An Underdetermined Blind Source Separation Method for Fault Diagnosis of Bearing Vibration Signals Based on an Improved Conv-TasNet Network

This paper proposes an improved Conv-TasNet network for bearing fault diagnosis in underdetermined blind source separation. By increasing the depth of the encoder and decoder, the network significantly enhances its feature extraction capability, enabling it to more effectively capture complex features and improve the model's generalization to different signal environments. Compared with the traditional Conv-TasNet, the improved network can learn richer features and more complex mapping relationships. Simulation experiments verify the effectiveness and superiority of this method in simulated bearing fault diagnosis, providing a new deep learning approach for machinery fault diagnosis.