EXPERIMENTAL AND DATA-DRIVEN PREDICTION OF POST-CURE DIMENSIONAL STABILITY AND COMPRESSION SET IN EXTRUDED HCR SILICONE PROFILES
DOI:
https://doi.org/10.46121/pspc.54.1.62Keywords:
HCR Silicone Rubber, Extrusion, Post-Cure, Compression Set, Dimensional Stability, Response Surface Methodology, Random Forest, Crosslink DensityAbstract
High-consistency rubber (HCR) silicone profiles are extensively used as sealing gaskets, glazing beads and structural seals in automotive, construction and appliance applications, where long-term dimensional stability and low compression set are essential to sustained sealing performance. This study combines a Box-Behnken response surface design with a data-driven Random Forest regression model to experimentally characterize and predict post-cure linear shrinkage and compression set in extruded, peroxide-cured HCR silicone profiles as a function of post-cure oven temperature (180-220 °C), post-cure duration (20-60 min) and extrusion line speed (2-8 m/min). Seventeen Box-Behnken experimental runs were produced, post-cured, and evaluated for linear dimensional shrinkage, compression set (ASTM D395, Method B, 22 h at 175 °C), Shore A hardness shift and tensile strength retention. The fitted quadratic regression model for compression set achieved R² = 0.984 on the 17-run design (p < 0.0001), and R² = 0.947 on held-out test batches, while a Random Forest model trained on an expanded dataset of 84 production and laboratory batches achieved a lower prediction error (RMSE = 1.42 versus 2.31 percentage points for the RSM model) on the same held-out test batches. Crosslink density, measured via equilibrium solvent swelling and the Flory-Rehner relation, correlated strongly with compression set (r = -0.86) and emerged as the third most influential predictor behind post-cure temperature and duration in the data-driven feature-importance ranking. The optimized post-cure condition (213 °C for 46 minutes) produced a verified compression set of 21.6 ± 1.3% and linear shrinkage of 2.4 ± 0.2%, both within specification for automotive sealing applications. These findings demonstrate that pairing classical design-of-experiments methodology with ensemble machine learning provides a practical, industrially deployable route to tightening post-cure process windows for extruded HCR silicone profiles without recourse to exhaustive full-factorial testing.

