2022 AI ATF SMR 2209.12146v1
2022 AI ATF SMR 2209.12146v1
2022 AI ATF SMR 2209.12146v1
Md. Shamim Hassan1, Abid Hossain Khan1, Richa Verma2, Dinesh Kumar3, Kazuma
Kobayashi4, Shoaib Usman4, Syed Alam4*
1
Institute of Nuclear Power Engineering, Bangladesh University of Engineering and Technology,
Dhaka 1000, Bangladesh
2
Department of Electrical Engineering, Indian Institute of Technology Delhi, Delhi 110016,
India
3
Department of Mechanical Engineering, University of Bristol, Bristol BS8 1TR, UK
4
Department of Nuclear Engineering and Radiation Science, Missouri University of Science and
Technology, Rolla, MO 65409, USA
*Corresponding author: Syed Alam (alams@mst.edu)
Abstract
The concept of small modular reactor has changed the outlook for tackling future energy crises.
This new reactor technology is very promising considering its lower investment requirements,
modularity, design simplicity, and enhanced safety features. The application of artificial
intelligence-driven multi-scale modeling (neutronics, thermal hydraulics, fuel performance, etc.)
incorporating Digital Twin and associated uncertainties in the research of small modular reactors
is a recent concept. In this work, a comprehensive study is conducted on the multiscale modeling
of accident-tolerant fuels. The application of these fuels in the light water-based small modular
reactors is explored. This chapter also focuses on the application of machine learning and
artificial intelligence in the design optimization, control, and monitoring of small modular
reactors. Finally, a brief assessment of the research gap on the application of artificial
intelligence to the development of high burnup composite accident-tolerant fuels is provided.
Necessary actions to fulfill these gaps are also discussed.
1. Introduction
Being one of the lowest carbon-emitting sources of power (Wang et al., 2011), the nuclear power
industry has become one of the leading global energy sources around the world. From 1950 to
2022, this industry extended its markets in more than thirty countries, accounting for around 10%
of the world's power from 440 nuclear reactors (Hore-Lacy, 2010). However, the industry is
experiencing slow growth after the events like Chernobyl disaster and the Fukushima disaster.
Uncorrected Proof
Although the enhanced safety features of new Gen III+ designs have mitigated the phobia of a
severe accident in a nuclear power plant to some extent (Taylor, Dessai and de Bruin, 2014; Sun,
Zhu and Meng, 2016), the nuclear era is far from being reincarnated. This is mainly because of
the increased investment costs associated with these extended safety systems; very few investors
are willing to invest such a huge amount of money on a project that can shut down prematurely if
something goes wrong. Therefore, the nuclear community is constantly in search of a safer yet
cost-effective substitute of the large-scale Gen III+ nuclear power plant (NPP) designs.
Many leading nations in the nuclear power industry have appeared intrigued by the concept of
small modular reactors (SMRs) lately. As a result, new and innovative research projects have
been initiated with the collaboration of scientists from top nuclear research laboratories across
the globe. Some countries have already planned to install them (Ricotti and Fomin, 2020). On the
other side of the table, some countries don’t see the commercial viability of SMRs because of its
lower efficiency and electrical power output compared to the ordinary light water reactors
(LWRs) (Ingersoll, 2010). Bingham and Mancini (2014) mentioned in his research paper that
the installation of SMRs is suitable for an installed capacity of 1 to 3 GWe. They also opined that
SMRs have the potential of creating new employment opportunities in the nuclear industry. He
compared the advantage of the small modular reactor with base-load technology after assessing
their life cycles. However, they did acknowledge that the economic and social boundaries would
be a major concern for the deployment of SMRs (Bingham and Mancini, 2014).
As an aftermath of the Fukushima disaster, serious questions were raised regarding the fuel
performance of light water reactors (LWRs). In the event of a serious nuclear accident, a
significant amount of radioactivity could leak into the environment. It became clear during the
Fukushima disaster that hydrogen explosions and the release of radionuclides could result in
significant health issues. (Koo et al., 2014). Due to the damage of the nuclear fuel rods, which
contain fuel pellets and fuel cladding, these two problems i.e., hydrogen explosions and
radionuclide release may occur. When exposed to a high-temperature steam environment, Zr-
based alloys rapidly oxidize in accident conditions. Hydrogen is generated, resulting in
explosions (Nikulina, 2004). The cladding of Zr-based alloys also balloons and opens in an
accident, releasing radionuclides into the environment (Kim et al., 2016). To enhance nuclear
power plant reliability and safety, fuel claddings are required to maintain their inherent
mechanical properties under normal and accident conditions. In addition, it is expected that the
fuel pellet itself will prevent release of radioactive fission products during a serious accident.
Therefore, recent LWR research is focused primarily on the development of accident-tolerant
fuels (ATFs) (Carmack et al., 2013; Kim et al., 2016).
The ATFs are characterized by their innovative design and material that prevent or delay
radionuclides from being released during an accident (Montgomery et al., 2013; Kim et al.,
2016). While ATFs are expected to tolerate a considerable period of coolant loss compared to
conventional UO2-Zr alloy fuels, they should also provide satisfactory performance during
normal operations and transient conditions. In addition, they should improve fuel safety for
events that are beyond the design basis (Koo et al., 2014). In other words, they must have more
oxidation resistance capability with strong mechanical strength under accident conditions
(Carmack et al., 2013). To diminish the outcome of the accident, various ATF concepts are
being explored (Koo et al., 2014).
Uncorrected Proof
As mentioned earlier, the primary objective of ATF is to increase the thermal conductivity and
contain the fission products inside the fuel pellets. Moreover, according to the ATF concept, it
must be ensured that the density of the uranium is high enough to compensate the burnup or
cycle length. (Ray, Johnson and Lahoda, 2013). Not only fuel pellets but also fuel cladding is the
main components of ATF. Ceramic composite claddings are being studied for use in ATFs (Kim,
Kim and Park, 2013). No-Zr-alloy cladding are also being considered (Terrani, Zinkle and
Snead, 2014). Researchers are working on the development of Mo–Zr claddings and iron-based
alloy for the midterm application in the reactor (Cheng et al., 2013; Terrani, Zinkle and Snead,
2014). Different types of cladding coating systems ease the production of ATFs (Idarraga-
Trujillo et al., 2013). Finally, silicon carbide (SiC) is appraised for long-term application in the
reactor (Stempien et al., 2013). During the fabrication of ATF, several factors must be kept in
mind such as technical challenges, economics, development schedule, and nonetheless the safety
(Carmack et al., 2013).
After installation and commissioning of any power plant, its proper operation and control are the
most crucial points to think about. There are numerous events where the accident in a power
plant is initiated due to human error rather than malfunction of the plant equipment. Therefore,
the researchers are suggesting autonomous control systems for operating the nuclear power
plants. Since these control systems must work satisfactorily under transient conditions, machine
learning (ML) has recently become the focus of research in this field. Richard T. Wood (2017)
believed that artificial intelligence (AI) can be used for controlling power plants to maintain safe
and stable operation (Wood, Upadhyaya and Floyd, 2017). Artificial neural network (ANN) is
the skeleton of computational intelligence techniques. Because of its diversity, ANN can be used
to solve different nuclear engineering problems (Manic and Sabharwall, 2011). Computational
intelligence (CI) has been applied in the nuclear industry for many years, although the lack of
real-life data of plant transient and the Blackbox nature of AI has been a matter of concern in
many situations (Suman, 2021). In his work, Bernard (1989) explored the applicability of ANN
for nuclear system. Uhrig (1991) wrote an encapsulated summary of neural network applications
in nuclear reactor monitoring. Using AI, Reifman (1997) investigated mechanisms of detecting
and locating faults in nuclear equipment. Uhrig and Tsoukalas (1999) reviewed how to use fuzzy
logic, neural networks, and genetic algorithms to operate, monitor, and diagnose nuclear
reactors. In a latter review articles by Uhrig and Wines (2005), it was pointed out that there was a
gradual shift in the focus towards implementation of CI in nuclear power plant design and
optimization. Recently, research is being carried out on designing methods for implementing AI
techniques for next-generation reactors and space reactors, which are under development (Uhrig
and Hines, 2005). advanced data-driven AI approaches to transient detection in reactors is also
becoming very common now-a-days (Moshkbar-Bakhshayesh and Ghofrani, 2013).
Furthermore, digital twin framework using constructive AI/ML surrogate model for ATF based
has been proposed by General Atomics Electromagnetic Systems (Jacobsen, 2022) in
collaboration with Idaho National Lab (INL) and Los Alamos National Lab (LANL).
From the above literature survey, it is evident that SMRs, ML and ATFs are three major topics of
research in recent times, but they are yet to cross paths. This chapter focused on the latest
progress in the research on machine learning and artificial intelligence for designing ATFs for
the LWR-based SMRs. After the informative introductory part, brief descriptions of ATF, ML
and AI are given. Also, the application of ATFs in light water-cooled SMRs is summarized.
Uncorrected Proof
Then, the current trend in the use of artificial intelligence in the field of nuclear engineering is
presented. Finally, the prospect of using Artificial Neural Network (ANN) in developing high
burnup composite ATFs for light-water cooled SMRs is explored. Finally, research gap and
scope for future studies are also discussed.
Accident-tolerant fuel (ATF) is a challenging area of research in the field of nuclear engineering.
Normally, the core damage frequency of any reactor is once in a million years (Purba et al.,
2020). But when it happens, the consequences are catastrophic. Thus, precautions are required to
prevent core damage in case of any form of accident, design basis or beyond design basis.
Currently, UO2-Zircaloy based fuel is the most familiar fuel in the nuclear industry. However,
there properties make them vulnerable to severe accidents (Alam, Goodwin and Parks, 2019a;
Almutairi et al., 2022).
The temperatures of the fuel and the moderator increase if a leak occurs in the coolant pipe or if
the pump stops working due to an accident in a conventional light water reactor (LWR). The
physical components, as well as chemical debasement, occur when the fuel temperature exceeds
800°C. For temperatures between 700°C to 1200°C, there is a probability of damage of the fuel
rod such as release of radioactive gases, oxidation of the material, and dangerous reactions
initiated between the Zr cladding and fuel (Purba et al., 2020). Without an emergency core
cooling system, there is a chance of the formation of hydrogen gas. In the presence of oxygen,
hydrogen gas is likely to explode. When temperatures rise above 1500°C, there is a greater
chance of a core meltdown (Purba et al., 2020). Fig.1 presents the possible damaging events of a
typical LWR fuel rod after an accident. Since the next generation reactors are expected to be
operational at higher power densities, there limitations must be overcome. Therefore, UO2-
Zircaloy based fuel is expected to be replaced by ATFs in future as they can withstand loss of
active cooling for a longer period without any damage (Zinkle et al., 2014). The major
advantages of ATFs over UO2-Zircaloy based fuel are (Ott, Robb and Wang, 2014; Zinkle et al.,
2014):
“The Consolidated Appropriations Act, 2012”, approved by the Department of Energy, Office of
Nuclear Energy, USA (Goldner, 2012), has given notable practical and scientific emphasis on the
development of ATFs. Multiple projects were sponsored to increase the safety and security
system of the current generation’s reactors. These projects aim at evaluation of different ATFs
and their potential use in LWRs. The projects also focus on successful implementation of the
design upgrades in the commercial nuclear power plants before 2021 (Goldner, 2012).
Uncorrected Proof
Fig.1. (a) A typical fuel assembly, (b) A fuel rod after accident (Zinkle et al., 2014)
Researchers are working on different ATF materials for prospective use in light water reactors
(LWRs) (Bragg-Sitton et al., 2016; Spencer et al., 2016; Terrani, 2018; Ebrahimgol, Aghaie and
Zolfaghari, 2021). Table 1 summarizes the recent research focused on the properties of ATF
materials and their responses to normal and transient conditions. From Table 1, it may be
observed that the researchers are considering multiple options; accident tolerant cladding
materials with conventional UO2 fuels, improved fuel material with conventional Zr-based alloy
or a completely new fuel-cladding system (Chen, He and Yuan, 2020). Numerous studies have
observed promising properties in Fully Ceramic Microencapsulated (FCM) (Powers et al., 2013;
He et al., 2019) and U3Si2 (Nelson et al., 2014) fuels. However, a few technical challenges such
as higher volumetric swelling, lower melting point, inferior neutronic behavior, etc. compared to
UO2 are yet to be resolved (Spencer et al., 2016; Chen, He and Yuan, 2020). Some researchers
have also suggested addition of BeO or Mo in UO2 fuels for improved thermal conductivity,
although the inclusions may cause changes in the neutronic behavior of the fuel (Chandramouli
and Revankar, 2014). On the other hand, high Chromium SS and SiC-based cladding materials
are slowly gaining attention of the nuclear community (Brown et al., 2017).
Uncorrected Proof
Small modular reactors (SMRs) are advanced Gen IV reactors with improved safety (Aydogan,
2016). There are many differences between the current light water reactor and small modular
reactors (SMRs). The two words, “small” and “modular” indicate the main differences between
the current light water reactor and SMRs. The name “small” refers to the low electric power
output of these reactors, typically below 300MWe. And the name “modular” indicates that these
reactors are constructed off-site. Although gas-cooled and metal-cooled SMRs are also being
considered, light water cooled SMRs (LW-SMRs) are the most likely reactor technologies to be
available commercially in near future (Aydogan, 2016).
In every typical LW-SMRs, the numbers of fuel assemblies are identical, that is 17×17. The
basic main differences between the commercial light water reactors and LW-SMRs are in their
sizes, lengths of the fuel cycle and the output power capacity. The reactivity control system of
LW-SMRs is identical to LWRs. Control rod assemblies are used for controlling the reactivity of
the LW-SMRs. Moreover, for controlling the reactivity, soluble and burnable absorbers are also
being considered in most designs (Aydogan, 2016). No revolutionary change in the fuel element
of commercial LW-SMRs has been proposed either. Table 2 indicates the proposed fuel
components in LW-SMRs.
From Table 2, it is observed that there is no difference in the fueling material of different types
of LW-SMRs; all of them are considering conventional UO2 fuel pins. Their enrichment level is
about five percent. However, the cladding materials are distinguishable for different LW-SMRs.
Nevertheless, almost all of them are Zirconium-based alloys. Only mPower has considered SS
cladding. The reason behind choosing the conventional fuel materials for the LW-SMRs is the
safety of these novel reactor designs. Since these reactors have lack of real-life data during
operational state, the transient behavior of these reactors can’t be predicted with sufficient
confidence (Aydogan, 2016). Therefore, the manufacturers are going for the proven technologies
of their predecessors i.e., the large-scale LWRs currently operational. However, the use of ATF
can significantly improve the safety of LW-SMRs (Awan et al., 2018; Tiang and Xiao, 2021).
Therefore, researchers are continuously exploring different fuel options for LW-SMRs. Table 3
summarizes some of the latest research on the application of ATFs in LW-SMRs.
Uncorrected Proof
From Table 3, it may be observed that ATFs have promising performances in LW-SMRs,
although the fuel and moderator temperature coefficients may increase depending on the fuel-
moderator combination (Pourrostam, Talebi and Safarzadeh, 2021). Nevertheless, the
unavailability of sufficient data makes it difficult to come to a definite conclusion. Further
research is required to access the prospective use of ATFs in LW-SMRs.
Scientists have worked hard for years to develop intelligent machines, and the pursuit of perfect
machines is seemingly never-ending. These machines are programmable and can help
accomplish any target goal. Intelligence is the capacity to perceive knowledge. It helps in making
decisions based on reasoning and learning. Many researchers think that machine learning (ML) is
a crucial part of artificial intelligence (AI) because they help to parse data. It also grasps this data
and applies the learning to make decisions. Reinforcement learning, supervised learning, and
unsupervised learning are the three main framework of ML, as shown in Fig.2. While these three
have their specific areas of application, supervised learning is one of the most adapted ML
frameworks. It is employed for the categorization or regression of data. Neural networks, support
Uncorrected Proof
vector machines, and naïve Bayes are examples of supervised learning (Suman, 2021). The use
of supervised learning is dependent on the availability of training data.
Neural network (NN), also known as artificial neural network (ANN), is the most common type
of supervised learning. The concept of NN comes from the biological working principle of the
animal brain (Wood, Upadhyaya and Floyd, 2017). The structural framework of NN is shown in
Fig.3. The input layer is the primary layer of the NN. The hidden layer is the second layer after
the input layer. And the output layer is the final stage. A neuron is the key element of the
Uncorrected Proof
network and connects these layers. The meeting points of the neurons can be referred to as a
node. This neural network is trained different learning algorithms. NN can be divided into a few
sub-branches. In recent years, researchers are working more on the recurrent neural network,
backpropagation neural network, radial basis neural network, and feed-forward neural network
because of their major application in the nuclear industry. But a preference is observed for
backpropagation neural networks because of their wide application in this sector (Suman, 2021).
The potential use of NN in the field of the nuclear power plant is expanding day by day (Suman,
2021). This has resulted from the nuclear industry's effort to stay on top of technological
advances. The research groups of the major manufacturing industries of nuclear reactors are
working on developing innovative reactor designs that are intrinsically safe and secured, clean
and reputable, inexpensive, workable, and well-grounded. They have changed and improved the
potential safety system of the reactor generation wise (Wood, Upadhyaya and Floyd, 2017;
Suman, 2021). Likewise, they are putting their effort to enhance the overall capacity and
availability factor by reducing the maintenance and operational cost of the current reactors. ANN
has emerged as a potential Fastlane to reach these goals (Suman, 2021) [1].
In terms of modeling and controlling the nuclear reactor to forecasting and pattern recognition,
the use of the NN is praiseworthy (Wood, Upadhyaya and Floyd, 2017). During any abnormal
condition, the operator needs to make a quick decision to handle the situation. NN-based
computational system can help the operator tackle this type of situation. It can also reduce the
volume of work of the operator as well as assist in problem-solving during any kind of unwanted
condition. Further, it can help to build up intelligible resources (Suman, 2021).
To reduce the overall cost, NN may be used to determine the optimum reactor core size,
necessary structural change of the core reload design, management of the fuel in the core, and
designing of the fuel pellets and lattices, etc. (Uhrig and Hines, 2005; Suman, 2021). NN can
enhance the capability of a NPP by diagnosis the fault condition of the plant, monitoring the
reactor operational status, detecting the failure of the sensors, actuator, and transducer condition.
During any transient condition, NN can help in managing the transient of the power plants by
classifying it according to safety criteria depending on its severity (Suman, 2021). In short, NN
may be used for optimization, monitoring and control of a NPP with high efficiency.
ML and AI have brought about a revolution in modern research. With this powerful tool, a
researcher can make a machine do the tasks on its own that would have needed human
intervention otherwise. And the thing is, machines are faster than men. So, artificial intelligence
has, to some extent, contributed to the rapid expansion of modern science, engineering and
technology. And since nuclear engineering is a complicated and sensitive branch of engineering,
no wonder researchers have employed artificial intelligence, especially ANN, in different
applications related to this field (Suman, 2021). The material science has gathered a lot of
attention in recent times. The future reactor designs ask for advanced materials with better
mechanical, chemical and neutronic properties [citation]. Machine learning and artificial
intelligence can contribute to the rapid development of this field (Morgan et al., 2022).
Uncorrected Proof
AI may be utilized in numerous possible ways in nuclear material research. The most common
application is prediction of mechanical properties of the structural materials in a nuclear facility
(Sharma, Bhasin and Ghosh, 2010; Sharma et al., 2011; Morgan et al., 2022). The structural
components may include Reactor Pressure Vessel, piping systems, containment building, etc.
(Morgan et al., 2022). NN is also found to be very efficient in fault detection of spent nuclear
fuel (Rossa, Borella and Giani, 2020). Perhaps the most sophisticated application of AI in
material science is the modeling of subatomic and interatomic behavior of different materials.
The interatomic potential of materials of Gen IV reactors such as molten salts (Lam et al., 2021;
Li et al., 2021; Sivaraman et al., 2021), high-grade stainless steels (Min et al., 2021), etc. are
being studied with the help of artificial intelligence. AI is also utilized for evaluation and
validation of nuclear data (Neudecker et al., 2020).
Nuclear safeguard has become a topic of sufficient debate and discussions within the nuclear
community. The international organizations are proactive to prevent proliferation of nuclear
materials, especially Plutonium and Uranium-233. AI is being employed at a regular basis in the
research related to identification of these Special Nuclear Materials (SNMs). Some of the recent
research proposed the use of gamma-spectroscopy (Curtis, 2016; Zhang et al., 2019) for
generating data for training the neural network while the others have suggested using a mixed
photon-neutron environment (Jinia et al., 2021; Zhang et al., 2021).
The use of AI has already become very common in optimization, monitoring and control of
modern Gen III+ reactors (Suman, 2021). The application of NN in improving the design and
safety of advanced and Gen IV reactors, including SMRs, is also being explored (Manic and
Sabharwall, 2011). In the early part of the century, the use of NN was observed for the thermal-
hydraulic design optimization of different components of a NPP (Ridluan, Manic and Tokuhiro,
2009; Manic and Sabharwall, 2011). The recent research works are somewhat more focused on
the core optimization, monitoring, transient behavior analysis and control of a nuclear reactor.
An example is the research carried out on the SMR test facility of Oregon State University
(Gomez Fernandez et al., 2017). It is a multi-disciplinary test facility used to generate data by
changing the core configurations. The behavior of the system during loss of coolant scenario is
also recorded. Using the generated data, NN was trained so that it can anticipate the best possible
way of behaving i.e., initiating the safety sequences with better precision in different scenarios
(Gomez Fernandez et al., 2017). The use of NN in designing the core and neutronic parameters
of the high conversion small modular reactor (HCSMR) was explored in another study (Janin,
2018). With the trained NN, some key parameters such as the type of fuel, reflector, and the
amount of plutonium in the reactor core were optimized using multi-objective optimization
techniques.
A nuclear power plant's safety is of prime importance. The safety of a NPP can be compromised
due to multiple reasons such as mechanical failure of plant components, fault propagation, power
peaking at undesired rate, etc. To ensure plant safety, these events must be predicted and
prevented beforehand. AI may be utilized for this purpose. The reactor power transient has a
direct relationship to the transient thermo-mechanical load on the structural components,
especially Reactor Pressure Vessel (RPV), nozzles and the piping systems. The use of ANN in
Uncorrected Proof
modeling the behavior of the structural components subjected to transient loading was proposed
by Santucho, Chimenti and Duo (2019). A finite element analysis-based thermo-mechanical
model was developed by Saeed et al. (2020) that utilized neural networks to diagnose fault in IP-
200 SMR. The result of the work suggested that the model is a reasonable strategy for the
diagnosis of the fault in NPPs (Saeed et al., 2020). Yao, Wang and Xie (2022) proposed adaptive
residual convolutional neural networks (ARCNNs) for SMRs to detect different kinds of faults
due to accidents. Chinese lead-based nuclear reactor (CLEAR) was used to generate
experimental data for training the neural network (Yao, Wang and Xie, 2022).
In a boron-free SMR, the reactor power level as well as reactivity is controlled only with the help
of control rods. Thus, the movement of rod rods has a direct link to the power peaking factor
(PPF) of the SMR. Sanchez and dos Santos (2021) demonstrated the use of Support Vector
Machine (SVM), a regression model of AI, to anticipate the PPF in a boron-free SMR. In case of
a natural circulation cooling-type SMR like NuScale, the main task of the safety systems is to
ensure adequate heat transfer from the core. Rahnama and Ansarifar (2021) used an advanced
ANN is used to anticipate the neutronic parameters, as well as thermal-hydraulic variables with
the Al2O3-water nanofluid in Nuscale. The change in safety performance of the reactor due to the
introduction of the nanofluid was investigated (Rahnama and Ansarifar, 2021).
In order to speed up the deployment of advanced nuclear energy technologies such as advanced
ATFs, the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program is a DOE-NE
(Office of Nuclear Energy) initiative (NEAMS, 2022a). There are scopes of multi-scale
modeling in NEAMS suited for ATF design in terms of neutronics, thermal-hydraulics, fuel
performance and structural performance.
Generally, NEAM code MPACT can be used for neutronics. MPACT (NEAMS, 2022b) can be
used effectively under the MOOSE Framework. DOE Fuel Cycle R&D expresses interest in
having a coordinating finding with NEAMS tools and therefore, MPACT code can be utilized.
MPACT uses the subgroup method to generate inline cross sections, with depletion performed
using the ORIGEN program. In order to obtain the higher burnup, deep learning-based
optimization can be performed using lattice geometry optimization (Alam, Goodwin and Parks,
2019b).
Deep Learning Surrogate Model Optimization: The authors are developing a deep learning
surrogate model optimization model, similar to the one referred and studied in a 2020 Cambridge
PhD thesis (Whyte, 2020; Whyte and Parks, 2021a, 2021b). Because fuel geometry optimization
is critical to extending burnup and increasing the power rating, thus enhancing the
competitiveness, this optimization receives special attention. Deep Learning Surrogate Model
Optimization (SMO) process (Whyte, 2020; Whyte and Parks, 2021b) uses Deep Multi Layer
Perceptrons and Convolutional Neural Networks with iterative robust optimization in order to
ensure simultaneous improvement of discharge burnup and power density of the reactor (Fig. 4).
Uncorrected Proof
Optimization method is as follows: (1) A training set will be generated; (2) Deep learning
techniques will be applied to the training set to generate a model; (3) The model will be used to
simultaneously optimize power uprating and burnup extension; and (4) The resulting optimized
solutions in terms of design spaces will be tested using the original accurate neutronics
simulation. Gaussian Process (GP)-based Bayesian optimization can also be implemented for
optimization, as shown in Fig.5.
Fig 4. Deep Learning Surrogate Model Optimization Framework (Whyte, 2020; Whyte and
Parks, 2021b)
Uncorrected Proof
Simulated Quantum Annealing (SQA) Optimization: Adiabatic quantum computers are used, as
is the technique of "quantum annealing," to produce a surrogate model that encodes the heuristics
for optimizing burnup and power density. This model can then be used in modern "quantum
annealers." Heuristic rules (Galperin, 1995) can be used by a quantum annealer (QA)
(Fingerhuth, Babej and Wittek, 2018; Whyte and Parks, 2021a). Quantum adiabatic theory is a
method for global optimizing which can be employed in time dependent Hamiltonian. A logical
qubit is assigned to each of the assembly's fuel pins for this optimization. Convergence is
achieved by converting Ising model connections into actual target QA architectures. The Ising
description allows for the connection qubits regardless of configuration (Whyte and Parks,
2021a), which is referred to as “minor embedding”. Individual qubit is coupled with two adjacent
groups and each group is connected to two adjacent qubits in the box labeled ‘Minor graph
embedding' in Fig. 6. It is necessary to couple several real qubits together in order to simulate the
connectivity of a logical qubit.
Uncorrected Proof
Fig.6: Quantum annealing optimization process (Fingerhuth, Babej and Wittek, 2018; Whyte and
Parks, 2021a).
(b) Thermal-Hydraulics:
The COBRATF (NEAMS, 2022b) thermal-hydraulics subchannel code has been updated as
CTF, and can be used effectively under the MOOSE Framework. It is fully integrated with
MPACT neutronics code. As addressed, since DOE Fuel Cycle R&D expresses interest in having
a coordinating finding with NEAMS tools and therefore, MPACT/CTF coupled code can be
utilized. Solid thermal conduction modeling for tubes, cylinders, and nuclear fuel rods is also
available from CTF. Burnup-dependent fuel temperatures are modelled by CTF Fuel, a nuclear
fuel rod solver. Also, wide ranges of neutronics and thermal-hydraulics codes can be also
coupled for this analysis, as shown in Fig.7 (Alam et al., 2019).
BISON (INL, 2022) is a finite element-based nuclear fuel performance code that can be used for
ATF with SMR. BISON can perform wide range of fuel performance calculations such as fission
gas release, axial fuel swelling, fuel porosity, cladding strain, crack propagation, thermal and
irradiation creep, fracture and so on.
Uncorrected Proof
BISON is now fully coupled to the mesoscale fuel performance code MARMOT. Because of its
foundation in the MOOSE framework, BISON is capable of running on both standard
workstations and high-performance computers.
As referred to our study (Kobayashi et al., 2022), the digital twin framework is proposed to
develop using constructive surrogate model. The authors are developing a new surrogate
modeling tool for the digital twin system for ATF based on the General Atomics Electromagnetic
Systems proposal for the DOE Nuclear Power Program Technologies (Jacobsen, 2022). The
method is proposed by George Jacobsen of General Atomics Electromagnetic Systems in
collaboration with Idaho National Lab (INL) and Los Alamos National Lab (LANL). This digital
twin framework results in various short-term and long-term ATF concepts and their
implementation. In the proposed work, a collaboration with the GE team, INL and LANL would
be important. This is also important to ensure the developed AI-driven and surrogate assisted
digital twin framework is accurate and validating it against experimental data available from
literature and industry. "Virtual twin" of ATF system is the goal of this proposed project.
According to General Atomics Electromagnetic Systems (Jacobsen, 2022), we are developing
digital twin framework following the route (which has also been addressed for hybrid energy
system in another study (Khan et al., 2022): (1) The goal is to build a ML/AI-driven surrogate
model to understand the response of ATF in LWR-based SMR environment, while using simple
equations to keep the model as simple as possible. (2) We can perform surrogate model
validation using a more empirical approach by incorporating relevant data and incorporating it
into existing ATF models/data. (3) Expand the ATF model to incorporate the ATF surrogate
model's behavior. (4) Use data from the system and experiments to figure out and limit model
uncertainties for all models. Fig.8. shows the framework (Kochunas and Huan, 2021; Kobayashi
et al., 2022) for interactions between digital twin and physical assets.
Uncorrected Proof
Fig.8. Framework for interactions between digital twin and physical assets (Kochunas and
Huan, 2021; Kobayashi et al., 2022)
ATF designs have certain uncertainties associated with its input and model parameters e.g.,
uncertainties in the thermal-hydraulics parameters, initial condition, boundary condition, etc.
Since these uncertainties are propagated to the system response, the influence of these
uncertainties must be estimated and described in the final solution of the proposed design to
ensure the quality and reliability of the results. To ensure reliable ATF design, it is necessary to
provide accurate input information in the models. In order to ensure a robust design, uncertainty
quantification (UQ) and sensitivity analysis (SA) (Pepper, Montomoli and Sharma, 2019;
Kumar, Alam, Vučinić, et al., 2020; Kumar, Koutsawa, et al., 2020; Kumar et al., 2021, 2022)
will be performed to ensure that all the design parameters and their variabilities are within the
safety margin. It is also important to consider sensitivity analysis to determine how the proposed
design is affected due to the variability in design variables (Kumar et al., 2019; Kumar, Alam,
Sjöstrand, et al., 2020).
The authors developed a non-intrusive Polynomial Chaos approach for the uncertainty
quantification to understand the variability of a core output parameters with respect to the
fluctuations in the core input values for the ATF design for SMR core response, as shown in
Fig.9. It would be practical to quantify the ‘combined’ effect of uncertainties simultaneously to
realize the real core condition for the proposed SMR concept. This requires a large number of
simulations for UQ and the classical polynomial chaos approach becomes ineffective for this
Uncorrected Proof
concept. To improve the efficiency of polynomial chaos, a sparse polynomial chaos expansion
(SPCE) is being adopted in our developed UQ model. Also, authors developed surrogate
modeling-driven Sobol' indices-based global SA to understand associated sensitivity.
Fig.9. Developed UQ and SA Method (Kumar et al., 2019; Kumar, Alam, Sjöstrand, et al.,
2020)
Overall, machine learning and artificial intelligence-driven multi-scale modeling framework for
high burnup accident-tolerant fuels for light water-based SMR applications is shown in Fig.10.
8. Discussions
Uncorrected Proof
From the preceding sections, the first thing that can be identified is the lack of sufficient data on
the neutronic and thermohydraulic behavior of ATFs when introduced in LW-SMRs. While most
of the studies on ATFs were conducted on the potential use of these fuels in conventional LWRs,
very few of the studies considered introducing ATFs in LW-SMRs. Most of the proposed LW-
SMR designs propose conventional UO2-Zr fuel system (Aydogan, 2016). And experimental
data for SMRs with ATF core is even rarer. Most of the research on the introduction of ATF in a
LW-SMR are based on computational tools. This was somewhat expected since SMRs are still
on the developmental stage and these reactors are yet to be installed to gather sufficient data.
Nevertheless, adequate test facilities for LW-SMRs should be able overcome this data gap.
Another important observation from the literature survey is that AI is yet to be employed for
developing high burnup ATFs. The recent studies employed AI in design optimization of reactor
core, reactor control and monitoring and failure study, but the use of AI in developing new ATFs
for next generation reactors is not observed in the available literature. This may or may not be
due to the lack of available data to train AI for accurate predictions since the concept of ATF has
come in the limelight just a few decades ago. Still, if AI is utilized to its full potential, it may
contribute to the drastic advancement of nuclear fuel technology; like what it has done for other
branches of material science such as degradation analysis, nanomaterial analysis, new material
discovery, etc. (Wei et al., 2019).
Finally, the literature review indicated that the application of ML and AI in nuclear industry is
quite versatile. However, there are certain research domains where AI is yet to be introduced.
One such domain is the prediction of thermohydraulic and neutronic safety parameters of a LW-
SMR fueled with accident tolerant materials. Lack of research and operational data for high-
burnup composite accident-tolerant fuels for LW-based SMR application is a major reason for
the absence of the AI-based research in the available literature. If sufficient data may be
generated for training, AI can be a strong but reliable computational tool to avoid costly
experiments with real-life reactors.
Conclusion
Safety concern is the main obstacle to the implementation of nuclear power in many countries of
the world. Although the possibility of major accident and radioactive contamination of the
surrounding area is very low in the Gen III+ reactors, the fresh wound caused by Fukushima
disaster on the minds of millions of people across the globe is a matter of concern to the nuclear
power industry. This concern is the driving force for the manufacturers of the NPPs to develop
safer reactors for future. A step towards safer Gen IV reactors is the development of accident-
tolerant fuels. Because of the superior resistance to degradation of fission product release, these
fuels are expected to replace conventional fuels soon. And since the small modular reactors are
believed to be the safer and more economic substitutes of their larger versions, the
implementation of ATFs in the SMRs. This literature review attempts to present the picture of
current trend in the research on ATFs and LW-SMRs. The work also discusses the use of
machine learning and artificial intelligence in the field of nuclear engineering and explores the
prospect of exploiting this powerful computational tool to accelerate the development of high
burnup ATFs. From the literature survey, it is identified that the research focused on the
development of ATFs for LW-SMRs are far from realizing the benefits AI because of the lack of
Uncorrected Proof
sufficient data. This data inadequacy can only be overcome through experiments in test SMR
facilities. Further research on high burnup ATFs is required to brough about revolution in the
fuel technology of LW-SMRs.
References:
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/f7623h05t?locale=en
(Accessed: 9 February 2022).
Ebrahimgol, H., Aghaie, M. and Zolfaghari, A. (2021) ‘Evaluation of ATFs in core degradation
of a PWR in unmitigated SBLOCA’, Annals of Nuclear Energy, 152, p. 107961.
Fingerhuth, M., Babej, T. and Wittek, P. (2018) ‘Open source software in quantum computing’,
PloS one, 13(12), p. e0208561.
Galperin, A. (1995) ‘Exploration of the search space of the in-core fuel management problem by
knowledge-based techniques’, Nuclear Science and Engineering, 119(2), pp. 144–152.
Goldner, F. (2012) Development strategy for advanced LWR fuels with enhanced accident
tolerance. USDOE Office of Nuclear Energy (NE).
Gomez Fernandez, M. et al. (2017) ‘Nuclear energy system’s behavior and decision making
using machine learning’, Nuclear Engineering and Design, 324, pp. 27–34. doi:
10.1016/J.NUCENGDES.2017.08.020.
He, Y. et al. (2019) ‘Fuel performance optimization of U3Si2-SiC design during normal, power
ramp and RIA conditions’, Nuclear Engineering and Design, 353, p. 110276.
Hore-Lacy, I. (2010) Nuclear Energy in the 21st Century: World Nuclear University Press.
Elsevier.
Idarraga-Trujillo, I. et al. (2013) ‘Assessment at CEA of coated nuclear fuel cladding for LWRs
with increased margins in LOCA and beyond LOCA conditions’, Top Fuel, 2, pp. 15–19.
Ingersoll, D. T. (2010) Passive safety features for small modular reactors. Oak Ridge National
Lab.(ORNL), Oak Ridge, TN (United States).
INL (2022) Bison: A Finite Element-Based Nuclear Fuel Performance Code. Available at:
https://bison.inl.gov/SitePages/Home.aspx (Accessed: 9 March 2022).
Jacobsen, G. (2022) On the Path to a Nuclear Fuel Digital Twin: Modeling and Simulation of
Silicon Carbide Cladding for Accelerated Fuel Qualification, US Department of Energy.
Available at: https://www.energy.gov/sites/default/files/2021-11/ne-abstract-silicon-112321.pdf.
Janin, D. (2018) ‘Core Design and Optimization of the High Conversion Small Modular
Reactor’. Available at: https://mediatum.ub.tum.de/1439748 (Accessed: 9 February 2022).
Jinia, A. J. et al. (2021) ‘Development of an Artificial Neural Network for Special Nuclear
Material Detection in a Mixed Photon-Neutron Environment’, pp. 1–3. doi:
10.1109/NSS/MIC42677.2020.9507886.
Karoutas, Z. et al. (2018) ‘The maturing of nuclear fuel: Past to Accident Tolerant Fuel’,
Progress in Nuclear Energy, 102, pp. 68–78.
Khan, A. H. et al. (2022) ‘Digital Twin and Artificial Intelligence-based Surrogate Modeling for
Hybrid and Sustainable Energy Systems’, in Handbook of Smart Energy Systems. Springer
Nature.
Kim, H.-G. et al. (2016) ‘Development status of accident-tolerant fuel for light water reactors in
Korea’, Nuclear Engineering and Technology, 48(1), pp. 1–15.
Kim, W.-J., Kim, D. and Park, J. Y. (2013) ‘Fabrication and material issues for the application of
SiC composites to LWR fuel cladding’, Nuclear engineering and technology, 45(4), pp. 565–
Uncorrected Proof
572.
Kobayashi, K. et al. (2022) ‘Digital Twin and Artificial Intelligence Framework for Composite
Accident-Tolerant Fuel for Advanced Nuclear Systems’, in Hnadbook of Smart Energy Systems.
Springer Nature.
Kochunas, B. and Huan, X. (2021) ‘Digital twin concepts with uncertainty for nuclear power
applications’, Energies, 14(14), p. 4235.
Koo, Y.-H. et al. (2014) ‘KAERI’s development of LWR accident-tolerant fuel’, Nuclear
technology, 186(2), pp. 295–304.
Kumar, D. et al. (2019) ‘Influence of nuclear data parameters on integral experiment assimilation
using Cook’s distance’, in EPJ Web of Conferences. EDP Sciences, p. 7001.
Kumar, D., Koutsawa, Y., et al. (2020) ‘Efficient uncertainty quantification and management in
the early stage design of composite applications’, Composite Structures, 251, p. 112538. doi:
10.1016/J.COMPSTRUCT.2020.112538.
Kumar, D., Alam, S. B., Sjöstrand, H., et al. (2020) ‘Nuclear data adjustment using Bayesian
inference, diagnostics for model fit and influence of model parameters’, in EPJ Web of
Conferences. EDP Sciences, p. 13003.
Kumar, D., Alam, S. B., Vučinić, D., et al. (2020) ‘Uncertainty quantification and robust
optimization in engineering’, in Advances in Visualization and Optimization Techniques for
Multidisciplinary Research. Springer, pp. 63–93.
Kumar, D. et al. (2021) ‘Quantitative risk assessment of a high power density small modular
reactor (SMR) core using uncertainty and sensitivity analyses’, Energy, 227, p. 120400.
Kumar, D. et al. (2022) ‘Multi-criteria decision making under uncertainties in composite
materials selection and design’, Composite Structures, 279, p. 114680.
Lam, S. T. et al. (2021) ‘Modeling LiF and FLiBe molten salts with robust neural network
interatomic potential’, ACS Applied Materials & Interfaces, 13(21), pp. 24582–24592.
Li, Q.-J. et al. (2021) ‘Development of robust neural-network interatomic potential for molten
salt’, Cell Reports Physical Science, 2(3), p. 100359.
Li, X. et al. (2019) ‘Assembly-level analyses of an innovative long-life marine SMR loaded with
accident tolerant fuel’, Annals of Nuclear Energy, 133, pp. 227–235.
Manic, M. and Sabharwall, P. (2011) ‘Computational Intelligence as a Tool for Small Modular
Reactors’, in Small Modular Reactors Symposium, pp. 299–310.
Min, H. et al. (2021) ‘Development of an interatomic potential for Fe-He by neural network’,
Computational Materials Science, 196, p. 110549.
Montgomery, R. et al. (2013) ‘Industry-valued design objectives for advanced LWR fuels and
concept screening results’, in Proc. of 2013 LWR Fuel Performance Meeting/TopFuel.
Morgan, D. et al. (2022) ‘Machine learning in nuclear materials research’, Current Opinion in
Solid State and Materials Science, 26(2), p. 100975. doi: 10.1016/J.COSSMS.2021.100975.
Moshkbar-Bakhshayesh, K. and Ghofrani, M. B. (2013) ‘Transient identification in nuclear
power plants: A review’, Progress in Nuclear Energy, 67, pp. 23–32.
NEAMS (2022a) About. Available at: https://neams.inl.gov/about-us (Accessed: 8 March 2022).
Uncorrected Proof
target—a discussion of low-carbon energy policies at province level’, Energy Policy, 39(5), pp.
2740–2747.
Wei, J. et al. (2019) ‘Machine learning in materials science’, InfoMat, 1(3), pp. 338–358. doi:
10.1002/INF2.12028.
Whyte, A. (2020) ‘Surrogate Model Optimisation for PWR Fuel Management’. University of
Cambridge.
Whyte, A. and Parks, G. (2021a) ‘Quantum annealing optimization of a heuristic surrogate
model for pwr fuel loading’, in EPJ Web of Conferences. EDP Sciences, p. 6028.
Whyte, A. and Parks, G. (2021b) ‘Surrogate model optimization of a ‘micro core’pwr fuel
assembly arrangement using deep learning models’, in EPJ Web of Conferences. EDP Sciences,
p. 12003.
Wood, R. T., Upadhyaya, B. R. and Floyd, D. C. (2017) ‘An autonomous control framework for
advanced reactors’, Nuclear Engineering and Technology, 49(5), pp. 896–904.
Wu, X. et al. (2015) ‘Preliminary safety analysis of the PWR with accident-tolerant fuels during
severe accident conditions’, Annals of Nuclear Energy, 80, pp. 1–13.
Yao, Y., Wang, J. and Xie, M. (2022) ‘Adaptive residual CNN-based fault detection and
diagnosis system of small modular reactors’, Applied Soft Computing, 114, p. 108064. doi:
10.1016/J.ASOC.2021.108064.
Zhang, C. et al. (2019) ‘Identification of SNM based on low-resolution gamma-ray
characteristics and neural network’, Nuclear Instruments and Methods in Physics Research
Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 927, pp. 155–160.
doi: 10.1016/J.NIMA.2019.02.023.
Zhang, C. et al. (2021) ‘An enhanced method of neural network algorithm with multi-coupled
gamma and neutron characteristic information for identifying plutonium and uranium’, Nuclear
Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors
and Associated Equipment, 996, p. 165128.
Zinkle, S. J. et al. (2014) ‘Accident tolerant fuels for LWRs: A perspective’, Journal of Nuclear
Materials, 448(1–3), pp. 374–379.