Intelligent Deep Learning Based Fault Classification Using Dual-Stream Transformer-CNN with Self-Supervised Feature Refinement for Industrial Applications

Fault classification plays a crucial role in industrial engineering, particularly in manufacturing and power generation, where accurate fault detection is essential to prevent system failures, reduce maintenance costs, and enhance operational safety.With the advancement of Industry 4.0 and 5.0, intelligent fault classification techniques leveraging real-time data processing have become increasingly important.This study proposes a deep learning-based fault classification model integrating Dual-Stream Transformer-CNN with Self-Supervised Feature Refinement (DSTC-SSFR) to improve classification accuracy and robustness.The core architecture consists of two parallel processing streams designed to effectively extract both spatial and temporal features from multivariate industrial sensor signals.The spatial stream uses multi-scale 1D Convolutional Neural Networks (1D CNNs) with varying kernel sizes to capture localized fault-related features at different frequency scales.The Bobcat is used for hyperparameter tuning, further enhancing model performance.The proposed approach achieves an overall accuracy of 95.07%, with a precision of 95.10%, recall of 95.07%, and F1-score of 95.07%.Additionally, the model attains a logarithmic loss of 0.1070, a Matthews correlation coefficient (MCC) of 0.9343, and an area under the ROC curve (AUC) of 99.54%.These results demonstrate the model's effectiveness in fault classification, offering a robust and efficient solution for industrial applications in smart manufacturing environments.