Deep understanding of phenomena and industrial implementation of what I have learned are my main goals, even if it takes a long time.
Industrial applications of Soft-computing as well as other types of calculation (e.g. analytical, Monte Carlo, numerical, etc.) are my interests.
My dream is to reach a level of scientific ability that can easily explain/illustrate the most complex issues to the most uninformed. This is what God has taught us.
Sharif University of Technology, Tehran
Sharif University of Technology, Tehran
Tabriz University, Tabriz
I am Assistant Prof. of Department of Energy Engineering since Jan. 2018. Industrial applications of soft computing, new methods especially analytical ones for dose calculation in comparison with older methods (e.g. Monte Carlo), study and design of new nuclear batteries, applications of radioisotopes in industry, wireless detectors, applications of aacelerators, dynamics of nuclear recators, safety of nuclear reactors especially fault detection and diagnosis (FDD) of nuclear power plants (NPPs) are my scientific interests.
My current research involves applications of soft computing for identification/prediction of NPPs transients/accidents. Moreover, study of new nuclear batteries, analytical calculation of dose, and wireless detectors are in progress.
Several reasons such as no free lunch theorem indicate that there is not a universal Feature selection (FS) technique that outperforms other ones. Moreover, some approaches such as using synthetic dataset, in presence of large number of FS techniques, are very tedious and time consuming task. In this study to tackle the issue of dependency of estimation accuracy on the selected FS technique, a methodology based on the heterogeneous ensemble is proposed. The performance of the major learning algorithms of neural network (i.e. the FFNN-BR, the FFNN-LM) in combination with the diverse FS techniques (i.e. the NCA, the F-test, the Kendall’s tau, the Pearson, the Spearman, and the Relief) and different combination techniques of the heterogeneous ensemble (i.e. the Min, the Median, the Arithmetic mean, and the Geometric mean) are considered. The target parameters/transients of Bushehr nuclear power plant (BNPP) are examined as the case study. The results show that the Min combination technique gives the more accurate estimation. Therefore, if the number of FS techniques is m and the number of learning algorithms is n, by the heterogeneous ensemble, the search space for acceptable estimation of the target parameters may be reduced from n×m to n×1. The proposed methodology gives a simple and practical approach for more reliable and more accurate estimation of the target parameters compared to the methods such as the use of synthetic dataset or trial and error methods.
In this study using real data of Bushehr nuclear power plant (BNPP) and by soft computing methods and without using the data of self-powered neutron detectors (SPNDs), the maximum linear heat rate of BNPP is estimated. The efficient learning algorithms of artificial neural network (ANN) including Levenberg-Marquardt (LM) and Bayesian regularization (BR) in combination with different features selection techniques including Pearson, Spearman, and Kendall’s tau are employed to estimate the target parameter. Results show that the proposed method is appropriate for estimation of the maximum linear heat rate. Given the importance of this parameter in terms of safety and the fact that its excessive increase actuates the shutdown signal of the reactor, use of the appropriated approaches such as the present study, can increase the safety of the plant and improve Defense-In-Depth (DID).
Several reasons such as no free lunch theorem indicates that any learning algorithm in combination with a specific feature selection (FS) technique may give more accurate estimation than other learning algorithms. Therefore, there is not a universal approach that outperforms other algorithms. Moreover, due to the large number of FS techniques, some recommended solutions such as using synthetic dataset or combining different FS techniques are very tedious and time consuming. In this study to tackle the issue of more accurate estimation of NPPs parameters, the performance of the major supervised learning algorithms in combination with the different FS techniques which are appropriate for parameters estimation is considered. The target parameters/transients of the Bushehr nuclear power plant (BNPP) are examined as the case study. By comparing three major supervised learning algorithms (i.e. the MLP-BR, the MLP-LM, and the SVM) in combination with six principal FS techniques (i.e. the NCA, the F-test, the Kendall’s tau, the Pearson, the Spearman, and the Relief) for estimation of three important parameters of NPP (i.e. FMT, CMT, and the DNBR), the BR learning algorithm gives the more accurate results. Therefore, the results show that if the number of FS techniques is m and the number of learning algorithms is n, the search space for more accurate estimation of the NPPs important parameters can be reduced from n × m to 1 × m.
In this study, accurate estimation of nuclear power plant (NPP) parameters is done using the new and simple technique. The proposed technique using the genetic algorithm (GA) in combination with the Bayesian regularization (BR) and Levenberg- Marquardt (LM) learning algorithms identifies the appropriate architecture for estimation of the target parameters. In the first step, the input patterns features are selected using the features selection (FS) technique. In the second step, the appropriate number of hidden neurons and hidden layers are investigated to provide a more efficient initial population of the architectures. In the third step, the estimation of the target parameter is done using different architectures of multilayer feed-forward neural network (MFFN) with LM learning algorithm in which the maximum number of hidden neurons and the maximum number of hidden layers have been limited. In the fourth step, the proposed technique using the GA in combination with the BR learning algorithm is proposed to determine the more appropriate number and the more appropriate distribution of hidden neurons. To study the performance of the proposed technique, Bushehr nuclear power plant (BNPP) transients are examined. The different important transients/parameters are estimated. The results of the estimations by the identified architecture in comparison with the other appropriate architectures show superiority of the proposed technique. Therefore, the proposed technique can be used reliably for accurate estimation of the important parameters and can be used as a support tool by the operators in confront with transients.
Up to now, different methods have been developed for estimation of buildup factor (BF). However, either expensive estimation or time-consuming estimation are major restrictions/challenges of these methods. In this study a new technique utilizing combination of Monte Carlo method and the Bayesian regularization (BR) learning algorithm of multilayer feed-forward neural network (FFNN) is employed for estimation of BFs. First, the BFs of the different elements (i.e. Al, Cu, and Fe) at different energies and different mean free paths (MFPs) are modeled by the MCNP code. The results show that the calculated BFs by MCNP code are in good agreement with the reported values of American nuclear society (ANS). Afterwards, the appropriate architecture of FFNN (i.e. the appropriate number of hidden neurons and hidden layers) and the appropriate input patterns features are investigated. The resulted FFNN is trained using the modeled BFs and the selected category of features. In the test process, the BFs of the master alloys (i.e. Fe-Al%50, Cu-Fe50%, Al-Cu50%) are estimated. To evaluate the performance of the proposed FFNN for training/estimation of the new elements/alloys, Si is added to the training process and the BFs of the Al-Si35% is estimated. Average mean relative error (AMRE) and cumulative distribution function (CDF) of the errors show the acceptable accuracy of the estimating the BFs of the alloys. The noticeable advantages of the proposed technique are: 1- The BFs of the different alloys are estimated only by using the BFs of the constituent elements of the alloys. 2- The time needed to estimate the new BFs by the proposed technique can be neglected versus the time needed to model the new BFs by Monte Carlo. 3- The proposed technique can generalize its ability for estimating the BFs of the new alloys. 4- Monte Carlo codes need the trained person to model the BFs of the alloys while the FFNN generates the new BFs easily.
Multilayer perceptron (MLP) neural networks have been used extensively for estimation/regression of parameters. Moreover, recent studies have shown that learning algorithms of MLP which are based on Gaussian function are more accurate. In this paper, the mass attenuation coefficient (MAC) of gamma radiation for light-weight materials (e.g. O-8), mid-weight materials (e.g. Al-13), and heavy-weight materials (e.g. Pb-82) is modelled using Gaussian function based regularization of MLP (i.e. Bayesian regularization (BR)) and by a modular estimator. The results are compared with the Reference results. To show better performance of the utilized algorithm, the results of the different supervised methods including support vector machine (SVM) with different kernel functions, decision tree (DT), and radial basis network (RBN) are given. Average mean relative error (AMRE) and cumulative distribution function (CDF) of errors of MACs estimation are calculated. Comparison of the results indicates that MLP-BR gives more accurate results (e.g. AMRE of O-8=0.0014, CDF of O-8 (0.0069) = 0.99, AMRE of Al-13=0.0015, CDF of Al-13 (0.0048) = 0.99, AMRE of Pb-82=0.0117, CDF of Pb-82 (0.0523) = 0.99).
This paper introduces the support system for nuclear power plants (NPPs) operators. Transient is identified by a supervised classifier combining auto-regressive integrated moving average (ARIMA) model and artificial neural network (ANN). Transductive support vector machine (TSVM) as a semi-supervised learning (SSL) is used to cluster the type of unknown transient. To forecast future states of NPPs, a hybrid network combining ARIMA model and ANN is developed. Bushehr nuclear power plant (BNPP) transients are probed to analyze the ability of the proposed system. Noticeable advantages are: clustering of unknown transients, sole dependency of identifier on sign of output signal, forecasting any quantifiable parameter without necessity to know its correlation with other, possibility for prediction of parameters in long temporal dependencies. Finally, modular decomposition of the developed support system is presented. The developed system will afterwards be completed by necessary interfaces to be installed on the BNPP full scale simulator to verify its applicability and performance.
In this study, the estimation of the uranium price as one of the most important factors affecting the fuel cost of nuclear power plants (NPPs) is investigated. Supervised learning algorithms, especially, multilayer feed-forward neural network (FFNN) are used extensively for parameters estimation. Similar to other supervised methods, FFNN can suffer from overfitting (i.e. imbalance between memorization and generalization). In this study, different regularization techniques of FFNN are discussed and the most appropriate regularization technique (i.e. Bayesian regularization) is selected for estimation of the uranium price. The different methods including different learning algorithms of FFNN, support vector machine (SVM) with different kernel functions, radial basis network (RBN), and decision tree (DT) are utilized for the prediction of the uranium price and are compared with FFNN-BR. Average mean relative error (AMRE) and cumulative distribution function (CDF) of the results indicate that FFNN-BR method is more accurate for the uranium price estimation (i.e. CDF (0.0720) = 0.99 and AMRE = 0.0533).
The motivation of this study is development of a technique to construct energy spectrum of higher price/high resolution detectors (e.g. NaI (Tl)) using spectrum of lower price/low resolution detectors (e.g. NE102). Since there is no explicit mathematical model between these type of detectors (i.e. organic and inorganic scintillator detectors), it is necessary to utilize model-free methods. Construction of mapping function to generate spectrum of inorganic scintillator using spectrum of organic scintillator can be done by supervised model-free methods. Different supervised learning methods including localized neural networks, statistical methods, feed-forward neural networks, and conditional methods are utilized for spectrum construction. Experimental spectrums of the different radioisotopes (i.e. Co-60, Cs-137, Na-22, Am-241) including 15 spectrums of NaI (Tl) detector and 15 spectrums of NE102 detector are respectively used as training data and test data in the supervised methods. Results demonstrate that localized network (i.e. radial basis network) is the more appropriate method for the spectrum construction. The results of statistical method (i.e. support vector machine) is acceptable while conditional method (i.e. decision tree) does not give acceptable results and multi-layer perceptron does not learn the spectrums. The developed technique can be applied with an interesting ratio of training set to test set (i.e. r/(2r-1-r)). In other words, constructing spectrums of all possible combinations of r radioisotopes (i.e. 2r-1-r) is possible only with training of single radioisotopes spectrums (i.e. r). The developed method for generation of spectrums is more appropriate for identification of the radioisotopes and is not so useful for spectrum tracking. Spectrum tracking can be done by training of supervised learning method using generated pulses of detector.
One of the most important challenges in target parameters estimation via model-free methods is selection of the most effective input parameters namely features selection (FS). Indeed, irrelevant features can degrade the estimation performance. In the current study, the challenge of choosing among the several plant parameters is tackled by means of the innovative FS algorithm named ranking of features with minimum deviation from the target parameter (RFMD). The selected features accompanied with the stable and the fast learning algorithm of multilayer perceptron (MLP) neural network (i.e. Levenberg-Marquardt algorithm) which is a combination of gradient descent and Gauss-newton learning algorithms are utilized for estimation of the target parameter. To evaluate the performance of the developed method, three transients of Bushehr nuclear power plant (BNPP) are examined. The results of estimation of departure from nucleate boiling ratio (DNBR) by the developed method in comparison with the final safety analysis report (FSAR) of BNPP show acceptable agreement. In general, RFMD shows outperformance (i.e. less estimation error) in comparison with the particle swarm optimization (PSO) algorithm and the Relief technique.
In this paper, a novel idea is developed to construct energy spectrum of inorganic scintillator detector (e.g. NaI(Tl)) using energy spectrum of organic scintillator detector (e.g. NE102) by means of a model-free method. For this purpose, support vector machine (SVM) accompanied with different kernel functions (i.e. linear, polynomial, and Gaussian) is applied. The spectra of NE102 and NaI(Tl) detectors of the single radioisotopes (i.e. Co60, Cs137, Na22, and Am241) are utilized for training of SVM. In other words, data of NE102 detector are input spectrums of training patterns and data of NaI(Tl) detector are target spectrums of training patterns. To construct an appropriate mapping function between spectrums of detectors, a kind of cross-correlation technique namely mapping of totality of channels to single channel (TSC) is utilized. In the test process, spectrums of different combinations of the target radioisotopes are constructed and the results are compared with the measured spectrums. Polynomial kernel function gives good results. Linear and Gaussian kernel functions do not give so appropriate results. The major advantages of the developed method are: 1- The NaI(Tl) spectrums of different combinations of the target radioisotopes are constructed only by training of single radioisotopes spectrums. 2- Energy spectrums of high price/ high resolution detectors (i.e. inorganic scintillator detectors) can be constructed using low price/ low resolution detectors (i.e. plastic scintillator detectors). 3- This method can be used to construct energy spectrum of any type of inorganic scintillator (e.g. BGO detector) using either plastic or liquid scintillators. Liquid scintillator detectors (e.g. NE213) are more appropriate for detection/identification of radioactive sources which emit both gamma and neutron radiations. However, these detectors give low resolution gamma spectrums. The developed method can be appropriate for construction of higher resolution gamma spectrums for this type of sources. This application of the developed method is discussed in the manuscript.
In this paper cross-correlation of measurable/unmeasurable parameters of nuclear power plants (NPPs) are detected. Correlation techniques including Pearson’s, Spearman’s, and Kendall-tau give appropriate input parameters for training/prediction of the target unmeasurable parameters. Fuel and clad maximum temperatures of uncontrolled withdrawal of control rods (UWCR) transient of Bushehr nuclear power plant (BNPP) are used as the case study target parameters. Different model-free methods including decision tree (DT), feed-forward back propagation neural network (FFBPNN) accompany with different learning algorithms (i.e. gradient descent with momentum (GDM), scaled conjugate gradient (SCG), Levenberg-Marquardt (LM), and Bayesian regularization (BR)), and support vector machine (SVM) with different kernel functions (i.e. linear and Gaussian functions) are employed to predict the target parameters. Comparison of the results indicates that BR learning algorithm of FFBPNN gives more precise results. Moreover, DT results is comparable with LM learning algorithm of FFBPNN. In addition, SVM results with Gaussian kernel function is better than SVM results with linear one. Results show that cross-correlation detection among the parameters has decisive effect on performance of learning algorithms. This claim is verified by prediction of the target parameters using either fewer number of correlated input parameters or uncorrelated input parameters. The input parameters without strong correlation have greater errors in forecasting. In addition, more parameters with strong correlation give better results. Prediction of unmeasurable parameters of NPPs can be used as a support system for the NPPs operators to perform more appropriate actions in confrontation with transients.
In this paper, a modular system is developed for estimation of mass attenuation coefficient (MAC) of different materials/energies using artificial neural network (ANN). Cascade feed-forward neural network (CFFNN) as a type of ANN constructs mapping function between input patterns and the targets (i.e. MAC). Performance of different learning algorithms of CFFNN including gradient descent (GD), gradient descent with momentum (GDM), scaled conjugate gradient (SCG), Levenberg-Marquardt (LM), and Bayesian regularization (BR) are compared. For training, different categories of input patterns features are utilized to show the more appropriate one. Average mean relative error (AMRE) and cumulative distribution function (CDF) of the results indicate that BR learning algorithm accompany with the selected category of features (i.e. atomic number, photon energy, and density) is more accurate for estimation of MAC (e.g. for Al, CDF(0.0069) = 0.99 and AMRE = 0.0017). The advantages of the present method are: 1- Estimation of MAC is done fast (i.e. in comparison with Monte Carlo methods) and is done at a lower cost (i.e. without need to extra experiments) 2- Modular system reduces the risk of incorrect estimation 3- It is possible to extend the number of estimators for more materials/mixtures without unfavorably affecting the existing system.
The field of any point source which is broadened equally in all directions without any limitation to its range is within category of the inverse square law (ISL). As a case study, the dose equivalent (DE) rate is calculated. For calculating the DE rate, the radiation source can be divided into multiple layers and each layer is fractionated to multiple rectangular surfaces. Each rectangular surface can be replaced with three types of sectors. The DE rate of a source on a target is then sum of DE rates of sectors. The developed method is independent of the target position relative to the source and is used for the dose calculation of any arbitrary arrangement of source and target. As an examples, the DE rate is calculated for the square/rectangular gamma emitter sources in comparison with MCNP code and numerical method. Results show very good agreement. Noticeable advantages are: (1) The developed method performs a very quick calculation (i.e. more than 100000 times faster) while Monte Carlo techniques usually take time to obtain adequate statistics on small regions (2) The developed method calculates the DE rate more than 100 times faster than numerical method (3) Usually the trained person calculates the dose with Monte Carlo codes while analytical calculation does not need the trained one. The proposed analytical method is basis of software package for the DE calculation of complex surfaces which is under development.
In this paper, an analytical method for calculation of the dose equivalent (DE) of coordinate surfaces of the Cartesian geometry is presented. DE of rectangular surfaces of gamma radiation emitters is calculated. The developed analytical method changes rectangular surface to multiple polar regions by dividing its surface into four types of sectors. By this method, the calculation of the dose is converted into calculation of simple mathematical series. The dose of rectangular shape sources for different gamma radiation emitters at different distances to target is calculated and the results are compared with MCNP code. Results show very good agreement. Advantages of the developed method are: 1—While Monte Carlo techniques usually take time to obtain adequate statistics on small regions, the developed method is independent of statics and therefore performs a very quick calculation (i.e. more than 100000 times faster) 2—Usually the trained person is required to calculate the dose with Monte Carlo codes while analytical calculation does not need the trained one. Moreover, analytical methods make possible to perform easily parametric analysis and to reach desired outlet of the dose. Since rectangles with proper size in sufficient numbers can completely reconstruct any surface, therefore, dose of complex surfaces can be calculated using the developed method.
In this paper, some important operating parameters of nuclear power plants (NPPs) transients are forecasted using different supervised learning methods including feed-forward back propagation (FFBP) neural networks such as cascade feed-forward neural network (CFFNN), statistical methods such as support vector regression (SVR), and localized networks such as radial basis network (RBN). Different learning algorithms, including gradient descent (GD), gradient descent with momentum (GDM), scaled conjugate gradient (SCG), Levenberg-Marquardt (LM), and Bayesian regularization (BR) are used in CFFNN method. SVR method is used with different kernel functions including Gaussian, polynomial, and linear. RBN is used with radial activation function. Comparison of the results indicates that learning algorithms based on Gaussian distribution function (i.e. BR algorithm) and Gaussian kernel/activation functions are, in general, more precise for time series prediction. Moreover, learning methods based on Gaussian function lead to acceptable results in prediction of complicated time series, such as core inlet flowrate of large break loss of coolant accident (LBLOCA) which are changed irregularly and drastically. In other words, Gaussian learning algorithms/kernel functions/activation functions are appropriate choices for NPPs parameters forecasting.
This study introduces a new identifier for nuclear power plants (NPPs) transients. The proposed identifier changes the representation of input patterns. Change of representation is a semi-supervised learning algorithm which employs both of labeled and unlabeled input data. In the first step, modified self-organizing map (MSOM) carries out an unsupervised learning algorithm on labeled and unlabeled patterns and generates a new metric for input data. In the second step, support vector machine (SVM) as a supervised learning algorithm classifies the input patterns using the generated metric of the first step. In contrast to unsupervised learning algorithms, the proposed identifier does not discard the collected information. The proposed identifier is examined by the Iris flower dataset and the Bushehr nuclear power plant (BNPP) transients. In overall, results show good performance of the developed identifier. Training with small fraction of labeled atterns, classification only by the sign of the classifiers outputs, and modular identification are main advantages of the proposed identifier. Results indicate that the developed MSOM is not able to cluster correctly quasi-static transients such as uncontrolled withdrawal of control rods (UWCR) in presence of steady state patterns. Quasi-static transients are very similar to steady state. Moreover, imposed noise on input data may eliminate minor differences among the patterns of them. This may cause wrong training of SVM. This is the most important challenge of semi-supervised learning. In other words, if the knowledge on density of unlabeled input patterns do not carry useful information for prediction of the targets of unlabeled patterns, then semi-supervised learning may degrade the prediction accuracy. Therefore, the proposed identifier is more appropriate for classification of transients in which either clusters have been apart or small noises have been imposed on data.
In this study, we propose a new method for identiﬁcation of nuclear power plants (NPPs) transients based on online dynamic quantum clustering (DQC). In this unsupervised learning algorithm, the Gaussian kernel is the eigen-state of the Schr¨odinger equation and the minimums of the Schr¨odinger potential are the cluster centers of patterns. For clustering of transients, data of each event are given to the DQC and form a cluster independent of other transients. This process is done for all target plant conditions. The formed clusters are labeled according to the name of their related transients. Afterwards, to test the proposed identiﬁer, as time goes by, new data points move toward the formed potential wells. Finally, each new datum falls into an appropriate cluster and, therefore, the type of transient is identiﬁed online. The DQC, unlike previously developed unsupervised learning algorithms, is not dependent on the geometric proximity of data. The developed identiﬁer is examined by the Iris ﬂower dataset and typical WWER-1000 plant transients. Results show a reasonable performance of DQC. We use singular value decomposition (SVD) and bipolar representation of real data to reduce the dimensions of data and to show explicitly positive and negative sides of information. The major novelty of this identiﬁer is the development of a technique for online transient identiﬁcation of NPPs utilizing the DQC without any preliminary information about the input patterns. The developed method is a step forward for practical recognition of NPPs transients.
I am skillful at using programming languages/ computer codes such as C#, Matlab, C, MCNP, DYNCO, SIGACE, ORIGEN, etc. I am eager to learn new programming languages/ codes. Moreover, I have acceptable dexterity at application of different kinds of soft computing methofs including ANN, SVM, DQC, DT, competitive and localized neural networks, fuzzy systems, and GA.
I have some work experiences including expert of transient analysis of NPPs, Consulting Engineer of Bushehr nuclear power plant, designer and developer of new research reactors for more than 7 years.
I would be happy to talk to you if you need my assistance in your research or anything else. I prefer to communicate through my Sharif email, but feel free to contact me however you like.
You can find me at my office located at Deaprtment of Energy Engineering, floor 3, south side.
I am ususally at my office, but you may consider an email to fix an appointment.