Results 1 - 10 of 84
Results 1 - 10 of 84. Search took: 0.016 seconds
|Sort by: date | relevance|
[en] As a part of nuclear safety activities, developed countries have performed Periodic Safety Review (PSR) to verify and improve the safety of operating Nuclear Power Plants (NPPs). In 1999, it was decided by the Korean Atomic Energy Safety Committee to adopt the PSR program. PSR is officially legislated in 2001 as a 10-year-basis safety evaluation process. Since the first tentative application of PSR for Gori Unit 1 in 2000, it is now progressing well. Generally PSR assesses the cumulative effects of plant ageing and plant modifications, operating experience, technical developments and site aspects. The reviews include an assessment of plant design and operation against current safety standards and practices. After reviewing activities, safety is enhanced by implementing the corrective actions and/or safety improvements. When a PSR was performed in Gori Units 3-4, several safety-related heat exchangers in the Reactor Coolant System (RCS) such as a letdown heat exchanger were pointed out as the components necessitating a corrective action which is the analysis of fouling resistance. The fouling resistance is used as an important parameter to evaluate the safety as well as the economics of heat exchangers. However it is difficult to develop a credible analysis procedure due to considerable discrepancy between normal operating conditions and design conditions. This issue was identified while we were conducting a study in KNICS (Korea Nuclear I and C System) R and D program. We might be able to guess other NPPs in Korea are likely to have the same issue. This paper involves the characteristics of the safety related heat exchangers and the methodology to develop the analysis procedure
[en] The first step of designing nuclear systems starts with the identification of the top-level requirements given by stake holders and regulatory authorities. A detailed design of structure, system and component then follows. Design is divided into two processes: 'synthesis' and 'analysis.' While synthesis is the process of decision making on parameters, analysis is the process of optimizing the parameters selected. It is known from experience that the mistakes made in the synthesis process, particularly of a conceptual stage, are never completely corrected in the analysis process, which is more serious in designing complex safety critical systems such as nuclear power plants. It should be also noted that we normally believe that synthesis is only driven by engineers' heuristic knowledge. This paper proposes the applications of Axiomatic Design (AD), which is a design management tool as slightly opposed to this conventional view. I hypothesize that the design management using design axioms reduces uncertainty and subjectivity particularly at a conceptual phase so that a safer nuclear system can be developed while reducing cost in view of the system's entire life cycle. I will describe the notion of AD and introduce a few case studies
[en] Automation system can be divided into 1) Rule-based automation and 2) Knowledge-based automation. Rule-based automation can make judgement based on rule which noted at operating plane and suggest necessary information or actions to operator. In this study, focused on knowledge-based automation and perform the fundamental study to ascertain the possibility to build base of application of knowledge-based automation to NPP. To simulate application of knowledge-based cooperative automation system in abnormal condition of NPP, generated virtual abnormal condition operation log. Performed process mining using virtual data to found heuristics. Commercial process mining tool, DISCO and ProM, were used. In this study, confirm the possibility of application of knowledge-based automation to support operator in NPP using virtual operation log data. confirmed two major following subject; 1) Identifying the timing of intervention of automation system by check the time step between each event. 2) suggestion of related actions by learning from past-operation log and real-time monitoring. It can make appropriate intervention of automation when a lot of signals and alarms were occurred at same time. Also automation system can provide operator with information, which expected to be needed, and help operator to make accurate and rapid decision.
[en] Nuclear power plants also recognize the need for automation. However, it is dangerous technology to have a significant impact on human society. In addition, due to the uncertain legal responsibility for autonomous operation, the application and development speed of nuclear energy related automation technology will be significantly decrease compared to other industries. It is argued that the application of AI and automation technology to power plants should not be prematurely applied or not based on the principle of applying proven technology since nuclear power plants are the highest level security operated facilities. As described above, the overall algorithm of the Test Bed is an autonomous operation algorithm (rulebased algorithm, learning-based algorithm, semiautonomous operation algorithm) to judge the entry condition of the procedure through condition monitoring and to enter the appropriate operating procedure. In order to make a test bed, the investigation for the heuristic part of the existing procedures and the heuristic part from the circumstance which is not specified in the procedure is needed. In the learning based and semi-autonomous operation algorithms, using MARS to extract its operating data and operational logs and try out various diagnostic algorithms as described above. Through the completion of these future tasks, the test bed which can compared with actual operators will be constructed and that it will be able to check its effectiveness by improving competitively with other research teams through the characteristics of shared platform.
[en] Many empirical methods are available for freeing signals from noise and outlier such as; Discrete Wavelet transform (DTW) which includes both frequency and time domain analysis using consecutive filters and decimators, Adaptive filtering which minimize the error using iterative computation to model the relationship between two signals, methods based on least-squares polynomial approximation like the Savity-Golay filtering, and other statistical data driven models like the Moving average and Moving Median filters. Data-driven techniques can widely simulate system behavior while being implemented quickly and cheaply. These models have ability to decrease the noise dimensions and transform it into lower dimensional information to be suitable for decision making. In this paper, a new filter called Cross Moving Median (CMM) is proposed as to tackle three main problems attached with sensors data; noise, outliers, and missing data. This filter is based on the running median that was improved to account for prior estimate as to update the moving median result. The CMM filter was validated using HANARO Cold Neutron Source data that was generated from a hydrogen transmitter. As many models, which are developed for nuclear reactor diagnostics and prognostics, are based on data generated from nuclear reactors computer, the need for high quality data is very important for developing reliable diagnostics and prognostics methods. In this paper, we proposed CMM filter that can deal with noise, outliers, and missing data providing reliable estimate of sensors' readings. The so called Cross Moving Median (CMM) with an optimal sliding window has the capability to attenuate noise, prune outliers, and reconstruct missing data.
[en] One of the applications using PSA is a risk monito. The risk monitoring is real-time analysis tool to decide real-time risk based on real state of components and systems. In order to utilize more effective, the methodologies that manipulate the data from Prognostics was suggested. Generally, Prognostic comprehensively includes not only prognostic but also monitoring and diagnostic. The prognostic method must need condition monitoring. In case of applying PHM to a PSA model, the latest condition of NPPs can be identified more clearly. For reducing the conservatism and uncertainties, we suggested the concept that updates the initiating event frequency in a PSA model by using Bayesian approach which is one of the prognostics techniques before. From previous research, the possibility that PSA is updated by using data more correctly was found. In reliability theory, the Bathtub curve divides three parts (infant failure, constant and random failure, wareout failure). In this paper, in order to investigate the applicability of prognostic methods in updating quantitative data in a PSA model, the OLM acceptance criteria from NUREG, the concept of how to using prognostic in PSA, and the enabling prognostic techniques are suggested. The prognostic has the motivation that improved the predictive capabilities using existing monitoring systems, data, and information will enable more accurate equipment risk assessment for improved decision-making
[en] It is a representative example when main steam mass flow of a steam generator is measured lower than main feedwater mass flow of outlet, or when efficiency of low pressure turbine is analyzed excessively low or exceeds 100 percent. Therefore, we need to obtain measured data minimizing uncertainty to calculate thermal efficiency as exactly as possible. In calculating the efficiency of an Nuclear Power Plant(NPP), measurement uncertainty is the most difficult to be solved technically and data reconciliation methodology is one method of ensuring to minimize uncertainty. In this paper, the case study on previous nuclear power plants was carried out by using redundancy of measured data from measuring instrument for plant operation, so as to calculate nuclear power plant efficiency accurately. As explain above, we performed the case study on data reconciliation methodology by using measurement redundancy and physical redundancy. The former comes up because of installing multiple measuring instruments for plant operation, and the latter is acquired based on the physical association like the first law of thermodynamics (the law of conservation of mass and energy). Through this case study, we got the reconciled data, which satisfies the constraint and minimizes data uncertainty measured in the nuclear power plant secondary system at the same time. The expected effects from data reconciliation methodology provided in VDI-2048, are considered totally four. First, this method can contribute to monitoring on-line efficiency in the operating nuclear power plant. Second, it can also improve the reliability of calculated results, minimizing the measurement uncertainty
[en] The demand for robust and resilient performance has led to the use of online-monitoring techniques to monitor the process parameters and signal validation. On-line monitoring and signal validation techniques are the two important terminologies in process and equipment monitoring. These techniques are automated methods of monitoring instrument performance while the plant is operating. To implementing these techniques, several empirical models are used. One of these models is nonparametric regression model, otherwise known as kernel regression (KR). Unlike parametric models, KR is an algorithmic estimation procedure which assumes no significant parameters, and it needs no training process after its development when new observations are prepared; which is good for a system characteristic of changing due to ageing phenomenon. Although KR is used and performed excellently when applied to steady state or normal operating data, it has limitation in time-varying data that has several repetition of the same signal, especially if those signals are used to infer the other signals. The convectional KR has limitation in correctly estimating the dependent variable when time-varying data with repeated values are used to estimate the dependent variable especially in signal validation and monitoring. Therefore, we presented here in this work a modified KR that can resolve this issue which can also be feasible in time domain. Data are first transformed prior to the Euclidian distance evaluation considering their slopes/changes with respect to time. The performance of the developed model is evaluated and compared with that of conventional KR using both the lab experimental data and the real time data from CNS provided by KAERI. The result shows that the proposed developed model, having demonstrated high performance accuracy than that of conventional KR, is capable of resolving the identified limitation with convectional KR. We also discovered that there is still need to further improve our model to make it more generalized as well for more robustness than the current performance
[en] The conventional way for tracking an accident during emergency operation is to manually track important safety related parameters whether they are approaching the unacceptable limits that are provided by vendors/designers in the technical specifications of the plant such as; observing the pressure, temperature, SG water level and so on. As the time of emergency in NPP, a fast estimation and decision of which procedure will be followed to assist and mitigate the accident; a computational aid system would be helpful for operators to identify the initiation time of accident and to help them identifying the type of accidents using classification or clustering methods. This computational engine is based on the RSM specifically the 1st RSM that can identify the increasing and decreasing pattern with their orientation angle and can capture the initiation time of accident.
[en] Being a safety-critical system, safety is extremely important for any nuclear power plant (NPP). Therefore, to maintain the safety of NPPs at an acceptable level, preventive measures are necessary to deal with potential issues. During plant operation, faults and failures can occur in sensors, equipment, and processes which can have impact on the performance of the plant. These faults are more prominent in aged NPPs because of their vulnerability to aging-related faults. Hence there is need to monitor the status of the plant during operation. Several data-driven methods has been developed and applied to detect faults and monitor the NPPs sensors and equipment. However, the applicability of deep learning, which is the current trend in the field of machine learning, has not been explored. In this work, we proposed, showed, demonstrated, and verified the deep learning architecture for fault detection and diagnosis of NPPs. The selected architecture in this work is deep belief network which is based on the restricted Boltzmann machine. To verify the proposed model, NPP simulation data is collected and used to train the model. Several plant accident simulations with normal operation are performed, and data is collected for each plant conditions. The proposed model gave the overall performance detection and classification accuracy of 98.9%, with error rate of 1.1% which is enough to monitor the status of the plant condition.