Improving Intelligent Fault Diagnosis With Semantic Segmentation for Industrial Applications

The rapid advancement of AI-based signal processing technologies and the exponential growth of sensor data have accelerated innovation in intelligent fault diagnosis (IFD) methodologies. In particular, the autonomous data analysis and end-to-end fault diagnosis capabilities of deep learning have marked a significant breakthrough compared to traditional manual feature extraction approaches. However, the increasing reliance on deep learning models has also heightened the demand for interpretability and transparency in results. The “black box” nature of deep learning models poses a significant challenge, potentially undermining the reliability of diagnostic outcomes and the transparency of decision-making processes, thus limiting their practical applicability in real-world systems. This paper identifies the key factors necessary to enhance the reliability of intelligent fault diagnosis systems and proposes a semantic segmentation-based approach to address these challenges. Experimental results using autonomous driving fault diagnosis data demonstrate that the proposed methodology achieves over 99.9% classification accuracy and over 98% segmentation accuracy in real-time environments, showcasing its exceptional utility and reliability. These findings highlight the potential of the proposed approach to significantly improve the practical applicability of IFD systems.