A convolutional neural network based on MSCAM for intelligent diagnosis of ball screws

Abstract

In response to the challenge of identifying fault types in ball screws of CNC machine tools, particularly under complex operating conditions with often low classification accuracy, we propose a convolutional neural network fault diagnosis model that incorporates multi-scale convolution and an attention mechanism (MSCAM). First, we collect fault data corresponding to various fault types of the ball screw and establish a comprehensive fault dataset. Next, we apply the S transform to the original data to generate time-frequency diagrams, which serve as input for the two-dimensional neural network. In this paper, we present a multi-scale convolutional layer integrated with an attention mechanism, designed to highlight key features in fault information and extract more comprehensive characteristics. Ultimately, the superior recognition and classification capabilities of the model are validated through experimental datasets, and its robustness is thoroughly analyzed.