Artificial Neural Network Based Prediction Models For Pneumatic Micro-deoplet Ejection And Their Applications

Pneumatic droplet ejection is a kind of liquid microloading technology.The working process is to make the high-pressure gas with pressure P0 at the front end of the solenoid valve rush into the liquid storage chamber in the form of pulse through the transient conduction?time for?t?solenoid valve,causing the air pressure oscillation in the liquid storage chamber.The pressure oscillation waveform P?t?forces the liquid sample to be ejected through the tiny nozzle to form microdroplets.This technology is widely used in various manufacturing fields,especially in the field of bio-printing.The survival rate of cells in the jet bio-ink is nearly 100%,which has great application potential.This paper designs and builds a set of pneumatic microdrop ejection device.This device can be set P0 and?t and reference for electromagnetic valve drive signal rise along the time,delay time taken ejection condition.At the same time,high speed air pressure sensor is used to collect P?t?in the fluid storage cavity of each ejection.By establishing a sound system of electrical equivalent model,illustrates in ideal condition,the relationship between the control variables P0 and?t of the device and the pressure oscillation waveform P?t?in the liquid storage chamber.It is concluded that,in practice,due to the impact of gas path effect and electromagnetic valve action inconsistencies,control variable P0 and?t contains large interference,poor consistency of P?t?.Since P?t?directly governs the ejection state of droplets,the consistency of the ejection state of droplets is greatly affected.The interference cannot be eliminated under the existing conditions,but P?t?can be measured in real time,so it is a useful attempt to establish a prediction model of ejection state based on P?t?.In this paper,the distance Hd between the droplet and the nozzle after a delay of ejection is used as the physical quantity to evaluate the ejection state of the droplet.In the experiment,two methods were used to predict the microdrop position Hd.One is the traditional statistical methods,this method under the same working conditions,by collecting a large number of droplets location Hd data,will this group of data of average???₍?9??as droplets location prediction.The standard deviation of the prediction error of this method is about 350?m,while the diameter of ordinary microdroplets is about500?m.The other method uses the actually measured P?t?in the liquid storage chamber as the input to establish the prediction model of the ejection position Hd of the microdroplets based on BP neural network.The standard deviation of the prediction error is only 70?m,which is about 5 times more accurate than the statistical prediction method.Meanwhile,the input of this method can be either the discrete sampling point P?ti?in the time domain or the frequency component P?fi?in the frequency domain.Another experiment shows that the BP neural network model can also effectively predict the amount of Nd produced by the droplets,with an accuracy of 99.9%.Pneumatic microdroplet ejection is mainly used in the printing of biological materials and other non-standard inks.Ink fluid characteristics often change over time,resulting in poor print quality.In the control variable P0 and?t unchanged,under the condition of droplet ejection state change is an important symbol of fluid characteristics change.However,the subtle changes in the ejection state of droplet caused by the change of fluid characteristics will be submerged under the influence of the interference of the device itself,which is difficult to be detected by the existing technology in real time.As the BP neural network prediction model is trained,P?t?as input already contains the influence of interference on the device.Therefore,compared with the prediction model based on statistics,the model combined with droplets image detection can more sensitively and accurately detect the change of fluid characteristics in biological ink under large interference.This innovative approach makes it possible for the pneumatic droplet ejection device to monitor the fluid characteristics in the bio-ink in real time.