Air pressure extrusion nozzle system with machine vision function and control method

Abstract

The invention discloses an air pressure extrusion nozzle system with a machine vision function and a control method, and is suitable for a direct ink writing process realized by air pressure extrusion. The system comprises a camera, an air pressure extrusion device, a TOF laser ranging module and the like, and can be mounted on corresponding additive manufacturing equipment. A self-adaptive extrusion algorithm is burnt in the nozzle, the current extrusion line width is measured by means of a visual sensor and a TOF ranging module carried by the carried nozzle, and self-adaptive extrusion parameters can be obtained and stored in cooperation with a preset test track. The problem of batch difference possibly existing when biological material ink is configured and replaced for multiple times can be effectively solved, the relatively stable printing effect is achieved, and the system reliability is improved.

Description

Technical Field

[0001] The present invention is applicable to the field of multi-material additive manufacturing, and more specifically, to a pneumatic extrusion nozzle system with machine vision function and a control method thereof. Background Art

[0002] Additive manufacturing technology, also known as 3D printing technology, constructs objects by layer-by-layer stacking of materials. Without the need for traditional molds or tools and being unaffected by the complexity of the model, it makes it possible to manufacture complex shapes, internal structures, and personalized customization. The technical characteristics of additive manufacturing are particularly suitable for highly customized, small-batch production, and rapid prototyping. With the continuous development of technology, additive manufacturing is no longer limited to plastic and metal materials, and more types of materials, such as ceramics, composite materials, and biological materials, have begun to be applied to 3D printing.

[0003] Multi-material 3D printing technology is an important high-tech for preparing new functional materials and achieving controllability of materials. Its basic principle is to achieve the combination of multi-materials in the same product by precisely controlling the distribution, transition, and gradient change of different materials, thereby endowing the product with some more excellent technical indicators. During the multi-material three-dimensional printing process, due to the differences in the properties of different materials, different printing parameters are required during the printing process. Therefore, it is necessary to regulate the flow rate, extrusion rate, temperature, etc. of each material. However, the current multi-material printing technology generally relies on the original open-loop control method, and multiple debugging and manual calibration are required before printing to ensure the matching of the flow rate and extrusion amount of each material. This process is not only time-consuming but also easily affected by irrelevant factors, easily leading to unstable printing quality and affecting the performance of the final product. This technology can achieve effective combination and functional integration between different materials during the ink direct writing printing process, meet the requirements of high-performance and complex structure biological three-dimensional printing, and improve the success rate of composite printing. Summary of the Invention

[0004] Aiming at the above problems, the present invention proposes a pneumatic extrusion nozzle system with machine vision function and a control method thereof. The present invention aims to solve several problems existing in the multi-material 3D printing technology under the existing direct ink writing process, specifically including: it is difficult to accurately model the blockage and flow of biological inks such as hydrogels in the nozzle, multiple debugging is required during open-loop printing to obtain a relatively stable printing effect, it is necessary to continuously and accurately adjust the extrusion amount and extrusion line width of different materials during multi-material printing, and batch differences that may occur during multiple configurations and replacements of biological inks, etc. The technical solution provided by the present invention aims to improve the extrusion accuracy and consistency of materials during the multi-material 3D printing process, optimize the control of material flow rate and flow velocity, overcome the influence of batch differences on printing quality, thereby improving printing efficiency and finished product quality, and ensuring the stability and repeatability of the printing process.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A pneumatic extrusion nozzle system with machine vision function, including power input, microcontroller, proportional regulating valve, soft rubber hose, camera, TOF ranging module, barrel, pressure transmitter, nozzle and additive manufacturing equipment. A high-pressure gas source that provides the pressure required for extrusion is introduced into the pneumatic extrusion nozzle. There are electrical connections and signal connections between the power input and the microcontroller and corresponding electrical equipment. There is a communication circuit between the microcontroller and the camera and TOF ranging module. There is an electrical connection between the microcontroller and the proportional regulating valve. The proportional regulating valve, barrel and nozzle form a complete gas path through the soft rubber hose, and a pressure transmitter is installed. The pneumatic extrusion nozzle is connected to the additive manufacturing equipment through mechanical and electrical interfaces.

[0007] As a further improvement of the system of the present invention, the pneumatic extrusion nozzle system with machine vision function further includes a temperature sensor, and the temperature sensor is adhered to the surface of the barrel.

[0008] As a further improvement of the system of the present invention, the camera is equipped with a ring-shaped LED light source.

[0009] As a further improvement of the system of the present invention, a three-way valve is connected into the complete gas path formed by the proportional regulating valve, barrel and nozzle through the soft rubber hose.

[0010] A magnetic adsorption movable material changing rack is installed on the 3D printing platform supporting the pneumatic extrusion nozzle system with machine vision function.

[0011] The present invention provides a control method for a pneumatic extrusion nozzle system with machine vision function, and the specific steps are as follows:

[0012] S1: Read the pre-stored test pattern information, use the motion control function provided by the additive manufacturing platform to move the nozzle along a predetermined trajectory. The TOF ranging module real-time detects the Z-axis distance between the nozzle and the printing plane and sends it back to the microcontroller, and use the equipped camera to detect the line width. At the same time, perform parameter and image acquisition, and synchronize using timestamps;

[0013] S2: Import the image into the classifier. The classifier is designed based on the ResNet and CNN models and labels it as three printing qualities: droplet, intermittent, and line;

[0014] S3: Measure the line width of the pattern marked as a line, use the triangulation scheme to obtain the extrusion line width, calculate the line width error rate, import the printing parameters into the Bayesian optimization model, and establish the relationship between the extrusion air pressure, printing speed, nozzle distance, nozzle diameter, nozzle shape coefficient, barrel temperature, hydrogel particle diameter and extrusion line width.

[0015] As a further improvement of the control method of the present invention, step S3 is specifically as follows:

[0016] First, train the surrogate model in the Bayesian optimization process. Use the Gaussian process regression model. The input features include adjustable parameters and fixed parameters, corresponding to the input vector x = [P, v, l, d, s, T, D]. The objective function uses the mean square error between the predicted extrusion width and the actual extrusion width. The formula is

[0017] Model the Gaussian process. The mapping relationship is:

[0018]

[0019] Considering that the optimization objective pays more attention to the relative change between errors, the mean function m(x) is assumed to be 0, and the kernel function k(x, x′) is a radial basis function kernel, which is used to describe the similarity between data points. Specifically:

[0020]

[0021] where σ 2 is the signal variance, which controls the function amplitude, and l is the length scale, which determines the correlation between features. is the noise variance;

[0022] Next, train the acquisition function in the Bayesian optimization process. Use the expected improvement algorithm to quantify the potential improvement that may be brought by sampling in the unexplored area of the adjustable parameters during printing, and balance the trade-off between exploring areas with high uncertainty and exploiting areas with known low objective values. Specifically, define the current minimum error as The expected improvement value is where the objective function follows a Gaussian distribution, that is

[0023] Introduce the expected improvement into the loss obtained by the Gaussian process regression model. The analytical solution formula:

[0024]

[0025] where Φ(Z) is the cumulative distribution function of the standard normal distribution, and φ(Z) is the probability density function of the standard normal distribution;

[0026] Finally, apply the trained surrogate model and acquisition function to the test set. The surrogate model predicts the extrusion width of the hydrogel based on the initially input extrusion air pressure, printing speed, nozzle distance, nozzle diameter, nozzle shape coefficient, barrel temperature, and hydrogel particle diameter. After comparing with the target extrusion width, substitute the error, extrusion air pressure, printing speed, and nozzle distance into the acquisition function to balance the influence of different variables on the error and return the adjustable parameters that can best fit the target extrusion width.

[0027] Beneficial effects:

[0028] 1. Improve extrusion accuracy: This system uses a camera to monitor the lines extruded by the nozzle in real time, compares them with the preset line width, calculates the percentage error of the extrusion line width, and can effectively detect the extrusion quality. This feedback mechanism significantly improves the accuracy of material extrusion in the multi-material 3D printing process. In the printing test, the adaptive parameter learning and adaptive line width adjustment control algorithm effectively reduce the line width standard deviation and improve the pore uniformity compared with the manual optimization method.

[0029] 2. Reduce material waste: Due to the function of adaptive control of the extrusion line width, the present invention effectively reduces the material waste caused by random factors such as material switching or uneven extrusion during the printing process.

[0030] 3. Easy to integrate and expand: The system has a modular design, can be integrated with existing 3D printing equipment, and has good scalability. The printing database established by adaptive parameter learning is easy to transplant, which can effectively reduce the R & D cost. Description of the drawings

[0031] Figure 1 is the physical structure diagram of the system of the present invention;

[0032] Figure 2 is the system connection block diagram;

[0033] Figure 3 is the physical structure diagram of the nozzle provided by the present invention;

[0034] Figure 4 is the block diagram of the extrusion width adaptive control algorithm;

[0035] Attachment name:

[0036] 1. Power input; 2. Microcontroller; 3. Proportional control valve; 4. High-pressure gas source; 5. Soft rubber hose; 6. Camera; 7. TOF ranging module; 8. Barrel; 9. Pressure transmitter; 10. Nozzle; 11. Temperature sensor; 12. Ring LED light source; 13. Additive manufacturing equipment. Detailed implementation manners

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0038] In the process of multi-material additive manufacturing using the inkjet direct writing process, for bioinks represented by hydrogels, problems such as the need for multiple manual debugging and optimization to obtain a suitable printing effect during current open-loop printing, the need to continuously adjust the extrusion amount and extrusion line width of different materials during multi-material printing, and batch differences that may occur during multiple configurations and replacements of bioinks. The adaptive extrusion nozzle and its control algorithm provided by the present invention can ensure the stability of printing and improve the printing quality under this process.

[0039] As Figure 1 shown, there is a mechanical and electrical connection between the nozzle provided in this embodiment and the additive manufacturing platform 13, forming a complete inkjet direct writing printing system.

[0040] As Figure 2 shown, this embodiment provides a function adaptive pneumatic extrusion nozzle with machine vision. The gas path is sequentially connected by a high-pressure gas source 4, a proportional regulating valve 3, a pressure transmitter 9, a soft rubber tube 5, a barrel 8, an extrusion nozzle 10, and related gas path connectors, which can provide adjustable and stable extrusion pressure for the barrel and the nozzle, and the current gas path pressure can be monitored by the pressure transmitter. For the proportional regulating valve of the voltage signal, there is a linear relationship between the output pressure and the control voltage.

[0041] P out =K p V input +V0

[0042] The visual measurement part consists of a camera 6, a TOF ranging module 7, a microcontroller 2, and a ring light source 12.

[0043] The power supply module has an electrical connection with peripherals such as the microcontroller 2 and the proportional regulating valve 3, and is used to provide a stable power supply voltage for the peripherals.

[0044] As Figure 3 shown, the principle of monocular camera line width detection in this embodiment is as follows. For a determined single camera hardware, its camera parameters are fixed, and the relationship between the world coordinate system and the camera coordinate system is obtained by the perspective projection relationship, where each matrix is: A is the camera internal parameter matrix, Π is the perspective projection matrix, c T w is the homogeneous transformation matrix for estimating the pose.

[0045]

[0046] After obtaining a relatively accurate Z-axis coordinate using the TOF ranging module, the actual size can be calculated based on the coordinates of each pixel point, and least squares fitting can be used. To synchronize the image captured by the camera and the movement position of the nozzle, it is stipulated that a calculation will be performed only when the distance between two waypoints is greater than the threshold.

[0047] The principle of the extrusion width adaptive algorithm provided in this embodiment is as follows. Define the extrusion line width error as

[0048]

[0049] Use an incremental discrete PI controller to dynamically adjust the extrusion pressure:

[0050] ΔP(t) = K p (E t -E t-1 ) + K i E t

[0051] Use a recursively generated fractal curve as the test trajectory for calibration. Taking the Hilbert curve as an example, a second-order Hilbert curve is used as the test trajectory. The second-order Hilbert curve has 16 waypoints. After discretizing the curve, each sub-segment is obtained and different widths are assigned. Using this filling method can reduce the error caused by the inclination of the printing plane.

[0052] As Figure 4 shown, this embodiment provides a learning method for adaptive printing parameters. First, use a vision measurement system to realize line width quantization. With the help of the vision measurement system, the actual effect of the printing parameters can be evaluated. Relevant research shows that using the canonical correlation analysis method (CCA), it can be known that the line width is positively correlated with the extrusion pressure, and negatively correlated with the nozzle distance and moving speed. Moreover, the influence of the extrusion pressure on the line width is higher than that of the nozzle distance and moving speed. Therefore, when designing the process parameters to be collected, select the extrusion air pressure (kPa), printing speed (mm / s), nozzle diameter (mm), nozzle distance (mm), nozzle shape factor (using One Hot encoding), barrel temperature (°C), and measure the line width.

[0053] In this example, the Bayesian optimization algorithm in machine learning is used for modeling. Bayesian Optimization (BO) is a global optimization method based on probability models, especially suitable for occasions where the calculation cost of the objective function is high (such as physical experiments or complex simulations) and efficient search for the optimal solution is required. Bayesian optimization mainly includes two parts, the surrogate model and the acquisition function. The surrogate model can predict the output based on the input by learning the training set data, and the acquisition function can balance the influence of multiple inputs on the output and select the optimal fitting method. In the Bayesian optimization of this example, the inputs include the extrusion air pressure P, printing speed v, nozzle distance l, nozzle diameter d, nozzle shape coefficient s, barrel temperature T, and hydrogel particle diameter D, and the output is the extrusion line width. Among them, the extrusion air pressure, printing speed, and nozzle distance are adjustable parameters during the printing process, and the nozzle diameter, nozzle shape coefficient, barrel temperature, and hydrogel particle diameter are fixed parameters after the printing starts.

[0054] First, we train the surrogate model in the Bayesian optimization process. Using the Gaussian Process Regression model (GP), the input features include adjustable parameters and fixed parameters, corresponding to the input vector x = [P, v, l, d, s, T, D]. The objective function uses the mean square error between the predicted extrusion width and the actual extrusion width, and the formula is

[0055] Model the Gaussian process, and the mapping relationship is:

[0056]

[0057] Considering that the optimization objective pays more attention to the relative change between errors, the mean function m(x) is assumed to be 0. The kernel function k(x, x′) is the radial basis function kernel, which is used to describe the similarity between data points. Specifically:

[0058]

[0059] Among them, σ 2 is the signal variance, which controls the function amplitude, l is the length scale, which determines the correlation between features, is the noise variance.

[0060] Next, we train the acquisition function in the Bayesian optimization process. Using the Expected Improvement (EI) algorithm, it quantifies the potential improvement that may be brought by sampling in the unexplored area of the adjustable parameters during printing, and balances the trade-off between exploring areas with high uncertainty and exploiting areas with known low objective values. Specifically, define the current minimum error as The expected improvement value is Among them, the objective function follows a Gaussian distribution, that is

[0061] Improve the desired loss obtained from the Gaussian process regression model, and the analytical solution formula:

[0062]

[0063] where Φ(Z) is the cumulative distribution function (CDF) of the standard normal distribution, and φ(Z) is the probability density function (PDF) of the standard normal distribution.

[0064] Finally, we apply the trained surrogate model and the acquisition function to the test set. The surrogate model predicts the extrusion width of the hydrogel based on the initially input extrusion air pressure, printing speed, nozzle distance, nozzle diameter, nozzle shape coefficient, barrel temperature, and hydrogel particle diameter. After comparing with the target extrusion width, the error, extrusion air pressure, printing speed, and nozzle distance are substituted into the acquisition function to balance the influence of different variables on the error, and the adjustable parameters that can best fit the target extrusion width are returned.

[0065] The above is only a preferred embodiment of the present invention, and it is not intended to limit the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. A pneumatic extrusion nozzle system with machine vision function, comprising a power input (1), a microcontroller (2), a proportional control valve (3), a soft rubber hose (5), a camera (6), a TOF distance measurement module (7), a barrel (8), a pressure transmitter (9), a nozzle (10) and an additive manufacturing device (13), characterized in that: A high-pressure gas source (4) is introduced into the pneumatic extrusion nozzle to provide the system with the pressure required for extrusion. There is an electrical connection and a signal connection between the power input (1) and the microcontroller (2) and the corresponding electrical equipment. There is a communication circuit between the microcontroller (2) and the camera (6) and the TOF ranging module (7). There is an electrical connection between the microcontroller (2) and the proportional control valve (3). A complete gas circuit is formed between the proportional control valve (3) and the barrel (8) and the nozzle (10) through a soft rubber hose (5), and a pressure transmitter (9) is installed. The pneumatic extrusion nozzle is connected to the additive manufacturing equipment (13) through a mechanical and electrical interface.

2. The pneumatic extrusion nozzle system with machine vision function according to claim 1, characterized in that: The pneumatic extrusion nozzle system with machine vision function also includes a temperature sensor (11), and the temperature sensor (11) is adhered to the surface of the barrel.

3. The pneumatic extrusion nozzle system with machine vision function according to claim 1, characterized in that: The camera (6) is equipped with a ring-shaped LED light source (12).

4. The pneumatic extrusion nozzle system with machine vision function according to claim 1, characterized in that: The proportional regulating valve (3), the barrel (8) and the nozzle (10) are connected to a three-way valve through a soft rubber tube (5) to form a complete gas circuit.

5. The pneumatic extrusion nozzle system with machine vision function according to claim 1, characterized in that: A magnetically movable material changing rack is installed on the three-dimensional printing platform matched with the pneumatic extrusion nozzle system with machine vision function.

6. A control method for a pneumatic extrusion nozzle system with a machine vision function according to any one of claims 1 to 5, characterized in that: The specific steps are as follows: S1: Read the pre-stored test pattern information, and use the motion control function provided by the additive manufacturing platform (13) to move the nozzle according to the predetermined trajectory. The TOF distance measurement module (7) detects the Z-axis distance between the nozzle and the printing plane in real time and sends it back to the microcontroller (2). The camera (6) is used to detect the line width, and the parameters and images are collected at the same time, and the timestamp is used for synchronization; S2: The image is imported into the classifier. The classifier is designed based on ResNet and CNN models and marks it into three types of print quality: droplets, intermittent, and lines. S3: Measure the line width of the pattern marked as a line, use the triangulation scheme to obtain the extrusion line width, and calculate the line width error rate. Import the printing parameters into the Bayesian optimization model to establish the relationship between the extrusion pressure, printing speed, nozzle distance, nozzle diameter, nozzle shape coefficient, barrel temperature, hydrogel particle diameter and the extrusion line width.

7. The pneumatic extrusion nozzle system and control method with machine vision function according to claim 6, characterized in that: Step S3 is as follows: First, the proxy model in the Bayesian optimization process is trained using a Gaussian process regression model. The input features include adjustable parameters and fixed parameters, corresponding to the input vector x = [P, v, l, d, s, T, D]. The objective function uses the mean square error between the predicted extrusion width and the actual extrusion width, and the formula is: Modeling the Gaussian process, the mapping relationship is: Considering that the optimization goal focuses more on the relative changes between errors, the mean function m(x) is assumed to be 0, and the kernel function k(x,x′) is a radial basis function kernel, which is used to describe the similarity between data points. Specifically: where σ 2 is the signal variance, controlling the function amplitude, l is the length scale, determining the correlation between features, is the noise variance; Next, we train the acquisition function in the Bayesian optimization process, using the expected improvement algorithm to quantify the potential improvement that can be achieved by sampling unexplored regions of the adjustable parameters at print time, balancing the trade-off between exploring regions with high uncertainty and exploiting regions with known low target values. Specifically, we define the current minimum error as The expected improvement is The objective function follows a Gaussian distribution, that is Improve the expectation to the loss obtained by the Gaussian process regression model, and solve the formula analytically: in Φ(Z) is the cumulative distribution function of the standard normal distribution, and φ(Z) is the probability density function of the standard normal distribution; Finally, the trained proxy model and acquisition function are applied to the test set. The proxy model predicts the extrusion width of the hydrogel based on the initial input extrusion pressure, printing speed, nozzle distance, nozzle diameter, nozzle shape factor, barrel temperature, and hydrogel particle diameter. After comparing with the target extrusion width, the error, extrusion pressure, printing speed, and nozzle distance are substituted into the acquisition function to balance the impact of different variables on the error and return the adjustable parameters that can best fit the target extrusion width.

Images