Integration of artificial intelligence with automation for predictive maintenance in sustainable hydrogen production plants

Abstract

Automation in modern industries is possible with the aid of sensors that measure signals needed for control, fault detection and decision making about a process. Such decisions include the time to perform predictive maintenance which is not possible when there is a failure in one of the sensors. Artificial intelligence techniques can be used to detect faults in a sensor and predict what its correct reading should be using signals from other sensors involved in the process. For accurate prediction, a signal from an alternative sensor, or a combination of signals from different sensors, should be selected that has a strong correlation with the signal to be predicted. In this study, to demonstrate the application of artificial intelligence in automation, an electrolyser operating in an automated process has been considered. A Deep Reinforcement Learning (DRL) algorithm was developed to select the best signal among others with the highest correlation coefficient of 0.99. The selected signal was then used in a long short-term memory (LSTM) to predict faulty temperature signals in the electrolyser. The root-mean-square error (RMSE) of the predicted signal was 0.1351.