Development of a planned preventive maintenance (PPM) model using a machine learning approach

Abstract

Planned preventive maintenance with some expert system is essential for appropriate planning and utilization of maintenance policy effectively and efficiently. A number of preventive maintenance model have been developed that have identified several factors which performed the models by subjective means. However, these models often lack robustness due to bias and variance. Now, the increased availability of data opens the scope of applying machine learning technique to predict the maintenance requirement more accurately and cost effectively.The aim of this research work is to develop a planned preventive maintenance model by using machine learning algorithms (SVM and SVR) that can forecast the maintenance requirements more accurately and cost effectively. To develop the model machine reliability is considered and the reliability depends on various subjective and objective measures which is a data driven approach. The subjective and objective features of Diesel Generator (DG) have been selected from literature and expert opinions and the data are collected from field survey. Two separate feature selection methods have been used to select the best feature set to improve the model accuracy. Wrapper method used correlation-based features selection to rank the features and generate eight different feature sets following backward elimination process. Filtering method eliminates the insignificant features by ANOVA test and selects the significant feature sets. All these feature sets aregenerated a total 54 number of different models with different accuracy level. Among them the best feature set have been selected with an accuracy of 92.5% from Wrapper method. Finally, a regression model has developed using Support Vector Regression (SVR) to determine the machine reliability value.