Thermo-mechanical formability, fracture mechanisms, and ANN-assisted multi-objective optimization in heat-assisted single point incremental forming of polycarbonate sheets

Heat-assisted single point incremental forming is a flexible, low-cost method for producing customized thermoplastic components, but its wider use for polycarbonate is still limited by a partial understanding of formability and failure evolution. This study presents an integrated experimental-computational investigation of thin polycarbonate sheets formed under thermo-mechanically assisted single point incremental forming conditions. Four characteristic failure modes were identified, namely peripheral cracking, oblique cracking, crazing, and twisting. Scanning electron microscopy revealed microvoid nucleation, fibrillar craze bridging, shear lips, and progressive crack propagation, indicating a multi-stage damage process. Artificial neural network models were developed using lubricant type, temperature, and step depth as inputs, with forming depth ( h ), surface roughness ( R a ), and angular deviation ( β ) as the target responses. Nested cross-validation selected final artificial neural network architectures of 7-logsig for h, 3-logsig for R a , and 3-tansig for β . The R a model showed the strongest predictive capability, with outer cross-validation root mean square error of 0.0950 ± 0.0289 and R 2 of 0.9176, while the corresponding full-fit model achieved root mean square error of 0.0705 and R 2 of 0.9686. Garson sensitivity analysis identified temperature as the dominant governing parameter, and artificial neural network-assisted NSGA-II optimization indicated that 5W30 lubricant at 180 °C and 0.2 mm step depth yielded the best compromise condition, with predicted h of 11.135 mm, R a of 0.341 μm, β of 5.199°, and overall desirability of 0.676.