Online, quasi-real-time analysis of high-resolution, infrared, boiling heat transfer investigations using artificial neural networks

Highlights: • We develop a machine learning methodology to analyze boiling experiments. • It is as accurate as image processing algorithms but much faster. • It can be used online and quasi-real-time. • Its capabilities are independent of the operating heat flux and surface morphology. -- Abstract: We present a machine learning methodology that can be used online and quasi-real-time (i.e., as fast as we can practically run an experiment) to accelerate the analysis of infrared, boiling heat transfer investigations. Precisely, we use feed-forward artificial neural networks with one layer of hidden neurons to measure bubble growth time, bubble period, and nucleation site density directly from the radiation recorded by the high-speed infrared camera. We test and validate the methodology against saturated pool boiling experiments with water, run on both plain and nanoengineered surfaces. Using such a technique, we have measurements of the quantities above within a few seconds from the moment the camera records the boiling surface radiation, with a regression coefficient of 0.95 or higher compared to reference measurements obtained by conventional, time-consuming, image processing techniques.