Investigation of acoustic characteristics in boiling phenomena and deep learning-based boiling monitoring system using acoustic signals

Abstract

This dissertation addresses the crucial challenge of enhancing nuclear power plant safety and efficiency by advancing the monitoring of boiling phenomena within reactors. Boiling, a fundamental process in nuclear reactors, plays a vital role in heat transfer and is essential for maintaining safe operational temperatures. However, traditional methods like temperature, pressure, and flow rate measurements often fall short in providing direct insights into reactor core conditions during boiling. The urgency of this research is driven by the need for more accurate and efficient monitoring methods to prevent critical issues such as Departure from Nucleate Boiling (DNB), which can lead to fuel damage and, in extreme cases, reactor meltdown. Understanding and controlling the boiling process is essential, as it directly influences heat and hydraulic conditions within the reactor core. The research comprises a series of experimental investigations in pool boiling, flow boiling, and quenching. In pool boiling experiments, meticulous analysis of AE signals during the nucleation and growth of boiling bubbles is conducted. Techniques like the Fast Fourier Transform (FFT) and Short- Time Fourier Transform (STFT) are used to interpret signal frequency and amplitude characteristics, revealing distinct AE signal patterns at various boiling stages. This analysis notes changes in AE signal characteristics with varying heat fluxes and identifies unique patterns at critical heat flux (CHF) phases, highlighting AE signals' potential as a diagnostic tool for different boiling regimes. In flow boiling experiments, a specialized experimental loop is employed to investigate different flow regimes and correlate AE signals with various boiling stages. This analysis provides novel insights into flow boiling phenomena, enhancing the understanding of two-phase heat transfer and its acoustic signatures. Quenching experiments are extended to observe the complete boiling regime, including film boiling, transient boiling, CHF, and nucleate boiling. The focus is on how subcooling influences vapor film collapse during quenching, with findings indicating that higher subcooling correlates with rapid vapor film collapse and intense AE signals, offering key insights into boiling dynamics and AE signal characteristics. This research introduces an approach in boiling heat transfer analysis using deep learning techniques, primarily employing Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs), such as ResNet models, to classify boiling regimes from AE signals. The effectiveness of a ResNet- based model in analyzing and predicting complex, nonlinear data patterns typical of boiling phenomena is demonstrated. Models are trained with extensive datasets from various boiling experiments, achieving high accuracy in predicting boiling heat transfer regimes, heat flux, and Heat Transfer Coefficients (HTCs). This study also explores the practical application of the developed AE-based boiling monitoring technology in two distinct case studies: the thermal-hydraulic integral effect test facility (ITE) and the high-pressure flow boiling facility. The technology's effectiveness in diagnosing two-phase fluid flow phenomena within these complex systems is assessed. In ITE, AE signal analysis combined with deep learning's capability to accurately distinguish between single-phase and two-phase flows is demonstrated. The high-pressure flow boiling facility case further validates the method's applicability in high-pressure environments, successfully detecting nucleate boiling at higher heat fluxes. These investigations highlight AE-based monitoring's potential as a tool for real-time analysis and safety assurance in nuclear power plants and other critical industrial applications. In summary, this research represents a significant advancement in nuclear reactor monitoring, merging Acoustic Emission (AE) signal analysis with deep learning. Key achievements include the development of precise deep learning models for boiling regime classification and heat flux prediction, and the effective application of AE signal analysis in complex environments. The study overcomes challenges in real-world applications, demonstrating this technology's potential in nuclear contexts. Future directions involve enhancing AE sensor capabilities, implementing robust noise cancellation, expanding deep learning datasets, exploring new deep learning architectures, conducting real-field testing, fostering interdisciplinary collaboration, and engaging with regulatory bodies. This work lays the groundwork for future exploration in AE signal analysis and deep learning in nuclear reactor monitoring, aiming to enhance the safety and efficiency of nuclear power plants.