Estimation of Vehicle Dynamic Response from Track Irregularity Using Deep Learning Techniques

To improve the quality of track maintenance work, it is a desire to estimate vehicle dynamic behavior from geometry irregularities. This paper proposes deep learning model predict responses (e.g., vertical wheel-rail forces, wheel unloading rate, and car body acceleration) using techniques. In proposed CA-CNN-MUSE model, convolutional neural networks (CNNs) are used learn features irregularities, multiscale self-attention mechanisms (MUSE) employed capture long-term short-term trends sequences. Coordinate attention (CA) introduced into CNN focus on important interchannel relationships spatial mileage points. The experiments were performed multibody simulation system measured data actual high-speed line. results show that has high prediction accuracy for fast computation speed. predicted time-domain waveforms power spectral densities (PSDs) agree well with responses. main lateral can also be captured by method, yet not as good ones.