Prognosis Smart System AI-based Applied to Equipment Health Monitoring in 4.0 Industry Scenario

In the age of IIoT - Industrial Internet of Things, data lake, data mining, big data, and cloud computing, the smart manufacturing enables to make more informed decisions in real-time by using the database extracted from sensors in its equipment. During an operational campaign, the Health Monitoring System (HMS) also allows an understanding of how component degradation is affecting the performance of the equipment. Through a structure supported by AI, as data lake and cloud computing, the HMS provides to monitored equipment a fault detection system, early warning alarms to prevent failures and a calculation of the remaining useful life (RUL).The purpose of this paper is to present a prognosis smart system based on AI applied to HMS to support decision-making regarding operational performance of equipment. A Recurrent Neural Network (RNN) procedure is developed to continuously analyze the mass of monitoring data generated during the machine operation. The ability to learn the behavior patterns of the collected signals and in this way to be able to make parameter predictions with high accuracy makes artificial neural networks a powerful tool to carry out an effective prognosis. Machine operational parameters are monitored simultaneously by the prognosis smart system. Then, this information is processed by the neural network and used to characterize the machine operational condition. Upon detecting a failure trend for one or more parameters monitored by recognizing deterioration patterns, the prognosis system calculates the remaining useful life (RUL) and allows maintainers to take early actions before the failure occurrence.The proposed methodology is applied as part of a HMS of a hydro generator based on parameters registered in operator inspections routes designed to identify critical equipment degradation. The registered data representing one operational year are used to train the neural network regarding normal and abnormal machine condition. After training, the neural network is able to predict failure trends for monitored temperature parameters of the hydro-generator lubricating system that is critical to support equipment performance. Comparing prediction data and data collected by the sensors, the developed neural network reached about 0,98 RMSE score. The remaining useful life prognosis proved to be an important tool to avoid hydro generator components unexpected failures which may affect power output and cause penalties to the power generation company.