Bearing faulty prediction based on knowledge distillation

This paper employs knowledge distillation to train teacher and student models using different motor bearing vibration datasets.The signal is transformed from the time domain to the frequency domain using Fast Fourier Transform (FFT), and a Convolutional Neural Network (CNN) model is used to recognize the bearing conditions.The teacher model is a deeper model trained with a larger dataset, while the student model is a shallower model trained with less data.The student model is guided by the soft labels provided by the teacher model.The results demonstrate that knowledge distillation improves the student model's recognition performance and enables knowledge transfer, allowing the student model to achieve good recognition accuracy even with limited training data.