A Data Mining Framework for Maintenance Prediction of Faulty Products Under Warranty

Abstract

Predicting the sequential patterns of maintenance activities (replacement or repair) with the needed spare parts for faulty products becomes the main challenge to maintenance engineers. This research, therefore, develops a data mining framework for predicting the integrated sequential patterns of maintenance activities and identifying the classification of the frequent components’ spare parts for faulty products under warranty. In this framework, a large data set was mined for products under warranty including product attributes, maintenance activities, and spare parts. Then, data mining techniques were performed to determine the frequent sequence pattern of maintenance activities, involving: the selection of monthly maintenance activities, generalized sequential patterns (GSP), generation of association rules, and rule-based classification with/without considering product attributes. The frequent sequential patterns were validated using testing data. Further, the GSP was applied to determine the frequent sequential patterns of spare parts. Finally, integrated sequential patterns were generated for maintenance activities and spare parts. A case study of water-cooled chiller products was deployed to illustrate the developed framework. The effectiveness of the framework was illustrated with the warranty data mining for a water-cooled chiller. In conclusion, the proposed framework allows maintenance engineers to extract hidden knowledge regarding sequential patterns of maintenance planning and provides valuable information for maintenance prediction.