Implementation Of Predictive Maintenance By Predicting The Remaining Useful Life Of Machine

Abstract

Machine maintenance is one of the most important aspects for maintaining productivity. Effective maintenance not only prevents damage, but also extends the service life of equipment and reduces maintenance costs. Therefore, this study aims to determine the accuracy of Remaining Useful Life (RUL) predictions using a hybrid Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) model, as well as to determine the optimal time to perform maintenance in order to minimize total maintenance costs. The data used consists of machine condition parameters as input and actual RUL values as the target. The hybrid LSTM-GRU model was developed with hyperparameter tuning using Keras Tuner to obtain the best architecture combination. Evaluation was conducted using the Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared (R2) metrics on the test data. After obtaining the RUL prediction results, maintenance cost optimization modeling was performed using the Weibull distribution approach and cost model. This model considers preventive costs and emergency breakdown repair costs based on the probability of failure at each point in time before the actual failure. The results of the study show that the hybrid LSTM-GRU model is capable of producing RUL predictions with excellent performance. In addition, the optimization model shows that performing maintenance before failure can significantly reduce total maintenance costs.