Inverse design of composite pipe fittings using deep learning for lightweight structural optimization

Abstract

The conventional design of composite pipe fittings is highly dependent on experimental testing and finite element simulations, which are both costly and time-intensive. To address these challenges, this study introduces a deep learning-based inverse design approach to optimize the stiffness characteristics of carbon fiber reinforced polymer (CFRP) composite pipe fittings. Two predictive models were developed: a Long-Short-Term Memory-Based (LSTMB) Model and a Multi-Head Attention-Based (MHAB) Model. Comparative evaluations revealed that the MHAB model outperformed the LSTMB model in terms of predictive accuracy and generalization capability. Based on this, a population-based optimization algorithm was integrated to achieve the inverse design of the composite pipe fittings, ensuring efficient structural optimization while satisfying design constraints. The proposed method was validated through two optimization case studies, demonstrating its effectiveness in improving the efficiency and precision of composite pipe fitting design. This study highlights the potential of deep learning, particularly the Transformer framework, to accelerate the design and optimization of composite materials.