An exploratory analysis of the industrial machine's data for predictive maintenance operations

Abstract

Modern industries nowadays heavily rely on hefty machineries which have lots of moving parts and contain sensor data. These sensor data are indexed in time order which are referred to as time series data. Industrial machines have a huge maintenance cost and failure risks involved with them. Sometimes, a lot is at stake for the companies for preserving the health of these machines. Also important machineries like airplanes need to be maintained on a regular basis in order to prevent any kind of disaster while in operation. E ective maintenance of these equipment are crucial to avoid several damage, downtime for repair and to prevent any mishap which is easily avoidable. Predictive Maintenance is a prominent strategy for dealing with maintenance issues given the increasing need to minimize downtime and associated costs. Time series data plays an important role in this eld. We have implemented SVM, Logistic Regression and Random Forest model for classi cation on which we got 94% accuracy on an average after using di erent metrics. Moreover, we used LSTM and ARIMA for forecasting future values where LSTM performed better with an accuracy of 38.7%. Due to the imbalance on the data,the accuracy for classifying failure rate is very poor. Our goal is to explore and analyze di erent approaches of dealing with the time series data of industrial machines for using them to train di erent types of machine learning models and compare the performances of each approach.