The connecting rod is one of the critical components of the engine which is subjected to complex cyclic loading, demanding precise evaluation of stress distribution and fatigue life. Conventional Finite Element Method (FEM)-based simulations, while accurate, are computationally expensive and time-intensive—especially when iterating through multiple design variations. To address this, we present an AI/ML-based predictive framework using Graph Neural Networks (GNNs) and Geometric Deep Learning (GDL) techniques to efficiently predict stress and fatigue life of the connecting rod.
The methodology involves representing FEM mesh data as graph structures, where nodes and edges capture geometric and physical relationships. GNN architectures, including MeshCNN and PointNet++, are trained on historical simulation data to learn spatial dependencies and nonlinear mappings between geometry, load conditions, and resulting stress/life outputs. The framework significantly reduces computation time while maintaining prediction accuracy close to high-fidelity simulations.
This novel application of GDL in Computer-Aided Engineering (CAE) showcases its potential to revolutionize the virtual validation process, offering a scalable solution for rapid design iterations and performance optimization in the automotive domain. Validation is performed by comparing AI-predicted stress and life results with full-scale FEA simulations. The proposed approach achieves high prediction accuracy (>90%), significantly reducing computational time by up to 80% while maintaining engineering precision. This framework enables rapid design iteration, optimization and real-time stress and fatigue life evaluation, making it a valuable tool for automotive industries seeking efficiency in structural component analysis.
Keywords: Geometric Deep Learning, Physics-Informed AI, Connecting Rod, Stress Prediction, Fatigue Life Estimation, Graph Neural Networks, Finite Element Analysis, MeshCNN, PointNet++, Deep Learning, Physics AI