Accurate control method for glue squeezing gap of grinding wheel screen cloth

Abstract

The invention relates to a grinding wheel screen cloth glue squeezing gap accurate control method, which comprises the following steps of S1, inputting parameters on a machine learning model trained by an upper computer, automatically calculating a target position value of an actuator, and issuing the target position value of the actuator to a machine controller as an instruction; s2, the controller controls an actuator to move according to the instruction, so that a movable roller gets close to a fixed roller, and the actuator moves to a target position; s3, the movable roller and the fixed roller rotate, glue extrusion is conducted on the gray fabric to be subjected to glue extrusion, and the glue extrusion pressure and distance are monitored in real time through a pressure sensor and a position sensor; the core of the control method is that a sensor data fusion and artificial intelligence machine learning method is utilized to measure and count proper process parameter characteristics to participate in feedback control; the relative position of the rubber extruding roller is accurately controlled by utilizing sensor fusion of position closed-loop control and pressure closed-loop control and influence factor characteristic data fusion, so that the rubber extruding gap control meets the process requirement.

Description

Technical Field

[0001] This invention relates to the field of adhesive control technology for abrasive wheel mesh fabric, and in particular to a method for precise control of the adhesive extrusion gap of abrasive wheel mesh fabric. Background Technology

[0002] The coating production of abrasive wheel mesh requires coating a certain thickness of adhesive onto the raw fabric to achieve a certain adhesive content, which is an important quality indicator in the production of abrasive wheel mesh.

[0003] During the coating process of abrasive mesh, the coated fabric passes between two extrusion rollers and is squeezed by the rollers to remove excess adhesive, leaving a suitable amount for the next process. Therefore, the gap between the two extrusion rollers is a crucial factor in controlling the adhesive content on the abrasive mesh. Currently, most methods for controlling the roller gap use servo motor positioning or pneumatic cylinder pistons to control the roller shafts, maintaining a certain distance between them. However, the roller bodies are often made of composite materials, which exhibit some expansion under prolonged immersion in organic solvents. This causes changes in the roller diameter over time, altering the gap between the rollers and resulting in deviations in the adhesive application, thus affecting the accuracy of the adhesive content on the abrasive mesh.

[0004] Therefore, this invention proposes a method for precise control of the gap between the extrusion of the grinding wheel mesh and the adhesive to solve the above problems. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for precise control of the extrusion gap of the abrasive wheel mesh, which takes into account both the distance between the extrusion roller shafts and the changes in the roller body size, and realizes closed-loop control of position and closed-loop control of pressure, so as to make the extrusion gap control more accurate.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: a method for precise control of the glue extrusion gap of a grinding wheel mesh cloth, which is accomplished by a grinding wheel mesh cloth glue control device. The grinding wheel mesh cloth glue control device includes a frame, two glue extrusion rollers and an actuator. The two glue extrusion rollers are a fixed roller and a movable roller, respectively. The two ends of the fixed roller are mounted on the frame through fixed blocks, and the two ends of the movable roller are movably mounted on the frame through movable blocks. The movable roller is driven by the actuator to move closer to or away from the fixed roller. The innovation is that a pressure sensor for feedback of the glue extrusion pressure between the two glue extrusion rollers is installed on the fixed block, and a distance sensor for detecting the position of the movable roller is installed on the frame. The precise control method includes the following steps: S1. Input parameters into the trained machine learning model on the host computer. The machine learning model automatically calculates the target position value of the actuator. Through industrial LAN bus communication, the target position value of the actuator is sent as an instruction to the machine controller. At the same time, the machine controller transmits the feedback values ​​of the ranging sensor and the pressure sensor to the host computer. S2. Pass the fabric to be extruded through the fixed roller and the moving roller. The controller controls the actuator to move according to the instruction so that the moving roller is close to the fixed roller. The actuator moves to the target position. At this time, the distance between the two extrusion rollers reaches the extrusion gap. The feedback value of the pressure sensor at this time is used as the extrusion pressure control value. S3. The moving roller and the fixed roller rotate to extrude glue into the fabric to be extruded. Pressure sensors and position sensors are used to monitor the extrusion pressure and the distance between the two extrusion rollers in real time. During the extrusion process, when the extrusion pressure changes beyond the required error range, the host computer sends a new actuator position value, and the machine controller controls the actuator to move and adjust the position of the moving roller until the pressure value detected by the pressure sensor returns to the extrusion pressure control value.

[0007] Furthermore, in step S1, the method for training the machine learning model is as follows: S1-1. Within the machine learning model, multi-source sensor data is fused using machine learning algorithms to construct a mapping model between sensor signal input and actuator target position value output, taking the thickness of the fabric as an example. The extrusion gap is fed back by the position sensor. The extrusion pressure fed back by the pressure sensor The extrusion temperature fed back by the temperature sensor Ambient humidity as reported by humidity sensor Usage time of extrusion rollers , finished product rubber content The independent variable is the actuator target position value fed back by the ranging sensor. As the dependent variable, conduct field experiments and record experimental data; S1-2. Input the feature vector into the mapping model, and the mapping model outputs the predicted actuator target position value. ,but: in, For the feature vectors of the mapping model, ; This is a mapping model function, which is a combination of the ensemble function of the random forest model and the activation function of the neural network regression model; For mapping model parameters; Train the mapping model by minimizing the prediction error: in, As the actual location label, i.e., the first one during the field test The target position value of the actuator fed back by the ranging sensor of each test sample; The number of test samples; ; S1-3. Put the trained machine learning model into the host computer program, and interact with the machine controller through the industrial LAN communication in conjunction with the operation interface.

[0008] Furthermore, in steps S1-2, when training the mapping model by minimizing the prediction error, a random forest ensemble learning method is used to make the final prediction by combining the prediction results of all decision trees. The detailed method includes the following steps: S1-2-1, Random Forest Model Decision Tree composition; in, It is the input feature vector of the random forest model; It is the first The mapping model parameters of each decision tree are randomly determined during the training process; It is the first Decision trees for input The prediction; This represents the total number of decision trees, set to 1000. Set the maximum number of features for each decision tree to 7, i.e., the number of features... Use all, set the maximum depth to 15, and set the minimum number of samples per leaf node to 5; S1-2-2. The coefficient of determination (R-squared) is selected as the evaluation metric for the mapping model. Combined with the mean squared error (MSE), the training results of the mapping model are evaluated. Using a R-squared of 0.9 as the selection criterion, the total number of decision trees and the maximum number of features per decision tree are adjusted. Then, with the R-squared reaching 0.9, the maximum depth of the decision trees and the minimum number of samples per leaf node are fine-tuned to bring the MSE close to 0.1 mm. 2 Finally, the importance of the features is output and the importance of the features is ranked. S1-2-3. A reference output is generated using a neural network regression model as a control. The neural network regression uses a multi-layer structure of input layer → hidden layer → output layer to learn the non-linear mapping between features and continuous target values. Backpropagation optimizes the loss function, which is the mean squared error. The input layer to the hidden layer uses 16 neurons, the hidden layer has 1 layer of 16 neurons, and the output layer has 1 neuron. The activation function is ReLU, and the learning rate is set to 0.001. By adjusting the parameters of the number of rounds and the batch size, the coefficient of determination is made to reach the target of 0.9. S1-2-4. In the gap distance generated by the decision model and the neural network regression model, the output of the random forest model is the main factor, and the output of the neural network regression model is the reference. When the output error between the decision model and the neural network regression model reaches 10%, a prompt is issued on the host computer interface, allowing for manual intervention or control.

[0009] Furthermore, in step S1, the input parameters include the thickness of the fabric. The extrusion gap is fed back by the position sensor. The extrusion pressure fed back by the pressure sensor The extrusion temperature fed back by the temperature sensor Ambient humidity as reported by humidity sensor Usage time of extrusion rollers , finished product rubber content .

[0010] Furthermore, in step S3, when adjusting the position of the moving roller, the machine controller determines the instruction position of the actuator according to the position instruction issued by the host computer, and after issuing a control instruction to the actuator, the actuator drives the moving roller to move, and uses a distance sensor to detect the actual position of the moving roller. The machine controller compares the actual position of the moving roller with the position of the moving roller corresponding to the instruction, generates an error signal, and the actuator continues to operate according to the error signal until the error is eliminated and the moving roller reaches the position corresponding to the instruction.

[0011] Furthermore, when the moving roller reaches the position corresponding to the instruction, the host computer determines whether the pressure value detected by the pressure sensor has recovered to the extrusion pressure target value. If it has not recovered, the host computer sends a new actuator position target value to the controller to control the actuator to continue to operate until the pressure value detected by the pressure sensor recovers to the extrusion pressure control value.

[0012] The advantages of this invention are: The core of the extrusion gap control method of the present invention is to use sensor data fusion and artificial intelligence machine learning to measure and statistically determine suitable process parameter characteristics and participate in feedback control; by using sensor fusion and influencing factor characteristic data fusion of position closed-loop control and pressure closed-loop control, the relative position of the extrusion roller is accurately controlled, so that the extrusion gap control meets the process requirements.

[0013] The extrusion gap control method of the present invention not only considers the distance between the shaft centers of the two extrusion rollers, using an actuator to control the shaft center positions of the two extrusion rollers, and using a distance sensor to feed back the position of the extrusion shaft, forming a position closed loop; it also considers the changes in the shaft itself, using a pressure sensor to feed back the pressure changes between the two extrusion rollers caused by changes in the outer diameter of the rollers, forming a pressure closed loop.

[0014] (3) The extrusion gap control method of the present invention uses sensor measurement and artificial intelligence big data analysis to train a machine learning model, calculates the target position value of the actuator using the model, and participates in feedback control, realizing the composite control of position closed-loop control and pressure closed-loop control, accurately controlling the relative position of the two extrusion rollers, making the extrusion gap control more reasonable and accurate. Attached Figure Description

[0015] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0016] Figure 1 This is a schematic diagram of the structure of the grinding wheel mesh adhesive control device of the present invention.

[0017] Figure 2 This is a schematic diagram of the method for precisely controlling the gap of the abrasive wheel mesh extrusion according to the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0019] Example This embodiment provides a method for precise control of the adhesive extrusion gap of an abrasive wheel mesh, which is accomplished using an abrasive wheel mesh adhesive control device, such as... Figure 1As shown, the abrasive wheel mesh adhesive control device includes a frame 1, a fixed roller 2, a movable roller 3, and a controller. The two ends of the fixed roller 2 are mounted on the frame 1 via fixing blocks 4. The two ends of the movable roller 3 are movably mounted on the frame 1 via movable blocks 5 and are driven by an actuator 6 to move closer to or away from the fixed roller 2. A pressure sensor 7 is installed on the fixing blocks 4 to provide feedback on the adhesive extrusion pressure between the movable roller 3 and the fixed roller 2. A distance sensor 8 is installed on the frame 1 to detect the position of the movable roller 3. In this embodiment, the fixed roller 2 is a steel roller, and the movable roller 3 is a rubber roller. The actuator 6 is a servo electric cylinder or a servo hydraulic cylinder.

[0020] like Figure 2 As shown, the precise control method includes the following steps: S1. Input parameters into the trained machine learning model on the host computer. The input parameters include the thickness of the fabric. The extrusion gap is fed back by the position sensor. The extrusion pressure fed back by the pressure sensor The extrusion temperature fed back by the temperature sensor Ambient humidity as reported by humidity sensor Usage time of extrusion rollers , finished product rubber content The machine learning model automatically calculates the target position value of the actuator and sends the target position value of the actuator as an instruction to the machine controller through industrial LAN bus communication. At the same time, the machine controller transmits the feedback values ​​of the ranging sensor and the pressure sensor to the host computer.

[0021] The method for training the machine learning model is as follows: S1-1. Within the machine learning model, multi-source sensor data is fused using machine learning algorithms to construct a mapping model between sensor signal input and actuator target position value output, taking the thickness of the fabric as an example. The extrusion gap is fed back by the position sensor. The extrusion pressure fed back by the pressure sensor The extrusion temperature fed back by the temperature sensor Ambient humidity as reported by humidity sensor Usage time of extrusion rollers , finished product rubber content The independent variable is the actuator target position value fed back by the ranging sensor. As the dependent variable, conduct field experiments and record the experimental data.

[0022] S1-2. Input the feature vector into the mapping model, and the mapping model outputs the predicted actuator target position value. ,but: in, For the feature vectors of the mapping model, ; This is a mapping model function, which is a combination of the ensemble function of the random forest model and the activation function of the neural network regression model; For mapping model parameters; Train the mapping model by minimizing the prediction error: in, As the actual location label, i.e., the first one during the field test The target position value of the actuator fed back by the ranging sensor of each test sample; The number of test samples; .

[0023] When training the mapping model by minimizing the prediction error, a random forest ensemble learning method is used. The final prediction is made by combining the predictions from all decision trees. The core principle is to output the result through voting among multiple decision trees. The detailed method includes the following steps: S1-2-1, Random Forest Model Decision Tree composition; in, It is the input feature vector of the random forest model; It is the first The mapping model parameters of each decision tree are randomly determined during the training process; It is the first Decision trees for input The prediction; This represents the total number of decision trees, set to 1000. The maximum number of features per decision tree controls the number of features randomly selected during splitting. The maximum number of features per decision tree is set to 7, i.e., the maximum number of features... Use all; The maximum depth of a decision tree determines the complexity of a single decision tree; the maximum depth of each decision tree is set to 15. To prevent excessive splitting of a single decision tree, the minimum number of samples per leaf node is set to 5, since samples can be collected over a long period of time. S1-2-2. The coefficient of determination (R-squared) is selected as the evaluation metric for the mapping model. Combined with the mean squared error (MSE), the training results of the mapping model are evaluated. Using a R-squared of 0.9 as the selection criterion, the total number of decision trees and the maximum number of features per decision tree are adjusted. Then, with the R-squared reaching 0.9, the maximum depth of the decision trees and the minimum number of samples per leaf node are fine-tuned to bring the MSE close to 0.1 mm.2 Finally, the importance of the features is output and ranked. This fully utilizes the good interpretability of the random forest model, providing an important reference for adjusting process parameters. S1-2-3. A neural network regression model is used to generate a reference output as a control. The neural network regression learns the nonlinear mapping between features and continuous target values ​​through a multi-layer structure of "input layer → hidden layer → output layer". Backpropagation optimizes the loss function, which is the mean squared error. The input layer to the hidden layer uses 16 neurons, the hidden layer has 1 layer of 16 neurons, and the output layer has 1 neuron. The activation function is ReLU, and the learning rate is set to 0.001. By adjusting the parameters of the number of rounds and the batch size, the coefficient of determination is made to reach the target of 0.9. S1-2-4. In the gap distance generated by the decision model and the neural network regression model, the output of the random forest model is the main factor, and the output of the neural network regression model is the reference. When the output error between the decision model and the neural network regression model reaches 10%, a prompt is issued on the host computer interface, allowing for manual intervention or control.

[0024] S1-3. Put the trained machine learning model into the host computer program, and interact with the machine controller through the industrial LAN communication in conjunction with the operation interface.

[0025] S2. Pass the fabric to be extruded through the fixed roller and the moving roller. The controller controls the actuator to move according to the instruction so that the moving roller is close to the fixed roller. The actuator moves to the target position. At this time, the distance between the two extrusion rollers reaches the extrusion gap. The feedback value of the pressure sensor at this time is used as the extrusion pressure control value. S3. The moving roller and the fixed roller rotate to extrude glue into the fabric to be extruded. Pressure sensors and position sensors are used to monitor the extrusion pressure and the distance between the two extrusion rollers in real time. During the extrusion process, when the extrusion pressure changes beyond the required error range, the host computer sends a new actuator position value, and the machine controller controls the actuator to move and adjust the position of the moving roller until the pressure value detected by the pressure sensor returns to the extrusion pressure control value.

[0026] When adjusting the position of the moving roller, the machine controller determines the position of the actuator according to the position command issued by the host computer. After issuing the control command to the actuator, the actuator drives the moving roller to move and uses a distance sensor to detect the actual position of the moving roller. The machine controller compares the actual position of the moving roller with the position of the moving roller corresponding to the command and generates an error signal. The actuator continues to operate according to the error signal until the error is eliminated and the moving roller reaches the position corresponding to the command.

[0027] When the moving roller reaches the position corresponding to the instruction, the host computer determines whether the pressure value detected by the pressure sensor has recovered to the extrusion pressure target value. If it has not recovered, the host computer sends a new actuator position target value to the controller to control the actuator to continue to operate until the pressure value detected by the pressure sensor recovers to the extrusion pressure control value.

[0028] The extrusion gap control method of this invention not only considers the distance between the axes of the two extrusion rollers, using actuators to control the axis positions of the two extrusion rollers and using distance sensors to provide feedback on the position of the extrusion shaft, forming a position closed loop; it also considers the changes in the shaft itself, using pressure sensors to provide feedback on the pressure changes between the two extrusion rollers caused by changes in the outer diameter of the rollers, forming a pressure closed loop; and it uses sensor measurement and artificial intelligence big data analysis methods to train a machine learning model, using the model to calculate the target position value of the actuator and participate in feedback control, realizing a composite control of position closed loop control and pressure closed loop control, accurately controlling the relative position of the two extrusion rollers, making the extrusion gap control more reasonable and accurate.

[0029] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for precisely controlling the glue extrusion gap of a grinding wheel mesh fabric, comprising a grinding wheel mesh fabric glue control device, the grinding wheel mesh fabric glue control device including a frame, two glue extrusion rollers and an actuator, the two glue extrusion rollers being a fixed roller and a movable roller, the two ends of the fixed roller being mounted on the frame via fixed blocks, and the two ends of the movable roller being movably mounted on the frame via movable blocks, and driven by the actuator to move closer to or away from the fixed roller, characterized in that: The fixed block is equipped with a pressure sensor for feeding back the extrusion pressure between the two extrusion rollers, and the frame is equipped with a distance sensor for detecting the position of the moving roller. The precise control method includes the following steps: S1. Input parameters into the trained machine learning model on the host computer. The machine learning model automatically calculates the target position value of the actuator. Through industrial LAN bus communication, the target position value of the actuator is sent as an instruction to the machine controller. At the same time, the machine controller transmits the feedback values ​​of the ranging sensor and the pressure sensor to the host computer. S2. Pass the fabric to be extruded through the fixed roller and the moving roller. The controller controls the actuator to move according to the instruction so that the moving roller is close to the fixed roller. The actuator moves to the target position. At this time, the distance between the two extrusion rollers reaches the extrusion gap. The feedback value of the pressure sensor at this time is used as the extrusion pressure control value. S3. The moving roller and the fixed roller rotate to extrude glue into the fabric to be extruded. Pressure sensors and position sensors are used to monitor the extrusion pressure and the distance between the two extrusion rollers in real time. During the extrusion process, when the extrusion pressure changes beyond the required error range, the host computer sends a new actuator position value, and the machine controller controls the actuator to move and adjust the position of the moving roller until the pressure value detected by the pressure sensor returns to the extrusion pressure control value.

2. The method for precisely controlling the gap of the abrasive wheel mesh extrusion according to claim 1, characterized in that: In step S1, the method for training the machine learning model is as follows: S1-1. Within the machine learning model, multi-source sensor data is fused using machine learning algorithms to construct a mapping model between sensor signal input and actuator target position value output, taking the thickness of the fabric as an example. The extrusion gap is fed back by the position sensor. The extrusion pressure fed back by the pressure sensor The extrusion temperature fed back by the temperature sensor Ambient humidity as reported by humidity sensor Usage time of extrusion rollers , finished product adhesive content The independent variable is the actuator target position value fed back by the ranging sensor. As the dependent variable, conduct field experiments and record experimental data; S1-2. Input the feature vector into the mapping model, and the mapping model outputs the predicted actuator target position value. ,but: in, For the feature vectors of the mapping model, ; This is a mapping model function, which is a combination of the ensemble function of the random forest model and the activation function of the neural network regression model; For mapping model parameters; Train the mapping model by minimizing the prediction error: in, As the actual location label, i.e., the first one during the field test The target position value of the actuator fed back by the ranging sensor of each test sample; The number of test samples; ; S1-3. Put the trained machine learning model into the host computer program, and interact with the machine controller through the industrial LAN communication in conjunction with the operation interface.

3. The method for precisely controlling the gap of the abrasive wheel mesh extrusion according to claim 2, characterized in that: In step S1-2, when training the mapping model by minimizing the prediction error, a random forest ensemble learning method is used. The final prediction is made by combining the prediction results of all decision trees. The detailed method includes the following steps: S1-2-1, Random Forest Model Decision Tree composition; in, It is the input feature vector of the random forest model; It is the first The mapping model parameters of each decision tree are randomly determined during the training process; It is the first Decision trees for input The prediction; This represents the total number of decision trees, set to 1000. Set the maximum number of features for each decision tree to 7, i.e., the number of features... Use all, set the maximum depth to 15, and set the minimum number of samples per leaf node to 5; S1-2-2. The coefficient of determination (R-squared) is selected as the evaluation metric for the mapping model. Combined with the mean squared error (MSE), the training results of the mapping model are evaluated. Using a R-squared of 0.9 as the selection criterion, the total number of decision trees and the maximum number of features per decision tree are adjusted. Then, with the R-squared reaching 0.9, the maximum depth of the decision trees and the minimum number of samples per leaf node are fine-tuned to bring the MSE close to 0.1 mm. 2 Finally, the importance of the features is output and the importance of the features is ranked. S1-2-3. A reference output is generated using a neural network regression model as a control. The neural network regression uses a multi-layer structure of input layer → hidden layer → output layer to learn the non-linear mapping between features and continuous target values. Backpropagation optimizes the loss function, which is the mean squared error. The input layer to the hidden layer uses 16 neurons, the hidden layer has 1 layer of 16 neurons, and the output layer has 1 neuron. The activation function is ReLU, and the learning rate is set to 0.

001. By adjusting the parameters of the number of rounds and the batch size, the coefficient of determination is made to reach the target of 0.

9. S1-2-4. In the gap distance generated by the decision model and the neural network regression model, the output of the random forest model is the main factor, and the output of the neural network regression model is the reference. When the output error between the decision model and the neural network regression model reaches 10%, a prompt is issued on the host computer interface, allowing for manual intervention or control.

4. The method for precisely controlling the gap of the abrasive wheel mesh extrusion according to claim 1, characterized in that: In step S1, the input parameters include the thickness of the fabric. The extrusion gap is fed back by the position sensor. The extrusion pressure fed back by the pressure sensor The extrusion temperature fed back by the temperature sensor Ambient humidity as reported by humidity sensor Usage time of extrusion rollers , finished product adhesive content .

5. The method for precisely controlling the gap of the abrasive wheel mesh extrusion according to claim 1, characterized in that: In step S3, when adjusting the position of the moving roller, the machine controller determines the instruction position of the actuator according to the position instruction issued by the host computer. After issuing the control instruction to the actuator, the actuator drives the moving roller to move and uses a distance sensor to detect the actual position of the moving roller. The machine controller compares the actual position of the moving roller with the position of the moving roller corresponding to the instruction and generates an error signal. The actuator continues to operate according to the error signal until the error is eliminated and the moving roller reaches the position corresponding to the instruction.

6. The method for precisely controlling the gap of the abrasive wheel mesh extrusion according to claim 5, characterized in that: When the moving roller reaches the position corresponding to the instruction, the host computer determines whether the pressure value detected by the pressure sensor has recovered to the extrusion pressure target value. If it has not recovered, the host computer sends a new actuator position target value to the controller to control the actuator to continue to operate until the pressure value detected by the pressure sensor recovers to the extrusion pressure control value.

Images