ATHLET BASED TRAINING OF NEURAL NETWORKS FOR THE ANALYSIS OF NUCLEAR POWER PLANT (NPP) SAFETY (original) (raw)
Related papers
Hybrid Identification of Nuclear Power Plant Transients With Artificial Neural Networks
IEEE Transactions on Industrial Electronics, 2004
Proper and rapid identification of malfunctions (transients) is of premier importance for the safe operation of nuclear power plants. Feedforward neural networks trained with the backpropagation (BP) algorithm are frequently applied to model simulated nuclear power plant malfunctions. The correct identification of unlabeled transients-or transients of the "don't-know" type-have proven to be especially challenging. A novel hybrid neural network methodology is presented which also correctly classifies the unlabeled transients. From this analysis the importance for properly accommodating practical aspects such as the drift of electronics elements of a simulator, the digitization of simulated and actual plant signals, and the accumulating errors during numerical integration became obvious. Beside the feedforward neural networks trained with the BP algorithm, many other types of networks and codes were used for finding the best (sensitive and robust) algorithms. Various neural network based models were successfully applied to identify labeled and unlabeled malfunctions of the Hungarian Paks nuclear power plant simulator. The BP and probabilistic methods have been proven as the most robust against the misleading recognition of unlabeled malfunctions.
A recent review of artificial intelligence applications in nuclear power plants (NPP) diagnostics and fault detection finds that mostly expert systems (ES) and artificial neural networks (ANN) techniques were researched and proposed, but the number of actual implementations in NPP diagnostics systems is very small. It lists the perceived obstacles to the ANN-based system acceptance and implementation. This paper analyses this list. Some of ANN limitations relate to "quantitative" difficulties of designing and training large-scale ANNs. The availability of an efficient large-scale ANN training algorithm may alleviate most of these concerns. Other perceived drawbacks refer to the "qualitative" aspects of ANN acceptance -how and when can we rely on the quality of the advice given by the ANN model. Several techniques are available that help to brighten the "black box" image of the ANN. Analysis of the trained ANN can identify the significant inputs. Calcula...
Nuclear Engineering and Design, 2009
The present work explores, through a comprehensive sensitivity study, a new methodology to find a suitable artificial neural network architecture which improves its performances capabilities in predicting two significant parameters in safety assessment i.e. the multiplication factor k eff and the fuel powers peaks P max of the benchmark 10 MW IAEA LEU core research reactor. The performances under consideration were the improvement of network predictions during the validation process and the speed up of computational time during the training phase.
Studies in Fuzziness and Soft Computing, 2000
In this chapter the problem of identifying events in dynamic processes (e.g. faults, anomalous behaviours, etc.) is tackled with soft computing techniques aimed at the classification of the process transients generated by such events. A review of previous wo rk is followed by a discussion of several alternative designs and models which employ both fuzzy and neural systems. These have been developed during an ongoing reserch program which was initiated by the need of finding new principled methods to perform alarm structuring/suppression in a nuclear power plant alarm system. This initial goal was soon expanded beyond alarm handling, to include diagnostic tasks in general. The application of these systems to domains other than NPPs was also taken into special consideration. A systematic study was carried out with the aim of comparing alternative neural network designs and models. Four main approaches have been investigated: radial basis function (RBF) neural networks and cascade-RBF neural networks combined with fuzzy clustering, self-organizing map neural networks, and recurrent neural networks. The main evaluation criteria adopted were: identification accuracy, reliability (i.e. correct recognition of an unknown event as such), robustness (to noise and to changing initial conditions), and real time performance. A series of initial tests on a small set of BWR transients was recently followed by more advanced tests on PWR transients corresponding to various occurrences of rapid load rejection events (plant islanding). The chapter is closed by a discussion of open issues and future directions for research and applications.
Application of Neural Networks in Reactor Diagnostics and Monitoring
Studies in Fuzziness and Soft Computing, 2000
This chapter gives an account of the use of neural network techniques in reactor diagnostics and monitoring through a few concrete examples of successful practical applications. Diagnostic problems require the solution of a so-called inverse task, namely to determine the normal or abnormal values of some system parameters ("noise sources ''), that cannot be directly measured, by observing their effect on other, measurable parameters ("reactor noise''). In the past, such inversion or urifolding techniques were possible to perform only if the direct task, i. e., calculation of the induced noise from the noise sources, could be made with a compact analytical solution. This condition hindered wide-spread practical applications. The use of artifiCial neural networks (ANN) presents a very powerful solution to the unfolding problem, since it does not require an analytical relationship between the cause and the reason. ANNs can be trained on simulated data. In the nuclear industry very powerful and accurate numerical methods exist to calculate process variables for operating plant, thus ANNs trained on simulated data can be used in real applications. This will be demonstrated through examples in this chapter.
Artificial neural network for predicting nuclear power plant dynamic behaviors
Nuclear Engineering and Technology, 2021
In order to control the plant operation under both normal and abnormal conditions, different systems in NPPs (e.g., the reactor core components, primary and secondary coolant systems) are usually monitored continuously, resulting in very large amounts of data. This situation makes it possible to integrate relevant qualitative and quantitative knowledge with artificial intelligence techniques to provide faster and more accurate behavior predictions, leading to more rapid decisions, based on actual NPP operation data. Data-driven models (DDM) rely on artificial intelligence to learn automatically based on patterns in data, and they represent alternatives to physics-based models that typically require significant computational resources and might not fully represent the actual operation conditions of an NPP. In this study, a feed-forward backpropagation artificial neural network (ANN) model was trained to simulate the interaction between the reactor core and the primary and secondary coolant systems in a pressurized water reactor. The transients used for model training included perturbations in reactivity, steam valve coefficient, reactor core inlet temperature, and steam generator inlet temperature. Uncertainties of the plant physical parameters and operating conditions were also incorporated in these transients. Eight training functions were adopted during the training stage to develop the most efficient network. The developed ANN model predictions were subsequently tested successfully considering different new transients. Overall, through prompt prediction of NPP behavior under different transients, the study aims at demonstrating the potential of artificial intelligence to empower rapid emergency response planning and risk mitigation strategies.
USING NEURAL NETWOK IN FAULT DIAGNOSIS of NUCLEAR POWER REACTORS
2000
In this paper, artificial neural networks (ANNs) are used in fault diagnosis of a nuclear power reactor (NPR). The Multilayer Neural Networks (MNNs) are used as a fault detection system, and this type of networks are learned using error back propagation training algorithm (EBPTA). The neural fault diagnosis system (NFDS) is used to identify reactor cooling pump (RCP) faults. To
Identification of nuclear power plant transients with neural networks
1997 IEEE International Conference on Systems, Man, and Cybernetics. Computational Cybernetics and Simulation
In order to help nuclear power plant operator reduce his cognitive load and increase his available time to maintain the plant operating in a safe condition, transient identification systems have been devised to help operators identify possible plant transients and take fast and right corrective actions in due time. In the design of classification systems for identification of nuclear power plants transients, several artificial intelligence techniques, involving expert systems, neuro-fuzzy and genetic algorithms have been used. In this work we explore the ability of the Particle Swarm Optimization algorithm (PSO) as a tool for optimizing a distance-based discrimination transient classification method, giving also an innovative solution for searching the best set of prototypes for identification of transients. The Particle Swarm Optimization algorithm was successfully applied to the optimization of a nuclear power plant transient identification problem. Comparing the PSO to similar methods found in literature it has shown better results.
Fault detection and analysis in nuclear research facility using artificial intelligence methods
AIP Conference Proceedings, 2016
In this article, an online detection of transducer and actuator condition is discussed. A case study is on the reading of area radiation monitor (ARM) installed at the chimney of PUSPATI TRIGA nuclear reactor building, located at Bangi, Malaysia. There are at least five categories of abnormal ARM reading that could happen during the transducer failure, namely either the reading becomes very high, or very low/ zero, or with high fluctuation and noise. Moreover, the reading may be significantly higher or significantly lower as compared to the normal reading. An artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) are good methods for modeling this plant dynamics. The failure of equipment is based on ARM reading so it is then to compare with the estimated ARM data from ANN/ ANFIS function. The failure categories in either 'yes' or 'no' state are obtained from a comparison between the actual online data and the estimated output from ANN/ ANFIS function. It is found that this system design can correctly report the condition of ARM equipment in a simulated environment and later be implemented for online monitoring. This approach can also be extended to other transducers, such as the temperature profile of reactor core and also to include other critical actuator conditions such as the valves and pumps in the reactor facility provided that the failure symptoms are clearly defined. Keywords: fault detection and analysis (FDA), transducer and actuator, artificial neural networks (ANNs), adaptive neuro-fuzzy inference system (ANFIS), area radiation monitor (ARM), online fault detection
ALADDIN: A Neural Classifier of Fast Transients for Alarm Filtering in Nuclear Power Plants
Events and faults in a NPP generate transients which can activate a large number of alarms presented in rapid sequence to the operator. A filtering of the less important alarms or a prioritization of the alarms can contribute to minimize the potential of human error in these stressful situations. In this work we have developed a neural network based system which aims at providing a fast classification of the occurring transient to an alarm handling system which can then use this information to perform event-driven alarm filtering. A systematic study was carried out with the aim of comparing alternative neural network designs and models. Four main approaches have been investigated: radial basis function (RBF) neural networks and cascade-RBF neural networks combined with fuzzy clustering, self-organizing map neural networks, and recurrent neural networks. The main evaluation criteria adopted were: identification accuracy, reliability (i.e. correct recognition of an unknown event as such), robustness (to noise and to changing initial conditions), and real time performance.