Data-driven machine learning and finite element-informed surrogate modeling for design optimization and failure prediction of reinforced thermoplastic pipes

Reinforced Thermoplastic Pipes (RTPs) offer a significant advantage as corrosion-resistant alternative to metallic pipelines, yet their mechanical performance under combined loading is not fully understood. This study proposes a comprehensive outline integrating finite element (FE) modeling with machine learning (ML) to predict RTP failure under multi-factorial conditions, including pressure, temperature, and bending. An LS-DYNA model of an experiment of flexural testing with 4.34% mean error validated the FE model and simulation of the RTP behavior with high accuracy. Through 135 parametric simulations, the diameter-to-thickness (D/t) ratio was identified as the most influential factor for bending moment capacity, while fiber orientation angle predominantly affected rotation angle at failure. A Random Forest regression model trained on this data achieved high predictive accuracy (R 2 = 0.93 for bending moment, R 2 = 0.97 for rotation angle). This ML-enhanced framework enabled the identification of optimal RTP designs, demonstrating significant potential for reducing computational costs and guiding future design optimization strategies for composite pipes.