Predictive Maintenance of Banbury Mixer Using Machine Learning Methods: A Case Study of a Tire Manufacturing Factory

This study investigates parameters of the Banbury mixer, focusing on Material Rubber Sheet (MRS) production duration, to develop a machine learning-based failure detection system. Proposed data preprocessing methods aim to uncover patterns of the relationship between current and previous batch parameters. While data differencing suffices for SVM and RF models, it minimally impacts ANN. Solely employing data cleaning renders RF suitable for model creation. SVM outperforms RF in failure detection but may produce occasional false alarms. Performance evaluation indicates RF accurately detects 13 out of 29 failures, surpassing SVM detecting 16 failures with 2 false alarms. The SVM model from data cleaning combined with differencing reduces waste by 55.17%, surpassing RF by 6.89%. Future research could explore advanced preprocessing for failure cause categorization and leverage sophisticated techniques.