PREDICTION OF HIGH CYCLE TIMES IN WHEEL RIM MOLDING WITH ARTIFICIAL NEURAL NETWORKS

Abstract

Purpose: Monitoring processes through real-time data collection is useful for businesses to understand their processes better, and deal with production problems. Predicting cycle-time allows identifying production delays, downtime, and productivity loss. Thereby, taking necessary actions is facilitated to eliminate detected losses and to prevent problems towards meeting customer due dates. This study proposes a two-stage approach to determine a cycle-time threshold and predict high cycle times by examining sample molding process data obtained from a wheel-rim manufacturer. Methodology: Our study firstly determines thresholds for high cycle times with two alternate approaches. Subsequently, data were labeled regarding the cycle-time threshold. Alternate models based on Artificial Neural Networks (ANNs) were developed in R to predict high cycle times. Findings: Our findings include a comparison of cycle-time threshold approaches through a distance-based metric. After labeling of high cycle times, our study presents the performance of alternate predictive models. The performance of models was compared in terms of accuracy, recall and precision. Originality: Process mining in wheel rim molding has been found meager in prior research, despite the abundance of process mining applications and cycle-time prediction models. Another distinctive aspect of the study is cycle-time threshold determination with multiple methods to eliminate manual labeling of processes.