Machine Learning Assisted Spraying Pattern Recognition for Electrohydrodynamic Atomization System

Abstract

In this work, two machine learning (ML) models for recognizing electrohydrodynamic (EHD) spraying patterns are developed to guide the process operation to achieve stable cone-jet mode directly. To this end, the EHD spraying patterns are first divided into three categories, namely, dripping, stable cone-jet, and unstable cone-jet. A database consisting of 86 140 EHD spraying patterns data are used to build the recognition models. The artificial neural network (ANN) and support vector machine (SVM) models are trained through the collected data to give excellent performance in prediction of EHD spraying patterns. With the testing set data, the accuracy of the EHD spraying patterns recognition is 99.611% for the ANN model, and the result is 99.867% for the SVM model. The performance of these two models are further evaluated by comparing the recognition rates of the EHD spraying patterns with experimental data from 22 literature references, and the results show that the SVM model gives a better performance. Lastly, the SVM model is employed to predict the full picture of EHD spraying patterns. Four pattern maps are drawn with the assistance of the SVM model to reveal the effect of volumetric flow rate (Q), tip-to-collector distance (L), polymer concentration (C), nozzle inner diameter (D in), and applied voltage (V).

A Review on Electrohydrodynamic (EHD) Pump

“…The high cost of commercial software has encouraged some researchers to use open-source software, although COMSOL Multiphysics is used to conduct the majority of simulations. In addition, in future research, due to the recent rapid development of machine learning and increasing open-source projects, the authors believe that neural networks can be used to explore the relationship between various parameters of the working fluid and the operation performance of the EHD pump [ 46 , 47 ]. After the vigorous development of computing resources, more research and applications have been produced on the EHD pump.…”

A Review on Electrohydrodynamic (EHD) Pump

“…Thermal management [44], flow analysis [45] ANSYS/FLUENT increasing open-source projects, the author believes that neural networks can be used to 192 explore the relationship between various parameters of the working fluid and the operation 193 performance of the EHD pump [46,47]. After the vigorous development of computing 194 resources, more research and applications exist on the EHD pump.…”

Review of Machine Learning for Hydrodynamics, Transport, and Reactions in Multiphase Flows and Reactors

“…203 In addition to bubble detection above, another increasingly popular trend is to detect particle and droplet characteristics by ML. 204,245 Li et al investigated the nonspherical biomass particles and spherical polyethylene particles in a lab-scale fluidized bed using PIV and PTV techniques. 205 The ML pixel-wise classification methodology was trained and used to acquire particle masks for PIV and PTV processing.…”

“…To be easier and more flexible for users, some tools modules such as a fuzzy inference system have been thereby developed for bubble mask extraction with a friendly graphic user interface . In addition to bubble detection above, another increasingly popular trend is to detect particle and droplet characteristics by ML. , Li et al investigated the nonspherical biomass particles and spherical polyethylene particles in a lab-scale fluidized bed using PIV and PTV techniques . The ML pixel-wise classification methodology was trained and used to acquire particle masks for PIV and PTV processing.…”