Measurement Uncertainty and Comparison in Practice —— A Comprehensive Measurement Quality Assurance System from Laboratory to Production Line

1. Introduction: Why Measurement Uncertainty is a Must for Quality Professionals

In a quality management system (QMS), there is a frequently emphasized statement: "Without measurement, there is no management." However, more dangerous than the absence of measurement is a measurement result that appears precise but is not reliable. When a dimension measurement result is 10.02mm, is its true range 10.01mm to 10.03mm, or 9.95mm to 10.09mm? This "possible range" is the core meaning of measurement uncertainty.

Many quality practitioners frequently encounter the concept of uncertainty in their daily work—control limits in SPC control charts need to consider measurement variation, MSA analysis results (GR&R) need to be judged using uncertainty, and laboratory test reports must include a statement of measurement uncertainty. However, few truly understand the methods for evaluating uncertainty and can apply them to actual quality decisions.

This article will systematically explain the basic concepts, evaluation methods, and applications in laboratory comparisons of measurement uncertainty, as well as how to integrate the concept of uncertainty into production line quality control. It aims to help quality managers establish a complete cognitive chain from "measurement data" to "scientific decision-making."

2. Five Key Understandings of Measurement Uncertainty

(1) Uncertainty is Not Error, but a Confidence Interval

This is the most fundamental and easily confused concept. Error (Error) is the difference between the measurement result and the true value. Since the true value is always unknown, error is a theoretical concept that cannot be precisely calculated. Uncertainty (Uncertainty), on the other hand, is a quantitative description of the dispersion of the measurement result. It answers the question, "How reliable is the measurement result within a certain range?"

To put it simply: Error is like an "unknown target deviation," while uncertainty is a "confidence circle" drawn around the measurement result. The center of the circle is the measurement value, and the size of the circle tells us where the true value is most likely to fall.

(2) Uncertainty is Everywhere, Absolute Precision Does Not Exist

Regardless of how precise the measurement equipment is or how skilled the operator is, measurement results always have dispersion. Factors such as environmental temperature fluctuations, non-uniformity of the sample, instrument digital resolution, and operator reading habits collectively ensure that measurement results cannot be absolutely precise.

It is crucial to recognize this: Uncertainty is not a sign of poor measurement capability. On the contrary, actively assessing and reporting uncertainty is a manifestation of a controlled measurement process and mature quality awareness.

(3) Two Methods for Evaluating Uncertainty—Class A and Class B

According to the GUM (Guide to the Expression of Uncertainty in Measurement, ISO/IEC Guide 98-3), uncertainty evaluation is divided into two categories:

Class A Evaluation: Standard uncertainty obtained through statistical analysis of repeated measurements. Simply put, it involves "measuring many times and observing how dispersed the data is," quantified using the experimental standard deviation. The core assumption of Class A evaluation is that the measurement process is in a statistically controlled state and the data follows a normal distribution.

Class B Evaluation: Standard uncertainty based on non-statistical information. This information may come from equipment calibration certificates, instrument precision specifications, historical experience data, standard material certificates, or even reasonable inferences about the measurement environment. Class B evaluation requires quality personnel to have a deep understanding of the measurement process and reasonable engineering judgment.

In actual uncertainty evaluations, Class A and Class B methods need to be combined to form a complete combined standard uncertainty.

(4) Synthesis and Expansion: From Components to the Whole

A complete measurement result is typically influenced by multiple sources of uncertainty. To combine these influences into a total uncertainty indicator, the "law of propagation of uncertainty" must be followed, which is based on the principle of variance synthesis from the Taylor series expansion.

The basic rule for synthesis is: For independent uncertainty components, use the "root sum square" (RSS) method. For related components, introduce covariance terms, which are more complex to handle.

The combined standard uncertainty (denoted as uc) is multiplied by a coverage factor k (usually k=2, corresponding to approximately 95% confidence level) to obtain the expanded uncertainty (denoted as U). This is the most common form of expression in test reports: Measurement result = x ± U, k=2.

(5) The Relationship Between Uncertainty and Tolerance: Not a Simple Comparison

Many quality engineers ask: What should be done if the measurement uncertainty exceeds one-tenth of the tolerance range? This is a typical "uncertainty and conformity assessment" issue.

According to ISO 14253-1 "Geometrical Product Specifications—Inspection of Workpieces and Measuring Equipment," when the measurement result is near the upper tolerance limit, the influence of uncertainty must be considered. Specifically, only when the measurement value plus the expanded uncertainty remains within the tolerance range can it be judged as "conforming"; conversely, if the measurement value minus the expanded uncertainty falls outside the tolerance range, it is judged as "nonconforming." The intermediate region is called the "uncertainty zone" and requires further precise measurement or consultation with the customer.

This criterion has significant compliance implications in actual quality inspections, especially in precision manufacturing industries such as automotive, aerospace, and medical devices.

3. Systematic Methods for Evaluating Measurement Uncertainty

(1) Step One: Define the Measurand and Measurement Process

The starting point for uncertainty evaluation is to clearly define "what is being measured" and "how it is measured." For example, when measuring the outer diameter of a batch of shaft parts, the measurand is "the outer diameter value measured at a specified cross-section and direction," and the measurement process includes the use of a micrometer, the specific operating method of the operator, environmental temperature conditions, and the sampling position for measurement.

The key in this step is: The object of uncertainty evaluation is the result of the measurement process, not a specific workpiece. Therefore, a clear description of the measurement process is the foundation for subsequent analysis.

(2) Step Two: Identify Sources of Uncertainty (Cause and Effect Analysis)

Using a fishbone diagram (cause and effect diagram) to systematically identify all factors that may affect the measurement result is the standard method for identifying sources of uncertainty. Common sources include:

• Equipment: Instrument indication error, resolution, sensor drift, wear
• Environment: Temperature deviation, humidity changes, vibration, lighting
• Operator: Reading habits, alignment deviation, differences in operating methods
• Sample: Surface roughness, shape deviation, material inhomogeneity
• Standard: Uncertainty of calibration standards, uncertainty of certified reference material values
• Method: Approximations in measurement principles, truncation errors in data processing algorithms

(3) Step Three: Quantify Each Uncertainty Component

For each identified source, determine its standard uncertainty value.

For Class A components, perform at least 10 repeated measurements (the more degrees of freedom, the better), calculate the standard deviation using the Bessel formula, and then divide by √n to get the standard uncertainty of the mean.

For Class B components, judge the probability distribution based on known information and calculate the standard uncertainty. Common methods include:

• Equipment Calibration Certificate: The given expanded uncertainty U divided by the coverage factor k
• Instrument Precision Specifications: Assuming a uniform distribution, the allowed error limit divided by √3
• Digital Resolution: The resolution value divided by 2√3 (assuming integerization error is uniformly distributed)
• Temperature Influence: Based on the expansion coefficient and temperature fluctuation range, assume a uniform or triangular distribution

(4) Step Four: Synthesis and Expansion

Under the assumption that each component is independent, synthesize the standard uncertainty using the root sum square method. Then, choose an appropriate coverage factor (k=2 for 95%, k=3 for 99%) to obtain the expanded uncertainty.

(5) Step Five: Report Uncertainty

The test report should clearly state the measurement result, the numerical value of the expanded uncertainty, the coverage factor k, and the confidence level. It should also include a note explaining the main contributions to the uncertainty, to help users assess the reliability of the measurement result.

4. Laboratory Comparison—The Golden Standard for Verifying Uncertainty Evaluation

Uncertainty evaluation is essentially a model-based estimation. No matter how rigorous the evaluation process is, the correctness of the evaluation result ultimately needs to be verified through comparison with "external references."

(1) Basic Forms of Laboratory Comparison

Laboratory comparison (Interlaboratory Comparison) refers to multiple laboratories measuring the same uniform and stable sample using the same measurement method, and then evaluating the consistency of each laboratory's measurement results with the reference value through statistical analysis.

Common forms of comparison include:

• Proficiency Testing (PT): Regular comparison activities organized by accreditation bodies
• Comparison Tests: Cross-comparisons spontaneously organized among several laboratories
• Reference Material Verification: Internal verification using certified reference materials (CRMs)

(2) Evaluation Indicators for Comparison Results—En Value

The most commonly used comparison evaluation indicator is the En value (normalized deviation), calculated using the formula:

En = (Xlab - Xref) / √(Ulab² + Uref²)

Where Xlab is the measurement result of the participating laboratory, Xref is the reference value (usually provided by authoritative laboratories or reference materials), and Ulab and Uref are the expanded uncertainties (k=2) of the two.

Evaluation standards:

• |En| ≤ 1: Satisfactory result, indicating that the laboratory's uncertainty evaluation is consistent with the measurement result
• 1 < |En| ≤ 1.5: Suspicious result, requiring investigation of systematic deviations or re-evaluation of uncertainty
• |En| > 1.5: Unsatisfactory result, requiring corrective action

The beauty of the En value lies in the fact that it not only evaluates the closeness of the measurement results but also considers the uncertainties of both parties. In other words, the more "conservative" (larger value) the uncertainty evaluation, the easier it is to be accepted, but this does not necessarily mean high measurement quality. Conversely, the more "confident" (smaller value) the uncertainty evaluation, the higher the requirement for measurement precision. Therefore, the En value drives laboratories to neither overestimate nor underestimate uncertainty, but to ensure it is realistic and reasonable.

(3) Systematic Path for Investigating Failed Comparisons

When the laboratory comparison results are unsatisfactory, the systematic investigation path includes:

1. Check Measurement Standards: Ensure calibration certificates are valid and that standard instruments are performing stably
2. Check Measurement Methods: Verify if the standard operating procedures are strictly followed and if there are any deviations
3. Check Environmental Conditions: Ensure temperature, humidity, vibration, etc., are within allowable ranges
4. Check Sample Handling: Ensure the representativeness, uniformity, and stability of the sample during transportation
5. Check Operator Procedures: Identify any operational deviations and retrain if necessary
6. Check Uncertainty Evaluation: Ensure no important sources of uncertainty are overlooked and that the estimation of certain components is reasonable

Each round of comparison is an opportunity for quality improvement. Truly excellent laboratories are not those that always pass comparisons, but those that can quickly identify and thoroughly resolve issues when comparisons fail.

5. Application of Uncertainty Thinking in Production Line Quality Control

Uncertainty is not just a technical term for laboratory technicians. For quality control on the production line, the concept of uncertainty also has significant practical value.

(1) Considering Uncertainty in Inspection Decisions

In incoming inspection, process inspection, and final inspection, when the measurement value is close to the specification limits, decision-makers must consider the impact of measurement uncertainty. A simple judgment principle is: If the measurement result plus the uncertainty is still within the specification, it can be confidently judged as conforming; if the measurement result minus the uncertainty exceeds the specification, it should be judged as nonconforming; if it falls between the two, a more precise measurement method or consultation with relevant departments is needed.

Many disputes over nonconforming product judgments in manufacturing enterprises often stem from the fact that measurement uncertainty has not been fully considered.

(2) Measurement Variation in SPC Control Charts

In the practical application of statistical process control (SPC), a commonly overlooked issue is that each data point on the control chart itself carries measurement uncertainty. When the measurement uncertainty is large, the fluctuations on the control chart may primarily come from the measurement system itself, rather than real changes in the process.

Therefore, before setting control limits, conducting an MSA analysis and ensuring that the measurement system's GR&R meets the requirements is a prerequisite for the effective operation of SPC. Generally, the process variation reflected on the control chart should be the superposition of process variation and measurement variation. When measurement variation is too large, the measurement system must be improved before process control can be discussed.

(3) Selection and Acceptance of Measuring Instruments

When purchasing and accepting new measurement equipment, attention should not only be paid to the resolution and range of the measuring instrument, but also to whether its uncertainty meets the tolerance requirements of the measurand. A practical rule of thumb is: The measurement uncertainty of the measuring instrument should not exceed one-tenth of the tolerance of the measurand (i.e., TUR≥10, where TUR is the Test Uncertainty Ratio). For precision measurements, TUR should be 4:1 or higher.

This criterion ensures that the measurement system has sufficient "credibility" in conformity assessment.

(4) Decision Support for Adjusting Calibration Cycles

By analyzing the trends in historical calibration data and the changes in measurement uncertainty, laboratories can scientifically adjust calibration cycles. If the calibration data of a measuring instrument is stable over multiple calibrations and the uncertainty has not significantly increased, the calibration cycle can be appropriately extended; conversely, if the uncertainty is increasing, the calibration cycle should be shortened, and the frequency of interim checks should be increased.

This data-driven calibration management approach is more efficient and cost-effective than simply "calibrating once a year."

6. Common Misconceptions and Practical Pitfall Avoidance Guide

Misconception One: The Larger the Uncertainty, the Better; the Smaller, the Better

Both extreme views are harmful. Overestimating uncertainty can lead to a large number of conforming products being judged as "pending confirmation" or nonconforming, causing unnecessary scrap and rework. Underestimating uncertainty can amplify the risk of conformity assessment, potentially leading to nonconforming products reaching the customer.

The correct approach is: Based on objective data from the actual measurement process, evaluate uncertainty truthfully, neither exaggerating nor underestimating.

Misconception Two: Uncertainty Evaluation Requires Complex Mathematical and Statistical Knowledge

Many quality professionals are intimidated by the formulas used in uncertainty evaluation. In fact, for most on-site measurement problems, the main contributions to uncertainty come from a few key factors—sometimes the uncertainty from the instrument calibration certificate alone accounts for more than 80% of the total uncertainty. As long as the basic root sum square synthesis method and reasonable estimates of the main components are mastered, a practical and valuable uncertainty evaluation can be made.

Misconception Three: No Need for Uncertainty Analysis After ISO/IEC 17025 Accreditation

ISO/IEC 17025 accreditation requires laboratories to establish and maintain uncertainty evaluation procedures, but accreditation itself is not the end. Measurement equipment and personnel can change, environmental conditions can fluctuate, and the complexity of the measurand can increase. All these changes can affect measurement uncertainty. Regularly reviewing and updating uncertainty evaluations is an inherent need for laboratories to remain in compliance.

Misconception Four: Treating MSA GR&R Results as Uncertainty

The results of GR&R analysis (%GR&R) focus on evaluating the precision of the measurement system relative to the tolerance, while uncertainty evaluation quantifies the dispersion of the measurement result itself. Although they have different focuses, they are complementary. In practice, GR&R analysis can provide important inputs for the A-class components of uncertainty evaluation, but it cannot replace a complete uncertainty evaluation.

7. Conclusion: Turning Uncertainty into the "Hard Currency" of Quality Management

Measurement uncertainty and laboratory comparison, on the surface, are technical knowledge in the field of metrology. However, delving deeper, they address the most fundamental questions in quality management— "How do we know if we are doing it right?"

When a company's quality culture can openly face and accurately express its measurement uncertainty, it means that the company's quality system has moved from "judgment based on experience" to "speaking with data." When a laboratory can verify its uncertainty evaluation through comparison, it means that every test report issued has technical confidence.

In an era where quality information is becoming increasingly transparent and customers are demanding higher data authenticity, measurement uncertainty has expanded from the professional domain of laboratory technicians to the entire quality value chain. From tolerance analysis in the product design phase, to conformity assessment in supplier incoming quality control (IQC), to monitoring and early warning in production process SPC, and to root cause tracing in customer complaints—uncertainty thinking runs through it all.

Learning the methods for evaluating uncertainty is not just to pass audits, but to ensure that every quality decision stands up to scientific scrutiny. When the reliability of measurement data becomes transparent, quality management truly achieves the leap from "roughly reasonable" to "precisely reliable."

This is the fundamental reason why the Quality Think Tank has set measurement uncertainty and comparison as independent knowledge nodes—it is not an isolated technical issue, but the cornerstone of the entire measurement quality assurance system. Every quality practitioner deserves to spend time solidifying this "basic skill" and making it a "hard currency" in their quality toolkit.

Measurement uncertainty is not a numbers game—it is the confidence foundation of quality decisions. Teams that dare to provide uncertainty are truly those that dare to take responsibility for quality.

Knowledge code: 11.2.2

Version: v20260728

Author: Quality Think Tank The Quality Think Tank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping enterprises continuously improve their quality capabilities.