Data-Driven Identification of Remaining Useful Life for Plastic Injection Moulds

Abstract

AbstractThroughout their useful life, plastic injection moulds operate in rapidly varying cyclic environments, and are prone to continual degradation. Quantifying the remaining useful life of moulds is a necessary step for minimizing unplanned downtime and part scrap, as well as scheduling preventive mould maintenance tasks such as cleaning and refurbishment. This paper presents a data-driven approach for identifying degradation progression and remaining useful life of moulds, using real-world production data. An industrial data set containing metrology measurements of a solidified plastic part, along with corresponding life-cycle data of 13 high production volume injection moulds, was analyzed. Multivariate Statistical Process Control techniques and XGBoost classification models were used for constructing data-driven models of mould degradation progression, and classifying mould state (early run-in, production, worn-out). Results show the XGBoost model developed using element metrology & relevant mould lifecycle data classifies worn-out moulds with an in-class accuracy of 88%. Lower in-class accuracy of 73% and 61% were achieved for the compared to mould-worn out less critical early run-in and production states respectively.KeywordsSmart manufacturingInjection mouldingData-driven modelMachine learningPredictive maintenance