Research on multi-sensor data fusion fault diagnosis method for rotating parts of injection molding machine

Abstract

In the field of fault diagnosis of rotating parts of injection molding machines, traditional methods often rely on single sensor data, which makes it difficult to comprehensively capture the complex dynamic characteristics of the system operation, limiting the accuracy and reliability of diagnosis. To overcome this challenge, this paper proposes a multi-sensor data fusion fault diagnosis method combining graph convolution and Transformer neural network. A constraint-based Bayesian network structure learning algorithm is adopted to fuse vibration and current multi-sensor data, and a graph structure reflecting the basic dependency relationship between sensor data is constructed; a multi-head self-attention mechanism for weight allocation of the basic dependency relationship between multi-sensor data is investigated; and a model architecture taking into account the global and local features of multi-sensor data is constructed by combining graph convolution and Transformer neural network. Experiments show that the fault diagnosis accuracy of the method proposed in this paper reaches 99.20%, which is 5.4% higher than that of single sensor data, proving the effectiveness and accuracy of the method proposed in this paper. Finally, it is applied in the actual scenario of an injection molding machine, and the experimental effect is good.