Unsupervised Learning and Digital Twin Applied to Predictive Maintenance for Industry 4.0

Abstract

The diagnosis of anomalies in industrial equipment is a vital research area, as the product quality to be manufactured is inextricably linked to the machine’s efficiency. To date, conventional statistics on the industrial equipment fault prevention are weak. The major concept of the future Industrial 4.0 framework is the integration of artificial intelligence (AI) and the implementation of digital twin (DT), which could avoid serious economic losses caused by unexpected equipment failures and significantly improve system reliability. DT is an emerging technology in the context of digital transformation that enables the monitoring, diagnosis, energy efficiency, and optimization of different systems. Numerous initiatives have shown how AI can enhance the performance of DT for industrial applications. This paper proposes a methodology based on the integration of the autoencoder (AE) and the long short-term memory (LSTM) networks by using DT architecture for monitoring and predictive maintenance (PdM) in manufacturing. Our methodology was put into practice on a real-life industry example using data collection, analysis and deep learning approach. The aim of the proposal is to implement a deep learning hybrid model combining LSTM with AE to perform anomaly detection tasks on a monofilament winding machine. This approach enables better quality results and more efficient management of the weaver’s workshop. The effectiveness of the proposed approach in a monofilament winding machine is demonstrating by a high accuracy (98%) of the model.