Learning and predicting flow boiling principles: A deep learning perspective - ONR Tracking #21-0000
Abstract
In pursuing high-performance electronics for DoD applications, it is crucial to advance the fundamental understanding of flow boilin,g related to these applications multiphase heat transfer, fluid dynamics, and surface phenomena. However, learning flow boiling pri,nciples has been a significant challenge as boiling involves complex interfacial phenomena between liquid and vapor phases, resultin,g in the formation of dynamic bubbles. Recent innovations in artificial intelligence have made it possible to incorporate machine le,arning models into research on flow boiling under a new paradigm. Above all, machine vision can demonstrate the precise and automati,c object prototyping of individual instances to extract physically meaningful features from dynamic bubbles, which has been nearly i,mpossible with conventional image processing methodologies. The graphical data collection from multiple inputs (i.e., conventional i,mages and neuromorphic event strings) will allow us to connect them with heat transfer performance and find hidden mechanisms that w,ould represent a game-changing innovation for thermofluidic engineering. Therefore, this projects overarching objective is to provi,de a holistic description of dynamic flow boiling physics and to push the knowledge boundaries of thermofluidic science. To achieve, this goal, we propose a data-driven framework that integrates experiments, neuromorphic imaging, machine vision, data processing, a,nd state-of-the-art machine learning methods. Our framework devotes special attention to collecting thermofluidic datasets using bot,h conventional and neuromorphic imaging sensors (Task 1: Experimental data collection) and autonomously extracting physically interp,retable features from live images and event strings (Task 2: Autonomous feature extraction). The proposed research activities will i,lluminate flow boiling mechanisms beyond previous boundaries by taking a data-centric analysis based on the visualization of flow bu,bbles characteristics such as instabilities (Task 3: Advancing flow boiling physics). The integrated research tasks will allow us t,o build a transferable, scalable, and computationally efficient deep bubble Map that predicts flow boiling under unseen conditions (,Task 4: Transferable deep learning framework). The success of the proposed framework will provide a guideline for designing input sp,aces that are suitable for DoD applications, even in the presence of flow boiling. The framework establishes general principles and, can be implemented in other phase change processes, such as condensation, flow condensation, and pool boiling, thereby advancing kn,owledge in the thermofluidic community.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 14, 2022
- Source ID
- N000142212063
Entities
People
- Yoonjin Won
Organizations
- Office of Naval Research
- United States Navy
- University of California, Irvine