Improved CNN Networks for Fault Diagnosis Based on Imaging Multi-dimensional Signals

This paper presents an innovative approach for intelligent fault diagnosis of equipment by integrating Multivariate Variational Modal Decomposition (MVMD) with Symmetric Dot Plot (SDP) to create a Multivariate Symmetric Dot Plot (M-SDP) for multi-channel time series data. Unlike previous studies focused solely on single-channel signals, this method processes multivariate signals using deep learning algorithms. MVMD extracts dominant subcomponents from multiple sensors, which are then represented within the M-SDP image. The structural similarity index (SSIM) is used to optimize M-SDP parameters for distinguishing between fault states. Additionally, the Convolutional Block Attention Module (CBAM) is incorporated into a Convolutional Neural Network (CNN), creating the CBAM-CNN approach. This proposed method, validated using gearbox data from Southeast University, demonstrates higher diagnostic accuracy and stability compared to other signal-to-image methods.