Abstract:
Efficient fault diagnostics and on-line condition monitoring is crucial for maintaining reliability and availability of engineering assets and infrastructures. Developing...Show MoreMetadata
Abstract:
Efficient fault diagnostics and on-line condition monitoring is crucial for maintaining reliability and availability of engineering assets and infrastructures. Developing diagnostic models capable of early detection of machinery's defects have always been a challenge due to dynamic operating and environment factors affecting performance of assets over the life cycle. Traditional condition monitoring strategies are often limited by strong statistical assumptions and their dependency on subject matter of expert knowledge. Recent advances in condition-based maintenance have elevated significance of machinery health management in various industries. However, the implementation of it in a dynamic industrial setting is an open problem due to the non-standard nature of asset condition data and quality restrictions of maintenance records. In this study, the authors investigate an application of random forest to develop an intelligent dynamic fault diagnostic model for high voltage water pump sets. Historical overall vibration data including (waveform and spectrum) collected from pump sets together with their maintenance event and repair data to develop the algorithm. To assess performance of this model, results have been compared to actual data considering 70% of datasets used for training and 30% for testing. The model achieves an average accuracy of 97% in distinguishing between the two conditions of healthy and unhealthy for misalignment. Finally, we demonstrate this model can be successfully used in real time industrial fault diagnostics and health management applications.
Date of Conference: 22-24 November 2023
Date Added to IEEE Xplore: 08 January 2024
ISBN Information: