A method and apparatus for dynamically compensating for polishing power

Abstract

The application provides a polishing force dynamic compensation method and device. The method provided by the application comprises the following steps: acquiring actual point cloud data of a polishing target object; constructing a theoretical model of the polishing target object based on polishing requirements, registering the actual point cloud data with the theoretical model, and generating a reference polishing track; inputting the reference polishing track into a pre-constructed variable impedance adaptive control model to calculate an expected polishing force; acquiring joint coordinates of a current polishing robot; inputting the expected polishing force and the joint coordinates into a pre-trained deep belief network to obtain a force error compensation value of the robot; compensating the expected polishing force based on the force error compensation value to obtain an updated polishing force, and controlling the robot to polish an aero-engine blade based on the updated polishing force. The polishing force dynamic compensation method and device provided by the application are used for improving the accuracy of polishing of the aero-engine blade.

Description

Technical Field

[0001] This application relates to the field of robotic blade grinding technology, and in particular to a method and apparatus for dynamic compensation of grinding force. Background Technology

[0002] The surface quality and profile accuracy of aero-engine blades directly affect the engine's aerodynamic performance, fuel efficiency, and service life. Therefore, high-precision grinding of aero-engine blades has become an important research direction.

[0003] In existing aero-engine blade grinding technologies, traditional CNC machine tool grinding methods can improve machining accuracy to a certain extent, but due to their poor versatility, the grinding environment and conditions vary, making it impossible to guarantee that each grinding operation will be performed according to the preset grinding force. This leads to problems such as uneven surface quality and difficulty in accurately controlling the machining allowance during the processing, thus affecting the grinding effect and processing efficiency. Therefore, there is an urgent need for a grinding optimization method that can adjust the grinding force in real time and adapt to different grinding conditions. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for dynamic compensation of grinding force to improve the accuracy of grinding aero-engine blades.

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

[0006] The first aspect of this application provides a method for dynamic compensation of grinding force, the method comprising:

[0007] Obtain the actual point cloud data of the target object to be polished;

[0008] A theoretical model of the target object for grinding is constructed based on the grinding requirements. The actual point cloud data is registered with the theoretical model to generate a reference grinding trajectory.

[0009] The reference grinding trajectory is input into a pre-built variable impedance adaptive control model to calculate the desired grinding force; wherein, the desired grinding force represents the grinding force required by the grinding robot to process the test workpiece to the grinding target under ideal conditions;

[0010] The construction of the variable impedance adaptive control model includes: establishing an initial impedance model; removing the target term from the initial impedance model based on the steady-state tracking error of the grinding robot to obtain a corrected impedance model; calculating the difference between the grinding wheel position and the end effector position of the grinding robot; determining the correlation between the damping coefficient and the force tracking error based on the difference and the corrected impedance model; using the material properties of the grinding target object and the robot processing properties as inputs to the fuzzy controller to calculate the update rate in the correlation; updating the correlation based on the update rate to obtain the variable impedance adaptive control model; and obtaining the joint coordinates of the current grinding robot.

[0011] The desired grinding force and the joint coordinates are input into a pre-trained deep belief network to obtain the robot's force error compensation value.

[0012] The desired grinding force is compensated based on the force error compensation value to obtain an updated grinding force, and the robot is controlled to grind the aero-engine blade based on the updated grinding force.

[0013] A second aspect of this application provides a dynamic compensation device for grinding force, the device comprising an acquisition module, a processing module, a calculation module, and a compensation module; wherein,

[0014] The acquisition module is used to acquire the actual point cloud data of the target object being polished;

[0015] The processing module is used to construct a theoretical model of the target object for grinding based on the grinding requirements, register the actual point cloud data with the theoretical model, and generate a reference grinding trajectory.

[0016] The calculation module is used to input the reference grinding trajectory into a pre-constructed variable impedance adaptive control model to calculate the desired grinding force; wherein, the desired grinding force represents the grinding force required by the grinding robot to process the test workpiece to the grinding target under ideal conditions;

[0017] The construction of the variable impedance adaptive control model includes: establishing an initial impedance model; removing the target term from the initial impedance model based on the steady-state tracking error of the grinding robot to obtain a corrected impedance model; calculating the difference between the grinding wheel position and the end effector position of the grinding robot; determining the correlation between the damping coefficient and the force tracking error based on the difference and the corrected impedance model; using the material properties of the grinding target object and the robot processing properties as inputs to the fuzzy controller to calculate the update rate in the correlation; and updating the correlation based on the update rate to obtain the variable impedance adaptive control model.

[0018] The acquisition module is also used to acquire the joint coordinates of the current grinding robot;

[0019] The processing module is also used to input the desired grinding force and the joint coordinates into a pre-trained depth belief network to obtain the robot's force error compensation value.

[0020] The compensation module is used to compensate the desired grinding force based on the force error compensation value to obtain an updated grinding force, and to control the robot to grind the aero-engine blade based on the updated grinding force.

[0021] The grinding force dynamic compensation method and apparatus provided in this application, in its first aspect, significantly improves the dynamic adjustment capability of grinding force during the grinding process by introducing a variable impedance adaptive control model. Compared with the traditional constant impedance control method, which often struggles to ensure precise control of grinding force in complex and ever-changing grinding environments, easily leading to force overshoot or undershoot, the variable impedance adaptive control model can dynamically adjust the damping coefficient according to the real-time state of the grinding target and grinding requirements, thereby ensuring the stability and consistency of force during the grinding process. Specifically, the variable impedance adaptive control model achieves precise control of grinding force by calculating the grinding force error in real time and updating the damping coefficient based on the error. During the grinding process, when the blade surface changes or external environmental disturbances occur, the model can respond quickly by adjusting the damping coefficient to suppress force overshoot, ensuring that the grinding force remains within the ideal range. This dynamic variable impedance adjustment capability not only improves the stability of grinding quality but also enhances the robustness and adaptability of the system, making the grinding process more stable and reliable.

[0022] Secondly, by employing a fuzzy control strategy, the generation of force overshoot is suppressed, thereby improving grinding accuracy. Fuzzy control processes real-time grinding parameters through fuzzification and calculates the update rate using fuzzy correspondence rules, thus achieving fine adjustment of the grinding force. When a force overshoot trend is detected, fuzzy control can respond quickly and suppress the generation of force overshoot by adjusting the update rate. This not only improves the control accuracy of the grinding force but also avoids problems such as blade surface damage or insufficient machining allowance caused by force overshoot.

[0023] In this way, the actual point cloud data is registered with the theoretical model to generate a high-precision reference grinding trajectory. This not only considers the geometric characteristics of the blade but also incorporates the requirements of the grinding process, ensuring the rationality and efficiency of the grinding path. This allows the planned reference grinding trajectory to more closely match the theoretical model, guaranteeing the accuracy of the grinding. The calculation of the expected grinding force provides a benchmark for subsequent force error compensation. By inputting the expected grinding force and joint coordinates into a pre-trained deep belief network, the system can quickly calculate the robot's force error compensation value. This fully utilizes the advantages of deep learning in handling complex data relationships, improving the accuracy and efficiency of force error compensation. Finally, although the expected grinding force can be calculated relatively accurately through the variable impedance adaptive control model, there will still be a certain gap between the calculated force and the actual force during the actual grinding process due to factors such as the uncertainty of the external environment and the limitations of equipment precision. To compensate for this gap, the expected grinding force and the joint coordinates of the current grinding robot are used as inputs. A multi-layer restricted Boltzmann machine is used to deeply analyze the complex relationship between the two, quickly and accurately calculating the force error compensation value. This compensation value accurately reflects the difference between the actual force and the expected force. This compensation mechanism not only enhances the robustness of the system but also makes the grinding process more stable and reliable. In summary, by using a dynamic compensation mechanism to adjust the grinding force in real time to adapt to changes in the blade surface and uncertainties in the external environment, not only is the grinding quality improved, but the robustness and adaptability of the system are also enhanced, ensuring the precision and efficiency of aero-engine blade grinding. Attached Figure Description

[0024] Figure 1 A flowchart of Embodiment 1 of the dynamic compensation method for grinding force provided in this application;

[0025] Figure 2 A flowchart illustrating dynamic compensation of grinding force as an exemplary embodiment of this application;

[0026] Figure 3 This is a schematic diagram of a three-dimensional model of an aero-engine blade, illustrating an exemplary embodiment of this application.

[0027] Figure 4 (a) A schematic diagram illustrating the grinding force result of an exemplary embodiment of this application;

[0028] Figure 4 (b) A schematic diagram illustrating the grinding force result in yet another exemplary embodiment of this application;

[0029] Figure 5 (a) is a schematic diagram illustrating the blade roughness before optimization, as shown in an exemplary embodiment of this application;

[0030] Figure 5(b) is a schematic diagram illustrating the optimized blade roughness as shown in an exemplary embodiment of this application;

[0031] Figure 5 (c) A comparison diagram of the effect of optimization on blade roughness before and after, as shown in an exemplary embodiment of this application;

[0032] Figure 6 This is a schematic diagram illustrating the blade edge shape detection result as an exemplary embodiment of this application;

[0033] Figure 7 This is a diagram illustrating a dynamic compensation device for grinding force, as shown in an exemplary embodiment of this application. Detailed Implementation

[0034] 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.

[0035] 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.

[0036] 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."

[0037] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0038] Figure 1 This is a flowchart of an embodiment of the dynamic compensation method for grinding force provided in this application. Please refer to... Figure 1 The method provided in this embodiment may include:

[0039] In the method provided by this invention, the multimodal data includes point cloud data of the target object to be polished, virtual model information of the target object to be polished, grinding wheel position data of the polishing robot, end position data of the polishing robot, and force data of the polishing robot, that is, data with multiple different positions, different time points, different amounts of information, and different dimensions.

[0040] S101. Obtain the actual point cloud data of the target object to be polished.

[0041] Specifically, the target object for polishing is the item that needs to be polished. In practice, the target object could be an aircraft engine blade to be polished.

[0042] Furthermore, to ensure the accuracy of point cloud data, a high-precision coordinate measuring machine can be used for measurement.

[0043] Furthermore, the laser tracker can be the Leica AT901-B, with an error of ±15μm+6μm / m.

[0044] It should be noted that the error of the laser tracker increases with the distance. For example, in one embodiment, combined with the above embodiment, the AT901-B uses an angle encoder to measure the angle and an absolute interferometer to measure the distance. The absolute interferometer in the AT901 integrates a helium-neon laser interferometer and an absolute rangefinder. These two lasers can work independently to measure the actual point cloud data of the aero-engine blade surface.

[0045] Furthermore, since the measurement process may be affected by factors such as environmental noise and equipment vibration, the actual point cloud data collected may contain some noise points. In order to improve the quality of the actual point cloud data, filtering algorithms (such as Gaussian filtering) can be used to remove noise points.

[0046] In practice, Gaussian filtering is used for noise reduction, and its mathematical expression is:

[0047] ;

[0048] in, Let the coordinates be the coordinates of each point in the point cloud. is the standard deviation of the Gaussian function.

[0049] ;

[0050] in, For the neighboring region The coordinates of each point The number of points in the neighborhood. For the first Gaussian weights of points For the first The distance between each point and the current point These are the coordinates of the filtered point.

[0051] S102. Construct a theoretical model of the target object for grinding based on the grinding requirements, register the actual point cloud data with the theoretical model, and generate a reference grinding trajectory.

[0052] Specifically, grinding requirements guide the grinding precision requirements for aero-engine blades. In practice, grinding requirements can include surface quality requirements, profile accuracy requirements, and the expected surface roughness after grinding.

[0053] In practice, before the grinding operation begins, a theoretical model is constructed based on the shape, size, and characteristics of the aero-engine blades to be ground.

[0054] Understandably, by registering actual point cloud data with theoretical models, a new and accurate model can be constructed to determine the parts of the target object that need to be polished.

[0055] The following is a specific example illustrating the process of generating a reference polishing trajectory:

[0056] (1) Based on the distance between each actual point cloud data and the data point in the theoretical model, determine the corresponding point of each actual point cloud data, and combine the actual point cloud data and the data point in the theoretical model into a neighboring point pair.

[0057] In practical implementation, the actual point cloud data can be traversed. For each actual point cloud data point, its distance to all data points in the theoretical model is calculated, and the point in the theoretical model that is closest to it is found as the corresponding point. For example, in one embodiment, the KD-tree algorithm can be used to quickly search for its corresponding point.

[0058] Furthermore, a neighboring point pair is the corresponding reference point in the theoretical model for the actual point cloud data point. Each actual point cloud data point and its corresponding point in the theoretical model can be combined into a neighboring point pair.

[0059] (2) When the distance from all actual point cloud data to the data points in the theoretical model is less than the preset threshold, the reference polishing trajectory is determined based on the theoretical model.

[0060] Specifically, the specific value of the preset threshold is set according to actual needs, and this embodiment does not limit it.

[0061] Furthermore, the distance values ​​from all actual point cloud data to the data points in the theoretical model are traversed and compared with a preset threshold. When all distance values ​​are less than the preset threshold, the theoretical model is determined to be successfully constructed.

[0062] Furthermore, based on the comparison between the boundary contour represented by the constructed theoretical model and the object to be polished, the influence of noise and outliers is eliminated through smoothing and other processes, and a reference polishing trajectory is generated. The reference polishing trajectory will serve as the benchmark for the movement of the polishing robot, ensuring that the polishing operation can proceed along the expected path.

[0063] The dynamic compensation method for grinding force provided in this embodiment performs high-precision registration between actual point cloud data and theoretical models, ensuring accurate alignment between the geometry of the actual blade and the theoretical model. This eliminates deviations caused by manufacturing errors, deformation, and other factors, ensuring high precision and efficiency in grinding operations.

[0064] The following is a specific embodiment to illustrate in detail the process of registering the actual point cloud data with the theoretical model:

[0065] A first point set is constructed based on the actual point cloud data, and a second point set is constructed based on the theoretical model.

[0066] Traverse each point in the first point set, search for matching points in the second point set based on the improved KD tree, establish a correspondence, and complete the registration.

[0067] The search for matching points in the second point set based on the improved KD tree includes:

[0068] Determine the position coordinates of each point in the first point set and the second point set;

[0069] Select the first target point that is not matched in the first point set, and determine the first position coordinates corresponding to the first target point;

[0070] The search neighborhood is calculated based on the contour shape determined by the actual point cloud data;

[0071] Within the search neighborhood, select the optimal second target point from the second point set and establish a correspondence.

[0072] In practice, high-precision coordinate measuring machines or laser trackers are used to acquire actual point cloud data of the surface of aero-engine blades. These data points constitute a three-dimensional point set representing the actual shape of the blade, which is called the first point set.

[0073] Furthermore, the theoretical model of the aero-engine blade constructed based on the grinding requirements is a three-dimensional digital model that includes the shape and size of the grinding target object. All data points are extracted from this model to form a second point set. The second point set is used to more accurately represent the shape of the theoretical model, especially the boundary contour and abrupt changes in the coordinates of the position points.

[0074] Furthermore, each point in the first point set is traversed. For each unmatched first target point, a matching point in the second point set is found through further searching. This further search process can be as follows: 1. Determine the point set coordinates: The three-dimensional coordinates of each point in the first and second point sets need to be clearly defined, which is the basis for spatial search and matching; 2. Select the first target point: Select a point from the unmatched first point set as the first target point to be processed, and determine its three-dimensional coordinates; 3. Calculate the search neighborhood: Determine the search neighborhood based on the contour shape of the actual point cloud data. For example, in one embodiment, contour point clouds are extracted from the actual point cloud data. These point clouds are connected according to their positional relationships to form a contour shape. Then, with the first target point as the center, the contour shape is scaled proportionally by a preset number to obtain an irregular search neighborhood. The shape and size of this search neighborhood are designed to cover the area that may contain a matching point; 4. Search for the optimal matching point within the search neighborhood: Within the determined search neighborhood, the nearest point is quickly searched in the second point set using an improved KD-tree data structure. The improved KD-tree accelerates the nearest neighbor search process by segmenting the space. For each point in the second set of points within the search neighborhood, calculate its Euclidean distance (or other suitable distance metric) to the first target point, and select the point with the smallest distance as the optimal second target point.

[0075] Furthermore, once the optimal second target point is found, a correspondence is established between the first target point and the second target point, and they are marked as matched. The above process is repeated until all points in the first point set have found their corresponding matching points.

[0076] Furthermore, once all points have been successfully matched, the correspondence between the first and second point sets is established, completing the registration of the actual point cloud data with the theoretical model.

[0077] The offline compensation method for absolute positioning accuracy of industrial robots provided in this embodiment constructs a first point set and a second point set, and uses an improved KD tree to efficiently search for the optimal matching point in the search neighborhood. This achieves high-precision registration between actual point cloud data and theoretical models, which not only improves the registration speed and accuracy, but also enhances the system's adaptability to complex shapes and boundary conditions. It provides more accurate basic data for subsequent grinding trajectory planning and grinding force control, thereby improving the accuracy and efficiency of aero-engine blade grinding.

[0078] S103. Input the reference grinding trajectory into the pre-constructed variable impedance adaptive control model to calculate the desired grinding force; wherein, the desired grinding force represents the grinding force of the grinding robot to process the test workpiece to the grinding target under ideal conditions.

[0079] The construction of the variable impedance adaptive control model includes: establishing an initial impedance model; removing the target term from the initial impedance model based on the steady-state tracking error of the grinding robot to obtain a corrected impedance model; calculating the difference between the grinding wheel position and the end effector position of the grinding robot; determining the correlation between the damping coefficient and the force tracking error based on the difference and the corrected impedance model; using the material properties of the grinding target object and the robot processing properties as inputs to the fuzzy controller to calculate the update rate in the correlation; and updating the correlation based on the update rate to obtain the variable impedance adaptive control model.

[0080] Specifically, the grinding force error is the error between the robot's actual grinding force and the expected grinding force.

[0081] Furthermore, by combining the characteristics of the object to be polished, the calculation process corresponding to the variable impedance adaptive control model is constructed, and the constructed variable impedance adaptive control model is obtained.

[0082] Understandably, the robot's current grinding position can be determined by the expected grinding force and the robot's current actual grinding force. For example, in one embodiment, the robot should be grinding position A according to the expected grinding force, but measurements determine that the robot is actually grinding with a grinding force b, so the grinding position is position B.

[0083] In practice, the reference grinding trajectory is input into the pre-trained variable impedance adaptive control model, and the model calculates the expected grinding force required at the current grinding position based on the input information.

[0084] The following is a specific example to illustrate the calculation process of the desired grinding force:

[0085] (1) Obtain the real-time grinding parameters during the grinding process and calculate the fuzzy parameters in combination with the corrected impedance model.

[0086] Specifically, real-time polishing parameters are used to characterize the parameters of the robot during polishing.

[0087] In a specific implementation, for example, in one embodiment, the real-time grinding parameters include blade stiffness. and robot feed speed .

[0088] Furthermore, a Gaussian membership function is used to fuzzify each real-time grinding parameter. The Gaussian membership function can map continuous values ​​to the interval [0, 1], indicating the degree to which the value belongs to a certain fuzzy set.

[0089] In practical implementation, for each real-time grinding parameter, the mean and standard deviation of the Gaussian membership function are determined based on its actual value range and the requirements of fuzzy control. The current value of each real-time grinding parameter is substituted into the corresponding Gaussian membership function to calculate the membership degree of the value to each fuzzy set. For example, in one embodiment, the grinding force is divided into three fuzzy sets: "small," "medium," and "large," with each set corresponding to a Gaussian membership function.

[0090] (2) Calculate the update rate corresponding to the real-time polishing parameters based on fuzzy parameters.

[0091] Specifically, the update rate is used to characterize the frequency of updates to real-time polishing parameters.

[0092] The following is a specific example to illustrate the calculation process of the update rate:

[0093] Step 1: The blade stiffness and the robot feed speed are fuzzed based on the Gaussian membership function to obtain fuzzy parameters.

[0094] In specific implementation, for example, in one embodiment, for blade stiffness, three fuzzy sets can be defined: "low stiffness", "medium stiffness" and "high stiffness"; for robot feed speed, three fuzzy sets can be defined: "slow speed", "medium speed" and "fast speed".

[0095] Furthermore, the actual measured values ​​of blade stiffness and robot feed speed are substituted into the corresponding Gaussian membership functions to calculate their membership degrees to each fuzzy set. Each actual value is then transformed into a fuzzy parameter, which consists of multiple fuzzy sets and their corresponding membership degrees.

[0096] Step 2: Process the fuzzy parameters based on the fuzzy correspondence rule to obtain the adjustment method and adjustment value corresponding to the fuzzy parameters; the fuzzy correspondence rule is used to characterize the correlation between different combinations of the fuzzy parameters and the adjustment method and adjustment value.

[0097] Specifically, fuzzy correspondence rules are formulated based on process experience and experimental data.

[0098] Understandably, the fuzzy correspondence rule describes the relationship between different combinations of fuzzy parameters (fuzzy sets of blade stiffness and robot feed speed) and adjustment methods (such as increasing, decreasing or keeping it unchanged) and adjustment values ​​(specific values ​​or ranges of values).

[0099] Furthermore, the fuzzy parameters are input into the fuzzy correspondence rules, and the corresponding adjustment method and adjustment value are determined according to the rules. For example, in one embodiment, combined with the above embodiment, the blade stiffness belongs to the "low stiffness" fuzzy set and the robot feed speed belongs to the "fast" fuzzy set. The corresponding fuzzy correspondence rule is: reduce the update rate to avoid overshoot.

[0100] Step 3: Defuzzify the adjusted value and calculate the update rate based on the adjusted method.

[0101] In practice, defuzzification can be performed based on methods such as centroid method and maximum membership degree method to obtain specific update rate values.

[0102] The dynamic compensation method for grinding force provided in this embodiment uses Gaussian membership function fuzzification to enable these parameters to adapt to complex and ever-changing grinding environments, thereby enhancing the system's adaptability and robustness. At the same time, by defuzzifying the adjustment values ​​and calculating the update rate in conjunction with the adjustment method, the control strategy can respond to changes in the grinding environment in real time and dynamically adjust the grinding force, thereby further improving grinding efficiency and accuracy while ensuring grinding quality.

[0103] (3) Update the current damping coefficient in real time based on the update rate;

[0104] In practice, the damping coefficient can be updated according to the update rate.

[0105] The following is a specific example to illustrate the calculation process of the damping coefficient:

[0106] Step 1: Based on the force sensor on the robot, obtain the actual force applied by the robot at the previous moment.

[0107] Specifically, in each control cycle, the force sensor transmits the actual force value from the previous moment to the control system.

[0108] Step 2: Calculate the robot's expected force at the previous moment by combining the robot's current position and the reference grinding trajectory.

[0109] Specifically, the control system calculates the expected force that the robot should have applied at the previous moment based on the pre-planned reference grinding trajectory and the robot's current position information.

[0110] Step 3: Calculate the difference between the expected force at the previous moment and the actual force at the previous moment, and calculate the first intermediate value by combining the update rate and the damping coefficient at the previous moment.

[0111] Specifically, the difference between the expected force and the actual force at the previous moment is calculated to obtain the force error.

[0112] Furthermore, the first intermediate value can be expressed as: ;

[0113] Where σ is the update rate and b is the damping coefficient of the previous time step. For the expectation of the previous moment, The actual force at the previous moment.

[0114] Step 4: Calculate the sum of the first intermediate value and the adaptive factor at the previous time step, and weight the damping coefficient at the previous time step to obtain the second intermediate value.

[0115] Specifically, the second intermediate value can be represented as: ;

[0116] in, The second intermediate value, The first median value, This is the adaptive factor from the previous time step.

[0117] Step 5: Calculate the current damping coefficient based on the second intermediate value and the position of the robot's grinding wheel.

[0118] Furthermore, the damping coefficient can be calculated based on the following formula:

[0119] ;

[0120] in, The second intermediate value, This indicates the position of the grinding wheel. Fixed value: 10 -8 .

[0121] (4) Calculate the desired grinding force based on the updated damping coefficient.

[0122] In practice, the desired grinding force is calculated based on the following formula:

[0123] ;

[0124] in, For force error, The actual force applied. To enhance the desired polishing power, The desired inertia coefficient, The second derivative of the force error. The damping coefficient is... The first derivative of the force error. The second derivative of the position error. The first derivative of the position error. This is the additional acceleration term caused by position error. This is the additional velocity term caused by position error.

[0125] The dynamic compensation method for grinding force provided in this embodiment enhances the adaptability and robustness of the system by acquiring grinding parameters in real time and performing fuzzification using a Gaussian membership function. The fuzzification process transforms continuous and complex grinding parameters into easily processed fuzzy parameters, providing a more flexible and reliable basis for subsequent force control. The update rate calculated based on these fuzzy parameters can more accurately reflect the real-time changes in the grinding environment, thereby achieving dynamic adjustment of the damping coefficient. This adaptive adjustment mechanism effectively addresses the uncertainties in the grinding process and ensures precise control of the grinding force.

[0126] The following is another specific embodiment to illustrate the process of dynamic compensation of grinding force:

[0127] In practice, the relationship between various computational quantities can be established first to obtain the corresponding formulas. Then, fuzzy calculations are performed using blade stiffness and robot feed speed. The update rate corresponding to the real-time grinding parameters is calculated based on the fuzzy parameters, thereby controlling the damping coefficient and establishing the relationship between the end force and position of the grinding robot. On this basis, a well-constructed deep belief network is added before the variable impedance adaptive control model, and the deep belief network and the variable impedance adaptive control model are integrated into a whole.

[0128] The following describes the process of "establishing the relationships between various computational quantities to obtain the corresponding formulas":

[0129] For the initial impedance model;

[0130] in, These are the inertial parameters of the initial impedance model. These are the damping parameters of the initial impedance model. This is the force error signal.

[0131] Furthermore, due to system communication delays and sensor measurement noise, the position of the grinding wheel used by the robot for polishing may have errors. Therefore, the grinding wheel position is estimated using... To indicate the position of the grinding wheel, the actual position is also indicated by... Therefore, the estimated position of the grinding wheel can be expressed as:

[0132] ;

[0133] in, This indicates the position of the grinding wheel. It is still used for the actual location. This refers to the position measurement error caused by uncertain factors.

[0134] Furthermore, the difference between the actual grinding wheel position and the actual position of the system actuator end can be simplified as follows:

[0135] ;

[0136] in, The estimated error after compensation is given, and e is the original force tracking error. This refers to the position measurement error caused by uncertain factors.

[0137] Furthermore, based on the above formula, the expression for the variable impedance adaptive control model is determined as follows:

[0138] ;

[0139] in, To address force tracking error, For actual polishing power, To refine the desired performance.

[0140] Furthermore, by adjusting the damping coefficient in real time This improves the force tracking accuracy of the system. The adaptive variable impedance control law can be expressed as:

[0141] ;

[0142] in, To address force tracking error, For actual polishing power, To refine the desired performance.

[0143] It should be noted that the damping coefficient Adjust force tracking error online in real time:

[0144] ;

[0145] in, This is an adaptive factor, related to the force tracking of the system, used to adjust the damping coefficient. . For the adaptive term at the current moment, The adaptive term from the previous time step is used in the initial time step. . To hone one's abilities to meet the expectations of the previous moment. The grinding force applied to the blade by the system based on feedback from the force sensor at the previous moment. This represents the system's sampling frequency. The update rate is the factor that determines the effectiveness of the system in compensating for force tracking errors and the stability of the system. .

[0146] The following is a specific embodiment to illustrate the construction process of the variable impedance adaptive control:

[0147] The construction of the variable impedance adaptive control model includes: establishing an initial impedance model; removing the target term from the initial impedance model based on the steady-state tracking error of the grinding robot to obtain a corrected impedance model; calculating the difference between the grinding wheel position and the end effector position of the grinding robot; determining the correlation between the damping coefficient and the force tracking error based on the difference and the corrected impedance model; using the material properties of the grinding target object and the robot processing properties as inputs to the fuzzy controller to calculate the update rate in the correlation; and updating the correlation based on the update rate to obtain the variable impedance adaptive control model.

[0148] Specifically, the initial impedance model is a fundamental mathematical model used to describe the influence of the dynamic characteristics of a grinding robot on contact forces when it interacts with its environment. In practice, the initial impedance model can be obtained through analysis and calculation based on robot dynamics theory and empirical data.

[0149] Furthermore, steady-state tracking error is the difference between the actual contact force and the expected contact force of the robot under steady-state operating conditions. In practice, during the robot's grinding process, the contact force can be monitored in real time by a force sensor and compared with the preset expected contact force to calculate the steady-state tracking error.

[0150] Furthermore, the modified impedance model is an optimized model obtained by removing objective terms (such as the desired stiffness term) related to the steady-state tracking error from the initial impedance model. In practice, the initial impedance model can be modified based on the analysis of the steady-state tracking error.

[0151] Furthermore, the positions of the grinding wheel and the end effector can be measured in real time using an encoder inside the robot or an external position sensor, and the difference between the two can be calculated to obtain the difference between the position of the grinding wheel and the position of the end effector of the grinding robot.

[0152] Furthermore, the correlation between the damping coefficient and the force tracking error characterizes how the damping coefficient affects the force tracking error. Through theoretical analysis and experimental data, the variation law of the force tracking error under different damping coefficients can be studied, thereby establishing the correlation between the damping coefficient and the force tracking error.

[0153] Furthermore, fuzzy controllers are controllers based on fuzzy logic, capable of handling uncertainty and fuzziness, and suitable for controlling complex systems. In practical implementation, fuzzy control rules and membership functions can be designed according to the system's control requirements, and the fuzzy controller can be implemented through programming.

[0154] Furthermore, the material properties of the target object to be polished can be the hardness and toughness of aircraft blades, while the processing properties of the robot can be the feed rate and spindle speed.

[0155] Furthermore, the update rate is a parameter used in fuzzy controllers to adjust the relationship between the damping coefficient and the force tracking error. The update rate affects the response speed and stability of the control strategy to errors.

[0156] In practical implementation, an initial impedance model is established as the starting point. Then, by analyzing the tracking error of the grinding robot in steady state, the target term in the initial impedance model is removed, resulting in a corrected impedance model. This process aims to reduce errors and improve model accuracy. Next, the difference between the grinding wheel position and the actuator end position is calculated, reflecting the positional offset during the actual grinding process. Subsequently, using the corrected impedance model and the positional difference, the correlation between the damping coefficient and the force tracking error is determined, i.e., the variation law of the force tracking error under different damping coefficients is clarified. To further optimize the control effect, the material properties of the grinding target object (such as hardness and toughness) and the robot's processing properties (such as feed rate and acceleration) are used as inputs to the fuzzy controller. The update rate in the correlation is calculated using a fuzzy control algorithm to dynamically adjust the damping coefficient. Finally, based on the calculated update rate, the correlation is updated in real time, thereby constructing a variable impedance adaptive control model that can adaptively adjust the grinding force.

[0157] The following is another specific embodiment to illustrate the construction process of the variable impedance adaptive control:

[0158] (1) Obtain the initial impedance model and convert the initial impedance model into a transfer function form model based on the Laplace transform.

[0159] Specifically, the relevant parameters and performance of the control strategy of the initial impedance model are extracted from historical data.

[0160] In practice, damping coefficients, stiffness coefficients, and desired forces can be obtained as historical data to determine the corresponding initial impedance model.

[0161] Furthermore, the transfer function form of the initial impedance model can be obtained using the following formula:

[0162] ;

[0163] in, For the transfer function form model, For the Laplace operator, , , These represent the three parameters of the desired impedance model: inertia, damping, and stiffness, respectively.

[0164] (2) Substitute the preset desired stiffness coefficient and the position information of the aero-engine blade into the transfer function form model to obtain the variable impedance adaptive control adapted to the aero-engine blade.

[0165] Specifically, the preset desired stiffness coefficient reflects the blade's ability to resist deformation when subjected to external forces.

[0166] In practice, a desired stiffness coefficient is preset based on the material properties and processing requirements of the aero-engine blades.

[0167] Furthermore, by substituting the preset desired stiffness coefficient and the real-time acquired blade position information into the transfer function model obtained through the Laplace transform, a variable impedance adaptive control strategy adapted to the aero-engine blade is obtained.

[0168] It should be noted that if the blade is modeled as a first-order spring, the actual applied force can be expressed as: Therefore, the force tracking error is:

[0169] ;

[0170] in, Indicates a reference position. Indicates the required location to be sent to the end of the system actuator. This indicates the actual position of the actuator end effector in the system. It is typically assumed that the position control system has good control performance. , Indicates the position of the grinding wheel.

[0171] Furthermore, it can be determined ;

[0172] Furthermore, the force tracking error when the system is stable is:

[0173] ;

[0174] Furthermore, as can be seen from the above equation, the steady-state force error of the system is not only related to the workpiece stiffness. It is related to the stiffness parameters of the desired impedance model, and also to the reference position trajectory. Relevant. To ensure that the force tracking error of the system approaches zero when it is stable, one of the following formulas must be satisfied:

[0175] ;

[0176] ;

[0177] Furthermore, if we want to reduce the steady-state error of the system's force... A value of 0 requires precise position and stiffness information of the blades to achieve the desired force tracking effect. This is because the blade stiffness coefficient... Since it is time-varying and its precise value is difficult to obtain, this paper will use the expected stiffness coefficient of the impedance model. Set to 0. Even with unknown blade position information, it is theoretically possible to achieve a force tracking error approaching 0 when the system reaches stability. Therefore, the desired stiffness term in the impedance model formula can be removed, resulting in a new impedance model as shown in the following formula: .in, , , These represent the three parameters of the desired impedance model: inertia, damping, and stiffness, respectively.

[0178] The dynamic compensation method for grinding force provided in this embodiment acquires historical variable impedance adaptive control strategies and converts them into transfer function models based on Laplace transform. This transforms complex time-domain differential equations into algebraic equations in the frequency domain, making the analysis and design of control strategies more intuitive and efficient. Furthermore, converting historical control strategies into transfer function form through Laplace transform not only simplifies the system model but also improves the efficiency of control strategy analysis and adjustment. This allows the control strategy to more flexibly adapt to various changes during blade grinding, thereby ensuring the stability and consistency of grinding quality.

[0179] The following is another specific embodiment to illustrate in detail the process of inputting the reference grinding trajectory into a pre-built variable impedance adaptive control model to calculate the desired grinding force:

[0180] Starting from the starting point of the reference polishing trajectory, determine the trajectory points to be calculated;

[0181] The coordinate information of the trajectory point to be calculated is input into the variable impedance adaptive control model to calculate the expected grinding force corresponding to the trajectory point to be calculated.

[0182] Determine the rate of change of the reference polishing trajectory within the first subsequent distance for the trajectory point to be calculated;

[0183] The adaptive period is calculated based on the product of the rate of change and the reference period;

[0184] The next sampling point is calculated based on the sum of the coordinates of the trajectory point to be calculated and the adaptive period.

[0185] The next sampling point is used as the trajectory point to be calculated. The process returns to the step of inputting the coordinate information of the trajectory point to be calculated into the variable impedance adaptive control model until the trajectory point to be calculated is the end point of the reference grinding trajectory.

[0186] Specifically, starting from the beginning of the reference polishing trajectory, the first trajectory point to be calculated is determined. In practice, the starting point of the reference polishing trajectory can be the first point of the reference polishing trajectory generated by registering the actual point cloud data with the theoretical model.

[0187] Furthermore, the coordinate information of the current trajectory point to be calculated (e.g., position, orientation, etc.) is input into the pre-trained variable impedance adaptive control model. Through a series of complex mathematical formulas and algorithms contained within the model, the expected grinding force to be applied at the trajectory point is calculated based on the input trajectory point information, the dynamic characteristics of the grinding robot (such as mass, inertia, etc.), and the material properties of the blades. This expected grinding force is intended to minimize force fluctuations and overshoot during the grinding process while ensuring grinding quality.

[0188] Furthermore, the rate of change of the reference grinding trajectory is determined within a first distance (the first distance is a preset parameter used to control the trajectory length considered when calculating the rate of change) of the trajectory point to be calculated. This rate of change can be obtained by calculating the changes in distance and angle between adjacent points on this segment of the trajectory, reflecting the degree of change in the shape and direction of the trajectory within this short distance.

[0189] Furthermore, the reference period is a fixed time interval used to control the sampling frequency during the polishing process.

[0190] In practice, the adaptive period is calculated by multiplying the calculated trajectory change rate by the preset reference period. The adaptive period is a dynamically adjusted parameter that adjusts the sampling frequency according to the complexity and change rate of the trajectory to ensure higher sampling accuracy in areas with large trajectory changes.

[0191] Furthermore, based on the coordinates of the current trajectory point to be calculated and the calculated adaptive period, the coordinates of the next sampling point are calculated through linear interpolation or other mathematical methods. This sampling point will serve as the starting point for the next calculation, and will be used to continue calculating the desired grinding force along the reference grinding trajectory.

[0192] Furthermore, the newly calculated next sampling point is used as the new trajectory point to be calculated, and the above steps are repeated (that is, its coordinate information is input into the variable impedance adaptive control model to calculate the desired grinding force, and then the trajectory change rate, adaptive period, and the next sampling point are calculated). The calculation is continued to form an iterative loop until the trajectory point to be calculated reaches the end point of the reference grinding trajectory.

[0193] Furthermore, the iteration process ends when the trajectory point to be calculated reaches the end point of the reference grinding trajectory. At this point, the system has calculated the expected grinding force at each sampling point during the grinding process based on the entire reference grinding trajectory and the variable impedance adaptive control model. These expected grinding forces will be sent as control signals to the grinding robot to guide it to perform grinding operations according to the predetermined trajectory and force.

[0194] It should be noted that by calculating the expected grinding force at each point on the reference grinding trajectory and dynamically adjusting the sampling period according to the trajectory change rate, the accuracy and stability of aero-engine blade grinding are improved. In areas with complex and variable trajectories or large variations, this technology can more accurately control the grinding force, effectively avoid force overshoot and fluctuations, and ensure the consistency and high quality of grinding results. At the same time, dynamically adjusting the sampling period not only improves grinding efficiency but also enhances the flexibility and adaptability of the system, making the grinding process more refined and controllable, ultimately achieving high-precision and high-efficiency blade grinding operations.

[0195] S104. Obtain the joint coordinates of the current grinding robot.

[0196] In practice, the joint coordinates of the current grinding robot can be obtained by measuring with a high-precision coordinate measuring machine.

[0197] S105. Input the desired grinding force and the joint coordinates into a pre-trained deep belief network to obtain the robot's force error compensation value.

[0198] Specifically, a pre-trained deep belief network is used to characterize the relationship between the expected grinding force and the force error compensation value.

[0199] It should be noted that the deep belief network consists of multiple restricted Boltzmann machines connected together and a regression layer. It is created by fine-tuning the resulting deep network through gradient descent and backpropagation to form the optimal model. The first layer of the restricted Boltzmann machine consists of the visible layer... and hidden layer Together, the visible layer and the hidden layer are composed of a weight matrix. , , These represent the biases of the visible and hidden layers, respectively. The second layer is the visible layer that confines the Boltzmann machine. The first layer is the hidden layer of the Boltzmann machine. ,Right now And so on. The last layer of the deep belief network is set up as a BP network, which receives the output feature vector of the restricted Boltzmann machine as its input feature vector.

[0200] Furthermore, the input layers of the deep belief network are the robot's end effector force and the corresponding six joint angles. The output layer represents the robot's force error. .

[0201] It should be noted that, in order to achieve the best training results, LSHADE (Linear Success history-based Adaptive Differential Evolution) can be used to optimize six dimensions of the deep belief network: the number of hidden layers, the number of nodes in the hidden layers, the learning rate, the momentum factor, the number of iterations of the restricted Boltzmann machine, and the number of fine-tuning iterations of the deep belief network.

[0202] The following is a specific example to illustrate the process of obtaining the force error compensation value:

[0203] (1) The correlation between the expected grinding force and the joint coordinates and the force error compensation value is analyzed by a multilayer restricted Boltzmann machine based on the deep belief network.

[0204] Specifically, the current expected grinding force and robot joint coordinates are used as input data and fed into a pre-trained deep belief network. The input data is propagated through each layer of the deep belief network through a restricted Boltzmann machine. Each layer of the restricted Boltzmann machine extracts and analyzes the features of the input data and predicts the feature vector of the force error compensation value corresponding to the input data of each layer. After receiving the output of the previous layer, the output layer performs linear combination and activation function processing again to finally obtain the force error compensation value.

[0205] (2) Output the force error compensation value based on the output layer of the deep belief network.

[0206] Specifically, the current expected grinding force and robot joint coordinates are input into the deep belief network. The feature vectors representing the force error compensation value of the output of the Boltzmann machine in each layer are merged and converted into the grinding force value through the output layer of the deep belief network, which is then determined as the force error compensation value.

[0207] The dynamic grinding force compensation method provided in this embodiment analyzes the complex relationship between the desired grinding force, robot joint coordinates and force error compensation value through deep belief networks and multi-layer restricted Boltzmann machines, and then accurately outputs the force error compensation value, which effectively improves the control accuracy of grinding force, reduces force fluctuations caused by factors such as position errors, and ensures the high quality and stability of grinding operations.

[0208] Figure 2 This is a flowchart illustrating dynamic compensation of grinding force as shown in an exemplary embodiment of this application. Please refer to... Figure 2First, actual point cloud data of the aero-engine blade surface is acquired using high-precision measurement equipment. Then, a theoretical model of the blade is constructed based on the grinding requirements, and the actual point cloud data is registered with the theoretical model to generate a high-precision reference grinding trajectory. Next, the reference grinding trajectory is input into a pre-trained variable impedance adaptive control model. Combined with real-time grinding parameters, the desired grinding force is calculated. This force is the ideal grinding force that the grinding robot should apply. Simultaneously, the joint coordinates of the current grinding robot are acquired, and the desired grinding force and joint coordinates are input into a pre-trained deep belief network. Utilizing the powerful processing capabilities of deep learning, the robot's force error compensation value is quickly calculated. Finally, the desired grinding force is compensated based on this force error compensation value to obtain an updated grinding force. The updated grinding force is then used to control the robot to perform the blade grinding operation. The entire process fully leverages the advantages of multimodal data fusion and intelligent control algorithms, ensuring high precision and efficiency in the grinding operation.

[0209] S106. Based on the force error compensation value, the desired grinding force is compensated to obtain an updated grinding force, and the robot is controlled to grind the aero-engine blade based on the updated grinding force.

[0210] In practice, the force error compensation value is a specific numerical value that represents the difference between the actual force applied by the robot to a point on the object to be polished at a specific location and state, and the expected force. The force error compensation value can be superimposed on the expected polishing force to obtain the compensated updated polishing force.

[0211] Furthermore, both the force error compensation value and the desired grinding force can be in numerical form. If the two forces are in the same direction, the force error compensation value can be directly added to the desired grinding force to obtain an updated grinding force value. If the force error compensation value and the desired grinding force are in different directions, the difference in direction between them needs to be considered when superimposing them. The force error compensation value should be inverted before being added to the desired grinding force to ensure that the updated grinding force is in the correct direction.

[0212] Furthermore, if the force error compensation value and the expected grinding force are vectors that include force and displacement, and have two attributes of magnitude and direction, then when superimposing them, it is necessary to ensure that the directions of the two forces are consistent or to correctly handle the situation where the directions are inconsistent. This process can be accomplished by vector addition, ensuring that the updated grinding force takes into account both the expected magnitude and corrects the actual error.

[0213] It should be noted that by compensating for the expected grinding force at each point of the grinding robot, the updated grinding force at each point of the robot can be determined, and the grinding can be achieved by the combined updated grinding forces at each point.

[0214] The grinding force dynamic compensation method and apparatus provided in this application register actual point cloud data with a theoretical model to generate a high-precision reference grinding trajectory. This not only considers the geometric characteristics of the blade but also incorporates the requirements of the grinding process, ensuring the rationality and efficiency of the grinding path. This allows the planned reference grinding trajectory to more closely resemble the theoretical model, guaranteeing grinding accuracy. Furthermore, through a variable impedance adaptive control model, the system can dynamically adjust the grinding force according to the real-time state of the blade and grinding requirements, ensuring the stability and consistency of the force during the grinding process. The calculation of the desired grinding force provides a benchmark for subsequent force error compensation. Finally, by inputting the desired grinding force and joint coordinates into a pre-trained deep belief network, the system can quickly calculate the robot's force error compensation value. This fully utilizes the advantages of deep learning in handling complex data relationships, ensuring the accuracy and reliability of aero-engine blade grinding.

[0215] Corresponding to the aforementioned embodiment of the dynamic compensation method for grinding force, this application also provides an experimental verification process for the dynamic compensation method for grinding force:

[0216] Taking a certain type of engine blade as the subject of the process test, Figure 3 This is a schematic diagram of a three-dimensional model of an aero-engine blade, illustrating an exemplary embodiment of this application.

[0217] Furthermore, a coordinate measuring machine (CMM) can be used. The Hexagon model is an example; this equipment is equipped with a high-precision probe capable of accurately measuring the surface of blades with complex geometries. By directly contacting the probe with the blade surface, the surface profile of blades with complex geometries can be accurately obtained, thus providing reliable data support for subsequent point cloud registration and grinding trajectory planning.

[0218] Furthermore, the scanned data was imported into MATLAB software, and Gaussian filtering was used to remove noise. The filtered point cloud data was then registered using the ICP algorithm. Through registration, the geometric relationship and machining allowance between the theoretical model and the actual blade edge were obtained. Based on the RIG software platform, a reference grinding trajectory was generated based on the theoretical model.

[0219] Furthermore, to verify the feasibility of the proposed algorithm, it was validated using a FANUC LR-Mate-200iD robotic grinding experimental platform. This platform consists of a robot body, grinding wheels, and sensors. The robot's repeatability is ±0.01mm, its radius of motion is ≥700mm, and its wrist can handle a weight of ≥5kg. The roughing and fine grinding wheels are nylon wheels with grit sizes of P80 and P240, respectively.

[0220] Furthermore, in the comparative experiment, the desired contact force was set at 15N for blade polishing. The experimental results are as follows: Figure 4 (a) Figure 4 As shown in (b). The surface roughness results are as follows. Figure 5 As shown in (a)-(c). Figure 4 (a) is a schematic diagram illustrating the grinding force result of an exemplary embodiment of this application. Figure 4 (b) is a schematic diagram illustrating the grinding force result in yet another exemplary embodiment of this application. Figure 5 (a)-(c) are schematic diagrams illustrating blade roughness in an exemplary embodiment of this application.

[0221] It should be noted that the experimental results show that the contact force of the traditional control strategy fluctuates within the range of 13.5~16.5 N, which deviates significantly from the desired contact force of 15 N, resulting in relatively high roughness. After optimization by incorporating the proposed dynamic compensation algorithm, the overall contact force becomes more uniform with smaller fluctuations, fluctuating within the range of approximately 14.5~15.5 N, close to the desired contact force of 15 N. The steady-state contact force tracking error is reduced by approximately 66.7%, and the roughness is low and uniform, with an average surface roughness reduction of approximately 15.6%. Measurements using a coordinate measuring machine and a roughness tester show that the profile of the polished blade is within the tolerance zone (±0.1 mm), and both the edge profile accuracy and surface roughness meet the process requirements. Figure 6 As shown. Figure 6 This is a schematic diagram illustrating the blade edge shape detection results of an exemplary embodiment of this application. The two innermost blue splines in the figure represent the upper and lower contour limit tolerances of the edge, the green spline represents the actual contour of the edge, and the black spline represents the theoretical contour of the edge.

[0222] Understandably, to address the issue of insufficient adaptive adjustment capability of robots in compliant constant-force grinding during the grinding process of aero-engine blades due to complex time-varying nonlinear coupling and uncertain disturbances, this paper aims to improve grinding accuracy and stability by real-time calibration of the grinding path. Specifically, the research results of this paper are reflected in the following aspects: 1. By accurately extracting the 3D model and high-precision point cloud data of the aero-engine blades, a reference grinding trajectory was constructed, providing a precise path basis for the robot grinding task. 2. Through a fuzzy adaptive variable impedance control strategy, combining pose deviation and dynamic model, the interaction between the robot and the environment was optimized, improving the real-time adjustment capability of the grinding force. 3. For the force difference caused by position error, a deep belief network algorithm was used to effectively compensate for the steady-state force error, thereby ensuring high-precision force tracking during the grinding process.

[0223] Corresponding to the aforementioned embodiment of a dynamic compensation method for grinding force, this application also provides an embodiment of a dynamic compensation device for grinding force.

[0224] An embodiment of the dynamic grinding force compensation device disclosed in this application can be applied to a dynamic grinding force compensation device. 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 dynamic grinding force compensation device loading the corresponding computer program instructions from the non-volatile memory into memory and executing them. From a hardware perspective, the dynamic grinding force compensation device in the embodiment typically includes other hardware depending on its actual function, which will not be elaborated further.

[0225] Figure 7 This diagram illustrates a dynamic compensation device for grinding force, as shown in an exemplary embodiment of this application. Please refer to... Figure 7 The apparatus provided in this embodiment includes an acquisition module 710, a processing module 720, a calculation module 730, and a compensation module 740; wherein,

[0226] The acquisition module 710 is used to acquire the actual point cloud data of the target object to be polished;

[0227] The processing module 720 is used to construct a theoretical model of the aero-engine blade based on the grinding requirements, register the actual point cloud data with the theoretical model, and generate a reference grinding trajectory.

[0228] The calculation module 730 is used to input the reference grinding trajectory into a pre-trained variable impedance adaptive control model to calculate the desired grinding force; wherein, the desired grinding force represents the grinding force of the grinding robot under ideal conditions, and the grinding robot can obtain the aero-engine blade corresponding to the theoretical model by grinding with the desired grinding force.

[0229] The acquisition module 710 is also used to acquire the joint coordinates of the current grinding robot;

[0230] The processing module 720 is also used to input the desired grinding force and the joint coordinates into a pre-trained depth belief network to obtain the robot's force error compensation value.

[0231] The compensation module 740 is used to compensate the desired grinding force based on the force error compensation value to obtain an updated grinding force, and to control the robot to grind the aero-engine blade based on the updated grinding force.

[0232] 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.

[0233] This application also provides a dynamic grinding force compensation device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods provided in the first aspect of this application.

[0234] 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.

[0235] 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.

[0236] 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.

[0237] 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 method for dynamic compensation of grinding force, characterized in that, The method includes: Obtain the actual point cloud data of the target object to be polished; A theoretical model of the target object for grinding is constructed based on the grinding requirements. The actual point cloud data is registered with the theoretical model to generate a reference grinding trajectory. The reference grinding trajectory is input into a pre-built variable impedance adaptive control model to calculate the desired grinding force; wherein, the desired grinding force represents the grinding force required by the grinding robot to process the test workpiece to the grinding target under ideal conditions; The construction of the variable impedance adaptive control model includes: establishing an initial impedance model; removing the target term from the initial impedance model based on the steady-state tracking error of the grinding robot to obtain a corrected impedance model; calculating the difference between the grinding wheel position and the end effector position of the grinding robot; determining the correlation between the damping coefficient and the force tracking error based on the difference and the corrected impedance model; using the material properties of the grinding target object and the robot processing properties as inputs to the fuzzy controller to calculate the update rate in the correlation; updating the correlation based on the update rate to obtain the variable impedance adaptive control model; and obtaining the joint coordinates of the current grinding robot. The desired grinding force and the joint coordinates are input into a pre-trained deep belief network to obtain the robot's force error compensation value. The expected grinding force is compensated based on the force error compensation value to obtain an updated grinding force, and the robot is controlled to grind the aero-engine blades based on the updated grinding force. The step of inputting the reference grinding trajectory into a pre-built variable impedance adaptive control model to calculate the desired grinding force includes: The real-time grinding parameters during the grinding process are obtained, and the fuzzy parameters are calculated by combining them with the corrected impedance model. The update rate corresponding to the real-time grinding parameters is calculated based on fuzzy parameters; wherein, the blade stiffness and robot feed speed are fuzzified using a Gaussian membership function to obtain fuzzy parameters; the fuzzy parameters are processed based on fuzzy correspondence rules to obtain the adjustment method and adjustment value corresponding to the fuzzy parameters; the fuzzy correspondence rules are used to characterize the correlation between different combinations of the fuzzy parameters and the adjustment method and adjustment value; the adjustment value is defuzzified, and the update rate is calculated in combination with the adjustment method; The current damping coefficient is calculated in real time based on the update rate. The desired grinding force is calculated based on the updated damping coefficient.

2. The method according to claim 1, characterized in that, The real-time calculation of the current damping coefficient based on the update rate includes: Based on the force sensor on the robot, the actual force applied by the robot at the previous moment is obtained; Based on the robot's current position and the reference grinding trajectory, calculate the robot's expected force at the previous moment; Calculate the difference between the expected force at the previous moment and the actual force at the previous moment, and calculate the first intermediate value by combining the update rate and the damping coefficient at the previous moment; Calculate the sum of the first intermediate value and the adaptive factor at the previous time step, and weight the damping coefficient at the previous time step to obtain the second intermediate value; The current damping coefficient is calculated based on the second intermediate value and the position of the robot's grinding wheel.

3. The method according to claim 1, characterized in that, The construction process of the variable impedance adaptive control includes: Obtain the initial impedance model, and convert the initial impedance model into a transfer function form model based on the Laplace transform; By substituting the preset desired stiffness coefficient and the position information of the aero-engine blade into the transfer function form model, the variable impedance adaptive control adapted to the aero-engine blade is obtained.

4. The method according to claim 1, characterized in that, The step of inputting the desired grinding force and the joint coordinates into a pre-trained depth confidence network to obtain the robot's force error compensation value includes: The multilayer restricted Boltzmann machine based on the deep belief network analyzes the correlation between the desired grinding force, the joint coordinates, and the force error compensation value. The force error compensation value is output based on the output layer of the deep belief network.

5. The method according to claim 1, characterized in that, The step of registering the actual point cloud data with the theoretical model to generate a reference polishing trajectory includes: Based on the distance between each actual point cloud data and the data point in the theoretical model, determine the corresponding point of each actual point cloud data, and combine the actual point cloud data and the data point in the theoretical model into a neighboring point pair; When the distance from all actual point cloud data to the data points in the theoretical model is less than a preset threshold, the reference polishing trajectory is determined based on the theoretical model.

6. The method according to claim 1, characterized in that, The registration of the actual point cloud data with the theoretical model includes: A first point set is constructed based on the actual point cloud data, and a second point set is constructed based on the theoretical model. Traverse each point in the first point set, search for matching points in the second point set based on the improved KD tree, establish a correspondence, and complete the registration. The search for matching points in the second point set based on the improved KD tree includes: Determine the position coordinates of each point in the first point set and the second point set; Select the first target point that is not matched in the first point set, and determine the first position coordinates corresponding to the first target point; The search neighborhood is calculated based on the contour shape determined by the actual point cloud data; Within the search neighborhood, select the optimal second target point from the second point set and establish a correspondence.

7. The method according to claim 1, characterized in that, The step of inputting the reference grinding trajectory into a pre-built variable impedance adaptive control model to calculate the desired grinding force includes: Starting from the starting point of the reference polishing trajectory, determine the trajectory points to be calculated; The coordinate information of the trajectory point to be calculated is input into the variable impedance adaptive control model to calculate the expected grinding force corresponding to the trajectory point to be calculated. Determine the rate of change of the reference polishing trajectory within the first subsequent distance for the trajectory point to be calculated; The adaptive period is calculated based on the product of the rate of change and the reference period; The next sampling point is calculated based on the sum of the coordinates of the trajectory point to be calculated and the adaptive period. The next sampling point is used as the trajectory point to be calculated. The process returns to the step of inputting the coordinate information of the trajectory point to be calculated into the variable impedance adaptive control model until the trajectory point to be calculated is the end point of the reference grinding trajectory.

8. An offline compensation device for absolute positioning accuracy of an industrial robot, characterized in that, The device includes an acquisition module, a processing module, a calculation module, and a compensation module; wherein, The acquisition module is used to acquire the actual point cloud data of the target object being polished; The processing module is used to construct a theoretical model of the target object for grinding based on the grinding requirements, register the actual point cloud data with the theoretical model, and generate a reference grinding trajectory. The calculation module is used to input the reference grinding trajectory into a pre-constructed variable impedance adaptive control model to calculate the desired grinding force; wherein, the desired grinding force represents the grinding force required by the grinding robot to process the test workpiece to the grinding target under ideal conditions; The construction of the variable impedance adaptive control model includes: establishing an initial impedance model; removing the target term from the initial impedance model based on the steady-state tracking error of the grinding robot to obtain a corrected impedance model; calculating the difference between the grinding wheel position and the end effector position of the grinding robot; determining the correlation between the damping coefficient and the force tracking error based on the difference and the corrected impedance model; using the material properties of the grinding target object and the robot processing properties as inputs to the fuzzy controller to calculate the update rate in the correlation; and updating the correlation based on the update rate to obtain the variable impedance adaptive control model. The acquisition module is also used to acquire the joint coordinates of the current grinding robot; The processing module is also used to input the desired grinding force and the joint coordinates into a pre-trained depth belief network to obtain the robot's force error compensation value. The compensation module is used to compensate the desired grinding force based on the force error compensation value to obtain an updated grinding force, and to control the robot to grind the aero-engine blades based on the updated grinding force. The step of inputting the reference grinding trajectory into a pre-built variable impedance adaptive control model to calculate the desired grinding force includes: The real-time grinding parameters during the grinding process are obtained, and the fuzzy parameters are calculated by combining them with the corrected impedance model. The update rate corresponding to the real-time grinding parameters is calculated based on fuzzy parameters; wherein, the blade stiffness and robot feed speed are fuzzified using a Gaussian membership function to obtain fuzzy parameters; the fuzzy parameters are processed based on fuzzy correspondence rules to obtain the adjustment method and adjustment value corresponding to the fuzzy parameters; the fuzzy correspondence rules are used to characterize the correlation between different combinations of the fuzzy parameters and the adjustment method and adjustment value; the adjustment value is defuzzified, and the update rate is calculated in combination with the adjustment method; The current damping coefficient is calculated in real time based on the update rate. The desired grinding force is calculated based on the updated damping coefficient.

Images