紀要論文

Implementation of Unsupervised Machine Learning Techniques for Monitoring Coal Mills at a South African Power Plant

eng

24368083

https://doi.org/10.15002/00030725

法政大学大学院理工学研究科

法政大学大学院紀要. 理工学研究科編

65

1

7

2024-03-24

Version of Record

This study uses advanced machine learning techniques to create and assess effective predictive maintenance strategies for coal mills. The focus is determining coal mills' safe and unsafe operational phases, predicting potential tripping events, and scheduling maintenance to prevent equipment failures. The study used Gaussian Mixture Models (GMM) to cluster operational phases and Long Short-Term Memory (LSTM) models, including the CNN+LSTM architecture, for predictive analysis. The study evaluated the models using various metrics such as accuracy, precision, recall, F1, and ROC_AUC score. The research findings show that LSTM models, especially the CNN+LSTM architecture, can accurately identify unsafe operational phases and predict coal mill tripping events. The CNN+LSTM architecture has an output value that increases towards the failure moment, indicating its effectiveness in capturing relevant temporal patterns. This can be useful for scheduling inspections before failure.

Article