Augmenting the Internet-based Machinery Health Monitoring System (MHMS) Through Integration of Long Short-Term Memory (LSTM) Model for Enhanced Prediction of Remaining Useful Lifetime (RUL)

Extensive research has been conducted to estimate the Remaining Useful Lifecycle (R UL) of components utilizing various approaches and models, including Machine Learning - Deep Learning (DL) techniques. This study aims to assess the efficacy of preselected DL methodologies in achieving superior prediction accuracy compared to conventional approaches when applied within the context of Predictive Maintenance (PdM) for turbomachinery equipment. Experimental findings reveal that the conventional model yields a Mean Absolute Percentage Error (MAPE) of 17.81%, whereas all preselected DL models consistently demonstrate improved prediction accuracy, with the lowest MAPE averaging 7.02%. Notably, the Long Short-Term Memory (LSTM) model emerges as the most accurate predictor, surpassing the naive model by 19.93 % at a 21-day horizon. Consequently, the LSTM model exhibits potential for enhancing the online Machinery Health Monitoring System (MHMS), showcasing its robust data analytic capabilities over extended forecasting horizons. Furthermore, it can provide recommendations for R UL estimation in turbo machinery components, facilitating PdM-based activities.