I-Corps: Non-Intrusive Cooling System Fault Detection Using Deep Learning of Acoustic Emissions

这个I-Corps项目更广泛的影响/商业潜力是开发一种数据驱动的方法，用于高性能冷却系统的可靠运行，这将突破高功率密度电子和能源系统的热限制。该技术不仅将对基础热传输研究产生影响，还将对电动汽车、混合动力汽车、电力电子、发电等的工业生产产生影响。这项技术将有利于经济和环境。这我-Corps项目的基础是开发准确可靠的技术，通过将高速成像、声发射（AE）与使用深度学习（DL）的多模态融合相结合，对高性能两相冷却系统进行故障检测。该技术具有快速，实时，在线监测和检测，提供极快的数据/图像处理，约为1-10 ns/帧。 创新点是非侵入性故障检测，使用连接到加热器的接触式AE传感器，浸入工作流体中的水听器，以及沸腾室外部的高速成像。该创新通过基于机器学习的降维和多模态信号的融合使用数据驱动的异常识别。我们的技术将利用优化的模型架构为每一种形式的信号，例如，该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is the development of a data-driven approach for reliable operations of high-performance cooling systems that will break through the thermal limitations for high-power-density electronic and energy systems. The proposed technology will not only make an impact on fundamental thermal transport research but also industrial production of electric vehicles, hybrid vehicles, power electronics, power generation, etc. By enabling more efficient cooling, reduced energy consumption, and reduced carbon oxide emissions, this technology will benefit both the economy and the environment.This I-Corps project is based on the development of accurate and reliable technology for fault detection of high-performance two-phase cooling systems by coupling high-speed imaging, acoustic emission (AE), with multimodal fusion using deep learning (DL). The technology has fast, real-time, online monitoring and detection providing extremely fast processing of data/images, in the order of 1-10 ns/frame. The innovation is non-invasive fault detection using contact AE sensor attached to the heater, hydrophones immersed in the working fluid, and high-speed imaging outside the boiling chamber. The innovation uses data-driven anomaly identification via machine learning-based dimensionality reduction, and fusion of multimodal signals. Our technology will leverage optimized model architectures for each modality of signals, e.g., convolutional neural networks for images, recurrent neural networks for time-series signals, and fuse the signals by concatenating the output layer of each channel.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.