Can GR&R Be Calculated When Test Pieces Break? —— Case Study of a Destructive Measurement System Analysis in an Automotive Parts Company

1. A "Excellent" Report of 7.1%, Resulting in Six Months of Fake Data

An automotive parts company supplies welded bracket assemblies for commercial vehicle manufacturers. The tensile force of critical welds is a mandatory inspection item upon shipment—blueprints require it to be no less than 15 kN. In 2024, the customer discovered three instances of bracket weld fractures in the after-sales market. Upon tracing back, the tensile force of the fractured pieces was found to be only 11.2 to 13.8 kN, significantly below the specification. However, the peculiar part was that all the outgoing sampling inspection records for this batch were qualified, with no piece falling below 15 kN.

The discrepancy between inspection data and customer complaints naturally pointed to the measurement system. The quality department organized a measurement system analysis (MSA) but encountered a challenge: the tensile force test is destructive, and once a test piece is pulled, it breaks and cannot be measured again. The conventional GR&R method, which involves multiple operators repeatedly measuring the same batch of parts, was not applicable. Therefore, the engineer took a shortcut: a single tester pulled and broke 10 consecutive test pieces from a shift, using the range of these 10 data points to estimate repeatability, and then dividing by the total variation to calculate a GR&R of only 7.1%—which, according to AIAG standards, is considered "excellent." The report was submitted to the customer, and the review passed smoothly.

However, over the next six months, the data began to behave erratically. The company used the tensile force data from three samples per shift to create an SPC control chart. The Xbar chart frequently went out of control, with Cpk fluctuating between 1.7 and 1.1, showing no consistency. The workshop suspected that the welding machines were aging and conducted two major overhauls and multiple adjustments of welding parameters, incurring direct costs of over 100,000 yuan, but the problem persisted, and the parameters became increasingly skewed.

It wasn't until the customer's annual process audit that a third-party auditor pointed out the flaw in the MSA report: "You are mixing the differences between different test pieces with the differences in repeated measurements of the same test piece." More critically, the 7.1% GR&R only covered one person, one machine, and one shift, excluding variations due to clamping, equipment, and shift differences. After redoing the analysis correctly, the true GR&R level was 31%—an "unacceptable" number.

2. Why Conventional GR&R Cannot Be Applied to Destructive Measurements

First, let's look at the foundation of conventional GR&R: it requires the same object to be measured repeatedly. In the classic crossed structure, several operators measure the same batch of parts two to three times each, with the parts remaining fixed. This allows the variation to be cleanly separated into "part-to-part variation" and "measurement system variation."

Destructive measurements, however, remove this premise. Once a test piece is pulled, crushed, or burned, it no longer exists, and the same object can only be measured once. The idea of "measuring the same part three times and taking the range" is simply not feasible. As a result, three typical shortcuts are often used on the shop floor:

The common consequence of these practices is a severe underestimation of GR&R and an overestimation of the measurement system. Once the measurement system is overestimated, a series of chain reactions follow: SPC treats measurement noise as process variation, leading to false alarms and missed alarms; Cpk is artificially high, and commitments to customers are built on sand; process improvements fail to identify the real culprits, and tens of thousands of yuan are wasted on major welding machine overhauls—all because of an incorrect MSA report.

3. Three Correct Approaches: How to Conduct MSA for Destructive Measurements

First Approach: Grouping and Nesting Method, Treating "Homogeneous Groups" as "the Same Part." Since the same part cannot be re-measured, a group of "almost identical" parts can be used as a substitute. The method involves: within a short period of continuous production on the same equipment and with the same parameters, grouping every three consecutive test pieces together, treating the pieces within each group as the same "part"; taking 12 to 20 groups, covering two shifts, 2 to 3 testers, and all in-use testing machines. Each tester measures one piece from each group, with each piece measured only once. The analysis uses nested ANOVA to separate "between-group variation" (real product variation) from "within-group variation + measurement error." GR&R is calculated based on the sum of personnel variation and residual error, with the same judgment criteria as conventional GR&R: less than 10% is excellent, 10% to 30% is acceptable, and over 30% requires corrective action.

Second Approach: Stable Sample Method, Monitoring Long-Term Drift. The biggest fear in destructive measurements is not a single measurement error but the gradual drift of testing machines, sensors, and fixtures over several months. The solution is to retain a "master piece": use a confirmed standard test block or a retained piece from the same batch, and measure it weekly. Plot the results on a control chart, and if it goes out of control, it indicates that the measurement is drifting and requires calibration—essentially, giving destructive measurements a "long-term blood pressure monitor."

Third Approach: Indirect Characteristic Linkage, Avoiding SPC Dependence on Destructive Inspections. Destructive sampling inspections are costly and have a small sample size, making them unsuitable for daily SPC. A more stable approach is to find non-destructive "stand-ins" for destructive indicators—weld seams can be checked for dimensions, hardness, and ultrasonic properties, while the process can monitor current and voltage curves. After confirming the relationship between the stand-ins and tensile force through correlation analysis, use the stand-ins for high-frequency monitoring and reduce destructive tests to periodic sampling and calibration comparisons, which is cost-effective on both ends.

In the case study, the company redid the analysis using the grouping and nesting method. The true GR&R level was 31%, with over 60% of the variation attributed to clamping. Upon investigation, it was found that the positioning blocks of the testing machine fixtures were worn, allowing the test pieces to move side to side; the two testers had different clamping habits, with one clamping closer to the front and the other closer to the back, resulting in a difference of several millimeters in the breaking point; there was also a system bias between the two testing machines. The corrective actions were straightforward—replacing the positioning blocks, creating quick-install positioning fixtures with engraved lines, standardizing the clamping SOP, and calibrating the two testing machines monthly with a standard force gauge. This cost a few thousand yuan and half a day of training, and the re-measured GR&R dropped to 12.4% (3.2% personnel variation, 4.1% equipment variation).

4. Five Lessons Learned from the Case

Lesson One: Sample Groups Must Be "Truly Homogeneous," or the Nesting Method Will Also Fail. The premise of the grouping and nesting method is that the pieces within each group are sufficiently similar to the "same part." Forcing pieces from different time periods, different equipment, or different mold cavities into the same group will inflate the measurement error, causing the GR&R to swing from one extreme to another. The selection of adjacent pieces within groups and the coverage of all process variations between groups must be precise.

Lesson Two: Destructive Measurements Do Not Mean Ignoring Personnel and Equipment. Many people believe that "the testing machine automatically records data, so there is no human error"—this case study proves the opposite, with the largest source of variation being clamping. Personnel, shift, and equipment differences are part of the measurement system, and the sampling design must include them.

Lesson Three: Don't Just Focus on the Overall GR&R Number; Look at the Variation Structure. In the 31% GR&R, 60% was due to clamping, pointing to issues with fixtures and SOPs. If the company had only replaced a more expensive testing machine, the money would have been wasted, and the problem would still persist. Breaking down the variance components helps identify where improvements are needed—this is where MSA is far more valuable than a single pass/fail rate.

Lesson Four: Completing MSA Is Not the End; Daily Monitoring Is Key. Without a "master piece" to monitor, drift in destructive measurements can quietly return. Establishing a routine mechanism of stable samples, equipment comparisons, and indirect characteristic linkage is essential to prevent GR&R from returning to 30% after six months.

Lesson Five: Supplier and Equipment Manufacturer Reports Can Never Replace Your Own MSA. The measurement system is a combination of "your equipment, your people, and your environment." Factory data only confirms that the equipment itself is qualified but does not ensure that your measurement process is qualified.

After six months of corrective actions: the tensile force SPC chart stabilized, with Cpk consistently ranging from 1.45 to 1.6, and no more machine stops due to "false variations." The customer's annual audit even cited this case as a positive example of supplier measurement management. All of this started with acknowledging a simple truth—once a test piece is pulled and broken, the measurement system's account cannot be "broken" along with it.

5. One-Sentence Summary

Even though test pieces break and cannot be re-measured, the measurement system's account can still be accurately calculated—through grouping, changing personnel, and nested analysis, reliable GR&R can be achieved for destructive inspections.

Destructive measurements have no "second chance," so the first measurement must be accurate.

Knowledge code: 6.2.1

Version: v20260904

Author: QTank QTank is dedicated to providing systematic knowledge, methodologies, and practical tools for quality management professionals, helping companies continuously improve their quality capabilities.