Visualization-based nucleate boiling heat flux quantification using machine learning
Visualization-based nucleate boiling heat flux quantification using machine learning
复制标题
DOI：
10.1016/j.ijheatmasstransfer.2018.12.170
复制
发表时间：
2019-05-01
影响因子：
5.2
通讯作者：
da Silva, Alexandre K.
中科院分区：
文献类型：
--
作者：
Hobold, Gustavo M.;da Silva, Alexandre K.
涉及复杂现象的过程在自然界和工业中无处不在，其中许多难以通过计算模拟。例如，有核沸腾传热有许多实际应用，而膜沸腾是一种不希望的操作制度。到目前为止，量化沸腾传热的大多数关联和计算机模拟依赖于热水力数据的直接测量，例如加热器温度，这通常是侵入性的。本文证明了基于神经网络的模型可以仅使用沸腾现象的直接和间接视觉信息来量化传热，而不需要任何控制方程的先验知识，这使得基于沸腾过程成像的热流密度非侵入式测量成为可能。结果表明，与精确的实验测量结果相比，神经网络可以编码气泡形态及其与热流密度的相关性，返回误差低至7%，比现有的沸腾传热预测方法有了显著改进。此外，研究表明，这些系统可以在廉价、紧凑的计算机中实现，如树莓派，从可视化中实时推断热通量。(C) 2019 Elsevier Ltd.版权所有。
Processes involving complex phenomena are ubiquitous in nature and industry, many of which are difficult to simulate computationally. Nucleate boiling heat transfer, for instance, has numerous practical applications, while the film boiling is an undesirable operation regime. So far, most correlations and computer simulations to quantify boiling heat transfer rely on direct measurement of thermohydraulic data, such as heater temperature, which is often invasive. Here it is demonstrated that neural network-based models can quantify heat transfer using only direct and indirect visual information of the boiling phenomenon, without any prior knowledge of the governing equations, which enables the non-intrusive measurement of heat flux based on boiling process imaging. It is shown that neural networks can encode bubble morphology and its correlation with heat flux returning errors as low as 7% when compared with precise experimental measurements, a significant improvement over current prediction methods of boiling heat transfer. Furthermore, it is shown that these systems may be implemented in inexpensive, compact computers, such as the Raspberry Pi, to infer heat flux in real time from visualization. (C) 2019 Elsevier Ltd. All rights reserved.