Email RIS file
Bookmark
Report an Issue

Towards Temperature Field Reconstruction, an Innovative Method for Advanced Reactors Monitoring

Author
Coppo Leite, Victor
Published
[University Park, Pennsylvania] : Pennsylvania State University, 2023.
Physical Description
1 electronic document
Additional Creators
Motta, A. T. (Arthur T.)
Access Online

Availability

I Want It
I Want It

Finding items...

Graduate Program
Restrictions on Access
Restricted (PSU Only).
Summary
The present research seeks to demonstrate the capabilities of a physics-informed Convolutional Neural Network (CNN) for reconstructing the temperature field from a limited set of measurements taken at the boundaries of physical domains. The proposed method opens new perspectives to critical limitations of current sensor technologies for deploying the next generation of nuclear reactors. The materials of these reactors are prone to degrading effects caused by high neutron flux and oxidation combined with elevated temperatures, leading to large thermal-mechanical loads that require attention. At the same time, industry experience has shown that direct measurements relying on traditional sensing technologies, such as thermocouples, can be unreliable due to the same adverse conditions. Hence, developing an indirect measurement technique is an existing demand. Here, datasets used to train and test the CNN are obtained upon numerical simulations performed with codes under the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program. The proposed methodology is geometry-agnostic and can be applied to a broad range of conditions, provided that the underlying numerical datasets are accurate. Training data should span a collection of actual operational conditions that define the scope of application of the developed model. To evaluate their performance, the models predict separate test cases not included in the training set but for which they are within their range. Once trained, a CNN model can reconstruct temperature fields in seconds. In comparison, a typical numerical model employed for data generation requires up to 6 minutes of wall time per second simulated. This difference in computation time clearly shows the benefits of using the CNN algorithm for monitoring purposes. Therefore, the proposed methodology is identified as a real-time diagnostics tool. As a proof of concept, the canonical problem of a heated channel was used for extensive testing of the CNN algorithm. Its performance was investigated by considering various coolant types, including liquid metals, water, gases, and molten salts. Sensitivity analyses were also conducted in this simple configuration to demonstrate that the CNN predictions are robust when noise is present in the measurement inputs, which is inevitable in a real application scenario. On top of that, results show that the most standard Multi-Layer Perceptron (MLP) algorithm incurs predictions that are orders of magnitude higher than the CNN. This fact demonstrates the advantages of using the physics-informed CNN algorithm in contrast to an architecture that does not account for any physics formulation, which is the primary motivation for its adoption in the present work. After that, this Dissertation presents how the field predicting capability of the CNN could benefit two promising candidates among the next generation of nuclear reactors. First, the algorithm is tested to reconstruct the temperature fields within the solid region of a High Temperature Gas Reactor (HTGR) fuel assembly. Second, predictions are made for the temperature distribution of the circulating fuel in a Molten Salt Fast Reactor (MSFR). These test cases demonstrate the potential of the CNN-based field reconstruction method to fill existing technological gaps and meet the demands of the industry for accurate temperature monitoring solutions. Ultimately, the corresponding nuclear system's Digital Twin (DT) may likely incorporate such models while they can offer useful information for operations and maintenance (O\&M) recommendations such as detecting temperature spatial derivatives and peaks.
Other Subject(s)
Genre(s)
Dissertation Note
Ph.D. Pennsylvania State University 2023.
Technical Details
The full text of the dissertation is available as an Adobe Acrobat .pdf file ; Adobe Acrobat Reader required to view the file.
View MARC record | catkey: 42248465