Academia.eduAcademia.edu

Figure 2 – uploaded by Sunaina Dixit

See full PDFdownloadDownload figure
Summary of reviews on application of artificial intelligence in nuclear reactor. Table 1

Table 1 Summary of reviews on application of artificial intelligence in nuclear reactor. Table 1

Related Figures (13)

Contents E-mail address: siddharthhuman@gmail.com. Nuclear industry is in crisis and innovation is the central theme of its survival in future. Artificial in- telligence has made a quantum leap in last few years. This paper comprehensively analyses recent advancement in artificial intelligence for its applications in nuclear power industry. A brief background of machine learning techniques researched and proposed in this domain is outlined. A critical assessment of various nuances of artificial intelligence for nuclear industry is provided. Lack of operational data from real power plant especially for transients and accident scenario is a major concern regarding the accuracy of intelligent systems. There is no universally agreed opinion among researchers for selecting the best artificial intelligence techniques for a specific purpose as intelligent systems developed by various re- searchers are based on different data set. Interlaboratory work frame or round-robin programme to develop the artificial intelligent tool for any specific purpose, based on the same data base, can be crucial in claiming the accuracy and thus the best technique. The black box nature of artificial techniques also poses a serious challenge for its implementation in nuclear industry, as it makes them prone to fooling. © 2020 Elsevier Ltd. All rights reserved.
Contents E-mail address: siddharthhuman@gmail.com. Nuclear industry is in crisis and innovation is the central theme of its survival in future. Artificial in- telligence has made a quantum leap in last few years. This paper comprehensively analyses recent advancement in artificial intelligence for its applications in nuclear power industry. A brief background of machine learning techniques researched and proposed in this domain is outlined. A critical assessment of various nuances of artificial intelligence for nuclear industry is provided. Lack of operational data from real power plant especially for transients and accident scenario is a major concern regarding the accuracy of intelligent systems. There is no universally agreed opinion among researchers for selecting the best artificial intelligence techniques for a specific purpose as intelligent systems developed by various re- searchers are based on different data set. Interlaboratory work frame or round-robin programme to develop the artificial intelligent tool for any specific purpose, based on the same data base, can be crucial in claiming the accuracy and thus the best technique. The black box nature of artificial techniques also poses a serious challenge for its implementation in nuclear industry, as it makes them prone to fooling. © 2020 Elsevier Ltd. All rights reserved.
Fig. 1. Typical stages of machine learning. Machine learning may be broadly classified into three main categories, as presented in Fig. 2, on the basis of methodology used for training the algorithm—supervised learning, unsupervised learning, and reinforcement learning. Supervised learning is that type of machine learning in which output value is explicitly used to train the algorithm. In other words, the algorithm is guided under the supervision of target or output. Usually supervised learning is used for classification or regression of data. Support vector ma- chine, neural network, naive Bayes are few popular examples of supervised learning. When an algorithm is used to form in- terpretations using database that consists of only input data, it falls under unsupervised learning. Clustering is the most common application of unsupervised learning method in which algorithm is used to find certain patterns or group in dataset. Reinforcement learning is a type of machine learning where the algorithm re- sponds to maximise the output within constrains of given scenario. The dissimilarity between reinforcement and supervised learning is the output signal that decides whether the step reserved by the algorithm is good or bad (Kaelbling et al., 1996).This section gives a brief overview about few popular machine learning techniques applied in nuclear energy sector.
Fig. 1. Typical stages of machine learning. Machine learning may be broadly classified into three main categories, as presented in Fig. 2, on the basis of methodology used for training the algorithm—supervised learning, unsupervised learning, and reinforcement learning. Supervised learning is that type of machine learning in which output value is explicitly used to train the algorithm. In other words, the algorithm is guided under the supervision of target or output. Usually supervised learning is used for classification or regression of data. Support vector ma- chine, neural network, naive Bayes are few popular examples of supervised learning. When an algorithm is used to form in- terpretations using database that consists of only input data, it falls under unsupervised learning. Clustering is the most common application of unsupervised learning method in which algorithm is used to find certain patterns or group in dataset. Reinforcement learning is a type of machine learning where the algorithm re- sponds to maximise the output within constrains of given scenario. The dissimilarity between reinforcement and supervised learning is the output signal that decides whether the step reserved by the algorithm is good or bad (Kaelbling et al., 1996).This section gives a brief overview about few popular machine learning techniques applied in nuclear energy sector.
Fig. 2. Classification of machine learning.
Fig. 2. Classification of machine learning.
Summary of research on application of artificial intelligence in fuel management. Table 2
Summary of research on application of artificial intelligence in fuel management. Table 2
Summary of research on application of artificial intelligence for fault diagnosis. Table 3 power plants are extremely complex system with possibility of different types of faults, a hybrid approach is better for correlating between a specific fault and its symptoms. On testing their fuzzy neural network model through simulation of different accident scenarios, improvement in fault diagnosis efficiency was verified. However, it was mentioned that it should be ultimately tested under actual operating conditions in nuclear power plants. Peng et al. (2018) applied a class of deep neural network to develop a fault diagnosis model for nuclear power plant. They trained the model using data obtained for seven different operating statuses of power plant—six different faulty conditions and the normal steady state. The developed deep neural network shown remarkable ac- curacy of above 99%, and it was found to be far superior than the performance of models developed using back propagation neural network and support vector machines in their study which have accuracy about 95% and 93%, respectively. Hatami et al., 2014, 2016 applied adaptive neuro-fuzzy technique to design a fault tolerant intelligent control system for the power plant during operational power change. They tested this fault tolerant control system under various failure scenario like loss of coolant accident, ejection of control rod etc. and found that it has satisfactory parameters tracking performance and robust stability against failures. Fantoni and Mazzola (1996) used artificial neural networks for identifica- tion of sensor failures in boiling water reactor while Ghazali and Ibrahim (2016) used neuro-fuzzy inference algorithm for online detection of sensor conditions, and the model showed good per- formances by promptly detecting the failing sensors.
Summary of research on application of artificial intelligence for fault diagnosis. Table 3 power plants are extremely complex system with possibility of different types of faults, a hybrid approach is better for correlating between a specific fault and its symptoms. On testing their fuzzy neural network model through simulation of different accident scenarios, improvement in fault diagnosis efficiency was verified. However, it was mentioned that it should be ultimately tested under actual operating conditions in nuclear power plants. Peng et al. (2018) applied a class of deep neural network to develop a fault diagnosis model for nuclear power plant. They trained the model using data obtained for seven different operating statuses of power plant—six different faulty conditions and the normal steady state. The developed deep neural network shown remarkable ac- curacy of above 99%, and it was found to be far superior than the performance of models developed using back propagation neural network and support vector machines in their study which have accuracy about 95% and 93%, respectively. Hatami et al., 2014, 2016 applied adaptive neuro-fuzzy technique to design a fault tolerant intelligent control system for the power plant during operational power change. They tested this fault tolerant control system under various failure scenario like loss of coolant accident, ejection of control rod etc. and found that it has satisfactory parameters tracking performance and robust stability against failures. Fantoni and Mazzola (1996) used artificial neural networks for identifica- tion of sensor failures in boiling water reactor while Ghazali and Ibrahim (2016) used neuro-fuzzy inference algorithm for online detection of sensor conditions, and the model showed good per- formances by promptly detecting the failing sensors.
Summary of research on application of artificial intelligence for transient identification. Table 4
Summary of research on application of artificial intelligence for transient identification. Table 4
Summary of research on application of artificial intelligence for identification of accident scenarios.
Summary of research on application of artificial intelligence for identification of accident scenarios.
Fig. 4. Sample crack detection results from the developed artificial intelligence framework based on naive Bayes and convolutional neural network. Fig. 3. Concept of the deep learning-based crack detection using naive Bayes anc convolutional neural network.
Fig. 4. Sample crack detection results from the developed artificial intelligence framework based on naive Bayes and convolutional neural network. Fig. 3. Concept of the deep learning-based crack detection using naive Bayes anc convolutional neural network.
Fig. 6. Accuracy of predicted directions for dispersion of radioactive materials during different season of a year using support vector machine algorithm. Fig. 5. Flow chart of support vector machine algorithm developed for predicting dispersion characteristics of radioactive materials.
Fig. 6. Accuracy of predicted directions for dispersion of radioactive materials during different season of a year using support vector machine algorithm. Fig. 5. Flow chart of support vector machine algorithm developed for predicting dispersion characteristics of radioactive materials.
Fig. 7. Fooling of state-of-the-art deep neural networks.
Fig. 7. Fooling of state-of-the-art deep neural networks.
Fig. 8. Deep neural network correctly labels the canonical poses of objects but fails tc recognize same image with a different pose. Each image is shown with predicted labe' and corresponding confidence level.
Fig. 8. Deep neural network correctly labels the canonical poses of objects but fails tc recognize same image with a different pose. Each image is shown with predicted labe' and corresponding confidence level.
Related topics:
Artificial Intelligence
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved