Real-time HTTR condition monitoring with neural networks

Nabeshima, Kunihiko  ; Nakagawa, Shigeaki  ; Makino, Jun*; Kudo, Kazuhiko*

Two types of neural networks have been utilized for real-time condition monitoring of High Temperature Engineering Tested Reactor (HTTR) in JAEA, Japan. Multi-Layer Perceptron (MLP) in auto-associative mode could model the whole plant dynamics and detect many kind of abnormal conditions. Another neural network with feedback connection can estimate the occurrence time and amount of helium leakage after auto-associative MLP detects the anomaly.