Wheel condition assessment of high-speed trains under various operational conditions using semi-supervised adversarial domain adaptation

Train wheels, among other components, are critical for the safety and ride comfort of high-speed rail systems. Various machine learning methods have been used together with onboard monitoring data to assess wheel health conditions. However, only in some well-controlled experiments or authorized circumstances (source domain) can well-labelled supervised be obtained. Even so, due difference operational conditions, directly applying model learned from this case interest (target is not reliable. Facing challenge, we propose an adversarial domain adaptation (DA) approach transfer knowledge a test one section interest. Since target domain, corresponding components that new after reprofiling labelled as “intact”, DA modified semi-supervised rather than unsupervised. Two-level marginal conditional conducted manner, which sufficiently eliminate distribution discrepancy induced by differences between two sections on train runs. Onboard collected Lanxin before study. Results demonstrate effectiveness well its superiority over three baseline models, underneath mechanisms visualized. The study expected provide thinking condition assessment key when runs under various