MetroNet: A Novel Data-Driven Fault Diagnosis Method Applied to Wheel Bearings of Metro Trains

The wheel bearing is a vital component of metro train, therefore, the fault diagnosis of wheel bearing is essential to ensure the reliability and safety. In recent years, some intelligent fault diagnosis algorithms have been proposed and achieved great success. However, to validate the performance of the proposed model, most researches only focus on several public datasets because of the limitation of data acquisition, which easily leads to inconsistent diagnosis after transferring the trained model to the real industrial scene. Therefore, this paper proposes an excellent method to implement data acquisition in industrial scenes. Based on data augmentation, we totally created the dataset with 40000 samples. Aiming at the training data, this paper proposes a novel data-driven fault diagnosis model named MetroNet that is applied to wheel bearings of metro trains. Notably, MetroNet is mainly constructed by CNN and RNN, and it can capture temporal correlation and spatial correlation of raw sensor data. Furthermore, the CNN adopts an innovation method of convolution over height and pooling over weight. The performance of MetroNet is evaluated on the testing dataset, and the optimal accuracy of fault diagnosis can be increased to 97.20%.