When moulded parts start cracking in the field, two explanations compete from the outset. Either the geometry concentrates stress in a way that makes failure inevitable in every part the tool produces, or the process drifted and some parts came out weaker than others. Both produce cracked parts that look similar and both are supportable from a single failed example, which is why a single failed example rarely settles it. What does settle it is how the failures distribute across the population that was made — and that analysis requires data most investigations do not ask for early enough.
The two hypotheses make different predictions
A geometric defect is present in every part from the tool, so failures should appear across the entire production history at a rate set by service conditions rather than by manufacturing date. A process defect is variable, so failures should cluster — by date range, by shift, by material lot, or by cavity in a multi-cavity tool. These predictions are testable, and testing them is considerably more informative than examining one more broken part.
Cavity identification is the fastest discriminator
Multi-cavity tools usually mark parts with a cavity number. If failures concentrate in one or two cavities while others are unaffected, the problem is that cavity — its cooling, its venting, its dimensions — and not the design, which is common to all of them. If failures distribute evenly across every cavity, a cavity-specific process explanation is substantially weakened. This single data point often redirects an entire investigation and costs nothing but reading the parts.
Date codes and lot traceability
Where parts carry date codes, plotting failures against production date against the number of parts produced in each period distinguishes a defect that started at a point in time from one that was always present. A step change coinciding with a documented material change, a tooling repair or a process adjustment is strong evidence. A flat distribution across years of production points the other way.
The denominator matters as much as the numerator
Failures clustering in a date range mean nothing if far more parts were made in that range, and clustering in a cavity means nothing without knowing that cavity's share of output. The analysis requires production volumes alongside failure counts. Investigations that work only from returned parts routinely misread ordinary volume variation as a defect signal, which is one of the more common analytical errors in this area.
Service exposure is the confounder to control
Parts do not all experience the same conditions. If the units that failed were disproportionately in hot climates, in heavy duty cycles or in a particular installation configuration, the clustering may reflect exposure rather than manufacturing. Separating manufacturing variables from service variables requires both to be recorded, and it is why a failure analysis that considers only the parts and not their deployment can reach a confident wrong answer.
What geometry contributes independently
Regardless of distribution, the geometry should be assessed on its own terms: internal radii at load-carrying transitions, rib-to-wall thickness ratios that produce sink and voids, bosses sized for the fastener load, and whether a weld line falls anywhere it carries stress. A design that concentrates stress at the observed initiation site is relevant even if a process factor triggered the failures, because it establishes how little margin existed.
What the process record contributes
For the periods in question: melt and mould temperatures, injection and packing pressures, cycle time, regrind percentage and its source, resin lot numbers with certificates of analysis, drying conditions, and any recorded deviations or tooling maintenance. Regrind and drying deserve particular attention, because both degrade material in ways that do not appear on the incoming resin certificate but do appear in the finished part.
Testing the parts closes the loop
Comparative testing on failed parts, unused parts from the same lot and unused parts from unaffected lots turns the distribution analysis into a mechanism. Thermal analysis and molecular weight measurement identify processing-induced degradation. Impact and tensile testing quantify how much strength differs between populations. Where the parts that failed measure materially weaker than parts that did not, the process explanation has direct support rather than inference.
What to secure early
Failed parts with cavity and date codes legible and recorded. Unused parts spanning multiple lots and cavities, including from periods with no failures. Production volumes by period. The full process and material record before retention schedules expire. And the field data — where each failed unit was deployed and how it was used — because without it the distribution analysis cannot control for exposure.
This article is general technical orientation, not a failure analysis, an engineering opinion, or advice on any specific matter. Determining the cause of a particular incident requires hands-on examination by a credentialed expert.