Stylianos Chatzidakis

Analysis of Spent Nuclear Fuel Imaging Using Multiple Coulomb Scattering of Cosmic Muons

Cosmic ray muons passing through matter lose energy from inelastic collisions with electrons and are deflected from nuclei due to multiple Coulomb scattering. The strong dependence of scattering on atomic number Z and the recent developments on position sensitive muon detectors indicate that multiple Coulomb scattering could be an excellent candidate for spent nuclear fuel imaging. Muons present significant advantages over existing monitoring and imaging techniques and can play a central role in monitoring nuclear waste and spent nuclear fuel stored in dense well shielded containers. The main purpose of this paper is to investigate the applicability of multiple Coulomb scattering for imaging of spent nuclear fuel dry casks stored within vertical and horizontal commercial storage dry casks. Calculations of muon scattering were performed for various scenarios, including vertical and horizontal fully loaded dry casks, half loaded dry casks, dry casks with one row of fuel assemblies missing, dry casks with one fuel assembly missing and empty dry casks. Various detector sizes (1.2 m ×1.2 m, 2.4 m ×2.4 m and 3.6 m ×3.6 m) and number of muons (105, 5 · 105, 106 and 107) were used to assess the effect on image resolution. The Point-of-Closest-Approach (PoCA) algorithm was used for the reconstruction of the stored contents. The results demonstrate that multiple Coulomb scattering can be used to successfully reconstruct the dry cask contents and allow identification of all scenarios with the exception of one fuel assembly missing. In this case, an indication exists that a fuel assembly is not present; however, the resolution of the imaging algorithm was not enough to identify exact location.

Chaotic neural networks for intelligent signal encryption

The non-linear capabilities of artificial neural networks coupled with the noise-like properties of chaotic systems are exploited to perform signal encryption. The proposed approach computes the signal digital envelope which consists of the encrypted signal, the secret key and the associated hash value. The methodology is demonstrated via the encryption and subsequent decryption of two frequently occurring radiation signals, Co-60 and Cs-137. The results obtained demonstrate the capability of the proposed methodology to couple artificial neural networks and chaos dynamics to produce the signal digital envelope and satisfy the security requirements of confidentiality, authentication, and non-repudiation.

Gaussian Processes for State Identification in Pressurized Water Reactors

Abstract A novel machine learning approach for nuclear power plant modeling and state identification is presented together with its test results using data from the Loss-of-Fluid Test experimental facility. The approach exploits Gaussian processes whose principal function is to tackle the temporal problem of forecasting the actual system state in the varying environment of a nuclear reactor facility that undergoes successive overcooling transients. The approach fuses independent Gaussian process expert predictions to provide a single recommendation to the plant operators in a form that is suitable to appear on a decision support system screen. A variety of test cases are developed to explore the validity and relevance of Gaussian processes. The proposed implementation is examined with various predictor variables under different conditions, and the results obtained are in accordance with model expectations.

Creep rupture forecasting for high performance energy systems

The non-linear capabilities of artificial neural networks to model the dynamics of creep rupture and failure mechanisms are exploited to achieve failure forecasting in high performance energy systems. The proposed approach forecasts the time to rupture due to creep mechanism and consists of the library construction, the experimental data and measurements necessary for the training process, the measurements gathered during operation and the artificial neural network. The methodology is demonstrated on experimental data gathered for this purpose, for two frequently applied high-temperature/high-load materials, namely Grade 91 steel and Hastelloy XR. The results obtained demonstrate the capability of the proposed methodology to apply artificial neural networks to forecast the time to rupture and improve safety and efficiency of high performance systems.

Creep Rupture Forecasting: A Machine Learning Approach to Useful Life Estimation

Creep rupture is becoming increasingly one of the most important problems affecting behavior and performance of power production systems operating in high temperature environments and potentially under irradiation as is the case of nuclear reactors. Creep rupture forecasting and estimation of the useful life is required to avoid unanticipated component failure and cost ineffective operation. Despite the rigorous investigations of creep mechanisms and their effect on component lifetime, experimental data are sparse rendering the time to rupture prediction a rather difficult problem. An approach for performing creep rupture forecasting that exploits the unique characteristics of machine learning algorithms is proposed herein. The approach seeks to introduce a mechanism that will synergistically exploit recent findings in creep rupture with the state-of-the-art computational paradigm of machine learning. In this study, three machine learning algorithms, namely General Regression Neural Networks, Artificial Neural Networks and Gaussian Processes, were employed to capture the underlying trends and provide creep rupture forecasting. The current implementation is demonstrated and evaluated on actual experimental creep rupture data. Results show that the Gaussian process model based on the Matern kernel achieved the best overall prediction performance 56.38%. Significant dependencies exist on the number of training data, neural network size, kernel selection and whether interpolation or extrapolation is performed.

An Algorithmic Approach for RELAP5/MOD3 Reactivity Insertion Analysis in Research Reactors

A systematic approach for performing a holistic reactivity insertion analysis in research reactors using the RELAP5/MOD3 code is proposed. The intention is to demonstrate, in an orderly manner, a method for determining the limiting reactivity insertion in a research reactor facility. Indispensable constituents of the algorithmic approach are the introduction of the “time-to-failure” parameter, the selection of the reactivity insertion duration, the evaluation of the control rod drop time, and the computation of engineering factors. The methodology is demonstrated through a RELAP5/MOD3 parametric study performed to determine the limiting reactivity insertion values for the Greek Research Reactor-1 (GRR-1). In the framework of this study, the core nodalization effect on reactivity limits and the degree of conservatism introduced by the engineering factors are discussed. The results obtained confirm the applicability of the approach and reveal the effect of the parameters mentioned above on the performance of reactivity insertion analysis.

Artificial Neural Networks and Chaos Dynamics for Radiation Signal Encryption

Abstract Governments are interested in radiation signal encryption in projects relating to international safeguards; however, the available algorithms do not suitably address the challenges presented by the increasing computational capability of various actors, which require recent encryption algorithms to be more robust against attack algorithms. Therefore, an algorithmic approach for performing radiation signal encryption employing the nonlinear capabilities of artificial neural networks with the noise-like properties of chaotic systems is proposed herein. The radiation signal digital envelope consists of the encrypted signal such as may be found through gamma spectroscopy, the secret key for the encryption, and the associated hash value. The presented algorithmic approach demonstrates, in an orderly manner, an integrated method for computing this radiation signal digital envelope. Indispensable constituents of encryption include both the construction of a time series with chaotic characteristics and the incorporation of a hash function generator to satisfy the security requirements of confidentiality, authentication, and nonrepudiation. The methodology is demonstrated via the encryption and subsequent decryption of two frequently occurring radiation signals, namely, gamma spectroscopy signals from 60Co and 137Cs. The results obtained demonstrate the capability of the algorithmic approach to integrate artificial neural networks and chaos dynamics to produce the radiation signal digital envelope (for given security requirements).

Data Driven Monitoring of Energy Systems: Gaussian Process Kernel Machine for Fault Identification with Application to Boiling Water Reactors

Energy production units are large complex installations comprised of several smaller units, subsystems, and mechanical components, whose monitoring and control to secure safe operation are high demanding tasks. In particular, human operators are required to monitor a high volume of incoming data and must make critical decisions in very short time. Although they are explicitly trained in such situations, there are cases that may not be able to identify a gradually developing crucial faulty state. To that end, automated systems can be used for monitoring operational quantities and detecting potential faults in time. The field of machine learning offers a variety of tools that may be used as the ground for developing automated monitoring and control systems for energy systems. In the current chapter, we present an approach that adopts a single Gaussian process learning machine in monitoring high complex energy systems. The Gaussian process is a data-driven model assigned to monitor a set of operational parameters. The values of the operational parameters at a specific instance comprise the system’s operational vector at that time instance. The operational vector consists the input to the individual Gaussian process machine whose task is to classify the operation of the system either as normal (or steady state) or match it to a faulty state. The presented approach is benchmarked on a set of experimentally data taken from the Fix-II test facility that is a representation of a Boiling Water Reactor. Obtained results exhibit the potential of Gaussian processes in monitoring highly complex systems such as nuclear reactors, by identifying with high accuracy the faults in system operation.