Long-Short Term Memory And Gradient Boosting Model For Hydraulic System Predictive Maintenance

A hydraulic system, a drive technology where a fluid is used to create force, is used in all kinds of large and small industrial settings, as well as buildings, construction equipment, and vehicles. Well-planned predictive maintenance is considered the most efficient maintenance strategy to maintain the performance of the system. A data-driven approach such as machine learning approaches while showing increasingly effective solutions in this domain, it has remained a challenge to adopt which method is fit, robust and provide the most accurate detection. This research proposes two Long-Short Term Memory (LSTM) models to predict the condition of each feature over time and various supervised algorithms to predict predicts the type of fault and the time fault that occur based on the condition of the features over time. The result shows the LSTM model which only considering one feature in the model provides higher accuracy than the model with all features. In predicting the fault, Gradient Boosting Classifier has the best performance among other models such as Logistic Regression, K-Nearest Neighbor, Support Vector Machine, Gaussian Naïve Bayes and the other ensemble models (extreme gradient boosting, random forest classifier, AdaBoost Classifier, extra tree classifier)