Deep Neural Network-Based Prediction of Temperature-Dependent Tensile Stress–Strain Behavior in Injection-Molded Impact-Modified Heat-Stabilized PA66 for Automotive Applications

Deep neural networks (DNNs) have emerged as promising tools for predicting temperature-dependent tensile responses in engineering polymers. In this study, a temperature-dependent DNN-based predictive framework was developed to estimate the stress–strain response of impact-modified heat-stabilized polyamide 66 (PA66 IM HS) across a wide thermal range relevant to automotive applications. The proposed model was trained and validated using experimentally obtained tensile data collected between −40 °C and 150 °C. The developed framework successfully captured the nonlinear tensile stress–strain response observed experimentally across the investigated temperature range. Experimental results demonstrated a strong dependence of stiffness, tensile strength, ductility, and post-yield deformation on thermal conditions, with distinct behavioral transitions associated with the glass transition region and thermally activated deformation processes. Compared with conventional interpolation approaches, the proposed DNN model provided stable and physically coherent stress–strain predictions at experimentally characterized temperatures and by interpolation at intermediate temperatures within the investigated thermal domain. The model accurately reproduced elastic behavior, yielding transition, and post-yield deformation across the evaluated thermal range, achieving excellent agreement with the experimental curves. Additionally, a fracture-aware interpolation strategy was implemented to preserve experimentally observed rupture behavior at low temperatures. The obtained results demonstrate the applicability of the proposed DNN-based methodology as a computational tool for estimating temperature-dependent monotonic tensile stress–strain behavior within the investigated thermal and experimental domain.