Convolutional-Transformer Model with Long-Range Temporal Dependencies for Bearing Fault Diagnosis Using Vibration Signals

Abstract

Fault diagnosis of bearings in rotating machinery is a critical task. Vibration signals are a valuable source of information, but they can be complex and noisy. A transformer model can capture distant relationships, which makes it a promising solution for fault diagnosis. However, its application in this field has been limited. This study aims to contribute to this growing area of research by proposing a novel deep-learning architecture that combines the strengths of CNNs and transformer models for effective fault diagnosis in rotating machinery. Thus, it captures both local and long-range temporal dependencies in the vibration signals. The architecture starts with CNN-based feature extraction, followed by temporal relationship modelling using the transformer. The transformed features are used for classification. Experimental evaluations are conducted on two datasets with six and ten health conditions. In both case studies, the proposed model achieves high accuracy, precision, recall, F1-score, and specificity all above 99% using different training dataset sizes. The results demonstrate the effectiveness of the proposed method in diagnosing bearing faults. The convolutional-transformer model proves to be a promising approach for bearing fault diagnosis. The method shows great potential for improving the accuracy and efficiency of fault diagnosis in rotating machinery.

Keywords

• bearing fault diagnosis
• vibration signals
• deep-learning architecture
• attention mechanism
• transformer model
• long-range temporal dependencies