Online fault monitoring based on deep neural network & sliding window technique

Nuclear Power Plants With Artificial Intelligence in Industry 4.0 Era: Top-Level Design and Current Applications—A Systemic Review

“…Content may change prior to final publication. [160] Colored Petri Nets [193], [197], [198] Tree Methods [50], [200] Unsupervised Learning [161], [181] Fuzzy Methods [159], [182]- [189], [192], [195], [201] Other Models/Algorithms Physical Plant-centered [39], [53], [86]- [88], [123], [127]- [129], [150], [175], [193] Cyber System Layer & Physical System Layer ANN [105], [160] Fuzzy Colored Petri Nets (FCPN) [106] Reinforcement Learning (RL) [107], [108], [134]- [137] Recurrent Neural Network (RNN) [95], [97], [138], [138], [139], [146] Convolutional Neural Network (CNN) [140]- [142] Deep Belief Network [90], [121] SVM [118], [119], [122] Tree Methods [124]-…”

“…Xue et al [145] propose an analytic hierarchy process based health condition diagnosis method scored by fuzzy comprehensive methods. Instead of direct detection and diagnosis, an on-line fault monitoring system with required validity is proposed based on CNN and sliding window [146] which makes it conducive for the realization of next-generation high accuracy fault monitoring and detection system. Anyway, there is nothing ambiguous about the fact that with deeper applications of AI in FDD, not only can the stress of human operators be greatly reduced, but also can the risk that a potential accident happens to be relieved on a great scale.…”

“…An interdisciplinary team from MIT has estimated that given the expected growth of nuclear power from 2005 to 2055, at least four serious nuclear accidents would be expected in that period [146]. There are always uncertainties during the operation of NPPs such as human operation error or sudden failure of some devices [168], [169] which makes emergency alarming greatly important as emergency Alarming itself is directly related to the safety of an NPP.…”

Prediction of the Loss of Feed Water Fault Signatures Using Machine Learning Techniques

“…Examples of these studies include a study done by [4] where two deep learning networks of convolution neural network (CNN) and long short-term memory (LSTM) were used simultaneously in the diagnosis of faults. Moreover, a deep convolution neural network (DCNN) was optimized by sliding window technique to diagnose faults [5]. LSTM model was applied in NPP to predict abnormal conditions [6].…”

Metamodeling and On-Line Clustering for Loss-of-Flow Accident Precursors Identification in a Superconducting Magnet Cryogenic Cooling Circuit

“…[57] a never-ending learning method (based on dendrograms and 1-nearest-neighbor classifiers) is developed for online diagnosis of different types of faults in a gas turbine oil system operating in dynamically evolving environments. In the nuclear field (which is of particular interest to the present paper), several techniques have been employed for the early identification and diagnosis of accidents in fission systems, including: classical neural network architectures and Bayesian statistics for identifying LOCA events in a pressurized heavy water reactor [58]; deep neural networks for the fault detection and remaining useful life prediction of solenoid operated valves [59] and for online monitoring of the (modular) Integrated Pressurized Water Reactor IP-200 [60]; (Kernel) Principal Component Analysis combined with clustering for anomaly detection and isolation in an advanced heavy water reactor [61] and for spotting pipe ruptures in the cooling system of a pressurized light-water reactor [62]; particle filters embedded with neural networks to detect very small-break LOCAs in pressurized water reactors [63]; Auto-Associative Kernel Regression for early warnings about the water level of a pressurizer, on the moisture separator and reheater temperature transmitters and on environmental influences in real nuclear power plants of the Korea Hydro & Nuclear Power Co., Ltd. (KHNP) (Central Research Institute, KHNP, 70, 1312-gil, Yuseong-daero, Yuseong-gu, Daejeon 34101, Republic of Korea) [64]; Bayesian Networks for the modelbased diagnosis in a single-phase heat exchanger [65]; Support Vector Machines combined with Gaussian Process Regression for the transient analysis of seven different (normal and accidental) conditions (LOCAs, load rejection, steam generator rupture, etc.) in a simulated nuclear plant [66]; incremental learning and reconciliation of different clustering approaches by unsupervised schemes applied to a fleet of nuclear power plant turbines during shut-down transients [67].…”