Electrical parts burr grinding control system

Abstract

The present application provides a burr grinding control system for electrical accessories, which relates to the field of grinding processing technology. The system includes: a first predicted roughness generation module, which is used to perform synchronous roughness prediction through a digital twin roughness prediction unit to generate a first predicted roughness; a second predicted roughness generation module, which is used to perform roughness prediction on a grinding acoustic emission pulse signal through a memory network roughness prediction unit to generate a second predicted roughness; and a termination module, which is used to terminate the grinding operation when the second predicted roughness meets the expected grinding roughness index. The present application can solve the technical problems of insufficient precision of grinding control, low grinding efficiency, and easy over-grinding due to the inability to accurately predict the surface roughness of ceramics during the grinding process. The accuracy and reliability of ceramic surface roughness prediction can be improved, and the technical effect of improving grinding efficiency and processing quality can be achieved.

Description

Technical Field

[0001] The present application relates to the technical field of grinding processing, and in particular to a burr grinding control system for electrical accessories. Background Art

[0002] With the rapid development of the electronics industry, the performance requirements for circuit board materials are increasing. Ceramic circuit boards have been widely used in aerospace, automotive electronics, electrical equipment and other fields due to their excellent electrical insulation, thermal stability, chemical stability and mechanical strength. However, the hardness and brittleness of ceramic materials make their processing relatively difficult, especially in the grinding process. Therefore, how to improve the grinding quality of ceramic circuit boards is very critical to their performance.

[0003] At present, the existing ceramic circuit board grinding processing methods cannot accurately predict the ceramic surface roughness during the grinding process, resulting in insufficient precision in grinding control, causing low grinding efficiency and prone to technical problems such as over-grinding. Summary of the invention

[0004] The purpose of this application is to provide a burr grinding control system for electrical accessories to solve the technical problem of existing ceramic circuit board grinding methods, which cannot accurately predict the surface roughness of the ceramic during the grinding process, resulting in insufficient accuracy of grinding control, low grinding efficiency, and easy occurrence of over-grinding.

[0005] In view of the above problems, the present application provides a burr grinding control system for electrical accessories. The system includes: an information acquisition module for acquiring the attribute information and expected grinding roughness index of the target ceramic circuit board; a digital twin roughness prediction unit construction module for generating a grinding process twin model based on the attribute information, grinding wheel abrasive parameters and grinding operation parameters based on digital twin technology, and building a digital twin roughness prediction unit in combination with the grinding prediction model, wherein the grinding prediction model is constructed based on a BP neural network; a first predicted roughness generation module for performing the grinding process of the target ceramic circuit board, and performing synchronous roughness prediction through the digital twin roughness prediction unit to generate a first predicted roughness; a grinding acoustic emission pulse signal A signal collection module is used to activate the sensor array to collect the grinding acoustic emission pulse signal when the first predicted roughness meets the first roughness threshold, and transmit the grinding acoustic emission pulse signal to the memory network roughness prediction unit through the preamplifier; a second predicted roughness generation module is used to predict the roughness of the grinding acoustic emission pulse signal through the memory network roughness prediction unit to generate a second predicted roughness, wherein the memory network roughness prediction unit is constructed based on a convolutional network and a long short-term memory network; a grinding operation termination module is used to terminate the grinding operation of the target ceramic circuit board when the second predicted roughness meets the expected grinding roughness index.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] By providing a burr grinding control system for electrical accessories, the existing ceramic circuit board grinding method can be solved. Since it is impossible to accurately predict the ceramic surface roughness during the grinding process, the grinding control accuracy is insufficient, resulting in low grinding efficiency and the technical problem of over-grinding. The system includes: an information acquisition module for acquiring the attribute information and expected grinding roughness index of the target ceramic circuit board; a digital twin roughness prediction unit construction module for generating a grinding twin model based on the digital twin technology according to the attribute information, grinding wheel abrasive parameters and grinding operation parameters, and building a digital twin roughness prediction unit in combination with the grinding prediction model, wherein the grinding prediction model is constructed based on the BP neural network; a first predicted roughness generation module for executing the target ceramic circuit board; a digital twin roughness prediction unit for generating a digital twin roughness prediction unit based on the attribute information, grinding wheel abrasive parameters and grinding operation parameters; a digital twin roughness prediction unit for generating a digital twin roughness prediction unit based on the grinding prediction model ... The invention relates to a grinding process of a road plate, and performs synchronous roughness prediction through the digital twin roughness prediction unit to generate a first predicted roughness; a grinding acoustic emission pulse signal acquisition module, which is used to activate the sensor array to collect the grinding acoustic emission pulse signal when the first predicted roughness meets the first roughness threshold, and transmit the grinding acoustic emission pulse signal to the memory network roughness prediction unit through the preamplifier; a second predicted roughness generation module, which is used to perform roughness prediction on the grinding acoustic emission pulse signal through the memory network roughness prediction unit to generate a second predicted roughness, wherein the memory network roughness prediction unit is constructed based on a convolutional network and a long short-term memory network; a grinding operation termination module, which is used to terminate the grinding operation of the target ceramic circuit board when the second predicted roughness meets the expected grinding roughness index. That is to say: by building a grinding twin model based on digital twin technology, and combining the grinding prediction model to form a digital twin roughness prediction unit; then during the grinding of the ceramic circuit board, the digital twin roughness prediction unit performs synchronous roughness prediction to obtain the first predicted roughness; further, when the first predicted roughness is greater than or equal to the first roughness threshold, the sensor array is activated to collect the grinding acoustic emission pulse signal; then the memory network roughness prediction unit constructed based on the convolutional network and the long short-term memory network performs roughness prediction on the grinding acoustic emission pulse signal to obtain the second predicted roughness; finally, when the second predicted roughness is greater than or equal to the expected grinding roughness index, the grinding operation of the target ceramic circuit board is terminated. The accuracy and reliability of the ceramic surface roughness prediction during the grinding process can be improved, the technical goal of accurately controlling the grinding of the ceramic circuit board can be achieved, and the technical effect of improving the grinding efficiency and processing quality can be achieved.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0010] Figure 1 A schematic diagram of the structure of a burr grinding control system for electrical accessories in this application;

[0011] Figure 2 A schematic diagram of the process of constructing a grinding prediction model in a burr grinding control system for electrical accessories in this application.

[0012] Description of reference numerals:

[0013] Information acquisition module 11, digital twin roughness prediction unit building module 12, first predicted roughness generation module 13, grinding acoustic emission pulse signal acquisition module 14, second predicted roughness generation module 15, grinding operation termination module 16. DETAILED DESCRIPTION

[0014] This application provides a burr grinding control system for electrical accessories, which solves the existing ceramic circuit board grinding method. Since the ceramic surface roughness during the grinding process cannot be accurately predicted, the grinding control accuracy is insufficient, resulting in low grinding efficiency and easy over-grinding. The accuracy and reliability of the ceramic surface roughness prediction during the grinding process can be improved, the technical goal of accurately controlling the grinding of ceramic circuit boards can be achieved, and the technical effect of improving the grinding efficiency and processing quality can be achieved.

[0015] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them. Example

[0016] Please see attached Figure 1 , the present application provides a burr grinding control system for electrical accessories, comprising:

[0017] The information acquisition module 11 is used to acquire the property information and expected grinding roughness index of the target ceramic circuit board.

[0018] Specifically, first, the property information and expected grinding roughness index of the target ceramic circuit board are obtained, and the property information includes information such as material properties, geometric dimensions, and surface state. The material properties include physical properties such as material type, composition, hardness, and toughness. The surface state includes the surface roughness of the surface to be ground of the target ceramic circuit board. The expected grinding roughness index refers to the target value expected to be achieved for the surface roughness of the ceramic circuit board during the grinding process, which can be obtained through the grinding process requirements of the target ceramic circuit board. By obtaining the property information and expected grinding roughness index of the target ceramic circuit board, data support is provided for the next step of grinding process control analysis of the target ceramic circuit board.

[0019] The digital twin roughness prediction unit building module 12 is used to generate a grinding process twin model based on the digital twin technology according to the attribute information, grinding wheel abrasive parameters and grinding operation parameters, and build a digital twin roughness prediction unit in combination with the grinding prediction model, wherein the grinding prediction model is constructed based on a BP neural network.

[0020] Specifically, digital twin technology is a technology that uses physical models, sensor updates, operation history and other data to complete mapping in virtual space, thereby reflecting the entire life cycle of the corresponding physical equipment. By integrating multi-disciplinary, multi-physical quantity, multi-scale, and multi-probability simulation processes, it creates digital twins of physical entities to achieve real-time monitoring, predictive maintenance, optimized design, and innovation management of physical equipment.

[0021] First, the grinding wheel abrasive parameters and grinding operation parameters of the grinding equipment are obtained, wherein the grinding wheel abrasive parameters are the abrasive characteristics of the grinding wheel, which have a significant impact on the grinding process and results; the grinding operation parameters include grinding speed, grinding feed, spindle speed, etc., which can be set according to actual conditions. Then, based on the digital twin technology, the target ceramic circuit board grinding process is simulated and modeled according to the attribute information, grinding wheel abrasive parameters and grinding operation parameters, and a grinding twin model is generated; by generating a grinding twin model, the grinding state of the target ceramic circuit board can be simulated in real time, while improving the authenticity and accuracy of the grinding simulation.

[0022] The BP neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm. It can be iteratively optimized and obtained through supervised training with sample data. It has many advantages such as strong adaptability and high fault tolerance. At the same time, it also has the disadvantage of large prediction error. Then, a grinding prediction model is constructed based on the BP neural network, and then a digital twin roughness prediction unit is established according to the grinding process twin model and the grinding prediction model. By constructing the digital twin roughness prediction unit, support is provided for the synchronous simulation monitoring of the target ceramic circuit board during the grinding process.

[0023] The first predicted roughness generation module 13 is used to perform the grinding process of the target ceramic circuit board, and perform synchronous roughness prediction through the digital twin roughness prediction unit to generate a first predicted roughness.

[0024] Specifically, the grinding process of the target ceramic circuit board is performed by the grinding equipment according to the preset grinding process steps; and during the grinding process, the surface roughness of the ceramic circuit board is predicted respectively by the grinding process twin model and the grinding prediction model in the digital twin roughness prediction unit to obtain the simulated roughness prediction result and the BP roughness prediction result; then the simulated roughness prediction result and the BP roughness prediction result are weightedly summed according to the preset weight ratio to obtain the first predicted roughness, wherein a roughness prediction is performed each time the grinding operation parameters change.

[0025] The grinding acoustic emission pulse signal acquisition module 14 is used to activate the sensor array to collect the grinding acoustic emission pulse signal when the first predicted roughness meets the first roughness threshold, and transmit the grinding acoustic emission pulse signal to the memory network roughness prediction unit through the preamplifier.

[0026] Specifically, a first roughness threshold is obtained, wherein the first roughness threshold is set based on the prediction time of the subsequent memory network roughness and the prediction accuracy of the digital twin roughness prediction unit, wherein the shorter the prediction time and the higher the prediction accuracy, the closer the first roughness threshold is to the expected grinding roughness index.

[0027] Then, the first predicted roughness is judged based on the first roughness threshold. When the first predicted roughness is greater than or equal to the first roughness threshold, the sensor array is activated, wherein the sensor array includes multiple AE sensors, and the output result is the mode value of the data collected by the multiple AE sensors, and the number of AE sensors can be set according to actual conditions; then, the grinding acoustic emission pulse signal in the grinding process is collected through the sensor array to obtain a grinding acoustic emission pulse signal, wherein in the ceramic grinding process, the interaction between the abrasive particles and the workpiece surface will generate various mechanical stresses and vibrations, and these stresses and vibrations will be released in the form of acoustic emission to form an acoustic emission pulse signal. The grinding acoustic emission pulse signal is correlated with the roughness of the object's grinding surface. By analyzing the characteristics of the grinding acoustic emission pulse signal, the ceramic surface roughness can be predicted. The grinding acoustic emission pulse signal is then transmitted to the memory network roughness prediction unit through a preamplifier, wherein the main function of the preamplifier is to amplify the weak signal captured from the acoustic emission sensor to an appropriate level for subsequent signal processing and analysis. Since the acoustic emission signal is usually weak, a low-noise, high-gain preamplifier is required to ensure the signal-to-noise ratio and accuracy of the signal; the acoustic emission signal amplified by the preamplifier is then transmitted to the memory network roughness prediction unit through an appropriate signal transmission line (such as a cable, optical fiber, etc.).

[0028] The second predicted roughness generation module 15 is used to perform roughness prediction on the grinding acoustic emission pulse signal through the memory network roughness prediction unit to generate a second predicted roughness, wherein the memory network roughness prediction unit is constructed based on a convolutional network and a long short-term memory network.

[0029] Specifically, a convolutional neural network is a type of feedforward neural network that includes convolution calculations and has a deep structure. It is usually used for feature extraction. A data recognition layer is constructed based on the convolutional neural network, and the data recognition layer is used to extract features from grinding acoustic emission pulse signals. The long short-term memory network is a special recurrent neural network, including three gated units (input gate, forget gate, output gate) and a memory unit, where the three gates control the flow of input information, forget information, and output information, respectively. At each time step, the long short-term memory network calculates the cell state and hidden state at the current moment based on the input, the cell state at the previous moment, and the hidden state at the previous moment, and passes it to the next moment. This mechanism enables the long short-term memory network to better process long sequence data and retain long-term state information. Then a feature analysis layer is constructed based on the long short-term memory network, and the feature analysis layer is used to predict grinding roughness based on the feature extraction results output by the data recognition layer.

[0030] A memory network roughness prediction unit is constructed according to the data recognition layer and the feature analysis layer, wherein the output layer of the data recognition layer is connected to the input layer of the feature analysis layer, the input data of the memory network roughness prediction unit is the grinding acoustic emission pulse signal, and the output data is the roughness prediction result. Then the grinding acoustic emission pulse signal is input into the memory network roughness prediction unit for roughness prediction to obtain a second predicted roughness. Since the prediction errors of the digital twin and the BP neural network are large, the roughness prediction is first performed through the digital twin roughness prediction unit. When the prediction result is close to the expected index; then the grinding roughness is accurately predicted based on the grinding acoustic emission pulse signal through the memory network roughness prediction unit, thereby saving the prediction resource consumption of the grinding acoustic emission pulse signal while meeting the accuracy requirements of the ceramic surface roughness prediction, while improving the practicality and reliability of the prediction.

[0031] The grinding operation termination module 16 is used to terminate the grinding operation of the target ceramic circuit board when the second predicted roughness meets the expected grinding roughness index.

[0032] Specifically, the second predicted roughness is judged according to the expected grinding roughness index. When the second predicted roughness is greater than or equal to the expected grinding roughness index, it indicates that the current grinding process meets the expected index, and the grinding equipment is controlled to terminate the grinding operation of the target ceramic circuit board; when the second predicted roughness is less than the expected grinding roughness index, the grinding acoustic emission pulse signal is collected at a fixed point by the sensor array based on the preset time interval to predict the roughness, wherein the preset time interval is short and can be set according to the actual situation, for example: the preset time interval is set to 0.5 seconds, and then the roughness prediction result is judged according to the expected grinding roughness index until the roughness prediction result is greater than or equal to the expected grinding roughness index, and the grinding operation of the target ceramic circuit board is terminated, and the grinding equipment is controlled to terminate the grinding operation of the target ceramic circuit board. The above method can improve the accuracy and reliability of the ceramic surface roughness prediction during the grinding process, achieve the technical goal of accurately controlling the grinding of ceramic circuit boards, and achieve the technical effect of improving the grinding efficiency and processing quality.

[0033] By providing a burr grinding control system for electrical accessories, the existing ceramic circuit board grinding method can be solved. Since the ceramic surface roughness during the grinding process cannot be accurately predicted, the grinding control accuracy is insufficient, resulting in low grinding efficiency and easy technical problems of over-grinding. A grinding twin model is constructed based on digital twin technology, and a digital twin roughness prediction unit is formed in combination with a grinding prediction model; then, during the grinding of the ceramic circuit board, a synchronous roughness prediction is performed through the digital twin roughness prediction unit to obtain a first predicted roughness; further, when the first predicted roughness is greater than or equal to the first roughness threshold, the sensor array is activated to collect the grinding acoustic emission pulse signal; then, the grinding acoustic emission pulse signal is predicted by the memory network roughness prediction unit constructed based on the convolutional network and the long short-term memory network to obtain a second predicted roughness; finally, when the second predicted roughness is greater than or equal to the expected grinding roughness index, the grinding operation of the target ceramic circuit board is terminated. It can improve the accuracy and reliability of ceramic surface roughness prediction during the grinding process, achieve the technical goal of accurately controlling the grinding of ceramic circuit boards, and achieve the technical effect of improving grinding efficiency and processing quality.

[0034] Furthermore, the digital twin roughness prediction unit building module 12 is also used for:

[0035] The image of the grinding wheel is collected by a microscopic image sensor to obtain a microscopic image of the grinding wheel; the abrasive feature is identified based on the microscopic image of the grinding wheel to determine the abrasive parameters of the grinding wheel, wherein the abrasive parameters of the grinding wheel include the number of effective abrasives, the coordinates of the distribution of abrasives and the height of the abrasives out of the edge.

[0036] Specifically, first, observe the surface characteristics of the grinding wheel and the required resolution, select an appropriate microscopic image sensor, including a high-magnification optical microscope or a matching sensor for a scanning electron microscope (SEM), and consider parameters such as the image sensor's sensitivity, dynamic range, pixel size, and frame rate to ensure that clear and detailed microscopic images of the grinding wheel can be captured; then, use an adapted microscopic image sensor to capture the image of the grinding wheel to obtain a microscopic image of the grinding wheel. Then, the abrasive feature recognition is performed on the microscopic image of the grinding wheel, wherein the abrasive feature extraction can be performed by constructing a convolutional neural network model, and analysis is performed based on the feature extraction results, wherein for each identified abrasive particle, it is judged whether it is an effective abrasive particle according to certain rules (such as area, shape, etc.), and the number of effective abrasive particles is counted; on the basis of identifying the abrasive particles, the position of each abrasive particle in the image is determined using image processing technology (such as centroid calculation, boundary detection, etc.), the position coordinates of the effective abrasive particles are recorded, and the grinding wheel abrasive particle parameters are determined, wherein the grinding wheel abrasive particle parameters include the number of effective abrasive particles, the abrasive particle distribution coordinates, and the abrasive particle edge height, and the abrasive particle edge height refers to the distance between the abrasive particle vertex and the grinding wheel base surface. In the microscopic image, the pixel distance between the abrasive particle vertex and the grinding wheel base surface in the image can be measured, and the pixel distance can be converted into the actual edge height value in combination with the magnification of the microscope and the image resolution. By determining the grinding wheel abrasive parameters through grinding wheel microscopic image analysis, the precision of grinding wheel grinding feature analysis can be improved, thereby improving the accuracy of grinding process twin model construction and further improving the accuracy of grinding process simulation.

[0037] Further, as attached Figure 2 As shown, the digital twin roughness prediction unit building module 12 is also used for:

[0038] Taking the attribute information and the grinding wheel abrasive parameters as constraints, the grinding process log is retrieved to obtain a grinding process record data set, wherein the grinding process record data includes grinding operation parameters, grinding process time, initial surface roughness and termination surface roughness; the grinding process record data set is divided into a training data set and a verification data set according to a preset division ratio; an initial grinding prediction model is built based on a BP neural network, with the grinding operation parameters, grinding process time and initial surface roughness as inputs, and the termination surface roughness as supervision, and the initial grinding prediction model is supervised and trained and feedback verified through the training data set and the verification data set, respectively, to obtain a grinding prediction model that meets the preset convergence index.

[0039] Specifically, the grinding log is retrieved with the attribute information and the grinding wheel abrasive parameters as constraints, and a grinding record data set is obtained based on the grinding log, wherein the grinding record data set includes multiple grinding record data, and each grinding record data includes grinding operation parameters, grinding duration, initial surface roughness and termination surface roughness, wherein the termination surface roughness refers to the surface roughness of the ceramic after the grinding process is completed. Then, the grinding record data set is divided into a training data set and a verification data set according to a preset division ratio, wherein the preset division ratio can be set according to actual conditions, for example: the training data ratio can be set to 80%, and the verification data ratio can be set to 20%.

[0040] An initial grinding prediction model is built based on a BP neural network, wherein the input data of the initial grinding prediction model are grinding operation parameters, grinding processing time and initial surface roughness, and the output data is the predicted surface roughness; then, the grinding operation parameters, grinding processing time and initial surface roughness are used as input, and the terminated surface roughness is used as supervision, and the initial grinding prediction model is supervised trained and feedback verified through the training data set and the verification data set, respectively, until a grinding prediction model that meets a preset convergence index is obtained, wherein the preset convergence index can be set according to actual conditions, for example: the preset convergence index is set to an output accuracy rate greater than or equal to 85%.

[0041] Furthermore, the first predicted roughness generating module 13 is also used for:

[0042] During the grinding process of the target ceramic circuit board, the grinding twin model is synchronously run, and the grinding operation parameters are obtained in real time, wherein the grinding twin model is communicatively connected to the control unit of the grinding equipment; whenever the grinding operation parameters change, the change node is recorded and the grinding prediction model is activated, and a prediction is made based on the grinding operation parameters, the initial surface roughness and the grinding process duration within the change time period, and the surface predicted roughness is output, wherein the change time period is the time interval between the current change node and the previous adjacent change node; based on a preset fusion strategy, the surface predicted roughness is fused with the simulated operation roughness under the change node to obtain the first predicted roughness, wherein the preset fusion strategy is a preset weight ratio.

[0043] Specifically, first, during the grinding process of the target ceramic circuit board, the grinding twin model is synchronously operated, wherein the grinding twin model is connected to the control unit of the grinding equipment in communication, and the grinding operation parameters are obtained in real time through the control unit. Then, whenever the grinding operation parameters change, the change time node is recorded and the grinding prediction model is activated, and then the grinding operation parameters, initial surface roughness and grinding duration within the change time period are input into the grinding prediction model for roughness prediction, wherein the change time period is the time interval between the current change node and the previous adjacent change node, and the surface prediction roughness is obtained.

[0044] Then, based on the grinding twin model, the simulated operation roughness under the variable node is obtained; a preset fusion strategy is configured, and the preset fusion strategy is a preset weight ratio, which can be set according to the prediction accuracy of the grinding prediction model and the grinding twin model, wherein the higher the accuracy, the larger the corresponding weight, and the first weight of the grinding prediction model and the second weight of the grinding twin model are obtained, wherein the sum of the first weight and the second weight is 1; then, based on the first weight and the second weight, the surface prediction roughness and the simulated operation roughness are weightedly calculated, and the weighted calculation result is used as the first predicted roughness to obtain the first predicted roughness. By performing a fusion analysis on the surface prediction roughness and the simulated operation roughness to obtain the first predicted roughness, the accuracy and reliability of the first predicted roughness can be improved.

[0045] Furthermore, the second predicted roughness generating module 15 is also used for:

[0046] A data recognition layer is built based on a convolutional neural network, wherein the input data of the data recognition layer is a grinding acoustic emission pulse signal within a sensitive frequency band, and the output data is an identification feature sequence; a feature analysis layer is built based on a long short-term memory network, wherein the feature analysis layer is composed of a forward long short-term memory network and a backward long short-term memory network, and the forward long short-term memory network and the backward long short-term memory network perform feature extraction based on positive order and reverse order respectively; the memory network roughness prediction unit is built based on the data recognition layer and the feature analysis layer, wherein the output layer of the data recognition layer is connected to the input layer of the feature analysis layer.

[0047] Specifically, a data recognition layer is built based on a convolutional neural network, the input data of the data recognition layer is a grinding acoustic emission pulse signal within a sensitive frequency band, and the output data is an identification feature sequence, wherein the sensitive frequency band refers to the frequency band in the grinding acoustic emission pulse signal that has a greater impact on the grinding roughness prediction result, that is, a sensitive frequency band with a higher degree of correlation, such as: the frequency is between 65 and 70KHZ. A feature analysis layer is built based on a long short-term memory network, wherein the feature analysis layer is composed of a forward long short-term memory network and a backward long short-term memory network, wherein the forward long short-term memory network and the backward long short-term memory network perform feature extraction based on forward and reverse order respectively, and can learn feature data bidirectionally, so the feature extraction efficiency and performance of the network are better than those of a single long short-term memory network.

[0048] Then, the memory network roughness prediction unit is built based on the data recognition layer and the feature analysis layer, wherein the memory network roughness prediction unit includes an input layer, a data recognition layer, a feature analysis layer and an output layer, wherein the input data of the input layer is a grinding acoustic emission pulse signal, and the output data is a roughness prediction result, wherein the output layer of the data recognition layer is connected to the input layer of the feature analysis layer.

[0049] Furthermore, the second predicted roughness generating module 15 is also used for:

[0050] A sample grinding acoustic emission pulse signal set and a sample grinding surface roughness set are collected and obtained, wherein the sample grinding acoustic emission pulse signal and the sample grinding surface roughness correspond one to one; the sample grinding acoustic emission pulse signal set and the sample grinding surface roughness set are analyzed based on the wavelet packet analysis principle to determine the main characteristic frequency band range and the corresponding grinding surface roughness set; a correlation evaluation function is constructed: ;in, is the correlation evaluation coefficient, is the characteristic value, is the grinding surface roughness, for and The covariance of for The variance of for variance; based on the correlation evaluation function, the main characteristic frequency band range is screened to obtain the sensitive frequency band range.

[0051] Specifically, first, a sample grinding acoustic emission pulse signal set and a sample grinding surface roughness set are collected, wherein the sample grinding acoustic emission pulse signal and the sample grinding surface roughness correspond one to one; then, the sample grinding acoustic emission pulse signal set and the sample grinding surface roughness set are analyzed based on the principle of wavelet packet analysis, wherein wavelet packet analysis is a very effective signal spectrum analysis method, which decomposes the signal into wavelet basis functions with different frequencies and time resolutions, and then automatically selects appropriate frequency bands to match the characteristics of the analyzed signal to ensure the accuracy and real-time performance of signal analysis and improve the reliability of feature extraction; for example: analyzing the correlation between the eigenvalues ​​within the main characteristic frequency band range and the grinding surface roughness, statistical methods (such as correlation coefficient, regression analysis, etc.) can be used to quantify this correlation, and then the characteristic frequency band range closely related to the grinding surface roughness is determined based on the results of the correlation analysis; the main characteristic frequency band range and the corresponding grinding surface roughness set are obtained.

[0052] Then construct the correlation evaluation function: ; In the correlation evaluation function, is the correlation evaluation coefficient, where the larger the correlation evaluation coefficient is, the higher the degree of correlation is; is the characteristic value, is the grinding surface roughness, for and The covariance of for The variance of for Further, the characteristic values ​​within the main characteristic frequency band are evaluated for correlation according to the correlation evaluation function, and a frequency band with a higher correlation is selected as the sensitive frequency band.

[0053] Furthermore, the second predicted roughness generating module 15 is also used for:

[0054] The main characteristic frequency band range is divided according to a preset step size to determine a plurality of characteristic frequency band intervals, and a first characteristic frequency band interval is selected from the plurality of characteristic frequency band intervals to obtain a first grinding surface roughness set corresponding to the first characteristic frequency band interval; according to the correlation evaluation function, a correlation evaluation is performed on the characteristic values ​​in the first characteristic frequency band interval and the grinding surface roughness in the first grinding surface roughness set to determine the correlation evaluation coefficient of each characteristic value, and a mean calculation is performed to obtain a first correlation evaluation coefficient mean, wherein the first characteristic frequency band interval and the first correlation evaluation coefficient mean have a corresponding relationship; a plurality of correlation evaluation coefficient means corresponding to a plurality of characteristic frequency band intervals are obtained in turn, and the characteristic frequency band interval corresponding to the maximum correlation evaluation coefficient mean is selected and set as the sensitive frequency band range.

[0055] Specifically, the method for screening the main characteristic frequency band range based on the correlation evaluation function is as follows: first, the main characteristic frequency band range is divided according to a preset step size to determine a plurality of characteristic frequency band intervals, and the preset step size can be set according to actual conditions, for example: selecting a preset step size of 3KHZ; then selecting a first characteristic frequency band interval from the plurality of characteristic frequency band intervals, wherein the first characteristic frequency band interval is any one of the plurality of characteristic frequency band intervals, and obtaining a first grinding surface roughness set corresponding to the first characteristic frequency band interval.

[0056] Then, based on the correlation evaluation function, the correlation evaluation is performed on the characteristic values ​​within the first characteristic frequency band and the grinding surface roughness in the first grinding surface roughness set, wherein the characteristic values ​​include maximum value, minimum value, peak-to-peak value, average value, standard deviation, variance, effective value, skewness, pulse factor, waveform factor, margin factor, peak frequency, etc., to obtain the correlation evaluation coefficient of each characteristic value; then, the mean of multiple correlation evaluation coefficients of multiple characteristic values ​​is calculated to obtain the first correlation evaluation coefficient mean, wherein the first characteristic frequency band and the first correlation evaluation coefficient mean have a corresponding relationship. Then, the mean values ​​of multiple correlation evaluation coefficients corresponding to multiple characteristic frequency bands are calculated in sequence, and the characteristic frequency band corresponding to the maximum correlation evaluation coefficient mean is selected as the sensitive frequency band range to obtain the sensitive frequency band range.

[0057] Furthermore, the second predicted roughness generating module 15 is also used for:

[0058] Acquire multiple eigenvalues ​​of the sensitive frequency band, wherein the multiple eigenvalues ​​include at least a maximum value, a minimum value, a peak-to-peak value, an average value, a standard deviation, and a variance; extract eigenvalues ​​whose correlation evaluation coefficients are greater than a predetermined coefficient threshold and set them as correlation eigenvalues ​​to form a correlation feature set; construct a feature matrix based on the correlation feature set, and use the feature matrix as input data of the memory network roughness prediction unit.

[0059] Specifically, first, a plurality of eigenvalues ​​of the sensitive frequency band are obtained, wherein the plurality of eigenvalues ​​at least include a maximum value, a minimum value, a peak-to-peak value, an average value, a standard deviation, and a variance, etc.; then, the correlation evaluation coefficient of each eigenvalue is judged according to the predetermined coefficient threshold, the predetermined coefficient threshold can be set according to the actual situation, and the eigenvalue whose correlation evaluation coefficient is greater than the predetermined coefficient threshold is selected as the correlation eigenvalue to obtain the correlation feature set; further, a feature matrix is ​​constructed based on the correlation feature set, that is, the quantized features are arranged in a certain order to construct a feature matrix, wherein each row of the feature matrix represents a sample, and each column represents a feature. And the feature matrix is ​​used as the input data of the memory network roughness prediction unit.

[0060] Furthermore, the second predicted roughness generating module 15 is also used for:

[0061] Taking the feature matrix as a constraint, feature extraction is performed on a sample grinding acoustic emission pulse signal set and a sample grinding surface roughness set to determine a sample data set, wherein the sample data includes a sample feature matrix and a sample grinding surface roughness; the sample data set is divided into a training set and a verification set according to a preset division ratio; the memory network roughness prediction unit is trained based on the training set, and when the output result tends to converge, the memory network roughness prediction unit is verified by the verification set to obtain a memory network roughness prediction unit that meets the expected accuracy index.

[0062] Specifically, the sample grinding acoustic emission pulse signal set is screened based on the sensitive frequency band range to determine the sensitive grinding acoustic emission pulse signal set that meets the sensitive frequency band range and the corresponding sensitive grinding surface roughness set; then, with the feature matrix as a constraint, the eigenvalues ​​in the sensitive grinding acoustic emission pulse signal set are extracted to obtain a sample feature matrix, and the sample grinding surface roughness corresponding to the sample feature matrix is ​​obtained; and a sample data set is further constructed based on multiple sample feature matrices and multiple sample grinding surface roughnesses.

[0063] Then, the sample data set is divided into a training set and a verification set according to a preset division ratio, and the preset division ratio can be set according to the amount of data; then, the training set is used to perform supervised training on the memory network roughness prediction unit, and when the output result of the memory network roughness prediction unit tends to converge, the verification training of the memory network roughness prediction unit is performed through the verification set until a memory network roughness prediction unit that meets the expected accuracy index is obtained.

[0064] In summary, the electrical accessories burr grinding control system provided by the present application has the following technical effects:

[0065] 1. By building a grinding twin model based on digital twin technology, and combining the grinding prediction model to form a digital twin roughness prediction unit; then during the grinding of the ceramic circuit board, the digital twin roughness prediction unit performs synchronous roughness prediction to obtain the first predicted roughness; further, when the first predicted roughness is greater than or equal to the first roughness threshold, the sensor array is activated to collect the grinding acoustic emission pulse signal; then the memory network roughness prediction unit constructed based on the convolutional network and the long short-term memory network performs roughness prediction on the grinding acoustic emission pulse signal to obtain the second predicted roughness; finally, when the second predicted roughness is greater than or equal to the expected grinding roughness index, the grinding operation of the target ceramic circuit board is terminated. The accuracy and reliability of the ceramic surface roughness prediction during the grinding process can be improved, the technical goal of accurately controlling the grinding of the ceramic circuit board can be achieved, and the technical effect of improving the grinding efficiency and processing quality can be achieved.

[0066] 2. Since the prediction errors of digital twin and BP neural network are large, the roughness prediction is first performed through the digital twin roughness prediction unit. When the prediction result is close to the expected index, the grinding roughness is accurately predicted based on the grinding acoustic emission pulse signal through the memory network roughness prediction unit. In this way, the prediction resource consumption of the grinding acoustic emission pulse signal can be saved while meeting the accuracy requirements of the ceramic surface roughness prediction, and the practicality and reliability of the prediction can be improved.

[0067] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0068] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. The electrical accessories burr grinding control system is characterized by: The system comprises: An information acquisition module, used to acquire property information and expected grinding roughness index of a target ceramic circuit board; A digital twin roughness prediction unit building module is used to generate a grinding process twin model based on the digital twin technology according to the attribute information, grinding wheel abrasive parameters and grinding operation parameters, and build a digital twin roughness prediction unit in combination with the grinding prediction model, wherein the grinding prediction model is built based on a BP neural network; A first predicted roughness generation module is used to perform a grinding process on the target ceramic circuit board and perform synchronous roughness prediction through the digital twin roughness prediction unit to generate a first predicted roughness; A grinding acoustic emission pulse signal acquisition module, used for activating the sensor array to acquire the grinding acoustic emission pulse signal when the first predicted roughness meets the first roughness threshold, and transmitting the grinding acoustic emission pulse signal to the memory network roughness prediction unit through the preamplifier; A second predicted roughness generation module, used for performing roughness prediction on the grinding acoustic emission pulse signal through the memory network roughness prediction unit to generate a second predicted roughness, wherein the memory network roughness prediction unit is constructed based on a convolutional network and a long short-term memory network; A grinding operation termination module, used for terminating the grinding operation of the target ceramic circuit board when the second predicted roughness meets the expected grinding roughness index; Among them, the digital twin roughness prediction unit building module is also used for: The image of the grinding wheel is collected by a microscopic image sensor to obtain a microscopic image of the grinding wheel; Based on the grinding wheel microscopic image, abrasive grain feature recognition is performed to determine the grinding wheel abrasive grain parameters, wherein the grinding wheel abrasive grain parameters include the number of effective abrasive grains, abrasive grain distribution coordinates, and abrasive grain edge height; Among them, the digital twin roughness prediction unit building module is also used for: Retrieving a grinding log based on the attribute information and the grinding wheel abrasive parameter to obtain a grinding record data set, wherein the grinding record data includes a grinding operation parameter, a grinding time, an initial surface roughness, and a termination surface roughness; Dividing the grinding process record data set into a training data set and a verification data set according to a preset division ratio; An initial grinding prediction model is built based on a BP neural network, the grinding operation parameters, grinding processing time and initial surface roughness are used as inputs, the terminated surface roughness is used as supervision, and the initial grinding prediction model is supervised and trained and feedback verified through the training data set and the verification data set, respectively, to obtain a grinding prediction model that meets the preset convergence index; Wherein, the first predicted roughness generation module is further used for: During the grinding process of the target ceramic circuit board, the grinding twin model is synchronously operated and grinding operation parameters are acquired in real time, wherein the grinding twin model is communicatively connected with a control unit of the grinding equipment; Whenever the grinding operation parameters change, the change node is recorded and the grinding prediction model is activated, and the prediction is performed based on the grinding operation parameters, the initial surface roughness and the grinding processing time within the change time period, and the surface prediction roughness is output, wherein the change time period is the time interval between the current change node and the previous adjacent change node; Based on a preset fusion strategy, the surface predicted roughness is fused with the simulated operation roughness under the change node to obtain the first predicted roughness, wherein the preset fusion strategy is a preset weight ratio; Wherein, the second predicted roughness generation module is further used for: A data recognition layer is built based on a convolutional neural network, wherein the input data of the data recognition layer is a grinding acoustic emission pulse signal within a sensitive frequency band, and the output data is a recognition feature sequence; Building a feature analysis layer based on a long short-term memory network, wherein the feature analysis layer is composed of a forward long short-term memory network and a backward long short-term memory network, and the forward long short-term memory network and the backward long short-term memory network perform feature extraction based on a forward order and a reverse order, respectively; The memory network roughness prediction unit is constructed based on the data recognition layer and the feature analysis layer, wherein the output layer of the data recognition layer is connected to the input layer of the feature analysis layer; Wherein, the second predicted roughness generation module is further used for: Acquiring a sample grinding acoustic emission pulse signal set and a sample grinding surface roughness set, wherein the sample grinding acoustic emission pulse signal and the sample grinding surface roughness correspond one to one; Based on the wavelet packet analysis principle, the sample grinding acoustic emission pulse signal set and the sample grinding surface roughness set are analyzed to determine the main characteristic frequency band range and the corresponding grinding surface roughness set; Construct a correlation evaluation function: ; in, is the correlation evaluation coefficient, is the characteristic value, is the grinding surface roughness, for and The covariance of for The variance of for The variance of Screening the main characteristic frequency band range based on the correlation evaluation function to obtain the sensitive frequency band range; Wherein, the second predicted roughness generation module is further used for: Dividing the main characteristic frequency band range according to a preset step length to determine a plurality of characteristic frequency band intervals, and selecting a first characteristic frequency band interval from the plurality of characteristic frequency band intervals to obtain a first grinding surface roughness set corresponding to the first characteristic frequency band interval; According to the correlation evaluation function, a correlation evaluation is performed on the characteristic values ​​within the first characteristic frequency band and the grinding surface roughness in the first grinding surface roughness set, a correlation evaluation coefficient of each characteristic value is determined, and a mean value is calculated to obtain a first correlation evaluation coefficient mean value, wherein the first characteristic frequency band and the first correlation evaluation coefficient mean value have a corresponding relationship; A plurality of correlation evaluation coefficient means corresponding to a plurality of characteristic frequency bands are obtained in sequence, and a characteristic frequency band corresponding to a maximum correlation evaluation coefficient mean is selected and set as the sensitive frequency band range.

2. The system according to claim 1, characterized in that The second predicted roughness generation module is further used for: Acquire multiple characteristic values ​​of the sensitive frequency band, wherein the multiple characteristic values ​​at least include a maximum value, a minimum value, a peak-to-peak value, an average value, a standard deviation, and a variance; Extracting feature values ​​whose correlation evaluation coefficients are greater than a predetermined coefficient threshold value and setting them as correlation feature values ​​to form a correlation feature set; A feature matrix is ​​constructed based on the associated feature set, and the feature matrix is ​​used as input data of the memory network roughness prediction unit.

3. The system according to claim 2, characterized in that The second predicted roughness generation module is further used for: Taking the characteristic matrix as a constraint, feature extraction is performed on a sample grinding acoustic emission pulse signal set and a sample grinding surface roughness set to determine a sample data set, wherein the sample data includes a sample characteristic matrix and a sample grinding surface roughness; Dividing the sample data set into a training set and a validation set according to a preset division ratio; The memory network roughness prediction unit is trained based on the training set, and when the output result tends to converge, the memory network roughness prediction unit is verified through the verification set to obtain a memory network roughness prediction unit that meets the expected accuracy index.

Images