Fault diagnosis method of hot die forging press based on short-time Fourier transform and optimized convolutional neural network

Abstract

Aiming at the vibration signals of rolling bearings of a hot die forging press, which are nonlinear, unstable, and easily disturbed by strong background noise, a fault diagnosis method based on a short-time Fourier transform and optimized convolutional neural network was proposed. First, the Fourier transform was applied to the vibration signals of rolling bearings of a hot die forging press to obtain complete time-frequency samples. After spectrum compression, it is input into the convolutional neural network model. By optimizing activation function and adjusting network parameters, the purpose of efficient and fast detection in the case of small samples can be achieved. Through the simulation test, the feasibility and effectiveness are verified. The results show that the proposed method has high recognition accuracy for fault diagnosis of rolling bearings of hot die forging press.