CN117634581A - Device and method for predicting pneumatic droplet ejection state based on BP neural network - Google Patents
Device and method for predicting pneumatic droplet ejection state based on BP neural network Download PDFInfo
- Publication number
- CN117634581A CN117634581A CN202311683612.0A CN202311683612A CN117634581A CN 117634581 A CN117634581 A CN 117634581A CN 202311683612 A CN202311683612 A CN 202311683612A CN 117634581 A CN117634581 A CN 117634581A
- Authority
- CN
- China
- Prior art keywords
- electromagnetic valve
- speed electromagnetic
- neural network
- droplet
- air pressure
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 63
- 238000000034 method Methods 0.000 title claims abstract description 53
- 239000007788 liquid Substances 0.000 claims abstract description 89
- 238000002347 injection Methods 0.000 claims abstract description 27
- 239000007924 injection Substances 0.000 claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 23
- 238000010606 normalization Methods 0.000 claims description 32
- 210000002569 neuron Anatomy 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 15
- 238000012544 monitoring process Methods 0.000 claims description 13
- 230000001105 regulatory effect Effects 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 238000009423 ventilation Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 230000001276 controlling effect Effects 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 230000005484 gravity Effects 0.000 claims description 4
- 238000005286 illumination Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 23
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 239000000976 ink Substances 0.000 description 4
- 230000000630 rising effect Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 239000002184 metal Substances 0.000 description 3
- 238000007639 printing Methods 0.000 description 3
- 210000004027 cell Anatomy 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000010146 3D printing Methods 0.000 description 1
- 238000005842 biochemical reaction Methods 0.000 description 1
- 230000003833 cell viability Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000007641 inkjet printing Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Landscapes
- Application Of Or Painting With Fluid Materials (AREA)
Abstract
The invention discloses a device and a method for predicting the pneumatic droplet ejection state based on a BP neural network. The method takes the time-varying signal I (t) of the solenoid current flowing through the high-speed solenoid valve as the input of the neural network and predicts the droplet ejection state parameter, namely the number of dropletsN D And a falling position H D The method comprises the steps that a neural network is built in a control and data processing system, wherein the control and data processing system comprises an upper computer and a lower computer; the invention uses the solenoid current I (t) of the high-speed electromagnetic valve, the liquid level height h of the liquid storage cavity and the air pressure P at the front end of the high-speed electromagnetic valve 0 As a feature, a new method of predicting injection state parameters through a neural network.
Description
Technical Field
The invention belongs to the field of droplet ejection, and particularly relates to a pneumatic droplet ejection system controlled by an electromagnetic valve and a droplet ejection state prediction method based on a BP neural network. The method can effectively predict the droplet ejection state by establishing a prediction model, and can be used for predicting the droplet state of the pneumatic droplet ejection device, and monitoring and controlling in real time.
Background
Droplet ejection techniques are widely used in many fields, such as in the fields of inkjet printing, printed electronics, 3D printing, etc. In addition, droplet ejection techniques are often used in the biomedical field. For precious or scarce samples, micro sample dispensing can greatly reduce sample usage. The droplet-based biochemical reaction is faster due to the larger surface area to volume ratio of the droplets. The main drop-on-demand jetting technologies mainly comprise two types of technologies, namely, thermal bubble driving and piezoelectric ceramic driving. Some non-conventional droplet ejection techniques have developed liquids in recent years. Pneumatic droplet ejection is simple to operate and is suitable for samples of various viscosity and temperature ranges, as compared to other modes of ejection. Pneumatic droplet ejection devices are applied to the ejection of metal droplets and enable 3D metal printing based on the metal droplets. Pneumatic droplet ejection is also applied to the printing of biological inks. Initially for cell printing, it achieves nearly 100% cell viability, making it a potential choice for dispensing cell-loaded biomedical samples.
Common pneumatic droplet ejection devices include droplet ejection systems, visual monitoring of droplet ejection status systems.
The droplet ejection system includes: the liquid storage cavity, the nozzle, the high-speed electromagnetic valve, the pressure regulating valve and the ventilation pipeline (comprising an air inlet channel, a ventilation pipe and a T-shaped joint).
The droplet machine vision monitoring system includes: the LED illumination is used for an industrial camera for machine vision monitoring; a high-speed air pressure sensor; control and data processing system (including upper computer, control software and lower computer). The control and data processing system drives the camera to record the droplet state image after a delay according to the user configuration parameters.
Droplet ejection requires the high speed solenoid to be turned on and off in milliseconds, with on-hold times also on the order of milliseconds. The high-speed solenoid valve cannot guarantee the repeatability of the switching process. This results in a certain randomness of the air pressure pulse waveform P (t) in the reservoir per injection. There is some inconsistency in the droplet ejection status.
Establishing a predictive relationship of drive and ejection status facilitates manipulation of the droplet ejection device. The most direct drive for droplet ejection is the time-varying air pressure P (t) in the reservoir. Common injection state parameters include: number of droplets N D Droplet drop position H D I.e. the distance of the droplet from the orifice at a certain delay (typically with the rising edge of the high-speed solenoid drive signal being the zero point of reference time).
The BP neural network is a multi-layer feedforward neural network trained according to an error back propagation algorithm, and is the most widely applied neural network at present. The BP neural network finishes the mapping of input and output by using the transfer function of the neurons, and trains the neural network through a learning algorithm of error back propagation and error gradient reduction. Because BP neural network has the characteristics of self-learning, strong self-adaptation capability, strong mapping capability of nonlinear relation and the like, the BP neural network is commonly used for researching the mapping relation of nonlinear problems, has mature network theory and performance, and is suitable for building a droplet state prediction model.
By combining a series of moments t i The lower air pressure value P i ＝P(t i ) As inputs to the neural network, the number of droplets and the droplet-to-nozzle distance are output as the neural network. P can be established i Relationship with the injection state. In the application example, the accuracy of the prediction of the number of droplets is higher than 99%. Compared with the statistical average drop position of the droplets obtained based on machine vision and image processing, the method comprises the following steps of (t) and (BP) neural network prediction model pair H D The prediction accuracy of (2) is higher.
The application of the predictive model includes the following aspects. 1: the device can replace an image acquisition device for monitoring and controlling the ejection state, thereby simplifying the droplet ejection device. 2: in the case that other injection parameters K may change, the model may more sensitively detect the change in K. Possible K include: liquid fluid characteristics, nozzle clogging status, nozzle hydrophobicity, etc.
Compared with the traditional machine visionThe neural network based on the air pressure signal P (t) is easier to realize rapid data analysis processing by the sense and image analysis technology. With this method, the droplet ejection state can be monitored and controlled in real time. However, in order to achieve the above function, a high-speed (dynamic) air pressure sensor is required and mounted on the wall of the reservoir. High-speed air pressure sensors are expensive and are susceptible to contamination and damage from splashed ink in practical use. Therefore, a signal convenient for acquisition is needed to replace the air pressure signal P (t i ) As input to the neural network.
Disclosure of Invention
As described in the background section, although the air pressure oscillation signal P (t) is a direct drive of the droplet ejection process. The droplet ejection status may be predicted as an input to the neural network. But the acquisition cost of the P (t) signal is high, and the acquisition module is easy to damage.
In order to solve the technical problems, the invention adopts the technical proposal that the pneumatic droplet ejection state prediction method based on BP neural network is that the time-varying signal I (t) of the solenoid current flowing through a high-speed electromagnetic valve replaces the P (t) signal, is used as the input of the neural network, and predicts the droplet ejection state parameter, namely the number N of droplets D And a falling position H D The method comprises the steps that a neural network is built in a control and data processing system, wherein the control and data processing system comprises an upper computer and a lower computer; the method specifically comprises the following steps:
step 1, collecting the injection state;
the control circuit of the lower computer generates an electrical switch control signal according to the parameters set in the control software of the upper computer, and controls the high-speed electromagnetic valve to be opened and closed. The lower computer generates a trigger signal according to the set time delay, controls the industrial camera to shoot the proper view field range at the nozzle, and returns the image of the droplet to the upper computer for storage.
Step 2, image processing and information extraction: the parameters of the ejection state are extracted for the droplet image using the OpenCV library function of python.
Step 3, collecting a solenoid current signal of a high-speed electromagnetic valve of the pneumatic droplet jetting device and the air pressure at the front end of the high-speed electromagnetic valve;
step 4, preprocessing a solenoid current signal of the high-speed electromagnetic valve:
(1) The effect of the current signal after droplet generation on the ejection state parameters is negligible based on a priori knowledge of droplet generation. According to experimental experience, the solenoid current signal of the high-speed electromagnetic valve is intercepted and reserved for the period of time before the breaking moment.
(2) And filtering the solenoid current signal of the high-speed electromagnetic valve, and storing the signal.
Step 5, normalizing the input characteristics and labels of the neural network; the input features include: solenoid current signal of high-speed electromagnetic valve, liquid level height h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 The method comprises the steps of carrying out a first treatment on the surface of the The tag includes a spray status parameter.
Step 6, establishing a BP neural network prediction model for predicting injection state parameters:
(1) BP neural network prediction model for predicting injection state parameters is built, and input variables of the prediction model are high-speed electromagnetic valve solenoid coil current signals I (t) i ) Liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 . The output variable is a jet state parameter including the number N of droplets D Droplet drop position H D 。
(2) The number of the neurons of the input layer, the number of the neurons of the output layer, the number of the hidden layers and the number of the neurons of each hidden layer are configured. The neuron number of the input layer is the air pressure waveform I (t) i ) The number of data points and the number of neurons of an output layer are 1;
(3) Setting the hidden layer number and the number of neurons in the hidden layer;
(4) Setting a neuron activation function in a neural network;
(5) Selecting a loss function;
(6) Setting an optimizer;
step 7, model training and verification:
a set of test samples is established that is independent of the set of training samples. A set of test samples is used to calculate a prediction error for the neural network. When the prediction error stops decreasing, training of the neural network is cut off. Thereby avoiding the occurrence of the overfitting phenomenon of the neural network.
Further, the implementation process of the step 2 is as follows: (1) Cropping a region of interest, i.e., a ROI region, in the droplet image;
(2) Binarization processing is carried out on the cut ROI image;
(3) And searching the contours of the images, calculating the m00 moment of each contour, corresponding to the two-dimensional area of the droplets, and screening out the corresponding contours of the droplets.
(4) Counting droplet profiles to yield N D . Drop position H of droplet D Obtained by measuring the distance of the center of gravity of the droplet from the orifice. H D And N D Is a parameter that characterizes the injection state.
Further, the implementation process of the step 3 is as follows: (1) collecting a solenoid current signal of a high-speed electromagnetic valve:
the pneumatic droplet jetting device collects a current signal I (t) flowing through a solenoid coil of a high-speed electromagnetic valve through an ammeter, wherein the sampling frequency is f S The sampling interval time is t S ＝1/f S . In the actual collection process, an electric switch opening signal in a high-speed electromagnetic valve driving circuit is taken as a time zero point, a high-speed electromagnetic valve solenoid current signal is collected, and n high-speed electromagnetic valve solenoid current signals I (t) at discrete moments are obtained i ). Finally, signal I (t i ) Transmitting to the upper computer and storing.
(2) Collecting liquid level information;
the liquid level height h in the liquid storage cavity is collected by a sensor and is sent to an upper computer for storage.
(3) Acquiring air pressure information at the front end of the high-speed electromagnetic valve; measuring the front end air pressure P of the high-speed electromagnetic valve by using an air pressure sensor 0 Sending to the upper computer and storing.
Further, the implementation process of the step 5 is as follows: (1) For each sample S, for the high-speed solenoid current signal I S (t i ) Global normalization is performed.
And carrying out global normalization on the waveform sample by using a normalization function, wherein the data range obtained after normalization is between [ -1,1 ]. The normalization function is:
wherein t is i Representing the sampling time of the ith discrete instant, I S Representing the S-th solenoid current signal.
(2) For each sample S, other input characteristics f (including liquid level height h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 ) Normalization was performed.
And normalizing the other characteristics f by using a normalization function, wherein the range of the normalized data is between [ -1,1 ]. The normalization function is:
where f represents other features including: liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 。
(3) For each sample S, the label, i.e. the ejection state parameters, are normalized.
Normalizing the injection state parameters by using a normalization function, wherein the range of the normalized data is between [ -1,1 ]. The normalization function is:
wherein p represents the ejection state parameter, representing the number of droplets N d Droplet drop position H d 。
Further, the device for realizing the pneumatic droplet ejection state prediction method based on the BP neural network comprises a droplet ejection system and a visual monitoring system of droplet ejection state; the implementing a droplet ejection system includes: the liquid storage device comprises a liquid storage cavity 1, a nozzle 2, a liquid level sensor 3, a high-speed electromagnetic valve 4, a pressure regulating valve 5, an air pressure sensor 6 at the front end of the high-speed electromagnetic valve and a ventilation pipeline 7.
The droplet machine vision monitoring system includes: LED lighting 8, industrial camera 9, control and data processing system 10.
The liquid storage cavity 1 is used for storing liquid to be sprayed; the nozzle 2 is used to control the geometry of the droplets produced; the liquid level sensor 3 is used for detecting the liquid level h and is used as the characteristic input of a neural network prediction model; the high-speed electromagnetic valve 4 is used for controlling high-pressure gas to enter the liquid storage cavity to force liquid to flow out of the nozzle to generate micro-droplets, the driving circuit of the high-speed electromagnetic valve controls the opening and closing of the high-speed electromagnetic valve, and the solenoid current measuring circuit of the high-speed electromagnetic valve is used for collecting solenoid current signals of the high-speed electromagnetic valve; the pressure regulating valve 5 is used for regulating the air pressure at the front end of the high-speed electromagnetic valve; the air pressure sensor 6 at the front end of the high-speed electromagnetic valve 4 is used for collecting the air pressure P at the front end of the high-speed electromagnetic valve 0 The method comprises the steps of carrying out a first treatment on the surface of the The ventilation pipeline 7 is used for connecting the high-speed electromagnetic valve 4, the liquid storage cavity 1 and the external atmosphere. The LED illumination 8 provides a backlighting environment for the industrial camera 9 to shoot. The control and data processing system 10 can control the driving circuit of the high-speed electromagnetic valve 4 according to user configuration parameters; driving the industrial camera 9 to record droplet ejection status images at set moments; collecting and storing solenoid current signal I (t), liquid level height h of liquid storage cavity and air pressure P at front end of high-speed solenoid valve 0 。
Compared with the prior art, aiming at the problems of high cost and easy damage of a high-speed air pressure sensor in the prediction of the injection state parameters in a pneumatic droplet injection device, the invention provides a method based on an air pressure generating mechanism in a liquid storage cavity and combining with the working principle of a high-speed electromagnetic valve by using the solenoid coil current I (t) of the high-speed electromagnetic valve, the liquid level height h of the liquid storage cavity and the air pressure P at the front end of the high-speed electromagnetic valve 0 As a feature, a new method of predicting injection state parameters through a neural network. The invention can be used for the prediction of the droplet state of a pneumatic droplet ejection device, and real-time monitoring and control.
Drawings
FIG. 1 is a schematic diagram of high speed solenoid drive and current sense functions.
Fig. 2 is a schematic diagram of a pneumatic droplet ejection device according to the present invention.
Fig. 3 is a schematic diagram of an image processing procedure for ejection state extraction.
Fig. 4 is a schematic diagram of a BP network structure.
Detailed Description
The present invention will be described in detail below with reference to the drawings and examples.
The invention provides a device and a method for predicting the pneumatic droplet ejection state based on BP neural network, which adopts a time-varying signal I (t) of the solenoid current flowing through a high-speed electromagnetic valve to replace a P (t) signal as the input of the neural network and predicts the droplet ejection state parameters (namely the number N of droplets D And a falling position H D )。
The mechanism of air pressure generation within the reservoir can be understood by a forced Helmholtz (Helmholtz) resonator. Under the condition that the structure of the liquid storage cavity is unchanged, the liquid storage cavity is completely formed by the air source pressure P at the front end of the high-speed electromagnetic valve 0 The volume of the cavity above the liquid level, the length and the inner diameter of the air release pipe and the switching process of the high-speed electromagnetic valve. Wherein the length and inner diameter of the bleed tube will generally remain the same in practical applications.
The switching process of the high-speed electromagnetic valve is mainly affected by three parameters. They include: valve on time, valve hold on time, valve off time. The core element inside the high speed solenoid valve is a solenoid actuator. The valve is normally maintained closed by a spring. When the current passing through the solenoid is higher than a certain threshold value, the magnetic force exceeds the restoring force of the spring in the electromagnetic valve, and the valve is opened. When the current is lower than a certain threshold value, the magnetic force is smaller than the restoring force of the spring, and the valve is closed. As described in the background, droplet ejection occurs on the order of milliseconds. The high-speed solenoid valve cannot guarantee the repeatability of the switching process. This results in a certain randomness of the current waveform I (t) of the pulse current flowing through the solenoid. The invention proposes to detect the current I (t) flowing through the solenoid and to construct a neural network model of the predicted injection state parameter as a driving signal.
The liquid level h is changed in consideration of ink consumption caused by ink injection and ejection, thereby changing the volume of a Helmholtz (Helmholtz) resonance cavity in the liquid storage cavity and affecting the air pressure pulse in the liquid storage cavity. Therefore, the liquid level h also needs to be inputted as a characteristic to the neural network. The liquid level h needs to be detected. Conventional level sensors can do this.
Air pressure P at front end of high-speed electromagnetic valve 0 Will also affect the air pressure pulse in the liquid storage cavity, thus P 0 The neural network also needs to be input as a feature. The air pressure at the front end of the high-speed electromagnetic valve needs to be detected. Conventional barometric pressure sensors may perform this function.
In order to realize a prediction model of injection state parameters based on the solenoid current I (t) of the high-speed solenoid valve, it is necessary to realize a function of driving the high-speed solenoid valve and detecting the solenoid current of the high-speed solenoid valve. The stabilized voltage power supply supplies power to the high-speed electromagnetic valve; the electric switch controls the opening and closing of the high-speed electromagnetic valve; an ammeter in series with the solenoid of the high speed solenoid measures the current flowing through the solenoid. There are various topologies and implementations for the driving of the high speed solenoid and the detection of the solenoid current of the high speed solenoid.
An example of a circuit is shown in fig. 1.
The high-speed electromagnetic valve driving circuit, the high-speed electromagnetic valve solenoid coil current detection circuit and the liquid level measurement belong to mature technologies and do not belong to the invention.
The pneumatic droplet ejection device of the present invention includes a droplet ejection system, a visual monitoring system of the droplet ejection state, as shown in fig. 2.
The droplet ejection system includes: a liquid storage cavity 1; a nozzle 2; a liquid level sensor 3; a high-speed solenoid valve (including a drive circuit and a high-speed solenoid current measurement circuit) 4; a pressure regulating valve 5; a pressure sensor 6 at the front end of the high-speed electromagnetic valve; a vent line (comprising an intake passage, a vent pipe and a T-joint) 7.
The droplet machine vision monitoring system includes: LED illumination 8; an industrial camera 9 for machine vision monitoring; control and data processing system (including upper computer, control software and lower computer) 10.
The liquid storage cavity is used for storing liquid to be sprayed; the nozzle is used to control the geometry of the droplets produced; the liquid level sensor is used for detecting the liquid level h and is used as the characteristic input of the neural network prediction model; high-speed electromagnetic valve for controlling high-pressure gas to enterThe liquid storage cavity forces liquid to flow out of the nozzle to generate micro-droplets, the high-speed electromagnetic valve driving circuit controls the high-speed electromagnetic valve to be opened and closed, and the high-speed electromagnetic valve solenoid current measuring circuit is used for collecting a high-speed electromagnetic valve solenoid current signal; the pressure regulating valve is used for regulating the air pressure at the front end of the high-speed electromagnetic valve; the air pressure sensor at the front end of the high-speed electromagnetic valve is used for collecting the air pressure P at the front end of the high-speed electromagnetic valve 0 The method comprises the steps of carrying out a first treatment on the surface of the The ventilation pipeline is used for connecting the high-speed electromagnetic valve, the liquid storage cavity and the external atmosphere. LEDs provide a backlighting environment for industrial camera photography. The control and data processing system controls the high-speed electromagnetic valve driving circuit according to the user configuration parameters; driving a camera to record a droplet ejection status image at a set time; collecting and storing solenoid current signal I (t), liquid level height h of liquid storage cavity and air pressure P at front end of high-speed solenoid valve 0 。
The neural network belongs to an upper computer software system. The process of building the neural network prediction model is as follows:
step 1, collecting injection state
The lower computer control circuit generates an electrical switch control signal according to the parameters set on the upper computer control software to control the high-speed electromagnetic valve to be opened and closed, as shown in figure 1.
Meanwhile, the lower computer generates a trigger signal according to the set time delay, controls the industrial camera to shoot the proper view field range at the nozzle, and returns the droplet image to the upper computer for storage. A typical droplet image is shown in fig. 3, comprising a jet, a liquid ribbon, a single droplet or a plurality of droplets.
Step 2, image processing and information extraction: the parameters of the ejection state are extracted for the droplet image using the OpenCV library function of python.
(1) Cropping a region of interest, i.e., a ROI region, in the droplet image;
(2) Binarization processing is carried out on the cut ROI image;
(3) The image is contour searched and the m00 moment for each contour is calculated, which corresponds to the two-dimensional area of the droplet. Since droplet diameter is typically 1-2 times the nozzle inner diameter, much larger than pi d can be ignored 2 The sum of/4 is much smaller than pi d 2 Profile of/4. Screen screenAnd selecting the outline corresponding to the droplet.
(4) Counting droplet profiles to yield N D . Drop position H of droplet D The distance between the center of gravity of the droplet and the nozzle is obtained by measuring the distance between the center of gravity of the droplet and the nozzle under a certain time delay (relative to the rising edge of the trigger signal of the electromagnetic valve). H D And N D Is a parameter that characterizes the injection state.
Step 3, collecting a solenoid current signal of a high-speed electromagnetic valve of the pneumatic droplet jetting device; collecting the air pressure at the front end of the high-speed electromagnetic valve; liquid level height acquisition
(1) Collecting a solenoid current signal of a high-speed electromagnetic valve:
the pneumatic droplet jetting apparatus collects the current signal I (t) flowing through the solenoid of the high-speed solenoid valve (also taking the rising edge of the solenoid driving signal as the reference time) by an ammeter, wherein the sampling frequency is f S The sampling interval time is t S ＝1/f S . In the actual collection process, an opening signal (usually a rising edge of an enabling voltage signal) of an electrical switch (shown in figure 1) in a high-speed electromagnetic valve driving circuit is taken as a time zero point, a high-speed electromagnetic valve solenoid current signal is collected, and high-speed electromagnetic valve solenoid current signals I (t) of n discrete moments are obtained i ). Finally, signal I (t i ) Transmitting to the upper computer and storing.
(2) Liquid level information acquisition
Before the electric switch is started, the liquid level height h in the liquid storage cavity is collected by a sensor and is sent to an upper computer for storage.
(3) High-speed electromagnetic valve front-end air pressure information acquisition
Before the electric switch is started, the air pressure sensor is used for measuring the air pressure P at the front end of the high-speed electromagnetic valve 0 Sending to the upper computer and storing.
Step 4, preprocessing a solenoid current signal of the high-speed electromagnetic valve:
(1) The effect of the current signal after droplet generation on the ejection state parameters is negligible based on a priori knowledge of droplet generation. According to experimental experience, the solenoid current signal of the high-speed electromagnetic valve is intercepted and reserved for the period of time before the breaking moment.
(2) And filtering the solenoid current signal of the high-speed electromagnetic valve, and storing the signal.
Step 5, inputting characteristics (including solenoid current signal of high-speed electromagnetic valve, liquid level height h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve) into neural network 0 ) And label (ejection state parameter) normalization:
(1) For each sample S, for the high-speed solenoid current signal I S (t i ) Global normalization is performed.
And carrying out global normalization on the waveform sample by using a normalization function, wherein the data range obtained after normalization is between [ -1,1 ]. The normalization function is:
wherein t is i Representing the sampling time of the ith discrete instant, I S Representing the S-th solenoid current signal.
(2) For each sample S, other input characteristics f (including liquid level height h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 ) Normalization was performed.
And normalizing the other characteristics f by using a normalization function, wherein the range of the normalized data is between [ -1,1 ]. The normalization function is:
where f represents other features including: liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 。
(3) For each sample S, the label (i.e., ejection state parameters) is normalized.
Normalizing the injection state parameters by using a normalization function, wherein the range of the normalized data is between [ -1,1 ]. The normalization function is:
wherein p represents the ejection state parameter, representing the number of droplets N d Droplet drop position H d 。
Step 6, establishing a BP neural network prediction model for predicting injection state parameters:
(1) BP neural network prediction model for predicting injection state parameters is built, and input variables of the prediction model are high-speed electromagnetic valve solenoid coil current signals I (t) i ) Liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 . The output variable being the ejection state parameter (number of droplets N D Droplet drop position H D )。
(2) And configuring parameters such as the number of neurons of an input layer, the number of neurons of an output layer, the number of hidden layers, the number of neurons of each hidden layer and the like. The neuron number of the input layer is the air pressure waveform I (t) i ) The number of data points and the number of neurons of an output layer are 1;
(3) Setting the hidden layer number and the number of neurons in the hidden layer;
(4) Setting a neuron activation function in a neural network;
(5) Selecting a loss function;
(6) Setting an optimizer;
step 7, model training and verification:
a set of test samples is established that is independent of the set of training samples. A set of test samples is used to calculate a prediction error for the neural network. When the prediction error stops decreasing, training of the neural network is cut off. Thereby avoiding the occurrence of the overfitting phenomenon of the neural network.
Claims (5)
1. The pneumatic droplet jetting state predicting method based on BP neural network features that the time varying signal I (t) of solenoid current flowing through high speed solenoid valve is used as the input of the neural network to predict the droplet jetting state parameter, i.e. the number of droplets N D And a falling position H D The method comprises the steps that a neural network is built in a control and data processing system, wherein the control and data processing system comprises an upper computer and a lower computer; concrete bagThe method comprises the following steps:
step 1, collecting the injection state;
the control circuit of the lower computer generates an electrical switch control signal according to the parameters set in the control software of the upper computer, and controls the high-speed electromagnetic valve to be opened and closed; the lower computer generates a trigger signal according to the set time delay, controls the industrial camera to take a picture of the proper view field range at the nozzle, and returns the image of the micro-droplet to the upper computer for storage;
step 2, image processing and information extraction: performing parameter extraction of the ejection state on the droplet image using the OpenCV library function of python;
step 3, collecting a solenoid current signal of a high-speed electromagnetic valve of the pneumatic droplet jetting device and the air pressure at the front end of the high-speed electromagnetic valve;
step 4, preprocessing a solenoid current signal of the high-speed electromagnetic valve:
(1) According to the prior knowledge of droplet generation, the influence of the current signal after droplet generation on the ejection state parameters is ignored; intercepting and retaining a solenoid current signal of the high-speed electromagnetic valve in the period before the breaking moment;
(2) Filtering the solenoid current signal of the high-speed electromagnetic valve, and storing the signal;
step 5, normalizing the input characteristics and labels of the neural network; the input features include: solenoid current signal of high-speed electromagnetic valve, liquid level height h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 The method comprises the steps of carrying out a first treatment on the surface of the The tag includes a jetting status parameter;
step 6, establishing a BP neural network prediction model for predicting injection state parameters:
(1) BP neural network prediction model for predicting injection state parameters is built, and input variables of the prediction model are high-speed electromagnetic valve solenoid coil current signals I (t) i ) Liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 The method comprises the steps of carrying out a first treatment on the surface of the The output variable is a jet state parameter including the number N of droplets D Droplet drop position H D ；
(2) Configuring the number of neurons of an input layer, the number of neurons of an output layer, the number of hidden layers and the number of neurons of each hidden layer; input layerThe neuron number is the air pressure waveform I (t) i ) The number of data points and the number of neurons of an output layer are 1;
(3) Setting the hidden layer number and the number of neurons in the hidden layer;
(4) Setting a neuron activation function in a neural network;
(5) Selecting a loss function;
(6) Setting an optimizer;
step 7, model training and verification:
establishing a test sample set independent of the training sample set; calculating a prediction error of the neural network using the set of test samples; when the prediction error stops decreasing, training the neural network is cut off; thereby avoiding the occurrence of the overfitting phenomenon of the neural network.
2. The method for predicting the pneumatic droplet ejection state based on the BP neural network according to claim 1, wherein the implementation process of step 2 is as follows: (1) Cropping a region of interest, i.e., a ROI region, in the droplet image;
(2) Binarization processing is carried out on the cut ROI image;
(3) Performing contour searching on the image, calculating m00 moment of each contour, corresponding to two-dimensional areas of the droplets, and screening out the corresponding contour of the droplets;
(4) Counting droplet profiles to yield N D The method comprises the steps of carrying out a first treatment on the surface of the Drop position H of droplet D Acquiring by measuring the distance between the gravity center of the droplet and the nozzle; h D And N D Is a parameter that characterizes the injection state.
3. The method for predicting the pneumatic droplet ejection state based on the BP neural network according to claim 1, wherein the implementation process of step 3 is as follows: (1) collecting a solenoid current signal of a high-speed electromagnetic valve:
the pneumatic droplet jetting device collects a current signal I (t) flowing through a solenoid coil of a high-speed electromagnetic valve through an ammeter, wherein the sampling frequency is f S The sampling interval time is t S ＝1/f S The method comprises the steps of carrying out a first treatment on the surface of the In the actual collection process, the high-speed electromagnetic valve is used for driving electricityThe on signal of the electrical switch in the road is a time zero point, the solenoid current signal of the high-speed electromagnetic valve is collected, and the solenoid current signals I (t) of the high-speed electromagnetic valve at n discrete moments are obtained i ) The method comprises the steps of carrying out a first treatment on the surface of the Finally, signal I (t i ) Transmitting to the upper computer and storing;
(2) Collecting liquid level information;
collecting the liquid level height h in the liquid storage cavity by using a sensor, and sending the liquid level height h to an upper computer for storage;
(3) Acquiring air pressure information at the front end of the high-speed electromagnetic valve; measuring the front end air pressure P of the high-speed electromagnetic valve by using an air pressure sensor 0 Sending to the upper computer and storing.
4. The method for predicting the pneumatic droplet ejection state based on the BP neural network according to claim 1, wherein the implementation process of step 5 is as follows: (1) For each sample S, for the high-speed solenoid current signal I S (t i ) Performing global normalization;
global normalization is carried out on the waveform sample by using a normalization function, and the data range obtained after normalization is between [ -1,1 ]; the normalization function is:
wherein t is i Representing the sampling time of the ith discrete instant, I S Representing the S-th solenoid current signal;
(2) For each sample S, for other input features f: liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 Normalizing;
normalizing other features f by using a normalization function, wherein the range of data obtained after normalization is between [ -1,1 ]; the normalization function is:
wherein f represents otherCharacterized in that it comprises: liquid level h of liquid storage cavity and air pressure P at front end of high-speed electromagnetic valve 0 ；
(3) Normalizing the label, namely the injection state parameter, for each sample S;
normalizing the injection state parameters by using a normalization function, wherein the range of data obtained after normalization is between [ -1,1 ]; the normalization function is:
wherein p represents the ejection state parameter, representing the number of droplets N d Droplet drop position H d 。
5. The BP neural network-based pneumatic droplet ejection state prediction method according to claim 1, wherein the device for realizing the BP neural network-based pneumatic droplet ejection state prediction method comprises a droplet ejection system and a visual monitoring system of droplet ejection state; the implementing a droplet ejection system includes: the liquid storage device comprises a liquid storage cavity, a nozzle, a liquid level sensor, a high-speed electromagnetic valve, a pressure regulating valve, an air pressure sensor at the front end of the high-speed electromagnetic valve and a ventilation pipeline;
the droplet machine vision monitoring system includes: LED lighting, industrial cameras, control and data processing systems;
the liquid storage cavity is used for storing liquid to be sprayed; the nozzle is used to control the geometry of the droplets produced; the liquid level sensor is used for detecting the liquid level h and is used as the characteristic input of the neural network prediction model; the high-speed electromagnetic valve is used for controlling high-pressure gas to enter the liquid storage cavity to force liquid to flow out of the nozzle to generate micro-droplets, the driving circuit of the high-speed electromagnetic valve controls the high-speed electromagnetic valve to be opened and closed, and the solenoid current measuring circuit of the high-speed electromagnetic valve is used for collecting solenoid current signals of the high-speed electromagnetic valve; the pressure regulating valve is used for regulating the air pressure at the front end of the high-speed electromagnetic valve; the air pressure sensor at the front end of the high-speed electromagnetic valve is used for collecting the air pressure P at the front end of the high-speed electromagnetic valve 0 The method comprises the steps of carrying out a first treatment on the surface of the The ventilation pipeline is used for connecting the high-speed electromagnetic valve, the liquid storage cavity and the outside bigA gas; LED illumination provides a backlighting environment for industrial camera shooting; the control and data processing system can control the driving circuit of the high-speed electromagnetic valve according to the user configuration parameters; driving an industrial camera to record a droplet ejection status image at a set time; collecting and storing solenoid current signal I (t), liquid level height h of liquid storage cavity and air pressure P at front end of high-speed solenoid valve 0 。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311683612.0A CN117634581A (en) | 2023-12-10 | 2023-12-10 | Device and method for predicting pneumatic droplet ejection state based on BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311683612.0A CN117634581A (en) | 2023-12-10 | 2023-12-10 | Device and method for predicting pneumatic droplet ejection state based on BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117634581A true CN117634581A (en) | 2024-03-01 |
Family
ID=90016211
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311683612.0A Pending CN117634581A (en) | 2023-12-10 | 2023-12-10 | Device and method for predicting pneumatic droplet ejection state based on BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117634581A (en) |
-
2023
- 2023-12-10 CN CN202311683612.0A patent/CN117634581A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109871614B (en) | Pneumatic type droplet ejection state prediction method based on BP neural network | |
| EP3429862B1 (en) | Systems and methods for precision inkjet printing | |
| Wang et al. | In-situ droplet inspection and control system for liquid metal jet 3D printing process | |
| CN109435473B (en) | A kind of flight ink droplet detection device and method suitable for inkjet printing | |
| Hutchings et al. | High speed imaging and analysis of jet and drop formation | |
| CN112373207B (en) | Ink jet testing method and device and terminal equipment | |
| CN109968671B (en) | 3D printing process fault diagnosis method and device | |
| EP1700700B1 (en) | Inkjet recording apparatus | |
| CN111070900B (en) | Piezoelectric ink-jet printing device for removing printing ink bubbles and control method thereof | |
| CN108734277A (en) | A kind of targeted drug spray method based on convolutional neural networks | |
| CN117634581A (en) | Device and method for predicting pneumatic droplet ejection state based on BP neural network | |
| Piovarci et al. | Closed-loop control of direct ink writing via reinforcement learning | |
| Fraters et al. | Meniscus oscillations driven by flow focusing lead to bubble pinch-off and entrainment in a piezoacoustic inkjet nozzle | |
| Zhou et al. | Experimental study of the piezoelectric drop-on-demand drop formation in a coaxial airflow | |
| Wang et al. | Controlling ejection state of a pneumatic micro-droplet generator through machine vision methods | |
| WO2021168174A1 (en) | Improved fluid dispensing process control using machine learning and system implementing the same | |
| CN112733418A (en) | Method for monitoring fluid characteristic change of ejected liquid in pneumatic droplet ejection process | |
| Castrejón-Pita et al. | Time-Resolved Particle Image Velocimetry within the Nozzle of a Drop-on-Demand Printhead. | |
| Huang et al. | Spatiotemporal fusion network for the droplet behavior recognition in inkjet printing | |
| Wang et al. | Prediction of Ejection State for a Pneumatic Valve-Controlled Micro-Droplet Generator by a BP Neural Network | |
| CN113324580B (en) | Static droplet jetting state detection method based on induced current and machine vision | |
| Luo et al. | Programmable high integration and resolution digital microfluidic device driven by thin film transistor arrays | |
| Conto | Breakup of liquid jets: the capillary retraction | |
| Mishra et al. | Control of high-resolution electrohydrodynamic jet printing | |
| Kawahara et al. | Development of on-chip automatic cell sensing and ejection system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |