Designing a lightweight 1D convolutional neural network with Bayesian optimization for wheel flat detection using carbody accelerations
Dachuan Shi, Yunguang Ye, Marco Gillwald, Markus Hecht
Abstract
Many freight waggons in Europe have been recently equipped with embedded systems (ESs) for vehicle tracking. This provides opportunities to implement the real-time fault diagnosis algorithm on ESs without additional investment. In this paper, we design a 1D lightweight Convolutional Neural Network (CNN) architecture, i.e. LightWFNet, guided by Bayesian Optimization for wheel flat (WF) detection. We tackle two main challenges. (1) Carbody acceleration has to be used for WF detection, where signal-to-noise ratio is much lower than at axle box level and thus the WF detection is much more difficult. (2) ESs have very limited computation power and energy supply. To verify the proposed LightWFNet, the field data measured on a tank waggon under operational condition are used. In comparison to the state-of-the-art lightweight CNNs, LightWFNet is validated for WF detection by using carbody accelerations with much lower computational costs.