A novel intelligent fault diagnosis method of rolling bearing based on two-stream feature fusion convolutional neural network

Previous bearing fault diagnosis models show either low accuracy or long iterations, which are not suitable for real-time production quality control scenarios lacking computing resources. In this paper, the Two-Stream Feature Fusion Convolutional Neural Network (TSFFCNN) is established. In-depth features are extracted from the proposed parallel multi-channel structure of 1D-CNN and 2D-CNN and then jointed by feature fusion strategy for a more reliable diagnostic effect. Besides, Particle Smarm Optimized-Support Vector Machine (PSO-SVM) is adopted for higher accuracy. Model?s structural parameters are well-configured for fewer iterations and less computational cost. The algorithm?s diagnostic effectiveness on the single and simulated compound fault is verified. Stationarity and synchronicity are conceptualized to prove the reliability. With accuracy, convergence iterations, and time consumption, the TSFFCNN-PSO-SVM model is comprehensively compared with other intelligent algorithms. The experimental results reveal that TSFFCNN-PSO-SVM can identify fault modes from vibration signals more accurately with fewer iterations at the same time.

Automated summary

This paper proposes a new bearing fault diagnosis model TSFFCNN-PSO-SVM, which extracts deep features using a parallel multi-channel structure of 1D-CNN and 2D-CNN, combined with feature fusion strategy to improve diagnostic accuracy and reduce iteration and computational cost.