MSiT: A Cross-Machine Fault Diagnosis Model for Machine-Level CNC Spindle Motors

Cross-machine fault diagnosis (CMFD) of complex equipment is necessary for modern intelligent manufacturing systems. Manufacturing and assembly errors lead to inherent individual differences in machine-level computer numerical control (CNC) spindle motors, resulting in more challenging diagnostic requirements. The verification of the CMFD task is essential to ensure the reliability and effectiveness of machine-level diagnosis, but is often ignored in current data driven approaches. The latest transformer architecture, known for its excellent global feature extraction ability, is an ideal solution but has not yet been applied in this scenario. This article proposes a novel multichannel signal transformer (MSiT) method specifically toward CMFD task of machine-level CNC spindle motors. Specifically, this article presents a special tokenizer that is suitable for processing multichannel signals as the inputs of transformer, namely the unidirectional patch (UDP). It performs on all the channels to capture channel correlation features without additional transformations. The effect of structural hyperparameters on fault diagnosis performance is analyzed in detail for engineering reference. The superiority of the proposed method is validated using real industrial motor signals in comparison with the benchmark models and some state-of-the-art methods. Besides, the bidirectional decision-making mechanism of MSiT is revealed based on t-SNE and heatmaps.

Automated summary

This article proposes a novel method for cross-machine fault diagnosis of complex CNC spindle motors. By using a special tokenizer and Transformer architecture, the method is able to effectively diagnose faults. Experimental results validate the effectiveness of the proposed method and analyze the impact of structural hyperparameters on diagnosis performance.