서브메뉴
검색
Impurity Gas Detection for Spent Nuclear Fuel (SNF) Canisters Using Ultrasonic Sensing and Deep Learning
Impurity Gas Detection for Spent Nuclear Fuel (SNF) Canisters Using Ultrasonic Sensing and Deep Learning
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152823
- ISBN
- 9798384033851
- DDC
- 621
- 저자명
- Zhuang, Bozhou.
- 서명/저자
- Impurity Gas Detection for Spent Nuclear Fuel (SNF) Canisters Using Ultrasonic Sensing and Deep Learning
- 발행사항
- [Sl] : University of Southern California, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 218 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-03, Section: B.
- 주기사항
- Advisor: Gencturk, Bora.
- 학위논문주기
- Thesis (Ph.D.)--University of Southern California, 2024.
- 초록/해제
- 요약The operation of nuclear power plants (NPPs) produces large amounts of high-level radioactive waste known as spent nuclear fuel (SNF). Currently, SNF is stored in dry cask storage systems (DCSSs) for extended temporary storage. During the extended temporary storage, the DCSS, particularly the SNF canisters, may degrade and abnormal conditions may occur. Therefore, non-destructive evaluation (NDE) and machine learning (ML) approaches are considered for inspection of SNF canisters.The SNF canisters are designed to maintain a helium-sealed environment throughout the extended dry storage. However, impurity gases may be mixed with helium due to internal material degradation or canister leakage. Therefore, the presence of impurity gases in helium indicates potential structural degradation. The objective of this study is to develop a robust NDE technique based on ultrasonic sensing and deep learning to detect impurity gases in SNF canisters. The method involves coupling ultrasonic excitation into the gas through the canister wall and capturing the signal with a receiver on the other side. However, due to the acoustic impedance difference between the steel and gas, a strong structural noise signal interferes with the gas signal, making it challenging to detect the gas-borne signal. This study conducted a comprehensive literature review on NDE and ML applications for SNF canisters, proposed an active noise cancellation method (ANC) and developed supervised and unsupervised ML frameworks for impurity gas detection. Contributions from this study hold promises in field applications to inspect thousands of canisters with SNF that are bound to be transported to temporary or permanent storage in the near future.Sixteen NDE methods were examined and compared, including visual inspection, ultrasonic guided waves, laser-based techniques, AE, non-invasive acoustic sensing, dynamic modal testing, ECT, cosmic ray muon tomography, neutron imaging, gamma ray detection, fiber optical sensors, through-wall communications, X-ray CT, vibrothermography, monoenergetic photon sources, and SAW sensors. The TRL for each NDE method was assessed and compared. Visual inspection, ultrasonic guided waves, laser-based approaches, AE, and ECT were found to have higher TRLs than other methods due to their demonstrations on actual or full-scale DCSSs.The proposed ultrasonic sensing concept was verified on a full-scale canister mock-up. The gas-borne signals were successfully detected by using damping materials and blocking and unblocking tests. Acoustic impedance matching (AIM) layers were proposed enhance the gas signals. The ultrasonic sensing method was reliable in detecting gas temperature and humidity variations.An efficient ANC method was developed to enhance the signal-to-noise ratio (SNR) of the gas signal. The ANC method was validated on small-scale specimens, and a full-scale canister mock-up. The proposed ANC method and its linear and time-invariant (LTI) assumption were tested and found to be valid on the steel plate and tube. The ANC method could cancel more than 90% of the noise signal amplitude in any time window of interest. The cancellation performance was found to be independent of the location of the canceling transducers and the selection of the silence window. This allows flexibility when placing the canceling transducers. The ANC method was verified on the full-scale canister mock-up. It was found that the ANC method can cancel the structural noise by 38.15% using two canceling transducers. The SNR was improved by 213.6% compared with the SNR of no active noise cancellation.This study established two experimental data sets for analyzing impurity gases (i.e., air and argon) in helium using ultrasonic sensing. Artificial neural networks (ANNs) and convolutional neural networks (CNNs) with model uncertainties were utilized to solve forward and inverse problems, respectively. The forward problem used ANNs to predict the response and time-of-flight given the excitation and gas concentrations. The inverse problem, on the other hand, employed probabilistic CNNs to predict the impurity concentrations based on the ultrasonic response and excitation. The proposed data-driven approaches can analyze three-component gas mixtures solely based on ultrasonic time series data. This eliminates the need for additional acoustic measurements such as acoustic attenuation.An ultrasonic data set was collected by sealing a 2/3-scaled canister mock-up and introducing up to 1.53% argon or 1.29% air into the helium background gas. Results showed that the time-of-flight (TOF) method had sufficient resolution to detect abnormal gas concentrations of less than 1.0%. The differential method demonstrated a periodic in- and out-of-phase behavior between the benchmark (i.e., pure helium) and abnormal (i.e., with argon or air) state signals. The variational auto-encoder (VAE) and the Wasserstein auto-encoder (WAE) were trained on the benchmark data and were applied directly to the abnormal state data. Both the VAE and the WAE were able to distinguish the benchmark and abnormal states of the canister mock-up based on the reconstruction error.A heater was sealed in the canister mock-up and sulfur hexafluoride (SF6) was used as the surrogate fission gas. Ultrasonic signals were collected non-invasively from the canister surface. A temperature-compensated TOF method and an envelope energy method were proposed. Research results showed that SF6 concentrations as low as 0.026% can be detected using the temperature-compensated TOF method. This resolution corresponds to xenon released from 58 % of a fuel assembly. However, the TOF method will break down for concentrations above 0.755 % due to TOF information loss. In contrast, the envelope energy method quantifies the amplitude reduction resulting from higher attenuation of fission gas and proved to be temperature independent. This method is effective in detecting the existence of higher concentrations, yet its ability to quantify concentrations above 0.866 % may be compromised due to the existence of structural noise.In conclusion, this study investigates the ultrasonic sensing and deep learning to characterize the internal gas environment for SNF canisters. The Matlab and Python codes developed from this dissertation are publicly available on GitHub (https://github.com/bozhouzh/PhD_Dissertation). Future research can focus on multi-modal NDE and data fusion, digital-to-real adaptation with digital twins, real-time and automated inspection with artificial intelligence (AI), and large database for the monitoring and diagnosis of DCSSs.
- 일반주제명
- Energy
- 일반주제명
- Nuclear engineering
- 키워드
- Deep learning
- 기타저자
- University of Southern California Civil Engineering
- 기본자료저록
- Dissertations Abstracts International. 86-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164033
■00520250211152823
■006m o d
■007cr#unu||||||||
■020 ▼a9798384033851
■035 ▼a(MiAaPQ)AAI31559827
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621
■1001 ▼aZhuang, Bozhou.
■24510▼aImpurity Gas Detection for Spent Nuclear Fuel (SNF) Canisters Using Ultrasonic Sensing and Deep Learning
■260 ▼a[Sl]▼bUniversity of Southern California▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a218 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-03, Section: B.
■500 ▼aAdvisor: Gencturk, Bora.
■5021 ▼aThesis (Ph.D.)--University of Southern California, 2024.
■520 ▼aThe operation of nuclear power plants (NPPs) produces large amounts of high-level radioactive waste known as spent nuclear fuel (SNF). Currently, SNF is stored in dry cask storage systems (DCSSs) for extended temporary storage. During the extended temporary storage, the DCSS, particularly the SNF canisters, may degrade and abnormal conditions may occur. Therefore, non-destructive evaluation (NDE) and machine learning (ML) approaches are considered for inspection of SNF canisters.The SNF canisters are designed to maintain a helium-sealed environment throughout the extended dry storage. However, impurity gases may be mixed with helium due to internal material degradation or canister leakage. Therefore, the presence of impurity gases in helium indicates potential structural degradation. The objective of this study is to develop a robust NDE technique based on ultrasonic sensing and deep learning to detect impurity gases in SNF canisters. The method involves coupling ultrasonic excitation into the gas through the canister wall and capturing the signal with a receiver on the other side. However, due to the acoustic impedance difference between the steel and gas, a strong structural noise signal interferes with the gas signal, making it challenging to detect the gas-borne signal. This study conducted a comprehensive literature review on NDE and ML applications for SNF canisters, proposed an active noise cancellation method (ANC) and developed supervised and unsupervised ML frameworks for impurity gas detection. Contributions from this study hold promises in field applications to inspect thousands of canisters with SNF that are bound to be transported to temporary or permanent storage in the near future.Sixteen NDE methods were examined and compared, including visual inspection, ultrasonic guided waves, laser-based techniques, AE, non-invasive acoustic sensing, dynamic modal testing, ECT, cosmic ray muon tomography, neutron imaging, gamma ray detection, fiber optical sensors, through-wall communications, X-ray CT, vibrothermography, monoenergetic photon sources, and SAW sensors. The TRL for each NDE method was assessed and compared. Visual inspection, ultrasonic guided waves, laser-based approaches, AE, and ECT were found to have higher TRLs than other methods due to their demonstrations on actual or full-scale DCSSs.The proposed ultrasonic sensing concept was verified on a full-scale canister mock-up. The gas-borne signals were successfully detected by using damping materials and blocking and unblocking tests. Acoustic impedance matching (AIM) layers were proposed enhance the gas signals. The ultrasonic sensing method was reliable in detecting gas temperature and humidity variations.An efficient ANC method was developed to enhance the signal-to-noise ratio (SNR) of the gas signal. The ANC method was validated on small-scale specimens, and a full-scale canister mock-up. The proposed ANC method and its linear and time-invariant (LTI) assumption were tested and found to be valid on the steel plate and tube. The ANC method could cancel more than 90% of the noise signal amplitude in any time window of interest. The cancellation performance was found to be independent of the location of the canceling transducers and the selection of the silence window. This allows flexibility when placing the canceling transducers. The ANC method was verified on the full-scale canister mock-up. It was found that the ANC method can cancel the structural noise by 38.15% using two canceling transducers. The SNR was improved by 213.6% compared with the SNR of no active noise cancellation.This study established two experimental data sets for analyzing impurity gases (i.e., air and argon) in helium using ultrasonic sensing. Artificial neural networks (ANNs) and convolutional neural networks (CNNs) with model uncertainties were utilized to solve forward and inverse problems, respectively. The forward problem used ANNs to predict the response and time-of-flight given the excitation and gas concentrations. The inverse problem, on the other hand, employed probabilistic CNNs to predict the impurity concentrations based on the ultrasonic response and excitation. The proposed data-driven approaches can analyze three-component gas mixtures solely based on ultrasonic time series data. This eliminates the need for additional acoustic measurements such as acoustic attenuation.An ultrasonic data set was collected by sealing a 2/3-scaled canister mock-up and introducing up to 1.53% argon or 1.29% air into the helium background gas. Results showed that the time-of-flight (TOF) method had sufficient resolution to detect abnormal gas concentrations of less than 1.0%. The differential method demonstrated a periodic in- and out-of-phase behavior between the benchmark (i.e., pure helium) and abnormal (i.e., with argon or air) state signals. The variational auto-encoder (VAE) and the Wasserstein auto-encoder (WAE) were trained on the benchmark data and were applied directly to the abnormal state data. Both the VAE and the WAE were able to distinguish the benchmark and abnormal states of the canister mock-up based on the reconstruction error.A heater was sealed in the canister mock-up and sulfur hexafluoride (SF6) was used as the surrogate fission gas. Ultrasonic signals were collected non-invasively from the canister surface. A temperature-compensated TOF method and an envelope energy method were proposed. Research results showed that SF6 concentrations as low as 0.026% can be detected using the temperature-compensated TOF method. This resolution corresponds to xenon released from 58 % of a fuel assembly. However, the TOF method will break down for concentrations above 0.755 % due to TOF information loss. In contrast, the envelope energy method quantifies the amplitude reduction resulting from higher attenuation of fission gas and proved to be temperature independent. This method is effective in detecting the existence of higher concentrations, yet its ability to quantify concentrations above 0.866 % may be compromised due to the existence of structural noise.In conclusion, this study investigates the ultrasonic sensing and deep learning to characterize the internal gas environment for SNF canisters. The Matlab and Python codes developed from this dissertation are publicly available on GitHub (https://github.com/bozhouzh/PhD_Dissertation). Future research can focus on multi-modal NDE and data fusion, digital-to-real adaptation with digital twins, real-time and automated inspection with artificial intelligence (AI), and large database for the monitoring and diagnosis of DCSSs.
■590 ▼aSchool code: 0208.
■650 4▼aEnergy
■650 4▼aNuclear engineering
■653 ▼aDeep learning
■653 ▼aNon-destructive testing
■653 ▼aSpent nuclear fuel
■653 ▼aUltrasonic sensing
■653 ▼aNuclear power plants
■690 ▼a0543
■690 ▼a0791
■690 ▼a0800
■690 ▼a0552
■71020▼aUniversity of Southern California▼bCivil Engineering.
■7730 ▼tDissertations Abstracts International▼g86-03B.
■790 ▼a0208
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164033▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
