Development Of Predictive Maintenance Methodology Utilizing Machine Learning Technology To Support Plant Health Management

To maintain availability and reliability in PT. Indonesia Power (IP) power plants at high level, proper maintenance strategy must be well established. IP utilizing Predictive Maintenance (PdM) to predict the equipment conditions from plant healthiness. PdM analyze data trends to predict future behavior of equipment. Recent development in PdM methodology is applying Machine Learning (ML) technology which is proven effective and improves the speed and accuracy in the Plant Health Management (PHM) decision making. The ML technology used utilize anomaly detection based on regressive models. The model utilizes historical data and trains using the linear regression trainer model then predict current value using its lagged past in the time series. The prediction results then evaluated using scoring model. Overall result shows high prediction accuracy indicates in high number from the scoring model. Further analysis performed to forecast equipment condition and prediction of remaining threshold reaching in certain period. The analysis results then fed to database software to be documented and visualized in web-based charts. By employing the analysis provided in the web-based charts, allows engineer to develop appropriate PHM execution strategy. The current development PdM Methodology is proven to be capable of equipment condition prediction by utilizing ML technology. The ongoing development will be aimed in processing more training data with higher granularity and deployment in production environment.