Parallel CNN small sample ball screw intelligent fault diagnosis method based on symmetric dot pattern and wavelet synchronous extraction transform

In the process of diagnosing faults in ball screws of CNC machine tools, deep learning-based methods have become widely adopted in actual industrial fault diagnosis scenarios. However, these learning methods typically require a substantial amount of training data. When the available data samples are insufficient, neural networks are prone to overfitting, resulting in low diagnostic accuracy. To address this issue, this paper proposes a parallel convolutional neural network (PCNN) for intelligent diagnosis of ball screw faults, utilizing symmetric dot pattern (SDP) and wavelet synchronous extraction transform (WSET). A SW-PCNN fault diagnosis model is established. First, SDP and WSET are employed to transform one-dimensional time series feature data into two sets of image data that encapsulate state features. These two sets of image data are then input into a dual-channel convolutional neural network for training. Finally, the model's recognition and classification performance are evaluated using principal component analysis (PCA) for dimensionality reduction and visualization, thereby verifying the model's robustness and generalization capabilities. Test results indicate that the proposed method can quickly and effectively identify fault types, achieving a fault classification accuracy of 98.91% on the test set.