Recurring Adhesive Bond Failures: Challenging an Operator-Error 8D
Over eight weeks a Tier-1 automotive supplier has seen adhesive-bond failures on a stamped steel bracket rise from 0.2% to 0.8% at end-of-line peel test. The 8D report lists root cause as 'operator did not follow the wipe procedure' and corrective action as 'retrained both operators on work instruction WI-447.' Three months later the failure rate is 1.1%. Production runs three shifts; the defect is concentrated on second shift (1.6% versus 0.3% on first). Only one of two adhesive lots is affected, and that lot was received before a supplier formulation-change notice that was filed but never reviewed. Ambient humidity logs show 68–74% RH on second shift in July versus 45–55% on first shift; the adhesive datasheet specifies 30–60% RH at cure. Fixture clamp torque is checked monthly and was last recorded out of specification in May. The customer requires a PPAP-level corrective action with effectiveness verification at 90 days.
1. Evaluate the 8D root cause and corrective action. What analytical errors are present, and what does the evidence suggest instead?
Read the worked answer
The 8D stops at 'operator error', which is a symptom, not a root cause. Three errors are visible.
First, the causal chain is not verified. A valid root cause must explain why the failure rate rose, why it is concentrated on second shift, and why only one adhesive lot is affected. Retraining explains none of these; it also predicts improvement, which did not occur (0.8% to 1.1%).
Second, the data were not stratified. Shift, adhesive lot, humidity, and clamp torque are four candidate variables, and two of them are confounded: second shift coincides with high July humidity. A simple is/is-not analysis (defect is on second shift, is with lot X, is not on first shift, is not with lot Y) points toward a cure-environment and material interaction, not a wiping habit.
Third, the change-notice review was skipped. If the formulation changed, the cure window, mix ratio, or surface-energy requirements may have shifted; the datasheet humidity limit of 60% RH is exceeded on second shift, which is a plausible mechanism for weak bonds.
Operator error can be a legitimate finding only when the error is predictable, the system permits the error, and the proposed action removes the possibility of recurrence. Here the system (no humidity control, no change-notice gate, intermittent clamp torque) drives the outcome. The finding should be reclassified as incomplete, with the investigation reopened around environment, material, and fixturing.
2. Design the investigation needed to confirm the mechanism before permanent action. Specify the data, the experiment, and the decision criteria.
Read the worked answer
The investigation should move from correlation to confirmation in three layers.
Retrospective data: pull 12 months of peel-test results joined to shift, humidity log, adhesive lot, clamp-torque record, and cure time. Compute failure rate by humidity band (for example <55%, 55–65%, >65% RH) holding lot and shift constant where possible. This quantifies the association but cannot prove causation because shift and humidity are confounded.
Structured experiment: run a 2³ factorial on the bracket line with factors humidity (45% versus 70% RH), adhesive lot (affected versus unaffected), and clamp torque (nominal versus low), three replicates, response peel strength. Randomize run order and verify humidity with a calibrated logger at the bond line. Also measure surface energy (dyne pens or contact angle) before bonding to test the mechanism directly. Hold oven temperature, cure time, and operator constant.
Decision criteria should be set in advance: an effect is confirmed if its ANOVA p-value is below 0.05, the effect size exceeds the practical threshold (for example 15% of mean peel strength), and the confirmation run at production settings reproduces the result within the replicate spread. If humidity is confirmed, the fix is environmental control and a cure-window interlock; if the lot is confirmed, the fix is incoming material verification and a supplier change-notice gate. Do not proceed to permanent action until the confirmation run passes.
3. Define containment, corrective and preventive actions, and the 90-day effectiveness verification. How do you distinguish a correction from an effective CAPA here?
Read the worked answer
Containment (immediate, before root cause is final): quarantine all product bonded with the affected lot, 100% peel or ultrasonic inspection of suspect WIP and finished goods, and flag any shipped product for customer notification. Add temporary humidity monitoring and a hold point on second shift; verify clamp torque at every shift start rather than monthly.
Correction: retrain, rework, and sort. This fixes the output but not the system.
Corrective action must address the verified mechanism. If humidity is confirmed: install dehumidification or a controlled cure room, add a humidity interlock that stops the line outside 30–60% RH, and log it. If the material change is confirmed: create a formal gate so supplier formulation notices are reviewed and approved by engineering before the lot is released, and update the PFMEA and control plan with the new failure mode and detection control.
Preventive action: read-across to other adhesive-bonded part numbers, add clamp-torque verification to the calibration and PM schedule, and audit the change-notice process.
Effectiveness verification differs from implementation. Define it in advance: after 90 days and at least 3,000 bonded parts, failure rate must be at or below 0.2% with the 95% upper confidence bound below 0.5%, control charts of peel strength must be stable, and an audit must confirm the interlock and change gate are functioning. Verification is against data, not against the presence of a completed action; the retraining-only 8D failed precisely because no such criterion was set.