MURI Machine learning Enabled Two-pHase flow metrologies, models, and Optimized DesignS (METHODS), ONR Tracking Number 23-000005245
Abstract
Thermal management systems that leverage liquid-to-vapor phase change have the potential for substantial size, weight, and power (SWaP) advantages compared to existing systems. However, the design and optimization of two-phase systems is hindered by the inabilityto model and predict flow and heat transfer with high certainty and generality. Established mechanistic models remain inherently empirical and have good prediction accuracy only for specific flow regimes, geometries, or fluid types. Furthermore, these models are developed and validated using a few global parameters, despite awareness that performance is governed by local spatiotemporal phenomena. Even when high-fidelity data are available, heuristic interpretation has limited physical insights and feedback into the model development process.To address these gaps, a multidisciplinary convergence of liquid-vapor phase change transport modeling domain knowledge, advanced fluid physics metrology, and emerging physics-informed (PhI) machine learning (ML) and computer vision techniques can provide a long-sought unified modeling #language# to predict two-phase flow and heat transfer. The goal of METHODS (Machine learning Enabled Two-pHase flow metrologies, models, and Optimized DesignS) is to leverage PhI ML approaches to enhance fundamental understanding and develop an end-to-end framework to predict phase change during two-phase flows.The project will address longstanding challenges to model framework development by applying PhI ML approaches in a cross-cutting manner across three thrust areas. Thrust 1novel optical metrology and advanced image processing will yield experimental datasets with unprecedented spatial resolution visualizing the dynamics of flow boiling. Essential for model validation and training, these data will include phase and temperature distributions at wall surfaces, thin slices of phase and liquid film thickness distributions, and velocity and temperature fields. The techniques will transform the ability of optical metrologies to probe interfaces, temperatures, and velocities in two-phase flows owing to novel structured illumination schemes, molecular tagging diagnostics, and 3D data augmentation techniques. Thrust 2 leverages deep learning models to detect features in first-of-their-kind empirical data to allow training of PhI ML-based two-phase models. Object detection models will extract detailed statistical information on two-phase flow features to improve understanding of phase-change behavior and its impact on performance. Gray-box models will discover enhanced, thoroughly validated formulations to eliminate empiricism from conservation equations. Training of interfacial transport relations using a physics informed neural network framework,coupled with numerical simulations, will deliver the ability to perform two-phase computational fluid dynamics. In Thrust 3, deep neural operator surrogates will predict two-phase flow at full complexity with high efficiency and uncertainty quantification. To ensure model accessibility, integration within a natural language design framework will offer a virtual assistant to optimize components and system designs. Reduced order models will enable robust two-phase flow forecasting, equipping the design assistant with capabilities to control thermal systems.The envisioned generalized framework for characterizing, modeling, and predicting two-phase flows will provide DoD a toolset that can transform the ability to engineer two-phase systems to their full potential. With access to generalized predictions across all operating conditions and transients, SWaP benefits of two-phase thermal management systems can be safely realized in applications of interest without requiring extensive hardware design build-test cycles. Impact will be equally significant for dual-use commercial applications to mature technical infrastructure that can be leveraged for defense applications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 09, 2024
- Source ID
- N000142412545
Entities
People
- Justin A Weibel
Organizations
- Office of Naval Research
- Purdue University
- United States Navy