Strength evaluation of CF/PEEK resistance welding based on improved artificial neural network: Interface failure mechanism study under extreme service temperatures

In this work, an improved genetic algorithm-optimized artificial neural network (GA-ANN) model was proposed to efficiently predict the lap shear strength (LSS) of carbon fiber-reinforced polyetheretherketone (CF/PEEK) composite resistance welding over a wide temperature range (293.15 K-473.15 K). Combined with variance analysis, the study further explored the interactions of service temperature, current, and time on the LSS during the welding process, which often demonstrated strong parameter sensitivity. Notably, the strength of the welding interface at high temperatures exhibited a unique temperature-dominant effect: as the service temperature increased, material-level temperature influence mechanisms led to a gradual decline in interface strength, showing clear dominance and weakening the synergistic interference of other factors on LSS. When the service temperature increased to 453.15 K, the average LSS increment was only 3.82 MPa as the welding current increased from 48 A to 52 A, which was just 35.11 % of the increment observed under the same parameters at 253.15 K (69.31 %). A micro-scale forming model of resistance welding was developed based on molecular dynamics simulations, systematically investigating the diffusion and entanglement behavior of polymer molecular chains at the interface under different temperatures. This indirectly confirmed the strong temperature sensitivity of polymer welding, where the intrinsic effects at the micro-scale significantly reduced the influence of process parameters. Through multi-scale interface damage morphology analysis, the interface failure modes of resistance welding under wide temperature range service conditions were discussed. The study revealed that at high temperatures (above 453.15 K), the failure mode of the welding interface was primarily fiber/matrix delamination, rather than the commonly observed wire mesh/resin interface debonding. This research not only expanded the multi-dimensional analytical approaches for thermoplastic composite resistance welding but also provided critical theoretical support for the design and optimization of jointed structural components under extreme high-temperature service conditions.