AI-Driven Real-Time Predictive Fatigue Assessment for EV Chassis Using In-Motion Structural Learning

Abstract

Fatigue durability assessment of EV chassis structures is traditionally based on static fatigue rigs and offline finite element analysis, limiting real-time adaptability under highly variable road excitations. This study proposes an AI-Driven In-Motion Structural Learning (IMSL) framework for real-time predictive fatigue assessment of an electric vehicle (EV) chassis using the structural learning-in-motion paradigm. The objective is to continuously infer fatigue severity and forecast Remaining Useful Life (RUL) on-vehicle without reliance on static laboratory durability cycles. The method integrates synchronized multi-sensor acquisition (foil strain gauges, tri-axial vibration sensors, IMU, GPS, and load cells), followed by digital filtering, normalization, and time–frequency feature extraction prior to neural structural learning. A physics-correlated FEA solver was used for stress validation, while neural models performed real-time fatigue inference on edge hardware. Results indicate repeated chassis vibration peaks of 25–30 g, and cyclic strain transients at critical welded interfaces reaching ≈3.6 MPa, while backbone regions remained ≈0.4–1.2 MPa. Stress-contour correlation confirmed fatigue hotspots spanning 2–18 MPa, with dominant concentration at 15–18 MPa. Neural training achieved stable convergence with final training loss ≈0.42 and validation loss ≈0.09, producing strong predictive generalization. Fatigue-life inference-maintained R² ≈0.95, with predicted fatigue cycles within ±5 cycles (40–80 cycles) and ±8–10 cycles (>80 cycles). Earliest measurable damage evolution appeared at ≈180 cycles (fatigue index ≈0.6), reaching saturation at ≈0.9 by 2,900 cycles, enabling implicit RUL intelligence. The study concludes that IMSL delivers a scalable, experimentally observable, and reviewer-defensible approach for real-time EV chassis fatigue durability learning and edge-capable predictive maintenance deployment.