PENERAPAN MACHINE LEARNING DALAM PREDICTIVE MAINTENANCE

Abstract

Predictive maintenance (PdM) is used to predict when equipment may fail and assist in providing adequate monitoring and maintenance planning before potential future failures. In this research there are several topics of discussion regarding predictive maintenance on the NKK-SEMT PIELSTICK 12PC4-2V-570 main engine turbo intercooler unit on passenger ships including how to model predictive maintenance using machine learning using supervised learning methods, namely decision tree and random forest (rf). Decision tree, then what models and methods have a high accuracy value and finally to get the operating hours of the equipment until it fails. The decision tree approach is used to predict category or class labels and construct knowledge based on training data and labels. Random forest is a collection of decision trees generated from random sample selection with different node splitting rules. This model utilises a subset of the features in each tree, then finds the best threshold to split the data. Through turbo system performance anomalies, the use of machine learning with supervised learning methods can process previously classified data to get output in the form of predictive maintenance. Based on the results of predictive maintenance modelling, the decision tree method is the most accurate method with a ratio of training data and testing data division of 60:40 resulting in the highest accuracy. With MAE value 0, RMSE 0, RAE 0%, TP rate 1, FP rate 0, recall 1, F-measure 1, and Accuracy 100%. By using a decision tree model with a data set comparison ratio of 60:40, predict fututre is performed, the equipment is predicted to fail after operating for the next 3112 hours.