Predictive Maintenance for Equipment Failure Analysis Using Machine Learning Approach

Abstract

Predictive maintenance has emerged as an effective strategy for enhancing equipment reliability, minimizing unplanned downtime, and optimizing maintenance resources in modern industrial systems. Conventional maintenance approaches, including reactive and preventive maintenance, remain insufficient for anticipating failures in complex machinery such as hydraulic press equipment. This study proposes a data‑driven predictive maintenance framework based on a Long Short‑Term Memory (LSTM) neural network for equipment failure prediction using sensor‑based operational data. A subset of 1,000 records comprising temperature, vibration, and usage‑related features was extracted from a publicly available hydraulic system dataset and pre-processed to ensure suitability for deep learning modelling. The proposed LSTM model was trained and optimized using two optimization algorithms, Adam and RMSprop, to examine their impact on convergence behaviour and predictive performance. Model evaluation was conducted using five widely accepted classification metrics: accuracy, precision, recall, F1‑score, and the area under the receiver operating characteristic curve (ROC‑AUC). Experimental results demonstrate that the LSTM model optimized with Adam achieved superior overall performance, recording an accuracy of 97%, precision of 97%, recall of 97%, F1‑score of 96%, and ROC‑AUC of 99%. The RMSprop‑optimized model exhibited comparable performance, achieving a perfect ROC‑AUC of 100%. Comparative analysis against benchmark machine learning models reported in prior studies including Random Forest, Gradient Boosting, and Support Vector Machine shows that the proposed LSTM‑based approach consistently outperforms conventional techniques. These findings demonstrate strong effectiveness of LSTM networks for predictive maintenance within the evaluated experimental setting in hydraulic press systems and highlight the role of optimizer selection in enhancing model performance.