Predictive Maintenance Framework for Production Environments Using Digital Twin

In this paper, we introduce an end-to-end IoT framework for predictive maintenance with machine learning. With this framework, all the processes for developing a learning-based predictive maintenance model such as data acquisition, data preprocessing, training the machine learning model and making predictions about the status of an equipment are automatically carried out in real-time. Independent modules for all of those processes can be arranged and connected on a visual environment which enables creating unique and specialized pipelines. This framework also provides a digital twin simulation of the production environment integrated with the real world and the machine learning models to evaluate the effect of different parameters such as the cost or the throughput rate. Furthermore, system modules can be controlled from a single dashboard which makes the use of the system easier even for a non-experienced user. Several open-source datasets are used to test the framework on different predictive maintenance tasks such as predicting turbofan engine degradation and predicting the stability of hydraulic systems. The effectiveness of the proposed framework is shown using metrics such as precision, recall, f1 score and accuracy.