Machine learning revealed force-stress-fatigue damage correlation of high-speed train bogies

Abstract

Clarifying the correlation of multi-level mechanical parameters of structures in complex dynamic systems is a prerequisite for determining the accruing fatigue damage. In this paper, we adopt the independent component analysis algorithm in unsupervised learning and tap the latent correlation between measured forces and stresses of high-speed train bogies. It is revealed that there exists a strong correlation between the vertical force and the stress at the junction of the transverse beam and the side frame, a site prone to fatigue. Stresses reconstructed by strongly correlated independent components account for more than 70% of the fatigue damage, which in turn supports the finding that the vertical forces are the main contribution to the fatigue damage at the junction of the transverse beam and the side frame. This strong correlation between vertical forces and stresses effectively reduce the error in fatigue damage prediction and provide insights into fatigue life enhancement of critical structures of dynamic systems beyond high-speed trains.