Predictive Maintenance of High-Velocity Oxy-Fuel Machine Using Convolution Neural Network

Abstract

Maintenance activities and programs have reached a critical factor for competitive advantage in a manufacturing firm. This is due to the increasing complexity of the interactions between different production activities as well with an increasingly extended manufacturing environment along with the increasing cost of machines’ maintenance. Existing practices namely preventive maintenance (PvM) which focused on routine scheduled maintenance is viewed as no longer economical. Therefore, with the advancement in the machine and artificial intelligence (AI) technology, predictive maintenance (PdM) is the future of machine maintenance. In the predictive maintenance, prognostic and health management (PHM) emerged as a technique that is acceded method used to anticipate machine breakdown through the Remaining Useful Life (RUL) determination. This study aims to apply a machine learning technique to predict the RUL to High-Velocity Oxyfuel machine (HVOF). As proposed by several recent researchers, the convolutional neural network (CNN), which primarily used in image processing has been used to predict machines breakdown due to its characteristics. In this study, the development of the CNN model is divided into 3 stages that serve the 1) pre-processing, 2) reference model development and 3) CNN development. At stage 1, missing data and imbalanced data were tackled using median imputation and synthetic over-sampling technique. In the second stage, clean data were fit into a conventional machine-learning algorithm namely Naive Bayes and result was remarked as the reference model and continued to stage 3 where the CNN is developed and clean data were then fed for prediction. Using the result from the reference model as a baseline, the CNN model is developed, evaluated and tested until the final accuracy of approximately 1.0 is obtained.