Data-driven optimization of moisture content in thermoset transfer moulding: an industrial case study on process stability and scrap reduction

This study investigates the influence of moisture content in an epoxy-based thermoset moulding compound on process stability and scrap rate during injection moulding, motivated by seasonal performance variations observed in an industrial setting. A controlled trial production series was conducted using six discrete moisture levels (1.38% to 2.76%), prepared from a single material batch. Data from 1 524 parts produced across 254 cycles were collected, including machine parameters and 100% automated camera inspection of eight surface defect types. Statistical analyses comprised Kruskal–Wallis testing, linear and quadratic logistic regression, and correlation modelling. A pronounced non-linear, quadratic relationship between moisture content and scrap rate was identified. At low moisture levels (< 2.0%), scrap rates surged to 22–26%, driven by excessive slip surface, protrusion, and void defects, while peak injection pressures exceeded 2 100 bar. In contrast, an optimal moisture window of 2.0%–2.6% yielded scrap rates as low as 3%, accompanied by a stable switchover volume of 6.0 cm 3 with zero variance and substantially reduced baric resistance. Quadratic logistic regression provided a significantly better fit than a linear model (ΔAIC > 10, where AIC is the Akaike Information Criterion) and confirmed the optimal operating range. Switchover volume was shown to be a more sensitive leading indicator of visual defects than pressure, and moisture levels above ≈2.7% introduced melt cushion instability without further quality gains. These findings challenge the common assumption that drier material always improves quality, instead demonstrating an optimal moisture window that maximizes yield. Practical recommendations include standardizing incoming material to 2.1% ± 0.2% moisture and using real-time switchover volume monitoring for predictive quality control. This work provides a data-driven framework for reducing scrap, lowering energy consumption, and extending tool life in thermoset injection moulding.