Session: 08-05-04 Industrial Reliability and AI Diagnostics

Submission Number: 157153

Algorithm Analysis on Rotating Machine Fault Diagnosis Using Explainable Deep Learning 

Rotating machines in ships and marine plants are important components that generate power and perform special functions. With the development of eco-friendly ships such as LNG ships, ammonia-powered ships, and offshoe wind turbines, the complexity and operating conditions of rotating machines are continuously becoming more severe. Therefore, CBM (Condition-Based Maintenance) and PHM (Diagnosis and Health Assessment) research are actively being conducted to predict future abnormalities from the current state and establish preemptive countermeasures. Recently, due to the characteristics of fault diagnosis research that requires the use of a large amount of data, deep learning has been introduced to efficiently handle data, increasing the accuracy of condition diagnosis, and the black box characteristic of the model, which was considered a fundamental limitation, without information on the classification criteria, is being overcome by explainable deep learning. Explainable deep learning is an algorithm that combines existing deep learning algorithms with an interpretation technique that can confirm the decision process of the model. It has been applied to the medical and legal fields, and has recently been expanded to machine fault diagnosis and is being used usefully. In this study, we apply the major deep learning algorithms and explainable deep learning algorithms known to have recently shown high performance to equipment simulating ship rotating machinery to analyze and diagnose vibration signals in defective situations of the equipment. The equipment can simulate defective situations such as mass imbalance, shaft misalignment, gear tooth defect, and bearing defect that may occur in rotating equipment. In this study, we analyze the basis for the latest algorithm to classify defective states targeting representative defective situations such as mass imbalance and gear tooth defect. This study will be able to contribute to selecting the optimal algorithm for diagnosing defective situations of rotating machinery inside ships and improving the algorithm performance by comparing deep learning classification methods used for defective diagnosis.

Presenting Author: Seong Hyeon Kim Chungnam National University

Presenting Author Biography: After graduating from Chungnam National University with an undergraduate degree, he is currently pursuing a master's degree in autonomous navigation system engineering at Chungnam National University. His area of ​​interest is fault diagnosis of rotating machinery.