TY - GEN
T1 - Mechanical Fault Classification and Root Cause analysis Based on Machine Learning and Explainable AI
AU - Lee, Ya Ting
AU - Wang, Hung Kai
N1 - Publisher Copyright:
© RQD 2023. All rights reserved.All right reserved.
PY - 2023
Y1 - 2023
N2 - This study utilizes machine learning and explainable AI (XAI) to construct accurate prediction models and root cause analysis for Prognostic and Health Management (PHM). The study proposes a novel research framework consisting of two stages: the first stage involves fault type classification, while the second stage involves fault severity classification and root cause analysis. In both stages, multiple machine learning classifiers are employed to identify faults in rotating machinery, utilizing vibration features associated with the rotation frequency and its harmonics for classification. XAI techniques are incorporated for feature extraction of important features and root cause analysis. The proposed approach is evaluated using the Machinery Fault Database (MAFAULDA), and experimental results show that the first stage achieves an average accuracy of 98.7%, better than results reported in the literature. The second stage achieves an average accuracy of 86.7%. The results indicate that XAI feature importance improves classification accuracy and reduces computational costs and time. Additionally, XAI-based root cause analysis provides a faster way to identify the type and severity of faults through important features and setting thresholds determining the root cause of process variations.
AB - This study utilizes machine learning and explainable AI (XAI) to construct accurate prediction models and root cause analysis for Prognostic and Health Management (PHM). The study proposes a novel research framework consisting of two stages: the first stage involves fault type classification, while the second stage involves fault severity classification and root cause analysis. In both stages, multiple machine learning classifiers are employed to identify faults in rotating machinery, utilizing vibration features associated with the rotation frequency and its harmonics for classification. XAI techniques are incorporated for feature extraction of important features and root cause analysis. The proposed approach is evaluated using the Machinery Fault Database (MAFAULDA), and experimental results show that the first stage achieves an average accuracy of 98.7%, better than results reported in the literature. The second stage achieves an average accuracy of 86.7%. The results indicate that XAI feature importance improves classification accuracy and reduces computational costs and time. Additionally, XAI-based root cause analysis provides a faster way to identify the type and severity of faults through important features and setting thresholds determining the root cause of process variations.
UR - http://www.scopus.com/inward/record.url?scp=85174254527&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85174254527&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85174254527
T3 - 28th ISSAT International Conference on Reliability and Quality in Design, RQD 2023
SP - 211
EP - 215
BT - 28th ISSAT International Conference on Reliability and Quality in Design, RQD 2023
PB - International Society of Science and Applied Technologies
T2 - 28th ISSAT International Conference on Reliability and Quality in Design, RQD 2023
Y2 - 3 August 2023 through 5 August 2023
ER -