Machine Learning-Based Predictive Maintenance System for Artificial Yarn Machines

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Predictive maintenance (PdM) has become a critical strategy for improving the efficiency and reliability of industrial machinery. Integrating machine learning methods into a PdM system provides a promising solution for optimizing maintenance strategies, preventing equipment failures on the production line, and reducing downtime. This research presents a data-driven approach for detecting faults in industrial machines using sensor data. The method aims to optimize system performance, resulting in economic savings including energy consumption, and maintenance costs. The approach outlined in this research includes the establishment of a PdM system designed for yarn production machines empowered by machine learning methods. The effectiveness of PdM applications depends on careful selection of machine learning methods. This study examines four machine learning algorithms and a deep learning algorithm for predictive modeling. The algorithms were trained on historical data to identify underlying patterns and correlations between operational parameters and failure events. The trained models were deployed in the PdM system to continuously monitor the health condition of industrial machines on ThingSpeak™ IoT interface platform in real-time. This research also presents a systematic process for developing a predictive maintenance framework. The process includes data acquisition from industrial machinery, preprocessing, feature selection, model training, and deployment. The effectiveness of the proposed system is validated through extensive experimentation and case studies conducted in an industrial setting. Evaluation metrics revealed that the deep learning algorithm outperformed the other approaches, achieving an accuracy score of 0.96 and a prediction validation score of 0.86. The results show significant improvements in predictive accuracy and enhanced operational efficiency compared to reactive maintenance approaches.

This publication has 34 references indexed in Scilit: