A review on predictive maintenance technique for nuclear reactor cooling system using machine learning and augmented reality

Abstract

Reactor TRIGA PUSPATI (RTP) is the only research nuclear reactor in Malaysia. Maintenance of RTP is crucial which affects its safety and reliability. Currently, RTP maintenance strategies used corrective and preventative which involved many sensors and equipment conditions. The existing preventive maintenance method takes a longer time to complete the entire system’s maintenance inspection. This study has investigated new predictive maintenance techniques for developing RTP predictive maintenance for primary cooling systems using machine learning (ML) and augmented reality (AR). Fifty papers from recent referred publications in the nuclear areas were reviewed and compared. Detailed comparison of ML techniques, parameters involved in the coolant system and AR design techniques were done. Multiclass support vector machines (SVMs), artificial neural network (ANN), long short-term memory (LSTM), feed forward back propagation (FFBP), graph neural networks-feed forward back propagation (GNN-FFBP) and ANN were used for the machine learning techniques for the nuclear reactor. Temperature, water flow, and water pressure were crucial parameters used in monitoring a nuclear reactor. Image marker-based techniques were mainly used by smart glass view and handheld devices. A switch knob with handle switch, pipe valve and machine feature were used for object detection in AR markerless technique. This study is significant and found seven recent papers closely related to the development of predictive maintenance for a research nuclear reactor in Malaysia.