Research On Recognition Algorithm Of Droplet Atomization Pattern Based On Convolutional Neural Network

The droplets formed by high-voltage electrostatic atomization have the advantages of controllable droplet size,small droplet size,single size,and good deposition performance,and are widely used in many engineering fields.There are high requirements for the use of liquid atomization modes in different fields.The formation and stability of droplet atomization modes are affected by many factors such as the characteristics of the liquid medium,the applied voltage and the external environment,and it depends on preset conditions.It is difficult to control the stability of the atomization mode and the effective generation of charged droplets.Aiming at the problems in the current droplet atomization mode control,this paper uses machine vision and neural network tools to build a droplet atomization pattern recognition model,which is real-time Provide reference for analysis and atomization mode control.In this work,a deep learning-based convolutional neural network droplet atomization pattern recognition model is proposed.Through image processing and analysis of a single frame image of the droplet,the Le Net-5 network is used to construct the droplet atomization pattern recognition model.Experiments show that,The model's recognition accuracy of the droplet atomization pattern image reached 94.9%.Aiming at the problem of continuous and periodic droplet information extraction in the change of droplet morphology under the action of high-voltage electric field,a method for constructing a convolutional neural network recognition model for multi-frame droplet atomization mode video sequences is proposed,using three volumes The construction method of the product neural network model is to extract the droplet image information characteristics under continuous frames to realize the droplet atomization mode recognition model of different network depths.After the droplet atomization mode video image data set under the two flow rates,three The recognition accuracy rates of the two methods are 99%,99.29% and 95.62% respectively.Finally,an electrostatic atomization droplet atomization model experimental analysis system is built,and the algorithm of the pattern recognition model is carried out based on the experimental data of the electrostatic atomization droplet atomization model.Test evaluation,optimize the matching of experimental parameters and network parameters to obtain an optimized identification model.The research results are useful for studying the relationship between the working parameters of electrostatic atomization and the droplet atomization mode,optimizing the operating parameters and improving the stability of the atomization mode,which lays the foundation for the next step of combining with the intelligent control system to adjust the operating parameters in real time and accurately.