Cross-machine reliability and fault diagnosis using CORAL feature alignment and enhanced deep extreme learning machine

Abstract

 Reliability and efficiency are vital components in modern manufacturing systems, especially for rotating machinery that operates continuously under a variety of environments. Recently, there has been growing interest among researchers in developing and deploying a cross-machine fault-diagnosis system in real-world industrial settings. It allows engineers to perform fault diagnosis across multiple machines without having to build a new intelligent model each time. Unlike most studies that utilized complex transfer learning architectures, this paper proposes a simple and efficient cross-machine bearing fault diagnosis framework based on Correlation Alignment (CORAL) and Deep Extreme Learning Machine (DELM). It aligned time-domain statistical features from two datasets, CWRU and the experimental dataset, to reduce the domain gap between them and improve generalization performance.  The aligned features were classified using a DELM model.  The experiment yielded reliable cross-machine generalization, with an average accuracy of 91.27%. These findings underscore the model’s capability to offer a simple, effective, and low-computational-demand solution for cross-machine applications, making it suitable and practical for real industrial use.