Advancing QE Capabilities (15) | When Capability is Insufficient, Should You Change the Process or the Specification: A Dialogue Between Capability and Specification

1. A Meeting Held Three Times Without a Conclusion

A car parts company's bracket hole size, after six months of mass production, had a Cpk of only 0.92, while the customer's key characteristic threshold was 1.33. In the first review meeting, the process team said the equipment accuracy was insufficient and new machines were needed; the equipment team said the process parameters were not properly adjusted; the quality department suggested, "This dimension has little impact on functionality, can we ask the customer to relax the tolerance?" Each party had its own rationale, but after three meetings, no conclusion was reached. The issue was ultimately resolved by "strengthening inspection and temporarily supplying." Four months later, the customer's sampling inspection found a batch of nonconforming products, leading to a complete return and rework and screening costs exceeding two million yuan.

The most awkward point in the post-mortem was that all three approaches could have been correct, but the company had no process to determine which one was right. Thus, everyone's suggestions remained at the level of "I think," with each side substituting their position for data. This article provides that process—first, identify the cause, then calculate the costs, and finally discuss the specification.

2. Key Principles: Four Sources of Insufficient Capability

The formula for Cpk is min((USL - μ) / 3σ, (μ - LSL) / 3σ). It has only two independent variables: the position of the process center μ relative to the tolerance center (offset) and the size of the process variation σ. All "insufficient capability" phenomena ultimately fall into these two variables, plus two often overlooked prerequisites—measurement inaccuracy and unreasonable specifications. These four sources require completely different handling methods:

• Measurement Type: The resolution of the measuring tool is insufficient, and the clamping differences are large, leading to measurement noise mixed into the calculated variation. Characteristics: GR&R ratio is high (over 30%), and repeated measurements of the same batch of parts show scattered results.
• Offset Type: Cp meets the standard (low variation), but Cpk is significantly lower than Cp. Characteristics: The difference between Cp and Cpk is greater than 0.3, and the mean deviates from the tolerance center by more than 0.5σ.
• Variation Type: Cp itself is low, and the process distribution is tight at both ends. Characteristics: Cp and Cpk are close and both do not meet the standard, and control charts often show jumps or cycles.
• Specification Type: The process capability is close to the physical limits of the equipment and process, but the specification is set according to theoretical values, without considering the total tolerance from stack-up and sensitivity analysis, or it significantly deviates from industry norms for similar products. Characteristics: Cp is consistently stable but does not meet the standard (e.g., stable around 1.0), and the dimension is not sensitive to downstream functionality.

Why is it important to classify first? Because the repair costs for the four types differ by an order of magnitude. For the offset type, usually only adjusting the baseline, correcting the tool setting point, and adding compensation parameters are needed, with immediate benefits; for the variation type, changes to fixtures, parameters, or even equipment are required, with a cycle measured in months; the measurement type has the lowest cost but is the easiest to overlook; and the solution for the specification type is not in the workshop or the quality department but in design and with the customer—spending three million yuan on a new machine might just be paying for an unnecessary tolerance precision.

3. Practical Steps: Five Steps to Turn Disputes into Criteria

Step 1: Fix the Ruler Before Discussing Capability. The data used to determine capability should have a measurement system GR&R target ≤10%, and it must not exceed 30%. Controlled data should be confirmed using a control chart and special cause points removed. The sample size should be no less than 25 subgroups and 100 data points, covering different shifts, machines, and material batches. Criterion: If the GR&R exceeds 30%, all subsequent analysis is invalid, and the measuring tool or method must be improved first. Skipping this step means all subsequent calculations are meaningless.

Step 2: Decompose Offset and Variation to Identify the Cause. Calculate both Cp and Cpk on the same clean data set. Criterion: If the difference between Cp and Cpk is less than 0.1 and both are low, it is classified as a variation type; if the difference is greater than 0.3, it is classified as an offset type; if Cp meets the standard (e.g., above 1.33) but Cpk does not, it is almost certainly a center management issue, and the tool setting baseline, mold positioning, program compensation, and measuring tool zero point should be checked. For the offset type, the specification must not be changed—any application to "relax the tolerance" is invalid until the center has been adjusted to the tolerance center.

Step 3: Determine the Ceiling on the Process Side. Adjust the mean to the tolerance center, and the theoretical optimal capability will be the value of Cp. The criterion line is very practical: if the predicted Cpk after center adjustment can reach 1.33, focus on the process side and avoid discussing the specification; if the prediction is between 1.0 and 1.33, it is "within reach," and parameter optimization and fixture improvements should be prioritized based on cost-benefit analysis, along with transitional controls; if the prediction is still below 1.0, it indicates that the current equipment and process parameters cannot achieve the required capability—only then is it appropriate to enter the fourth step of specification discussion. When calculating the predicted value, subtract the measurement variance: the true process standard deviation is approximately the square root of the difference between the observed standard deviation and the measurement standard deviation.

Step 4: Specification Side Argument, All Four Types of Evidence Are Required. Relaxing the tolerance is not just changing a number but moving the "failure boundary," which must be supported by a chain of evidence:

• ① Statistical Evidence: The relationship between the process distribution and the specification, the proportion and destination of current nonconforming products, and the loss quantified in PPM.
• ② Functional Evidence: Conduct a specialized functional verification to find the "functional failure boundary" of the dimension, i.e., the extent to which the dimension can deviate before downstream functional or assembly issues arise. The new specification must remain within the functional boundary and retain a 30% to 50% safety margin.
• ③ Downstream Evidence: Perform a tolerance stack-up and sensitivity analysis to confirm that relaxing the dimension will not cause the cumulative tolerance of the assembly or the entire machine to exceed control limits, thereby transferring risk to assembly or the final customer.
• ④ Boundary Evidence: Safety characteristics, regulatory characteristics, and customer drawings explicitly marked characteristics must not be relaxed. Any change to the customer's specification must follow the formal engineering change request (ECR/ECN) process and not be unilaterally modified internally.

Among these four types of evidence, the sample size for functional verification is the easiest to skimp on—using one sample machine or one batch to determine the "functional boundary" is essentially a guess. Generally, at least three different dimension levels near the boundary should be covered, with no fewer than 5 samples per level, and this should be clearly documented in the verification report.

Step 5: Communication and Implementation Verification. When communicating with the customer, avoid writing, "We cannot achieve it, please relax the specification," and instead write, "Based on functional verification, the functional boundary for this dimension is X, the current specification Y does not match the actual process center Z; we recommend adjusting the specification to W, with attached verification data, improvement plan, and transitional control scheme." A request based on hardship will be rejected, while a request with data and a plan has a chance of being approved. When communicating with the design department, use sensitivity analysis to speak: point out that the dimension is not sensitive to the overall functionality but consumes most of the tolerance budget, while truly sensitive areas lack sufficient margin. This helps the design team see the fact of "misaligned tolerance allocation" rather than "production trying to take shortcuts." If the specification is relaxed and approved, three follow-up actions must be completed: a new capability study (with a Cpk target of 1.33 or higher under the new specification, not declaring it qualified just because the specification is wider), updating the drawings, PFMEA, control plan, and supplier agreements, and implementing enhanced monitoring for the first year to validate the initial functional verification conclusions with actual failure data.

4. Five Common Misconceptions

First, Discussing Relaxing the Specification at the First Sign of Insufficient Capability. Specifications are design outputs, not a dumping ground for capability issues. Problems of the offset type and measurement type can be resolved internally, but changing the drawings is equivalent to using document changes to cover up production management deficiencies.

Second, Adjusting the Center Without Considering Variation. To make Cpk look good, the mean is forced to the tolerance center, pushing both sides close to the limits. When Cp itself is very low, the short-term index may look better, but the long-term Ppk will be worse. The correct sequence is to reduce variation first, then adjust the center.

Third, Using Small Samples to Argue for the Functional Boundary. Testing three samples to show that "nonconforming parts can still be used" and then applying for a relaxation is a typical case of generalizing from a small sample. Functional verification should cover multiple dimension levels near the boundary and clearly document the sample batches and pre-treatment conditions.

Fourth, Using "Full Inspection" as a Reason to Relax the Specification. Full inspection addresses the issue of nonconforming products flowing out but does not address whether the specification itself is reasonable. Mixing these two issues in one application will only make the customer think you are avoiding capability improvement.

Fifth, Not Updating Documents After Changing the Specification. If the drawings are changed but the control plan is not, and the supplier drawings are not synchronized, the "dimension nonconformance affecting assembly" failure mode in the PFMEA remains unresolved—this will be a nonconformity in document consistency during audits and is often more fatal than insufficient capability.

5. Self-Check List

• Before conducting a capability study, has the measurement system GR&R been confirmed to be within 10% (maximum 30%) and does the sample size cover at least 25 subgroups and major sources of variation?
• Have both Cp and Cpk been calculated simultaneously, and has the difference between the two been used to determine whether the issue is "offset-dominated" or "variation-dominated"?
• Has the predicted capability value after "center alignment" been calculated, and have the 1.33/1.0 criteria lines been used to decide whether to focus on the process side or the specification side?
• When applying for a specification adjustment, has the functional verification covered at least three dimension levels near the boundary, with no fewer than five samples per level, and retained a 30% to 50% safety margin?
• Once the specification is changed, have the drawings, PFMEA, control plan, and supplier agreements been updated simultaneously, and has a new capability study been conducted?

Identify the cause, then calculate the costs, and finally discuss the specification.

Knowledge code: 6.3.2

Version: v20260925

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