Train Bogie Bearings Fault Diagnosis Model Based on Multimodal Signal Features and Physics Knowledge Learning

Acoustic and vibration signals share consistent principles in their generation and transmission mechanisms. Studying the similarities and complementarities between the two types of signals can effectively improve the accuracy of fault identification. Fault diagnosis methods incorporating domain knowledge can significantly enhance the recognition precision and efficiency of models in tasks with limited samples. Therefore, a two-stage fusion-based fault diagnosis framework has been developed based on the characteristics of acoustic and vibration signals. First, according to the respective signal characteristics, multiscale multilayer perceptron models were constructed to fuse the physical features of vibration and acoustic signals separately. Then, a second-level fusion model was established based on Dempster–Shafer evidence theory, realizing the joint operation of the acoustic–vibration system. This proposed model had been validated on a bogie fault simulation test dataset, achieving an identification accuracy of over 95<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> even in experiments with small sample sizes. Compared to other methods, this approach demonstrated its effectiveness and feasibility.