Annals of Nuclear Energy 38 (2011) 1418–1426
Contents lists available at ScienceDirect
Annals of Nuclear Energy journal homepage: www.elsevier.com/locate/anucene
An efficient Neuro-Fuzzy approach to nuclear power plant transient identification Rafael Gomes da Costa a, Antônio Carlos de Abreu Mol a,b, Paulo Victor R. de Carvalho a, Celso Marcelo Franklin Lapa a,b,⇑ a
Instituto de Engenharia Nuclear – CNEN, Programa de Pós-Graduação em Ciência e Tecnologia Nucleares, Via Cinco, s/no, Cidade Universitária, Rua Hélio de Almeida, 75, Postal Box 68550, Zip Code 21941-906 Rio de Janeiro, Brazil b Instituto Nacional de C&T de Reatores Nucleares Inovadores, Brazil
a r t i c l e
i n f o
Article history: Received 27 October 2010 Accepted 14 January 2011
Keywords: NPP transient identification Neuro-Fuzzy system Human cognition
a b s t r a c t Transient identification in nuclear power plants (NPP) is often a computational very hard task and may involve a great amount of human cognition. The early identification of unexpected departures from steady state behavior is an essential step for the operation, control and accident management in NPPs. The bases for the transient identification relay on the evidence that different system faults and anomalies lead to different pattern evolution in the involved process variables. During an abnormal event, the operator must monitor a great amount of information from the instruments that represents a specific type of event. Recently, several works have been developed for transient identification. These works frequently present a non reliable response, using the ‘‘don´t know’’ as the system output. In this work, we investigate the possibility of using a Neuro-Fuzzy modeling tool for efficient transient identification, aiming to helping the operator crew to take decisions relative to the procedure to be followed in situations of accidents/ transients at NPPs. The proposed system uses artificial neural networks (ANN) as first level transient diagnostic. After the ANN has done the preliminary transient type identification, a fuzzy-logic system analyzes the results emitting reliability degree of it. A validation of this identification system was made at the three loops Pressurized Water Reactor (PWR) simulator of the Human-System Interface Laboratory (LABIHS) of the Nuclear Engineering Institute (IEN/CNEN/Brazil). The obtained results show the potential of this new transient identification system to be used in an operational NPP in order to assist the operators to take decisions during transients/accidents. Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction Between a third and a quarter of human population does not have access to electricity, relying on burning wood or whatever materials found available for burning, as its only energy source. Conservatives previsions indicate that in next 50 years the world electric energy demand will increase around of 80%. Considering the actual alternatives energy generation, the emission of carbon dioxide to the atmosphere will increase substantially. This situation is a contradiction with international initiatives, such as the Kyoto treaty, attempt to reduce carbon dioxide emissions aiming to minimize the possibilities of climatic changes from a global warming. In view of the considered scenario, the nuclear energy will have an important role, either in the generation of electricity in next half
⇑ Corresponding author at: Instituto de Engenharia Nuclear – CNEN, Programa de Pós-Graduação em Ciência e Tecnologia Nucleares, Via Cinco, s/no, Cidade Universitária, Rua Hélio de Almeida, 75, Postal Box 68550, Zip Code 21941-906 Rio de Janeiro, Brazil. Fax: +55 21 21733909. E-mail addresses:
[email protected] (Antônio Carlos de Abreu Mol), paulov@ ien.gov.br (Paulo Victor R. de Carvalho),
[email protected] (C.M.F. Lapa). 0306-4549/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.anucene.2011.01.027
century. In fact, in last years, a considerable amount of the new NPPs have been build around of the world, especially in third world countries. Despite the nuclear electricity generation to be emission-free and does not contribute for global warming, the plant safety improvement remain been the main objective of this new NPPs projects. In order to attempt this fundamental challenger of the nuclear engineering: the improvement of the safety and reliability of the new NPPs, researchers have focused its studies in two basic points: ‘‘Passive Systems and Safety By Design’’ (Guimarães and Lapa, 2004, 2006, Guimaraes et al., 2006; Cunha et al., 2007; Lapa et al., 2004) or ‘‘Transient Diagnostic to Operation Support’’ (Alvarenga et al., 1997; Mol et al., 2003; Pereira et al., 1998). This article presents a methodological contribution to the state of the art to transient diagnostic in NPPs. The difficult with the diagnostic of events in nuclear power plants, and especially the need to deal with hundreds of variables simultaneously in case of accidents, has motivated the development of many diagnostic systems based on artificial intelligence (AI) to support operators work. Several researchers have used artificial neural networks (ANN), fuzzy logic (FL) and genetic
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426
algorithms (GA) to solve problems related to monitoring a nuclear power plant, especially regarding the problem of events identification. Alvarenga et al. (1997) used artificial neural networks, fuzzy logic and genetic algorithms for the diagnostic of postulated accidents of the ANGRA II nuclear power plant, using a set of 17 variables of the plant. This work also did not consider the treatment of unknown events. More recently, the system developed by Mol et al. (2003) explored the good performance in multi-ANNs with backpropagation training algorithm for the event identification, even when they added noise in the input data. The system also presented a procedure for validation of the diagnosis in order to obtain an output ‘‘I do not know’’ to the events outside the scope of ANNs’ training. This system characteristic has been considered at now an important contribution to control room team in a incident occurrence. However, cognitive studies reveled recently that the ‘‘do not know’’ may be inadequate to plant operation in incident situations (Carvalho et al., 2007, 2006). In this context, the main objective of this work is to develop an operator support system for the diagnostic of accidents in nuclear power plants, using artificial neural networks together with fuzzy logic to indicate the possibility of events occurrence, with their degrees of confidence. Using the proposed system, an operator can direct his/her attention and anticipate actions to deal with the situation. The system developed recognizes that operators deal with uncertain situations and make decisions without a complete set of information about the state of the systems (Carvalho et al., 2007). The support system aims to help one of the main objectives of the operation team: to update and validate individual and collective situation awareness that allows a resilient and safety operation (Vidal et al., 2009). The support system uses a set of RNAs to identify the events, and a fuzzy logic structure to provide the confidence level of each identification made by the RNAs. The system aims to direct the operator(s) attention, indicating the type of accident that may be starting out in the plant, together with the probability of the indication correctness, allowing the operator to plan their actions in the near future, seeking the information and the support necessary to deal with the accident, before that the traditional indications, coming from the conventional system alarm, occur.
2. Methodology 2.1. A new paradigm for failure management in complex systems The main problem of the current work system in nuclear power plants is the assumption that there is always a correct way to do the job (task management), because this approach limits the possibilities of operators to cope with system complexity due: The task management approach, based on instructions that are supposed to be strictly followed, has limitations due to the difficult of the system designers to provide the entire set of actions that are needed in new situations and because of constraints imposed by the environmental variability. The limitation of the question 1 means that the operators of nuclear plants have difficult to determine, where the procedures, they are supposed to follow, are not best suited to dealing with a new situation (Carvalho et al., 2007). Operating teams do not have the necessary help to determine whether (and when) procedures could be modified, briefly and without jeopardize safety. In uncertain, unfamiliar situations, in which operators do not have the support of written procedures and completely correct (unambiguous) information such as FAIL or NOT FAIL,
1419
RIGHT or WRONG, operators have faced situations in which they have no support at all, and have to improvise, searching for ad hoc re-configurations in the plant systems (Carvalho et al., 2007). These factors justify the use of support systems that provide partial information about situations that may be evolving in the plant, to support the cognitive strategies used by humans. Several factors contributing to the difficulties in actual fault management systems based on information provided by alarm systems currently used in nuclear power plants had already been identified: meaningless alarms, unclear or underspecified alarm messages, alarm inflation, alarms indicating the state of the system rather than abnormalities are only a small part of these difficulties (Woods, 1995). The temporal dynamics is also relevant. Because the close correlation between the variables of NPPs processes, the period when a lot of alarms occur is during the beginning of the accident. It is precisely during this period of high workload that technological artifacts should provide the necessary assistance to the operators assess the situation. However, it is in this period that occurs most of meaningless alarms, coming from systems that are not important to solve the problem. Therefore, the alarm system and inadequate diagnostic system distract operators and disrupt their activities, making diagnosis more difficult and hindering the activities of information prioritization. These factors constitute the so-called alarm systems problems (Woods, 1995). To help accident diagnosis in this complex operational situation, the system, developed in this work, assumes the role of an agent trying to anticipate problems in order to direct the attention of the human observer for potentially more interesting events that are occurring, such as, a situation with a huge number of data, limited time to make decisions and multiple action selection possibilities. The direction of attention, as early as possible, to the type of event that is occurring in the plant is paramount for the operator plan their actions in the near future, reaching a proper situation awareness (Vidal et al., 2009). The control of attention allows an anticipation of action plans in a cognitive system, and occurs when attention driven signals provide important information for action selection, or when the attention driven signals show information that can be ignored, or may be delayed, safely, in accordance with the situation. In general, an attention driven signal says ‘‘there is something I believe you will find interesting or important, so you should check it out’’. The criteria for an effective attention direction requires that the attention driven signals can be received by the operator in parallel with their line of reasoning and activities in progress, including partial information about the problem solution. The goal is to allow the operator to decide when the interrupt signal endorses an authorization for a change (or not) in the attention focus. The concepts presented above were used to develop the system to help nuclear power plant in the diagnosis of accidents described here. The system indicates, in advance, the type of accident that might be occurring in the plant, as well as the confidence level of this statement without do not know response, allowing the operator to plan his/her future, seeking the information and support needed to deal with the event, before indications coming from traditional alarm system occur.
2.2. The events diagnosis description The diagnostic system consists of an identification system of events, based on ANNs and one nebulous system, that informs confidence degree of the event identification.
1420
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426
2.2.1. Events identification using ANNs For the events identification, we use the multi-type ANN jump model. Fig. 1 shows the network with an input layer, two intermediate layers and an output layer. For this type of ANN, a neuron, in any layer of the network, is connected to all other neurons in later layers (no feedback). The signal flowed through this network is propagated ‘‘forward’’, from left to right, going through all the layers. Fig. 2 shows a part of the jump, where two types of signals are identified: (1) the ‘‘sign stimulus’’, which is the input signal, coming from the first network layer, spreading ‘‘forward (neuron per neuron) through the network, emerging at the output of the network as output signal and (2) ‘‘error signal’’, indicating that the error is originating from one neuron from the network and propagated back (neuron by neuron) through the network. The ‘‘backpropagation’’ method, in this ANN type, results as a general update of the connections (Dwji(n)) using the following equations:
Dwji ðnÞ ¼ g dj ðnÞ yi ðnÞ; dFðaj ðnÞÞ X dk ðnÞ wkj ðnÞ; dj ðnÞ ¼ daj ðnÞ
ð1Þ
modules provide output A equals zero and the output B equals 1 (indicating that the event in progress is not their responsibility). Thus, each module IIM is capable of identifying up to four different events. The final identification structure/events selection is composed of many modules IIM. They are needed to cover all the events you want to identify. Fig. 5, below, shows the final structure for the identification stage.
ð2Þ
where k represents any neuron of the subsequent layers to layer neuron j, d(n) is the local gradient, yi is the output of neuron i F(.) is the activation function, aj(n) is the activation of neuron j and wki the synapse between the neuron k and i. 2.2.2. Identification system events To identify what type of event is occurring in the plant, it was used a modular structure consisting of several types of ANNs jump, which form the different identification modules, called Identification Independent Module (IIM). Each of these is composed of four Basic Neural Modules (BNM) based on an ANN Jump. Each BNM is responsible to identify a specific event among any others. To this end, each BNM has process variables as input and only two exits, one to represent the event by which the module is responsible (class A) and the other to represent all other events (class B). Thus, it was necessary that the ANN of each BNM were trained with two sets of standards, a set representing the class A and another set representing the class B. To make the selection, was assigned a coding of 1 for output (the class that represents the input pattern in progress) and zero encoding to another. Fig. 3 shows a selection of BNM. Each Independent Identity Module is composed of a stack up to four BNMs, Fig. 4, where each of these modules is responsible for identifying the event for which it was trained. During the system operation, it is expected that, for a given event X, the BNM responsible for the event X presents the output A equals 1 (indicating that the event X is running) and the output B equals zero, and all other
Fig. 1. Neural network jump.
Fig. 2. Illustration of the two directions flow signal.
2.2.3. Events Identification in Noisy Environments and Dynamic In order to increase the network generalization and robustness and to noise simulating the uncertainties of the operation real environment, it is necessary that a large pattern number is presented to the network during the training phase. This is done by adding noise patterns with overlapping and forcing through the training, the network recognizes these new patterns as belonging to their original classes (no noise). The figures presented in the outputs of ANNs in response to input patterns that represent the same event with noise, floating around the discrete value expected. Thus, it is necessary to associate the continuous values presented at the output of the network to discrete values that represent each of the events. For this purpose, it determines each event deviation (Dev). This deviation is defined as the difference between the expected value and the value displayed in the network output to the respective event. ^
Dev ¼ jyev yev j;
ð3Þ
where ev indicates the event, yev discrete value is expected and ev is the value obtained by continuous ANN. This deviation will be used later on, to get the confidence level of each module. To treat dynamical systems (which vary in time), we adopted the first-time mobile window ‘‘that is, joined to the external network architecture, the time dependence by sequential presentation in the recent time history of variables state, used in the events identification. Fig. 6 shows the block diagram of the identification system, after the temporal window mobile inclusion. In this diagram, the values recent history, of each variable, is displayed at the neural network entrance. With each new acquisition, the windows oldest sample is discarded, the elements are displaced and the new value is entered. The confidence level is determined using the difference Dev. The fuzzy structure of each module takes as input Dev deviations set up for each event. 2.2.4. Cloudy system for determining the confidence degree in event The elements that make up the lifecycle of a case foggy, or fuzzy logic module are: (1) fuzzification, which converts the input variables (crisp or exact measures) on fuzzy sets to represent uncertainty, (2) Base of the Rules directs the knowledge system through the rules governing the variables relations, and (3) inference, that is, the inference mechanism evaluates the control relevance, at a given time, and decides which exit should the process takes (4) aggregation which are techniques used to obtain a output fuzzy set from one set of rules and the inference 5) defuzzification, that converts the decision taken by the inference engine into a
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426
1421
Fig. 3. Basic neural module with two outputs: class A and class B.
Fig. 4. Stack neural core module with outputs A and B.
crisp value (numerical value), transforming qualitative into quantitative information, through a process specification. In the system developed to provide the degree of ANNs diagnostic confidence that identify events, we used a system based on fuzzy logic, as described in Fig. 7. For each BNM II is a basic module cloudy. To determine the event confidence level presented by the BNM, the module compares the output fuzzy submitted by the BNM with their respective outputs presented by the other BNM that make the same IIM. In the system developed to provide the degree of ANNs diagnostic confidence that identify events, we used a system based on fuzzy logic, as described in Fig. 7. For each BNM, IIM is a basic cloudy module. To determine the event confidence level presented by the BNM, the module compares the output fuzzy submitted by the BNM with their respective outputs presented by the other BNM that make the same IIM. To make the comparison between the outputs of the MNB, we define the variable input linguistic variable Module_Event, to test the relevance degree of the current event in the BMN, and linguistic variables Event_1, Event_2 and Event_3 to test the relevance of the same event in other BMN same IIM. As output variable, it was defined linguistic variable Confidence_Degree. Then, to illustrate the variable definition process, we present the variable definition event module and confidence degree. Module_Event – this variable checks the current event relevance in five fuzzy sets: Very_Module_Event (VME), Module_Event (ME), Medium_Module_Event (MME), Weak_Module_Event (WME) and Not_Module_Event (NME). To this end, it was used as input the calculated deviation to the responsibility of the event module, Dev (Eq. (3)). Fig. 8 shows the fuzzy sets for the linguistic variable Module_Event. The other variables nebulae of type Event are defined similarly. Confidence degree – this variable determines the current event degree of confidence in five fuzzy sets: not confident (NC), little confident (lC), medium confidence (MC), confident (C) and very
confident (VC). Fig. 9 shows the fuzzy sets for the linguistic variable confidence degree. The fuzzy logic rules for determining the identification confidence degree are empirical, based on logic. Table 1 illustrates the formation of these rules.
3. Realistic study of case We use some of the accidents postulated to project a conventional PWR nuclear power plant (typical second NPP generation) such as: Main Feed Water (MFW) – A break from the power line cut off the main water supply to the steam generators. Loss Of Coolant Accident (LOCA) – A break in the pipes that cool the reactor in the primary or secondary. Steam Generators Tube Rupture (SGTR) – Occurrence of leaks in pipes in U of steam generators, there will be a transfer of radioactive coolant circuit for the water–vapor due to the high pressure difference between the primary side and secondary side. Turbine Trip (TRIPTUR, automatic turbine shutdown) – The largest load rejection of a nuclear power plant is the result of the turbine shutdown, hindering the withdrawal of the steam generated by the ballast. In the case of unavailable alternative lines for the steam, the vapor pressure will increase rapidly, with consequent increase in temperature and pressure of the primary side. To test the system, these accidents were generated by three loops PWR simulator at the IEN Laboratory of Human Interface System simulator (LABIHS), for power operation of 100%. The total simulation time was 120 s (where the first 10 s represent the condition of normality). The variables chosen were those that most contribute to the accident characterization, in question such as:
1422
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426
Fig. 5. Sets of modules MII, MII each adds 4 MNB.
Flow in the core temperature in hot leg, cold leg temperature, level in the steam generator – broadband, level in the steam generator – close range, pressure in the steam generator, flow of feed water flow rate of steam pressure in the pressurizer – range narrow temperature margin of the coolant, the pressurized level. For the diagnostic system, it is required three stages: training the neural network; fuzzy logic development and operation. The main objective of the training phase is the ANNs synapses adjustment. In the operation phase, the system, already trained, is used to make sample classifications that would be provided by the real plant instrumentation or a simulator at intervals of 1 s. Table 2 presents the operating conditions used to test the system.
6848 standards were used in total. To set the window size and all ANNs parameters, it took several tests, since there is a general criterion, well-defined choice for these parameters. After these attempts, the ANNs that provided the best results showed the following configuration: input layer consisting of 60 neurons with linear activation functions (12 ‘‘time windows mobile’’ with five members each); 1 tier with 103 neurons with logistic activation functions; output layer with 2 neurons with activation function type logistics.
3.2. Fuzzy logic setting 3.1. ANNs training and system parameters determination The ANNs trained how to identify the MFW accident, LOCA, SGTR, with the plant operating at 100% power. Accident TRIPTUR was chosen as unknown event and thus not part of the ANNs training set. To form the ANNs training set, 428 standards were set without noise, the noise standards in 2140 to 1%, 5% and 10% each.
For each accident trained (MFW, SGTR and LOCA), and the situation Normal, it was created a fuzzy-logic system, amounting to a total of four fuzzy sets. Thus, the variables used in fuzzy logic module: LOCA, MFW, NORMAL and SGTR (TRIPTUR for the accident that was not part of the ANN training is not designed fuzzy logic module).
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426
1423
Fig. 6. Identification dynamic system scheme.
Fig. 7. Presents MII, each color represents the training specialist module with their respective confidence degrees.
The membership function was chosen according to the authors’ experience in accidents analysis in nuclear reactors. It was chosen the triangular and trapezoidal curves. There were partitioned areas of input and output fuzzy interference system, so that the partitions number must be equal to the number of linguistic terms used to assess the accident reliability level. After implementing the domain partition, membership functions and linguistic terms of input and output variables, the next step – to complete the fuzzy modeling – was to create a set of rules
between the dependent and independent variables, based on empirical analysis. Through the experience of a specialist operator, it has resulted in 40 rules of the type If ... Then, which were incorporated into the rules bank. 3.3. Operation and test of the system There were generated for the accidents group, included in the training phase (LOCA, SGTR and MFW) and provided (NORMAL)
1424
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426 Table 2 Operating conditions at the plant.
Fig. 8. Fuzzy sets for variable module event.
Fig. 9. Fuzzy set for variable confidence degree.
power plant 535 new standards representing the events (adding, for each accident, new noises to the state variables). In addition to 535 new patterns representing the TRIPTUR accident, which was not included in the training phase. In each incident, we calculated the maximum deviation (Dev) for each of the 535 standards, the value was incorporated into the fuzzy system to provide the confidence degree of the accidents identity. The Dev value was achieved by performing a values calculation provided by the basic neural module (BNM), after testing the accident standards set. It was subtracted from, the expected value output BNM – which is always (1) – the output value obtained by the basic neural module. 4. Results and discussion The system was tested with different noise levels, 1%, 15% and 20%. Table 3 presents the results obtained in the normal situation and the identification of LOCA accidents, MFW and STGR and their confidence degree, with noise level of 15%. Table 4 presents the result for the TRIPTUR accident, unknown to the IIM, it was not part of the training set of ANNs. To evaluate the response time of the system we tested the transition between the normal and the accident conditions. In these
Table 1 Rules formation. IF Event is AND Is not strong event 1 IF Event is AND Is weak event 1 IF Is not AND Is not event event IF Event
AND Is not event 1 IF Event is AND Is not weak event 1
AND Is not event2 AND Is not event 2 AND Is not event 2 AND Is not event 2 AND Is event 2
AND Is not event 3 AND Is not event 3 AND Is not event 3 AND Is event 3 AND Is not event 3
THEN Very confidant THEN Medium confident THEN Confident
THEN Little confident THEN No confident
Nomenclature
Operation Condition
NORMAL LOCA SGTR MFW TRIPTUR
Condition normal power Loss of coolant of the primary Break tube steam generator Break the main power line Turbine shutdown
tests, the first five standards represent the normal condition and the last 530 patterns represent the accident evolution condition. Tables 5–8 present the transition results from NORMAL to LOCA, MFW, SGTR and TRIPTUR respectively. The system output, presented in Table 3, shows the correct event identification, even in situations of high uncertainty at the input (simulated by 15% noise level added in the input). In all situations, the correct accident was identified correctly after 15 s (more than 50% of certainty). Based on that information, an operator may quickly direct (15 s after the start of the event) his/her attention to the most likely event, even in a noisy background, reducing his/her search field. Therefore, an operator has more time to test and validate his/her action options, resulting in a faster and more effective way to cope with the events. Table 4 shows that the independent identification module was able to identify the TRIPTUR accident, as an unidentified accident (this accident was not trained), indicating that the system presents an adequate response for an event that did not belong to the training scope of neural modules. The results presented in Tables 5–7 show that the accident identification module was able to successfully handle the transition from the normal condition for the accident condition. The system identifies the LOCA accident in 2 s and the other accidents in 6 s. This rapid accident identification, directing operators’ attention to the most probable situations, can improve the operators’ dynamic failure management during accidents, because during such moments (beginning of accidents) dozens of alarms occur simultaneously competing for the operators’ attention. An early and reliable indication about what is happen in the plant, provided by the diagnostic system, will facilitate operators’ choices and direct their course of actions during these important moments of the plant operation. Table 3 Identification of NORMAL situation and the accidents LOCA, MFW and STGR. Largest deviation (Dev)
Fuzzy reliability
NORMAL condition with noise of 15% at time t = 73 s LOCA network 1.132 MFW network 1.213 NORMAL network 0.094 SGTR network 1.214
0% LOCA accident 0% MFW accident 96.5% NORMAL condition 6.32% SGTR accident
LOCA accident with noise of 15% at time t = 15 s LOCA network 0.162 MFW network 0.804 NORMAL network 0.906 SGTR network 0.842
72.5% LOCA accident 3% MFW accident 4% NORMAL condition 7.57% SGTR accident
MFW accident with noise of 15% at time t = 34 s LOCA network 1.249 MFW network 0.488 NORMAL network 1.052 SGTR network 1.065
4% LOCA accident 52.5% MFW accident 5% NORMAL condition 6.9% SGTR accident
SGTR accident with noise of 15% at time t = 13 s LOCA network 1.128 MFW network 1.200 NORMAL network 1.060 SGTR network 0.429
0% LOCA accident 0% MFW accident 96.5% NORMAL condition 6.32% SGTR accident
1425
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426 Table 4 Accident unknown MII (TRIPTUR) at time 98 s.
Table 6 Transition from NORMAL to MFW in the moments of time 2 s, 4 s and 6 s.
Largest deviation (Dev)
Fuzzy reliability
TRIPITUR accident with noise of 10% at time t = 98 s LOCA network 1.302 MFW network 1.247 NORMAL network 1.463 SGTR network 1.344
0% 0% 0% 0%
LOCA accident MFW accident NORMAL condition SGTR accident
Table 5 Transition from NORMAL to LOCA in moments of time 1–3 s. Transition from normal condition to the accident LOCA with noise of 1%
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
Dev at time 1 s
Fuzzy reliability
0.998 0.991 0.026 0.991
0% LOCA accident 0% MFW accident 100% NORMAL condition 6.32% SGTR accident
Dev at time 2 s
Fuzzy reliability
0.361 1.291 1.119 1.720
72.5% LOCA accident 2.5% MFW accident 2% NORMAL condition 6.85% SGTR accident
Dev at time 3 s
Fuzzy reliability
0.118 1.344 1.136 1.645
94.5% LOCA accident 0% MFW accident 0% NORMAL condition 6.32% SGTR accident
Table 8 shows the transition from normal condition to the TRIPTUR accident that was not part of the neural modules training scope. The system, in the first few seconds, tries to classify this event as an accident pertaining to the identification module training set. After 6 s the system shows that this possibility is close to zero. Even in a situation of unknown event, the system indicate signs of possible abnormalities in the plant (4.5% LOCA, 6.9% SGTR, different from the 0% indication expected in the normal condition). These weak abnormalities indications would be enough to alert operators that something unusual (and different from the accidents that the system can identify) is already be happening in the plant. An important feature of the system is its independence from a sign that indicates the start of the accident such as the REACTOR TRIP event in most of event identification systems. The process to identify accidents independent from initiator signals improves the response time of the system and has been achieved due to the robustness of the system in relation to noise; allowing the system to distinguish between a noisy condition of normality and a condition outside the normal operation range, and due the use of the time window, making the system able to identify the event considering its dynamic nature. 5. Conclusions We present an operator support system which aims to direct the attention of the operators during the diagnosis of accidents in nuclear power plants using techniques and concepts of Artificial Intelligence, particularly artificial neural networks and fuzzy logic. The system objective is to help the operator during the assessment of accidents, indicating in advance and reliably way, what type of accident may be occurring in the plant, and allowing the operators to direct their attention, narrowing the information search field in the noisy background of the operation during accident situations in nuclear power plants. Focusing their attention in the most likely event contributes to reducing the cognitive overload of the opera-
Transition from normal condition to the accident MFW with 1% noise
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
Dev at time 2 s
Fuzzy reliability
0.771 0.876 0.373 1.702
2% LOCA accident 7% MFW accident 51% NORMAL condition 6.7% SGTR accident
Dev at time 4 s
Fuzzy reliability
0.735 0.658 0.895 1.816
24% LOCA accident 25.5% MFW accident 5.5% NORMAL condition 6.32% SGTR accident
Dev at time 6 s
Fuzzy reliability
0.973 0.075 1.020 0.987
0% LOCA accident 99.5% MFW accident 0% NORMAL condition 6.32% SGTR accident
Table 7 Transition to NORMAL SGTR moments in time 2 s, 4 s and 6 s. Transition from NORMAL condition to the accident SGTR with 1% noise
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
Dev at time 2 s
Fuzzy reliability
0.984 0.980 0.081 0.927
0% LOCA accident 0% MFW accident 98% NORMAL condition 6.48% SGTR accident
Dev at time 4 s
Fuzzy reliability
0.927 1.034 0.696 0.337
1% LOCA accident 1.5% MFW accident 5% NORMAL condition 65.4% SGTR accident
Dev at time 6 s
Fuzzy reliability
0.998 0.965 0.963 0.045
0% LOCA accident 0% MFW accident 0% NORMAL condition 93.4% SGTR accident
Table 8 Transition from NORMAL to TRIPTUR the moments of time 2 s, 4 s and 6 s. Transition from NORMAL condition to the accident TRIPTUR with 1% noise
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
LOCA network MFW network NORMAL network SGTR network
Dev at time 2 s
Fuzzy reliability
0.556 1.175 0.629 1.960
47.5% LOCA accident 3.5% MFW accident 51% NORMAL condition 6.68% SGTR accident
Dev at time 4 s
Fuzzy reliability
0.336 1.053 1.004 1.295
72% LOCA accident 1.5% MFW accident 1.5% NORMAL condition 6.6% SGTR accident
Dev at time 6 s
Fuzzy reliability
0.744 0.814 0.955 0.829
0.5% LOCA accident 4.5% MFW accident 1% NORMAL condition 6.9% SGTR accident
tors during accident diagnosis, increasing their availability for the execution of appropriate corrective actions to bring the plant to a safe operating condition. The method uses artificial neural networks to identify the accident which are occurring in the plant, based on the correlation among selected process variables and uses fuzzy logic to identify the degree of reliability of the identification.
1426
R.G. da Costa et al. / Annals of Nuclear Energy 38 (2011) 1418–1426
By observing the results presented in the training phase, it was found that the jump type ANN, with backpropagation training is able to quickly diagnose the accidents that have been postulated for a PWR nuclear reactor, even with addition of noise that simulates the noisy background of actual operation. The system developed and evaluated in the LABIHS–IEN simulator to test the proposed method for event identification was able to provide reliable results allowing a rapid and accurate identification of accidents, and can be easily implemented in a real nuclear power plant, towards the addition of more identification accidents modules, always following the same implementation method. Acknowledgements This research has been supported by National Council of Research and Development (CNPq), Research Support Foundation of State of the Rio de Janeiro (FAPERJ) and Financier of Studies and Project (FINEP). References Alvarenga, M.A.B., Martinez, A.S., Schirru, R., 1997. Adaptive vector quantization optimized by genetic algorithm for real-time diagnosis through fuzzy sets. Nuclear Technology 120, 188–197. Carvalho, P.V.R., Vidal, M., Santos, I.L., 2006. Safety implications of some cultural and cog-nitive issues in nuclear power plant operation. Applied Ergonomics 37 (2), 211–223.
Carvalho, P.V.R., Vidal, M.C., Carvalho, E.F., 2007. Nuclear power plant communications in normative and actual practice: a field study of control room operators’ communications. Human Factors in Ergonomics and Manufacturing 17, 43–78 (vol. 6, p. 38–45). Cunha, J.J., Alvim, A.C.M., Lapa, C.M.F., 2007. Reduced scale PWR passive safety system designing by genetic algorithms. Annals of Nuclear Energy 34 (12), 1014–1023. Guimarães, A.C.F., Lapa, C.M.F., 2006. Hazard and operability study using approximate reasoning in light–water reactors passive systems. Nuclear Engineering and Design 236 (12), 1256–1263. Guimarães, A.C.F., Lapa, C.M.F., 2004. Fuzzy inference system for evaluating and improving nuclear power plant operating performance. Annals of Nuclear Energy 31 (3), 311–322. Guimarães, A.C.F., Cabral, D.C., Lapa, C.M.F., 2006. Adaptive fuzzy system for degradation study in nuclear power plants’ passive components. Progress in Nuclear energy 48 (7), 655–663. Lapa, C.M.F., De Sampaio, P.A.B., Pereira, C.M.N.A., 2004. A new approach to designing reduced scale thermal-hydraulic experiments. Nuclear Engineering and Design 229 (2–3), 205–212. Mol, A.C.D., Martinez, A.S., Schirru, R., 2003. A neural model for transient identification in dynamic processes with ‘‘do not know’’ response. Annals of Nuclear Energy 30 (13), 1365–1381. Pereira, C.M.N.A., Schirru, R., Martinez, A.S., 1998. Learning an optimized classification system from a data base of time series patterns using genetic algorithms. In: Ebecken, N.S.S. (Ed.), Data Mining, South Hampton. WIT Computational Mechanics Publication, United Kingdom. Vidal, M.C.R., Carvalho, P.V.R., Santos, M.S., Santos, I.J.L., 2009. Collective work and resilience of complex systems. Journal of Loss Prevention in the Process Industries 22 (4), 516–527. Woods, D., 1995. The alarm problem and direct attention in dynamic fault management. Ergonomics 38 (11), 2371–2393.
