Symptom
The symptom is not a failing part; it is a failing process conversation. A customer receives a lot with a tensile result slightly below the previous lot, asks for a root cause analysis, and the factory answers with a single test value and a promise to "strengthen inspection". No chart exists, so nobody can tell whether the change is a real shift or ordinary variation, and the corrective action becomes an increase in sampling that consumes labour without changing the process.
Tensile strength on a zipper is a destructive test, so it is expensive to measure and tempting to measure rarely. That combination produces the classic failure mode of quality control: too few data points to see a trend, and a reaction to every individual result as though it were a signal.
- Lot-to-lot tensile variation above 8 percent without an assignable cause.
- Individual results below the specification limit while the average remains acceptable.
- Successive lots drifting in one direction over five or more shipments.
- Different operators obtaining materially different results from the same lot.
- Control limits that were copied from a drawing rather than calculated from data.
Root Causes
When tensile measurement behaves erratically, the cause is usually in the measurement system or in the sampling plan rather than in the product. Before attributing variation to the process, confirm that the process is being measured consistently.
- Inadequate gauge repeatability. Clamp slippage, inconsistent specimen mounting and varying pull speed contribute more variation than most factories assume.
- Unrepresentative sampling. Taking five pieces from one carton measures one carton, not one lot. Rational subgroups must span the production stream.
- Wrong subgroup size. Subgroups of two or three cannot detect small shifts and inflate false alarms; subgroups of ten stabilise the estimate but slow response.
- Limits calculated from specification. Using the drawing tolerance as a control limit guarantees that the chart will not signal until the process is already producing defects.
- Mixed populations on one chart. Different sizes, tape constructions or dye lots on a single chart produce bimodal data and meaningless limits.
- Reaction to common-cause variation. Adjusting the machine after every low point increases variation, a well-documented effect that turns a stable process into an unstable one.
Troubleshooting Sequence
Establish the measurement system first, then the chart. A chart built on an unreliable gauge will generate signals that cannot be acted on, and the team will stop trusting it within a month.
- Run a gauge repeatability and reproducibility study with three operators, ten samples and three trials.
- Confirm that pull speed, clamp type and specimen length match the test standard exactly.
- Collect at least 25 subgroups before calculating limits; fewer than that gives unstable estimates.
- Calculate limits from the average moving range or within-subgroup range, not from the specification.
- Plot the range chart before the average chart, because a range out of control invalidates the average limits.
- Mark every process change, material lot change and shift change on the chart so signals can be matched to events.
- Recalculate limits only after a verified permanent process change, never after removing an inconvenient point.
Corrective Actions
When the chart shows a point outside the limits, treat it as a signal and investigate the specific event: a tape lot change, a new needle, a different operator, a machine repair. When the chart shows a run of seven points on one side, treat it as a shift and look for a gradual cause such as tool wear, temperature drift or supplier lot change.
When the process is stable but not capable, do not widen the limits. Reduce variation by changing the inputs: tighter tape tensile specification, improved stitch density control, or a more consistent tooth moulding process. Capability improvement is an engineering project, and the chart's role is to prove whether the project worked.
| Signal | Interpretation | Response |
|---|---|---|
| One point outside limit | Special cause | Investigate the specific event |
| Seven points one side | Process shift | Look for gradual drift |
| Increasing range | Measurement or material instability | Audit gauge and lots |
| Stable but Cpk below 1.33 | Common cause | Engineering improvement |
Prevention
Make the chart part of the shift routine rather than a monthly report. An operator who plots the previous hour's result and sees the point land can connect the number to the machine state; a chart reviewed at month end can only produce explanations, not corrections.
Donglong Zipper calculates tensile capability per product family and reviews it at each management meeting, with the specification limit and the calculated limit displayed side by side so that the gap between customer requirement and process reality is visible. YCC buyers should ask for the range chart as well as the average chart, because a stable average with a widening range is the earliest signal that a supplier's process is about to fail, and it appears well before the first rejected lot.
Keep the raw data. A chart image without the measurements cannot be re-analysed when the specification changes, and re-analysis is often what turns a vague complaint into a precise engineering finding.
Do not stop at variable data. Slider function, stop presence and visible defects are attribute characteristics, and they need a p-chart rather than an average and range chart. Plot the proportion defective per subgroup, and treat a run above the upper limit as a signal even when every individual part looked acceptable to the inspector. Combining a variables chart for tensile with an attributes chart for function gives a complete picture: the first predicts mechanical margin, the second predicts what the customer will actually complain about, and neither substitutes for the other.