A robotic polishing process multi-objective optimization method and apparatus

Abstract

The application provides a robot polishing process multi-objective optimization method and device. The method provided by the application comprises the following steps: determining polishing parameters according to characteristic information of a target object and robot dynamics characteristics; screening polishing effect evaluation factors; polishing a test product according to the polishing parameters, and constructing a parameter matching table comprising pairs of historical polishing parameters and historical polishing effects; processing the historical polishing parameters and the historical polishing effects based on a principal component analysis method to obtain reduced polishing parameters after reduction; calculating a weighted grey correlation degree; training a deep belief network to obtain a trained target deep belief network; obtaining real-time polishing parameters, processing the real-time polishing parameters based on the deep belief network to obtain a predicted polishing effect, and optimizing the real-time polishing parameters based on the predicted polishing effect. The robot polishing process multi-objective optimization method and device provided by the application are used to accurately determine a polishing process and improve a polishing effect.

Description

Technical Field

[0001] This application relates to the field of robotic polishing technology, and in particular to a multi-objective optimization method and apparatus for robotic polishing processes. Background Technology

[0002] Blades are critical components of engines, accounting for approximately one-third of the total number of blades, and their machining quality directly affects the overall performance of the engine. Blades with high geometric precision and excellent surface quality play a vital role in improving the overall performance and reliability of aero-engines. With the continuous development of engine technology, the design and machining precision requirements for blades are constantly increasing, especially the machining quality of blade edges, which has become a key factor affecting engine performance. Therefore, researching and optimizing blade edge machining technology to improve its machining precision and quality is particularly important.

[0003] Currently, traditional blade grinding processes mainly rely on historical experience to select processing parameters. While this can improve efficiency to some extent, it struggles to adapt to changes in new materials, processes, or workpiece shapes, lacking flexibility and adaptability. Therefore, accurately and efficiently selecting and optimizing processing parameters has become an urgent problem to be solved. Summary of the Invention

[0004] In view of this, this application provides a multi-objective optimization method and apparatus for robotic polishing processes, which can accurately determine the polishing process and improve the polishing effect.

[0005] Specifically, this application is implemented through the following technical solution:

[0006] The first aspect of this application provides a multi-objective optimization method for a robotic polishing process, the method comprising:

[0007] The grinding parameters are determined based on the feature information of the target object and the dynamic characteristics of the robot. The grinding parameters are used to represent the robot control parameters for grinding the object to the target object. There are multiple grinding parameters.

[0008] Factors for evaluating the polishing effect are selected based on the characteristic information of the target object;

[0009] The test specimen was polished according to the polishing parameters, and the values ​​of the polishing effect evaluation factors corresponding to each set of polishing parameters were obtained. A parameter matching table containing pairs of historical polishing parameters and historical polishing effects was constructed. The polishing effect evaluation factors were used to evaluate the polishing quality of the target object after polishing. There were multiple polishing effect evaluation factors.

[0010] Principal component analysis is used to process the historical polishing parameters and the historical polishing effects to obtain simplified polishing parameters; the number of simplified polishing parameters is less than that of historical polishing parameters.

[0011] Calculate the contribution rate of each sanding parameter in the reduced sanding parameters, and calculate the weighted grey relational degree based on the reduced sanding parameters, the contribution rate, and the grey relational coefficient; each weighted grey relational degree characterizes the relationship between all sanding parameters in the reduced sanding parameters and the corresponding sanding effect;

[0012] A deep confidence network is trained based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network; wherein, the target deep confidence network contains the correlation between polishing parameters and polishing effect;

[0013] Real-time grinding parameters are generated based on the target object and the test piece to be processed. The real-time grinding parameters at each point are input into the depth confidence network to obtain the predicted grinding effect. The real-time grinding parameters are then optimized based on the predicted grinding effect.

[0014] A second aspect of this application provides a multi-objective optimization device for robotic polishing processes, the device comprising:

[0015] The module comprises an acquisition module, a processing module, a calculation module, and a training module; among which,

[0016] The acquisition module is used to determine the polishing parameters based on the feature information of the target object and the dynamic characteristics of the robot. The polishing parameters are used to represent the robot control parameters for polishing the object to the target object. There are multiple polishing parameters.

[0017] The processing module is used to filter polishing effect evaluation factors based on the feature information of the target object;

[0018] The acquisition module is also used to polish the test sample according to the polishing parameters, obtain the value of the polishing effect evaluation factor corresponding to each set of polishing parameters, and construct a parameter matching table containing pairs of historical polishing parameters and historical polishing effects. The polishing effect evaluation factor is used to evaluate the polishing quality of the target object after polishing. There are multiple polishing effect evaluation factors.

[0019] The processing module is further configured to process the historical polishing parameters and the historical polishing effect based on principal component analysis to obtain simplified polishing parameters; the number of simplified polishing parameters is lower than that of historical polishing parameters.

[0020] The calculation module is used to calculate the contribution rate of each sanding parameter in the reduced sanding parameters, and to calculate the weighted grey relational degree based on the reduced sanding parameters, the contribution rate and the grey relational coefficient; a single weighted grey relational degree represents the correlation between all sanding parameters in the reduced sanding parameters and the corresponding sanding effect;

[0021] The training module is used to train a deep confidence network based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network; wherein, the target deep confidence network contains the correlation between polishing parameters and polishing effect;

[0022] The processing module is further configured to acquire real-time polishing parameters, process the real-time polishing parameters based on the deep belief network to obtain a predicted polishing effect, and optimize the real-time polishing parameters based on the predicted polishing effect.

[0023] The multi-objective optimization method and apparatus for robotic grinding processes provided in this application have two main aspects. First, by adaptively determining grinding parameters and grinding effect evaluation factors based on the characteristic information of the target object, it selects evaluation factors closely related to the grinding effect, accurately adapts them to the target object, and obtains multiple key grinding parameters for the target machine during the process optimization, thereby improving the optimization effect. Compared with the traditional method of fixing grinding parameters and grinding effect evaluation factors based on experience, this method ensures the accuracy and reliability of the evaluation results. Second, considering the complexity of the processing environment, pure simulation may not accurately reflect the actual processing situation. Therefore, a combination of experimentation and simulation is used to comprehensively analyze the optimal processing method, ensuring the reliability and accuracy of the grinding process. Thirdly, principal component analysis (PCA) is used to reduce the dimensionality of the collected historical polishing parameters and effects, extracting the simplified polishing parameters that have the greatest impact on the polishing effect. This effectively reduces the dimensionality of the data, lowers the complexity of subsequent analysis, and retains the main information in the data. By calculating the contribution rate of each parameter in the simplified polishing parameters and combining it with the grey relational coefficient, a weighted grey relational degree is calculated. This not only quantifies the impact of each parameter on the polishing effect but also transforms the multi-objective optimization problem (i.e., optimizing multiple polishing parameters and multiple polishing effects) into a single-objective optimization problem (i.e., optimizing only the weighted grey relational degree). This transformation simplifies the optimization process and avoids the non-uniqueness of solutions and computational complexity that may occur in multi-objective optimization. By using the weighted grey relational degree and historical polishing parameters as training data, a deep belief network is trained, enabling the network to capture the complex nonlinear relationship between polishing parameters and polishing effects, and accurately predict the polishing effect. Thus, by combining PCA and weighted grey relational degree, the deep belief network can accurately optimize multi-objective polishing parameters. Attached Figure Description

[0024] Figure 1 A flowchart of an embodiment of the multi-objective optimization method for robot grinding process provided in this application;

[0025] Figure 2 This is a schematic diagram of a polishing robot shown in an exemplary embodiment of this application;

[0026] Figure 3 This is a schematic diagram illustrating the grinding parameter optimization process in an exemplary embodiment of this application;

[0027] Figure 4 This is a schematic diagram illustrating the grey relational analysis result as an exemplary embodiment of this application.

[0028] Figure 5 A schematic diagram illustrating a comparison of experimental results for an exemplary embodiment of this application;

[0029] Figure 6 This is a schematic diagram of the structure of Embodiment 1 of the multi-objective optimization device for robot grinding process provided in this application. Detailed Implementation

[0030] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0031] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0032] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0033] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0034] Figure 1 This is a flowchart of an embodiment of the multi-objective optimization method for robot grinding process provided in this application. Please refer to... Figure 1 The method provided in this embodiment may include:

[0035] S101. Determine the polishing parameters based on the feature information of the target object and the dynamic characteristics of the robot. The polishing parameters are used to represent the robot control parameters for polishing the object to the target object. There are multiple polishing parameters.

[0036] Specifically, the target object is the best-finished body after polishing. In this embodiment, the target object can be an ideal aero-engine blade after polishing.

[0037] Furthermore, the object being polished is made of the same material as the target object, and by polishing the object, its shape is made to resemble that of the target object.

[0038] Furthermore, the feature information of the target object is data related to the robot and the polishing process. For example, in one embodiment, the feature information of the target object may include material (such as metal / composite material), surface shape (planar / curved surface), hardness, etc.

[0039] Furthermore, robot dynamics involves the robot's joint torques, speed range, force control precision, and so on.

[0040] Figure 2 This is a schematic diagram of a polishing robot illustrating an exemplary embodiment of this application. Please refer to... Figure 2 It can be done Figure 2 The robot shown is polishing the blades.

[0041] Furthermore, when obtaining polishing parameters, experiments and simulations can be conducted based on the feature information of the target object and the dynamic characteristics of the robot. Multiple sets of different polishing parameters can be obtained through experiments and simulations, and finally, a set of polishing parameters with the best effect can be determined by comprehensive analysis.

[0042] In a specific implementation, for example, in one embodiment, based on the above example, the feature information of the target object and the dynamic characteristics of the robot are constructed based on experiments and simulations, and multiple sets of grinding parameters are determined. Finally, the combination of grinding speed, pressure and path trajectory is selected as the grinding parameters.

[0043] It should be noted that the final number of polishing parameters is multiple, in order to control the polishing process for multiple targets.

[0044] The following is another specific embodiment to illustrate in detail the process of determining the polishing parameters:

[0045] Determine the geometric and material characteristics of the target object;

[0046] Determine the target contour trajectory of the target object after polishing based on the geometric features of the target object;

[0047] The grinding allowance is determined based on the difference between the target contour trajectory and the contour of the test specimen.

[0048] Based on the material characteristics, the interference torque of the robot when polishing the target object is determined;

[0049] Determine the robot's dynamic characteristics, and based on the robot's dynamic characteristics, determine the real-time initial grinding parameters corresponding to the grinding margin.

[0050] The interference parameter quantity of the real-time initial grinding parameters at each time point is calculated based on the interference torque.

[0051] The real-time initial polishing parameters are corrected based on the interference parameter values ​​to obtain the polishing parameters.

[0052] In specific implementation, for example, in one embodiment, taking the blade as the target object, the target contour trajectory to be polished is determined according to the geometric characteristics of the blade. Then, the difference between the target contour trajectory and the test sample is determined as the margin to be polished. Then, the force required to polish the alloy blade is determined as the disturbance torque. Combined with the robot's model and performance, appropriate real-time initial polishing parameters are determined to achieve polishing of the polishing margin. At the same time, since the robot polishes at different positions at different times, it will produce different degrees of influence. These influences are corrected as disturbance parameters to obtain the polishing parameters.

[0053] Furthermore, when extracting geometric features, key parameters such as the surface shape (planar / curved), size (length, width, and height), and radius of curvature of the target object can be obtained through 3D scanning or machine vision.

[0054] Furthermore, when conducting material characteristic analysis, equipment such as hardness testers and spectrometers can be used to determine the material's hardness, elastic modulus, ductility, and other properties.

[0055] Furthermore, based on key combinations of geometric features, a path planning algorithm is used to generate the target contour trajectory. It is understandable that, since the target object is a standard shape after polishing, the edge contour of the target object can be determined as the target trajectory contour.

[0056] It should be noted that a unique target contour trajectory is obtained through all geometric and material features, representing the ideal shape that needs to be polished.

[0057] Furthermore, when determining the polishing margin, a laser scanner can be used to obtain the actual contour of the test specimen, and the iterative nearest point algorithm can be used to align it with the target contour trajectory to determine the initial positional deviation.

[0058] Furthermore, the test specimen is rotated within a preset angle range, and the volume removal amount at different angles is calculated to select the optimal matching scheme that minimizes the grinding volume. For example, in one embodiment, the preset angle is 30°. The test specimen is rotated within ±30°, and it is determined that the amount to be removed is minimized at -8°. The portion to be removed at this point is defined as the initial position deviation.

[0059] Furthermore, a force-torque conversion model is established based on the material hardness and coefficient of friction. For example, in one embodiment, wood, due to its porous texture, experiences relatively small frictional interference torque during sanding, but its fluctuation range is large. In another embodiment, alloy, due to its hard texture, experiences relatively large frictional interference torque during sanding, but its fluctuation range is small.

[0060] Furthermore, when establishing dynamic modeling, a multibody dynamic model can be established based on parameters such as robot joint stiffness and inertia to predict motion stability under different grinding forces.

[0061] In practice, during the initial grinding, the initial force and time are calculated using a pre-designed material removal rate model based on the grinding direction and allowance thickness. For example, for an aluminum alloy part with a thickness of 0.5mm, the friction force needs to be set to 80N and the speed to 200mm / s.

[0062] Furthermore, when adjusting subsequent polishing parameters: for example, in one embodiment, the time points are iterated, and when the parameters at time n need to be adjusted, the deviation between the predicted position and the target position is compared based on the parameters at time n-1, and the robot's end effector posture and force direction are adjusted through PID or model predictive control. In another embodiment, assuming polishing is performed at time n based on the parameters at time n-1, the parameters at that time are adjusted according to the deviation between the polishing result and the target position to determine the actual polishing parameters at time n.

[0063] Furthermore, the material disturbance torque is decomposed into the robot's machining directions (such as normal and tangential), and the effective disturbance parameters are calculated using vector projection. For example, in one embodiment, the normal disturbance of the curved trajectory is converted into force disturbance parameters in the robot's base coordinate system through coordinate transformation.

[0064] It should be noted that the direction of the disturbance torque may be different at different times. It is necessary to project the force onto the direction of robot processing to determine whether the disturbance torque is positive (gain grinding parameter) or negative (loss grinding parameter).

[0065] Furthermore, the real-time initial polishing parameters are corrected based on the amount of interference parameters to obtain the polishing parameters. For example, in one embodiment, the amount of interference parameters can be added to the real-time initial polishing parameters to obtain the polishing parameters.

[0066] The multi-objective optimization method for robot grinding provided in this embodiment significantly improves the accuracy and stability of robot grinding through precise geometric-material feature modeling and dynamic parameter compensation mechanism: First, the grinding margin is optimized based on the matching of the target contour trajectory and the test piece to reduce the amount of material removed. Through real-time projection of material disturbance torque and parameter correction, the surface size deviation is greatly reduced, while avoiding over-cutting or under-grinding. Combining robot dynamic characteristics and the first-point parameter iterative generation strategy, adaptive path planning for complex curved surfaces is realized, reducing tool wear and enabling the parameter matching table to have dynamic expansion capability, supporting rapid process transfer in multi-material (metal / composite material) scenarios, and improving the overall grinding performance and grinding accuracy.

[0067] S102. Select polishing effect evaluation factors based on the characteristic information of the target object.

[0068] In practice, the corresponding grinding effect evaluation factors are determined based on the correlation between the characteristic information of the target object and the grinding effect evaluation factors. For example, in one embodiment, the target object is an aero-engine blade. Based on the characteristic information, the blade material is found to be an alloy with a hardness of level III. Therefore, surface roughness and material removal rate are selected as the grinding effect evaluation factors.

[0069] The following is a specific example illustrating the process of selecting factors for evaluating polishing effects:

[0070] Determine the candidate effect index;

[0071] Calculate the contribution of the candidate effect index to the geometric features of the target object;

[0072] Candidate effect indices with contribution values ​​higher than a preset contribution value are selected as evaluation factors for the polishing effect.

[0073] Specifically, the candidate performance index is used to evaluate the results of polishing.

[0074] In practice, the corresponding candidate effect index can be determined by the correspondence between the feature information of the target object and the candidate effect index. For example, in one embodiment, the candidate effect indices are determined as arithmetic mean roughness, maximum height difference, material removal rate, and dimensional accuracy error.

[0075] Furthermore, the contribution of each candidate effect index is calculated during the polishing process to obtain the target object. In specific implementation, for example, in one embodiment, the contribution of surface roughness is determined to be 40%, the contribution of material removal rate is determined to be 45%, etc.

[0076] Furthermore, the specific value of the preset contribution level is set according to actual needs, and this embodiment does not limit it. For example, in one embodiment, the preset contribution level is set to 40%.

[0077] In a specific implementation, for example, in one embodiment, all effect indices are determined as candidate effect indices, and then the contribution of each candidate effect index is calculated: candidate effect index a is 55%, candidate effect index b is 65%, candidate effect index c is 45%, candidate effect index d is 63%, and candidate effect indices with a contribution of more than 50% are selected, that is, candidate effect index a, candidate effect index b, and candidate effect index d are used as the polishing effect evaluation factors.

[0078] The multi-objective optimization method for robot polishing provided in this embodiment analyzes the contribution of different candidate effect indices to the polishing process. It can select the most valuable candidate effect index based on the geometric features of the target object, preventing effect indices that cannot characterize the target object from interfering with it, and ensuring the reliability and accuracy of the polishing effect evaluation factors.

[0079] S103. Polish the test sample according to the polishing parameters, obtain the value of the polishing effect evaluation factor corresponding to each set of polishing parameters, and construct a parameter matching table containing pairs of historical polishing parameters and historical polishing effects. The polishing effect evaluation factors are used to evaluate the polishing quality of the target object after polishing. There are multiple polishing effect evaluation factors.

[0080] Specifically, the test specimen and the target object to be polished have the same characteristic information. For example, in one embodiment, the target object is a blade of type 1, and the test specimen is also a blade of type 1.

[0081] In practice, the test specimen is polished according to the generated polishing parameters, and the values ​​of the polishing effect evaluation factors under each set of polishing parameters are recorded.

[0082] Furthermore, the numerical values ​​of the polishing environment, polishing parameters, and polishing effect evaluation factors are paired and stored to construct a structured parameter matching table, which shows the numerical values ​​of the polishing effect evaluation factors that can be achieved under different polishing environments and with different polishing parameters.

[0083] In practice, simulation experiments are conducted based on multiple polishing parameters, and the values ​​of polishing effect evaluation factors corresponding to all polishing parameters are recorded. For example, in one embodiment, data is analyzed using MATLAB scripts to establish a mapping relationship between the values ​​of polishing parameters and polishing effect evaluation factors, which serves as a parameter matching table.

[0084] It should be noted that establishing the relationship between grinding parameters and the numerical values ​​of grinding effect evaluation factors through experiments and simulations allows for thorough analysis of different parameters, different test samples, and different grinding environments during simulation and experimentation. This enables the acquisition of parameter sensitivity under complex working conditions. Compared to obtaining corresponding values ​​based on experience, the correspondence in the parameter matching table is more accurate. When facing complex environmental scenarios, there is data support to ensure the accuracy and reliability of obtaining grinding parameters.

[0085] The following is a specific example to illustrate in detail the process of extracting historical sanding parameters and historical sanding effect evaluation values:

[0086] (1) Based on the influence of the polishing parameters on the historical polishing effect, the polishing parameters are divided into gain parameters and loss parameters.

[0087] Specifically, the gain parameter is the polishing parameter whose value is directly proportional to the historical polishing effect, while the loss parameter is the polishing parameter whose value is inversely proportional to the historical polishing effect.

[0088] In practice, for each type of polishing parameter, their impact on the polishing effect is analyzed. For example, in one embodiment, the larger the value of parameter a, the better the polishing effect, so parameter a is determined to be a gain parameter; the smaller the value of parameter b, the better the polishing effect, so parameter b is determined to be a loss parameter.

[0089] (2) Calculate the negative of the reduction parameter and combine it with the gain parameter to generate intermediate polishing parameters.

[0090] Specifically, for all polishing parameters classified as loss parameters, their negative values ​​are calculated.

[0091] In practice, the intermediate polishing parameters can be calculated based on the following formula:

[0092] A= ;

[0093] in, For gain parameters, A is the loss reduction parameter, and A is the intermediate polishing parameter.

[0094] (3) The intermediate grinding parameters are standardized based on the Z-score method to obtain the historical grinding parameters.

[0095] In practice, standardization is performed based on the Z-score method, and each parameter is converted into a dimensionless Z score. These Z scores constitute the final historical polishing parameters.

[0096] In practice, the standardization process can be achieved based on the following formula to calculate historical polishing parameters:

[0097] ;

[0098] in, Refining parameters for historical purposes For gain parameters, For loss parameters, This represents the sample standard deviation.

[0099] The multi-objective optimization method for robotic polishing provided in this embodiment analyzes the impact of each type of polishing parameter on historical polishing results, clearly classifying the parameters into gain parameters and loss parameters. This allows for a more accurate understanding of the contribution of each parameter to the objective function during subsequent optimization, avoiding blind optimization and mutual interference between parameters. At the same time, Z-score standardization converts the parameter values ​​into dimensionless Z-scores, eliminating possible deviations between different parameters due to different dimensions, and improving the accuracy of the optimization results.

[0100] S104. Based on principal component analysis, the historical polishing parameters and the historical polishing effect are reduced to obtain the reduced polishing parameters; the number of the reduced polishing parameters is lower than that of the historical polishing parameters.

[0101] In a specific implementation, for example, in one embodiment, historical polishing parameters and historical polishing effects are processed based on principal component analysis, and polishing parameters with higher contribution are selected and retained to obtain simplified polishing parameters.

[0102] It is understandable that by using principal component analysis to reduce parameters and remove those with lower contributions, the reduced polishing parameters contain fewer parameters than the historical polishing parameters.

[0103] The following is a specific embodiment to illustrate in detail the process of obtaining simplified polishing parameters:

[0104] The process of processing the historical polishing parameters and historical polishing effects using principal component analysis to obtain the reduced polishing parameters includes:

[0105] (1) Calculate the correlation coefficient matrix between each historical polishing parameter and the corresponding historical polishing effect to obtain the feature vector and feature value corresponding to the correlation coefficient matrix; each feature vector corresponds to a polishing parameter.

[0106] Specifically, the correlation coefficient matrix reflects the linear correlation between various polishing parameters and between polishing parameters and polishing results.

[0107] In practice, the correlation coefficient between each historical polishing parameter and the corresponding historical polishing effect is calculated to form a correlation coefficient matrix.

[0108] Furthermore, the correlation coefficient matrix is ​​decomposed into eigenvalues ​​to obtain a series of eigenvalues ​​and corresponding eigenvectors.

[0109] It should be noted that the feature vector represents the polishing parameters and polishing effect in the data, while the corresponding feature value reflects the contribution of these polishing parameters to the polishing effect.

[0110] In practice, the correlation coefficient matrix can be calculated based on the following formula:

[0111] ;

[0112] in, Reflection indicators and The degree of relevance, D is the covariance, and D() is the variance, which is used to measure the dispersion of the data of a single indicator.

[0113] (2) Sort the feature vectors based on the feature values ​​and calculate the contribution rate of each polishing parameter to the polishing effect.

[0114] Specifically, the feature values ​​are sorted from largest to smallest. The larger the feature value, the stronger the contribution of the corresponding feature vector to the polishing effect. In other words, the sorted data represents the importance of the polishing effect.

[0115] In practice, the contribution rate can be calculated based on the following formula:

[0116] ;

[0117] in, To calculate the correlation coefficient matrix eigenvalues, The contribution rate is denoted by n, where n is the total number of polishing parameters. This is the sum of the eigenvalues ​​of the correlation coefficient matrix.

[0118] (3) Remove the feature vectors whose contribution rate is lower than the preset threshold to obtain the reduced polishing parameters.

[0119] Specifically, the exact value of the preset threshold is set according to actual needs, and this embodiment does not limit it. For example, the preset threshold can be set to 25%.

[0120] In practice, feature vectors below a preset threshold have lower importance in the data. The remaining feature vectors and feature values ​​after removal are used to construct the reduction and polishing parameters.

[0121] The multi-objective optimization method for robotic polishing provided in this embodiment achieves data dimensionality reduction through principal component analysis. This not only reduces the dimensionality of the data but also retains the main information in the data, providing a more concise and effective dataset for subsequent parameter optimization. At the same time, by removing feature vectors with a contribution rate lower than a preset threshold, the method focuses on factors that have a significant impact on the polishing effect, making the optimization results more accurate and reliable, which helps to improve the quality and efficiency of robotic polishing.

[0122] S105. Calculate the contribution rate of each polishing parameter in the reduced polishing parameters, and calculate the weighted grey relational degree based on the reduced polishing parameters, the contribution rate and the grey relational coefficient; the single weighted grey relational degree characterizes the relationship between all polishing parameters in the reduced polishing parameters and the corresponding polishing effect.

[0123] Specifically, the grey relational coefficient is an indicator that measures the degree of correlation between two sequences. It is used to characterize the relationship between reduced polishing parameters and polishing effect.

[0124] It should be noted that the grey relational coefficient can characterize the correspondence between all grinding parameters and the grinding effect in the reduced grinding parameters. In other words, the original grinding parameters, as well as the two grinding effects of surface roughness and material removal rate, need to be considered. By converting to the grey relational coefficient, only the value of the grey relational coefficient needs to be considered. This realizes the process of converting multi-objective optimization into single-objective optimization. While ensuring accuracy, it reduces the difficulty of calculation and prevents errors caused by mutual interference between grinding parameters.

[0125] The following is a specific example to illustrate the calculation process of weighted grey relational degree:

[0126] (1) Calculate the absolute difference between the historical polishing effect and the ideal polishing effect corresponding to each of the historical polishing parameters.

[0127] In a specific implementation, for example, in one embodiment, a certain index value of the historical polishing effect is x, and the corresponding index value of the ideal polishing effect is x1. Then the absolute difference between them is |x1 – x|.

[0128] In practice, the absolute difference can be calculated based on the following formula:

[0129] ;

[0130] in, The absolute difference Polishing the effect for history, For the ideal polishing effect.

[0131] (2) Determine the maximum absolute difference and the minimum absolute difference based on the absolute difference, and calculate the grey relational coefficient based on the absolute difference, the maximum absolute difference and the minimum absolute difference.

[0132] Specifically, different groups of polishing parameters correspond to different historical polishing effects, thus resulting in multiple absolute differences. The absolute difference with the largest value is determined as the maximum absolute difference, and the absolute difference with the smallest value is determined as the minimum absolute difference.

[0133] In practice, the grey relational coefficient can be calculated based on the following formula:

[0134] ;

[0135] in, 11 is the grey relational coefficient, and n is the number of polishing parameters. Polishing the effect for history, For the ideal polishing effect.

[0136] (3) Based on the reduced polishing parameters, the contribution rate and the gray relational coefficient, determine the correlation between each polishing parameter and the corresponding polishing effect, and obtain the weighted gray relational degree.

[0137] In practice, the weighted grey relational coefficient can be calculated based on the following formula:

[0138] ;

[0139] in, For the weighted grey relational coefficient, The contribution rate is denoted by n, where n is the number of polishing parameters. Polishing the effect for history, For the ideal polishing effect.

[0140] The multi-objective optimization method for robotic polishing provided in this embodiment accurately quantifies the gap between the actual polishing effect and the expected target by calculating the absolute difference between the polishing effect corresponding to historical polishing parameters and the ideal polishing effect. This allows the optimization process to be more targeted and focused on narrowing this gap. At the same time, the weighted grey relational degree not only comprehensively reflects the combined influence of multiple polishing parameters on the polishing effect, but also effectively transforms the multi-objective optimization problem into a single-objective optimization problem. This transformation not only simplifies the complexity of the optimization process, but also improves the accuracy and reliability of the optimization results.

[0141] S106. Based on the weighted grey relational degree and the historical polishing parameters, train a deep confidence network to obtain a trained target deep confidence network; wherein, the target deep confidence network contains the correlation between polishing parameters and polishing effect.

[0142] Specifically, the weighted gray relational degree and historical polishing parameters are used as inputs to the deep belief network, and the polishing effect is used as the output of the deep belief network. The network learns the correlation between polishing parameters and polishing effect to complete the training of the model and obtain the target deep belief network.

[0143] The following is a specific example to illustrate the training process of a target deep belief network:

[0144] (1) Input the historical polishing parameters into the first layer of the restricted Boltzmann machine of the depth belief network, and output the desired polishing effect through the first layer of the restricted Boltzmann machine.

[0145] Specifically, in a deep belief network, the visible layer of a restricted Boltzmann machine (RBM) is used to receive external input data, and there can be multiple RBMs. In practice, the desired pose data is input into the visible layer of the first RBM.

[0146] Furthermore, the Boltzmann machine is restricted to the visible layer. and hidden layers This consists of two layers. The visible layer receives training data, specifically the desired end effector position in the robot coordinate system and the current robot joint angle. The hidden layer takes the output of the visible layer as its input and is used to extract features. There is no intra-layer connection but full inter-layer connection between the neurons in the two layers.

[0147] In practice, historical polishing parameters are used as input data and fed into the first layer of the Restricted Boltzmann Machine (RBM). The input data is received through the visible layer of the first layer of the RBM and learns the preliminary feature representation of the data by interacting with the weight matrix and bias of the hidden layer, and outputs the desired polishing effect.

[0148] (2) Based on the correlation between the historical polishing parameters and the desired polishing effect, the weight and bias of the Boltzmann machine are adjusted by combining the weighted gray relational feedback.

[0149] In practice, the desired polishing effect is compared with the actual historical polishing effect. Based on the difference between the two, the weights and biases of the first-layer restricted Boltzmann machine are adjusted using the backpropagation algorithm.

[0150] It should be noted that when adjusting the weights and biases, it is necessary to combine the association relationship represented by the weighted grey relational degree so that the new weights and biases can simultaneously reflect the weighted grey relational degree.

[0151] (3) Use the data output of the first layer of restricted Boltzmann machine as the input of the second layer of restricted Boltzmann machine, and repeat the feedback adjustment steps of the weights and biases of the restricted Boltzmann machine.

[0152] In practice, the output of the first-layer restricted Boltzmann machine is used as the input of the second-layer restricted Boltzmann machine. The above input, output and feedback adjustment steps are repeated to train each layer of restricted Boltzmann machine layer by layer.

[0153] (4) After the number of layers of the trained restricted Boltzmann machine reaches the preset number of layers, the trained target depth confidence network is obtained.

[0154] Specifically, the exact number of preset layers is set according to actual needs, and this embodiment does not limit it.

[0155] In practice, the number of layers is preset to ensure that the deep belief network has sufficient depth to capture the complex features of the data, while avoiding overfitting and excessive increase in computational complexity.

[0156] The multi-objective optimization method for robotic polishing processes provided in this embodiment restricts Boltzmann machines (SBMs) to automatically extract effective features from input data through unsupervised learning, providing more abstract and meaningful feature representations for subsequent network layers. Simultaneously, based on the correlation between historical polishing parameters and desired polishing effects, and combined with weighted grey relational analysis for feedback adjustment, the weights and biases of the SBMs can be dynamically adjusted, thereby optimizing the performance of the entire deep belief network. Through layer-by-layer training and repeated feedback adjustment steps, each layer of the SBM can better learn the inherent patterns and features of the data. The resulting well-trained target deep belief network exhibits stronger robustness and generalization ability, ensuring the robustness and accuracy of the target deep belief network.

[0157] The following is another specific example to illustrate the training process of a deep belief network:

[0158] Specifically, the training process involves two steps: unsupervised pre-training and fine-tuning of the deep belief network. During pre-training, a greedy algorithm is used, employing the result of training the previous Restricted Boltzmann Machine (RBM) as the input to the next RBM, continuing until all RBMs have been trained, and the initial parameters for each RBM are obtained.

[0159] Furthermore, fine-tuning refers to training the entire network using the backpropagation (BP) algorithm after pre-training to bring the deep belief network to its optimal state, avoiding drawbacks such as critical points and long training times. Assume y and These are the actual output and the expected output of the deep belief network, respectively. The loss function of the output layer is:

[0160] ;

[0161] Where τ is the number of iterations and N is the number of training samples.

[0162] Furthermore, the weights between the hidden and output layers of the last layer of the network are iterated through an update function:

[0163] ;

[0164] Where τ is the number of iterations, It is the output of the hidden layer.

[0165] Furthermore, Figure 3 This is a schematic diagram illustrating the polishing parameter optimization process in an exemplary embodiment of this application. Please refer to... Figure 3 Based on simulation and experiment, the grinding parameters and grinding effects corresponding to the feature information of different environments and target objects are determined. In this way, after processing by principal component analysis, weighted grey relational degree calculation and deep confidence network, the predicted grinding effect can fit the target object. At the same time, it can fully combine the influence of all environments included in the simulation and experiment, ensuring the accuracy and reliability of grinding.

[0166] S107. Generate real-time grinding parameters based on the target object and the test piece to be processed, input the real-time grinding parameters of each point into the depth confidence network to obtain the predicted grinding effect, and optimize the real-time grinding parameters based on the predicted grinding effect.

[0167] Specifically, during the robot polishing process, the current polishing parameters corresponding to the target object and the test piece to be processed are acquired in real time through sensors or control systems.

[0168] Furthermore, the acquired data can be normalized.

[0169] Furthermore, the acquired real-time polishing parameters are used as input and fed into the previously trained target depth belief network. Through its internal multi-layer structure, the input parameters are nonlinearly transformed and features are extracted, and finally a predicted polishing effect is output, which comprehensively reflects the optimization of multiple indicators such as surface roughness and material removal rate.

[0170] Furthermore, by combining the predicted polishing effect, the real-time polishing parameters are adjusted to achieve a better polishing result.

[0171] The following is another specific embodiment to illustrate in detail the process of obtaining the predicted polishing effect:

[0172] Obtain the changing state of the polishing environment corresponding to the real-time polishing parameters;

[0173] The empirical variation value of the polishing effect is calculated based on the aforementioned changing state;

[0174] The predicted polishing effect is corrected based on the empirical change value.

[0175] Specifically, it involves acquiring information about changes in the grinding environment. In practice, this can be achieved by determining changes in the grinding environment's temperature (which affects material thermal deformation), humidity (which affects dust adhesion), tool wear (detected by visual sensors to change the diameter of the grinding wheel), and vibration amplitude (monitored by accelerometers to monitor the stability of the robot's joints).

[0176] It should be noted that the change in the polishing environment refers to the numerical change in the polishing environment from the base environment, rather than the current environment value. By observing the change in the environment, the impact of the current environment on polishing can be quickly determined based on the existing environment, thus improving the efficiency of the calculation.

[0177] Furthermore, the empirical variation value of the polishing effect is the specific numerical value of the change in polishing effect caused by environmental changes.

[0178] In practice, environmental variables and effect indicators can be established based on experimental data (such as polishing experiments under different temperatures / humidities) through random forest regression or grey relational analysis. For example, in one embodiment, it was determined that an increase of 10°C in temperature would increase the surface roughness Ra of titanium alloy by 0.2μm, with an empirical coefficient K=0.02μm / °C, and the empirical change value of the polishing effect was 8.

[0179] Furthermore, preset parameters can be directly adjusted based on real-time empirical changes. For example, in one embodiment, if a 20% increase in humidity is detected, the polishing pressure is increased from 80N to 85N according to an empirical database to counteract the effect of dust adhesion.

[0180] The multi-objective optimization method for robotic grinding process provided in this embodiment captures environmental variables such as temperature, vibration, and tool wear in real time, and combines them with a disturbance influence model constructed from historical data to accurately quantify the cumulative deviation of environmental changes on the grinding effect. Through a feedforward-feedback dual-loop correction strategy, the predicted values ​​of parameters (such as pressure and speed) are dynamically adjusted to offset the over-cutting or under-grinding problems caused by environmental interference, thus realizing a highly robust intelligent manufacturing process.

[0181] The following is a specific example to illustrate in detail the process of optimizing real-time polishing parameters based on predicted polishing results:

[0182] Candidate data bars are calculated from the parameter matching table based on the real-time polishing parameters, wherein the candidate data bars are data bars in the polishing parameter matching table whose similarity to the real-time polishing parameters is greater than a threshold;

[0183] Match the candidate data bars with the processing environment corresponding to the real-time grinding parameters, and filter the target data bars from the candidate data bars;

[0184] By comparing the historical polishing parameters in the target data bar with the predicted polishing effect, the polishing parameter corresponding to the value with the best polishing effect is selected as the final polishing parameter.

[0185] Specifically, candidate data bars are used to characterize the polishing parameters that better match the real-time polishing parameters.

[0186] Furthermore, a similarity algorithm can be used to process the real-time polishing parameters and the historical polishing parameters in the parameter matching table to calculate their similarity value. For example, in one embodiment, the difference can be calculated for each polishing parameter to determine the similarity of each polishing parameter, and the similarity with the real-time polishing parameters can be determined by weighted summation.

[0187] Furthermore, the target candidate data bar is used to characterize the grinding parameters that still match the real-time grinding parameters when combined with environmental factors.

[0188] Furthermore, environmental label classification can be performed: the environmental state of candidate data bars can be marked as discrete labels (such as "high temperature-low wear", "normal temperature-high vibration"), and matched with the current environmental label.

[0189] Furthermore, similarity can be used to expand the matching: if the environmental parameters are continuous values, the closest candidate bars can be selected by calculating environmental similarity (such as Mahalanobis distance).

[0190] Furthermore, single-index ranking can be performed: the target data bars are arranged in ascending order according to the core evaluation factors, and the polishing parameters with the highest ranking are selected as the final polishing parameters. Alternatively, a comprehensive multi-index scoring can be performed: the comprehensive effect value is calculated using a weighted scoring method (such as roughness weight 0.6, efficiency 0.3, energy consumption 0.1), and the polishing parameters with the highest ranking are selected as the final polishing parameters.

[0191] The multi-objective optimization method for robotic polishing provided in this embodiment filters highly correlated candidate data based on the similarity of multi-dimensional sub-parameters to ensure the comprehensiveness and representativeness of parameter matching; it further locks the target data bar by combining real-time environmental conditions (such as temperature and tool wear) to effectively eliminate interference from inapplicable historical working conditions; and finally selects the optimal parameters by comparing the effects of multiple indicators (such as surface quality and efficiency) to achieve dynamic optimization of polishing effect. This closed-loop decision-making mechanism not only strengthens the robustness of parameters under complex working conditions, but also comprehensively improves the accuracy and efficiency of polishing by continuously learning from historical experience and real-time feedback.

[0192] The following is another specific embodiment to illustrate in detail the process of optimizing real-time polishing parameters based on predicted polishing effects:

[0193] (1) Compare the predicted polishing effect with the ideal polishing effect obtained in advance, and determine whether the predicted polishing effect is better than the ideal polishing effect.

[0194] Specifically, multiple indicators such as surface roughness and material removal rate can be compared between the predicted polishing effect and the ideal polishing effect obtained in advance to comprehensively judge which one is better.

[0195] (2) If the predicted polishing effect is better than the ideal polishing effect, keep the real-time polishing parameters unchanged.

[0196] In a specific implementation, for example, in one embodiment, the predicted polishing effect corresponding to the real-time polishing parameter a is better than the ideal polishing effect, and the real-time polishing parameter a is still used for polishing.

[0197] (3) If the predicted polishing effect is not better than the ideal polishing effect, the real-time polishing parameters are updated to the ideal polishing parameters corresponding to the ideal polishing effect.

[0198] In a specific implementation, for example, in one embodiment, if the predicted polishing effect corresponding to the real-time polishing parameter a is worse than the ideal polishing effect corresponding to the polishing parameter b, the real-time polishing parameter is replaced with the polishing parameter b, and polishing is performed.

[0199] Understandably, by comparing the predicted polishing effect with the pre-set ideal polishing effect, the effectiveness of the current polishing parameters can be evaluated in real time. If the predicted polishing effect is better than or equal to the ideal polishing effect, it indicates that the current parameter settings are reasonable and no adjustment is needed, thus avoiding unnecessary parameter adjustments and improving polishing efficiency. If the predicted polishing effect does not meet the ideal state, the real-time polishing parameters are immediately adjusted to the ideal polishing parameters to ensure optimal polishing quality. This real-time feedback and adjustment mechanism ensures the high efficiency and high quality of the polishing operation.

[0200] The multi-objective optimization method for robotic polishing processes provided in this embodiment has two aspects. First, it uses principal component analysis to reduce the dimensionality of collected historical polishing parameters and effects, extracting the reduced polishing parameters that have the greatest impact on the polishing effect. This effectively reduces the dimensionality of the data, lowers the complexity of subsequent analysis, and retains the main information in the data. Second, by calculating the contribution rate of each parameter in the reduced polishing parameters and combining it with the grey relational coefficient, a weighted grey relational degree is calculated. This not only quantifies the impact of each parameter on the polishing effect but also effectively addresses the multi-objective optimization problem (i.e., multiple polishing parameters and multiple...). The optimization of polishing effect is transformed into a single-objective optimization problem (i.e., optimizing only the weighted grey relational degree). This transformation simplifies the optimization process and avoids the non-uniqueness of solutions and computational complexity that may occur in multi-objective optimization. Thirdly, by using the weighted grey relational degree and historical polishing parameters as training data, the deep belief network is trained, enabling the network to capture the complex nonlinear relationship between polishing parameters and polishing effect, and accurately predict the polishing effect. In this way, by combining principal component analysis and weighted grey relational degree, the deep belief network can accurately achieve the optimization of multi-objective polishing parameters.

[0201] Corresponding to the aforementioned embodiment of a multi-objective optimization method for robotic polishing process, this application also provides a verification method for multi-objective optimization of robotic polishing process.

[0202] Specifically, to verify the feasibility and accuracy of the proposed algorithm, it was validated using the SIASUN GCR20-1100 robotic grinding experimental platform. This platform consists of a robot body, grinding discs, and sensors. The robot's repeatability is ±0.05mm, its effective payload is 20kg, its working space is 1100mm, and the force sensor used is ACF / 110 / 04 with a maximum force of 100N and a stroke of 35.5mm. Figure 4 As shown in Table 1, the experimental workpiece was designed as a 400×250mm stainless steel plate. This experiment mainly considered the influence of feed speed Vf and grinding force Fn on the grinding effect. The grinding process parameters are shown in Table 1. A coordinate measuring machine (e.g., Hexagon GLOBAL STATUS 9128) and a surface roughness tester (e.g., Mitutoyo SJ-210) were used to detect the single-pass grinding removal amount and surface roughness of the machined workpiece surface.

[0203] Table 1. Grinding process test parameters

[0204] Furthermore, the weights of surface roughness and material removal rate were calculated to be 0.568 and 0.432, respectively, using principal component analysis. Grey relational analysis was then performed on the experimental results in Table 1, and the results are as follows: Figure 4 As shown in Table 2, Figure 4 This is a schematic diagram illustrating the grey relational results of an exemplary embodiment of this application. The grey relational degree of group 7 is the highest, i.e., the grey relational degree is best, indicating that the process parameters of this group best meet the multi-objective requirements. The process parameter combination is forward tilt / lateral tilt angle of -30° and 6°, feed speed of 120mm / s, initial grinding force of 10N, final grinding force of 40N, and row spacing of 2.5mm.

[0205] Table 2 Results of Grey Relational Analysis

[0206] After obtaining the grey relational degree, a mapping relationship between the grey relational degree and process parameters is established using a deep belief network. By training on experimental data, the deep belief network can effectively learn the nonlinear relationship between process parameters and grey relational degree. Figure 5 This is a schematic diagram illustrating a comparison of experimental results for an exemplary embodiment of this application. Please refer to... Figure 5 The results show that the deep belief network model can accurately predict the grey relational degree value, and the results are highly consistent with the actual experimental results. When a new grinding parameter is received, the model can instantly calculate the corresponding grey relational degree value, compare this value with the known grey relational degree best, and determine the optimal combination of grinding parameters, thereby realizing the dynamic optimization and adjustment of the robot grinding operation process.

[0207] Furthermore, experiments confirmed that using principal component analysis to optimize grey relational analysis successfully transformed multi-objective optimization into a single-objective optimization problem. Deep belief networks established an accurate process parameter optimization model, effectively predicting grey relational degrees and enabling intelligent adjustment of grinding process parameters, thus resolving the contradiction between grinding quality and efficiency caused by the coupling of multiple process parameters.

[0208] Furthermore, the proposed optimization method not only experimentally verified its effectiveness in balancing surface roughness and material removal rate, but also improved the stability and reliability of the grinding process. This method provides strong support for the application of robotic grinding technology in improving the surface quality of civil aircraft fuselages and has broad application prospects.

[0209] Corresponding to the aforementioned embodiment of a multi-objective optimization method for robotic polishing process, this application also provides an embodiment of a multi-objective optimization device for robotic polishing process.

[0210] An embodiment of the multi-objective optimization device for robotic grinding processes disclosed in this application can be applied to a multi-objective optimization device for robotic grinding processes. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the multi-objective optimization device for robotic grinding processes loading the corresponding computer program instructions from the non-volatile memory into memory and executing them. The multi-objective optimization device for robotic grinding processes in the embodiment may also include other hardware depending on its actual function, which will not be elaborated further.

[0211] Figure 6 This is a schematic diagram of the structure of Embodiment 1 of the multi-objective optimization device for robotic grinding process provided in this application. Please refer to... Figure 6 The apparatus provided in this embodiment includes an acquisition module 610, a processing module 620, a calculation module 630, and a training module 640; wherein,

[0212] The acquisition module 610 is used to acquire historical polishing data of the robot, and extract historical polishing parameters and historical polishing effect evaluation values ​​from the historical polishing data; the historical polishing parameters include multiple parameters.

[0213] The processing module 620 is used to process the historical polishing parameters and the historical polishing effect based on principal component analysis to obtain simplified polishing parameters; the number of simplified polishing parameters is lower than the number of historical polishing parameters.

[0214] The calculation module 630 is used to calculate the contribution rate of each type of sanding parameter in the reduced sanding parameters, and to calculate the weighted grey relational degree based on the reduced sanding parameters, the contribution rate and the grey relational coefficient; a single weighted grey relational degree represents the relationship between all sanding parameters in the reduced sanding parameters and the corresponding sanding effect;

[0215] The training module 640 is used to train a deep confidence network based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network; wherein, the target deep confidence network contains the correlation between polishing parameters and polishing effect;

[0216] The processing module 620 is further configured to acquire real-time polishing parameters, process the real-time polishing parameters based on the deep belief network to obtain a predicted polishing effect, and optimize the real-time polishing parameters based on the predicted polishing effect.

[0217] The apparatus of this embodiment can be used to perform... Figure 1 The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0218] This application also provides a multi-objective optimization device for robotic polishing processes, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the methods provided in the first aspect of this application.

[0219] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods provided in this application.

[0220] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0221] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0222] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A multi-objective optimization method for robotic grinding process, characterized in that, The method includes: The grinding parameters are determined based on the feature information of the target object and the dynamic characteristics of the robot. The grinding parameters are used to represent the robot control parameters for grinding the object to the target object. There are multiple grinding parameters. Factors for evaluating the polishing effect are selected based on the characteristic information of the target object; The test specimen was polished according to the polishing parameters, and the values ​​of the polishing effect evaluation factors corresponding to each set of polishing parameters were obtained. A parameter matching table containing pairs of historical polishing parameters and historical polishing effects was constructed. The polishing effect evaluation factors were used to evaluate the polishing quality of the target object after polishing. There were multiple polishing effect evaluation factors. Based on principal component analysis, the historical polishing parameters and historical polishing effects are reduced to obtain the reduced polishing parameters; the number of the reduced polishing parameters is less than the number of the historical polishing parameters. Calculate the contribution rate of each sanding parameter in the reduced sanding parameters, and calculate the weighted grey relational degree based on the reduced sanding parameters, the contribution rate, and the grey relational coefficient; each weighted grey relational degree characterizes the relationship between all sanding parameters in the reduced sanding parameters and the corresponding sanding effect; A deep confidence network is trained based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network; wherein, the target deep confidence network contains the correlation between polishing parameters and polishing effect; The process of training a deep confidence network based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network includes: The historical polishing parameters are input into the first layer of the restricted Boltzmann machine of the depth belief network, and the desired polishing effect is output through the first layer of the restricted Boltzmann machine. Based on the correlation between the historical polishing parameters and the desired polishing effect, the weights and biases of the Boltzmann machine are adjusted by combining the weighted grey relational feedback. The data output from the first-layer restricted Boltzmann machine is used as the input to the second-layer restricted Boltzmann machine, and the steps of feedback adjustment of the weights and biases of the restricted Boltzmann machine are repeated. After the number of layers of the trained restricted Boltzmann machine reaches the preset number, the trained target depth belief network is obtained; Real-time grinding parameters are generated based on the target object and the test piece to be processed. The real-time grinding parameters at each point are input into the depth confidence network to obtain the predicted grinding effect. The real-time grinding parameters are then optimized based on the predicted grinding effect.

2. The method according to claim 1, characterized in that, The calculation of the weighted grey relational degree based on the reduced polishing parameters, the contribution rate, and the grey relational coefficient includes: Calculate the absolute difference between the historical polishing effect and the ideal polishing effect corresponding to each of the historical polishing parameters; The maximum absolute difference and the minimum absolute difference are determined based on the absolute difference, and the grey relational coefficient is calculated based on the absolute difference, the maximum absolute difference, and the minimum absolute difference. Based on the reduced polishing parameters, the contribution rate, and the grey relational coefficient, the correlation between each polishing parameter and the corresponding polishing effect is determined, and the weighted grey relational degree is obtained.

3. The method according to claim 1, characterized in that, The process of processing the historical polishing parameters and historical polishing effects using principal component analysis to obtain the reduced polishing parameters includes: Calculate the correlation coefficient matrix between each historical polishing parameter and the corresponding historical polishing effect to obtain the feature vector and feature value corresponding to the correlation coefficient matrix; each feature vector corresponds to a polishing parameter. The feature vectors are sorted based on the feature values, and the contribution rate of each polishing parameter to the polishing effect is calculated. Remove the feature vectors whose contribution rate is lower than a preset threshold to obtain the reduced polishing parameters.

4. The method according to claim 1, characterized in that, The step of selecting polishing effect evaluation factors based on the feature information of the target object includes: Based on the influence of polishing parameters on the historical polishing effect, the polishing parameters are divided into gain parameters and loss parameters; Calculate the negative of the reduction parameter and combine it with the gain parameter to generate intermediate polishing parameters; The intermediate polishing parameters are standardized using the Z-score method to obtain the historical polishing parameters.

5. The method according to claim 1, characterized in that, The process of determining grinding parameters based on the feature information of the target object and the dynamic characteristics of the robot includes: Determine the geometric and material characteristics of the target object; Determine the target contour trajectory of the target object after polishing based on the geometric features of the target object; The grinding allowance is determined based on the difference between the target contour trajectory and the contour of the test specimen. Based on the material characteristics, the interference torque of the robot when polishing the target object is determined; Determine the robot's dynamic characteristics, and based on the robot's dynamic characteristics, determine the real-time initial grinding parameters corresponding to the grinding margin. The interference parameter quantity of the real-time initial grinding parameters at each time point is calculated based on the interference torque. The real-time initial polishing parameters are corrected based on the interference parameter values ​​to obtain the polishing parameters.

6. The method according to claim 1, characterized in that, The step of selecting polishing effect evaluation factors based on the feature information of the target object includes: Determine the candidate effect index; Calculate the contribution of the candidate effect index to the geometric features of the target object; Candidate effect indices with contribution values ​​higher than a preset contribution value are selected as evaluation factors for the polishing effect.

7. The method according to claim 1, characterized in that, The step of inputting the real-time polishing parameters of each point into the depth confidence network to obtain the predicted polishing effect includes: Obtain the changing state of the polishing environment corresponding to the real-time polishing parameters; The empirical variation value of the polishing effect is calculated based on the aforementioned changing state; The predicted polishing effect is corrected based on the empirical change value.

8. The method according to claim 1, characterized in that, The optimization of the real-time polishing parameters based on the predicted polishing effect includes: Candidate data bars are calculated from the parameter matching table based on the real-time polishing parameters, wherein the candidate data bars are data bars in the parameter matching table whose similarity to the polishing parameters and the real-time polishing parameters is greater than a threshold; Match the candidate data bars with the processing environment corresponding to the real-time grinding parameters, and filter the target data bars from the candidate data bars; By comparing the historical polishing parameters in the target data bar with the predicted polishing effect, the polishing parameter corresponding to the value with the best polishing effect is selected as the final polishing parameter.

9. A multi-objective optimization device for robotic grinding process, characterized in that, The device includes: The acquisition module is used to determine the polishing parameters based on the feature information of the target object and the dynamic characteristics of the robot. The polishing parameters are used to represent the robot control parameters for polishing the object to the target object. There are multiple polishing parameters. The processing module is used to filter polishing effect evaluation factors based on the feature information of the target object; The acquisition module is also used to polish the test sample according to the polishing parameters, obtain the value of the polishing effect evaluation factor corresponding to each set of polishing parameters, and construct a parameter matching table containing pairs of historical polishing parameters and historical polishing effects. The polishing effect evaluation factor is used to evaluate the polishing quality of the target object after polishing. There are multiple polishing effect evaluation factors. The processing module is used to reduce the historical polishing parameters and the historical polishing effect based on principal component analysis to obtain the reduced polishing parameters; the number of the reduced polishing parameters is lower than the number of historical polishing parameters. The calculation module is used to calculate the contribution rate of each sanding parameter in the reduced sanding parameters, and to calculate the weighted grey relational degree based on the reduced sanding parameters, the contribution rate and the grey relational coefficient; a single weighted grey relational degree represents the relationship between all sanding parameters in the reduced sanding parameters and the corresponding sanding effect; The training module is used to train a deep confidence network based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network; wherein, the target deep confidence network contains the correlation between polishing parameters and polishing effect; The process of training a deep confidence network based on the weighted grey relational degree and the historical polishing parameters to obtain a trained target deep confidence network includes: The historical polishing parameters are input into the first layer of the restricted Boltzmann machine of the depth belief network, and the desired polishing effect is output through the first layer of the restricted Boltzmann machine. Based on the correlation between the historical polishing parameters and the desired polishing effect, the weights and biases of the Boltzmann machine are adjusted by combining the weighted grey relational feedback. The data output from the first-layer restricted Boltzmann machine is used as the input to the second-layer restricted Boltzmann machine, and the steps of feedback adjustment of the weights and biases of the restricted Boltzmann machine are repeated. After the number of layers of the trained restricted Boltzmann machine reaches the preset number, the trained target depth belief network is obtained; The processing module is further configured to generate real-time grinding parameters based on the target object and the test piece to be processed, input the real-time grinding parameters of each point into the depth confidence network to obtain the predicted grinding effect, and optimize the real-time grinding parameters based on the predicted grinding effect.

Images