST-calibrated standards. In this case, the comparison is carried out on the Automated Universal Calibrating Machine (AUCM). Two load cell standards, both calibrated directly by NIST, are compared under controlled conditions.

For this evaluation:

• Reference (NIST):

  • Expanded measurement uncertainty = 0.001 % of applied force.

  • The ASTM Lower Limit Factor (LLF) is included as a statistical estimate of expected device errors. This LLF serves as an “expected performance” benchmark for the load cell when used under similar conditions.

• Morehouse Laboratory Standard:

  • Claimed expanded uncertainty = 0.01 % of applied force.

  • Force range covered: 50,000 lbf through 1,000,000 lbf.

By substituting these values into the En formula, the ratio quantifies whether the Morehouse claimed capability (CMC) aligns with the NIST reference. In practice, an En ≤ 1 validates that the laboratory’s CMC statement is supported by objective inter-comparison data.

Figure 5: Data Series Showing the Error Bars at Each Force Point.

Data Series

1. Participant (blue diamonds, with error bars):

   • Shows the measured deviations of the participant’s device from the nominal force at various applied forces.

   • Error bars indicate the participant’s measurement uncertainty at each test point.

   • The deviation starts slightly negative (–20 to –40), trends toward zero around mid-range (~500,000–600,000 lbf), and then increases positively up to +50 by 1,000,000 lbf.

   • Uncertainty grows as the applied force increases (larger error bars at higher loads).

2. Reference (orange squares, with error bars):

   • Represents the reference lab (likely NIST) results, with the ASTM llf, and Resolution of the UUT included.

   • Deviations are consistently near zero across the entire force range.

   • Error bars are smaller and relatively stable compared to the participant’s, reflecting the tighter uncertainty of the reference, which makes sense as NIST is using primary standards known to better than 0.001 % from 50,000 lbf to 1,000,000 lbf.

Interpretation

• The Participant’s measurements are consistently close to the reference but show a bias: slightly negative at lower forces, crossing through zero mid-range, and slightly positive at higher forces.

• The Reference is treated as the “true” value (deviation ~0).

• The overlap of the error bars indicates that, despite the bias trend, the participant’s results are statistically consistent with the reference—the differences are within the combined uncertainty.

Why the En Ratio Matters

1. Verification of CMC Claims

   • Laboratories accredited to ISO/IEC 17025 are required to demonstrate the validity of their stated measurement uncertainties. En ratio outcomes provide evidence of compliance and technical competence.

2. Complement to Control Charts

   • Control charts monitor ongoing stability and drift within the lab’s processes.

   • En ratios demonstrate external comparability to national standards, addressing traceability and uncertainty claims.

3. Risk Reduction in Decision Rules

   • When applied together with decision rule frameworks (e.g., guardbanding or shared risk methods), En ratios reduce the risk of false acceptance (PFA) and false rejection (PFR).

Morehouse has a CMC capability worksheet that also includes an En worksheet to calculate En ratios and the plots shown above.

In summary:
The En ratio is an indispensable complement to control charts. While control charts provide continuous assurance of measurement stability, En ratio analysis demonstrates that a laboratory’s uncertainties are realistic and defensible when benchmarked against NIST or other NMIs. At Morehouse, these evaluations reinforce both measurement confidence and compliance with ISO/IEC 17025 requirements, strengthening the credibility of the laboratory’s force calibration services.

Morehouse ILC Force Calibration Rental Kit

The Morehouse ILC Force Calibration Rental Kit is designed to help laboratories satisfy ISO/IEC 17025:2017 interlaboratory comparison requirements by providing a comprehensive solution for force calibration, including a load cell, adapters, indicator, and calibration certificate. This kit enables labs to validate their calibration and measurement capability (CMC) claims with very low overall uncertainty, supporting confidence in measurement results and compliance with recognized standards. With robust performance and an easy-to-use analysis worksheet, the kit offers an objective way to benchmark results and improve measurement processes. More information here.

Understanding the Normal Distribution

The normal distribution, sometimes called the bell curve or Gaussian distribution, is one of the most important concepts in statistics. It describes how values of a variable are spread, and it often appears in nature, science, and everyday life.

What is the Normal Distribution?

At its core, the normal distribution is a probability distribution that is symmetric about its mean. Most of the data points cluster around the central value, while fewer and fewer occur as you move farther from the mean. This creates the familiar bell-shaped curve.

Key characteristics include:

• Mean (μ): The center point of the distribution.

• Standard Deviation (σ): A measure of how spread out the data is. Smaller σ means the curve is narrower; larger σ means it’s wider.

• Symmetry: The left and right sides of the curve are mirror images.

Why is it Important?

The normal distribution is so helpful because:

1. Many real-world phenomena follow it. Human heights, measurement errors, and IQ scores are just a few examples.

2. It underpins statistical methods. Techniques like confidence intervals, hypothesis testing, and regression often assume normality.

3. The 68-95-99.7 Rule. Approximately 68 % of the data lies within 1σ of the mean, 95 % within 2σ, and 99.7 % within 3σ. This makes it easy to estimate probabilities.

Metrology Specific Examples of the Normal Distribution

1. Calibration of Measuring Instruments

When a micrometer or pressure gauge is calibrated, repeated readings of a stable reference standard are taken. These repeated measurements typically scatter around the true value in a way that follows a normal distribution.

• The mean value gives the best estimate of the true measurement.

• The spread (standard deviation) quantifies the repeatability uncertainty.

2. Environmental Monitoring

In a controlled laboratory, parameters like temperature or humidity are monitored. Small, random fluctuations in these quantities often follow a normal distribution around the setpoint.

• Example: A lab maintained at 20 °C might show a bell-shaped variation between 19.8 °C and 20.2 °C.

3. Gauge Repeatability and Reproducibility (Gage R&R) Studies

In metrology quality control, repeated measurements of a part by different operators and instruments are analyzed. The resulting spread of data often approximates a normal distribution, enabling analysts to distinguish between repeatability error and reproducibility error.

4. Weighing and Mass Standards

When repeatedly weighing a 100 g calibration mass, the balance will not give exactly the same reading every time. Instead, results like 100.002 g, 99.999 g, and 100.001 g form a normal distribution around the true mass.

5. Uncertainty Budgets

When building an uncertainty budget (per ISO/IEC Guide 98-3, the “GUM”), contributors such as repeatability, reference standards, or drift are often modeled as normal distributions. These uncertainties are then combined statistically.

Final Thoughts

The normal distribution is more than just a neat mathematical curve—it’s a practical tool that helps us understand uncertainty, variation, and probability in the world around us. Whether you’re analyzing scientific data or making everyday decisions, chances are the bell curve plays a role behind the scenes. It also explains why, when I received a 45 % on my economics exam and thought I had to drop the college course, I ended up with a B+, because that 45 % was higher than the majority of the population.

Subscribe now