Intelligent optimization design method for constant-force gripper of robot

Abstract

According to the intelligent optimization design method for the constant-force gripper of the robot provided by the invention, the black-wing algorithm is improved, and the BP neural network algorithm is combined, so that the parameters of the whole gripper and key components are subjected to multi-dimensional optimization, and the comprehensive performance of the gripper can be remarkably improved. Finite element analysis is carried out based on an optimization result, an optimal design scheme meeting industrial requirements can be quickly generated in a virtual environment, a theoretical basis and practical guidance can be provided for production of similar products with light weight and high strength indexes in cooperation with optimization effect experimental verification of an entity prototype, the design time and cost are greatly shortened, and the design efficiency is improved. And the repeated experiment process in the traditional design is also avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot gripper design, and particularly relates to an intelligent optimization design method for a robot constant-force gripper based on swarm intelligence algorithms, finite element analysis, neural network prediction, and response surface optimization. Background Art

[0002] As an important part of common robot end effectors, the design and performance of robot grippers play a decisive role in the efficiency and product quality when robots are applied in production. At present, the problems of insufficient gripping force and stability of existing robot grippers are relatively common, especially when facing objects with complex shapes or different materials, showing insufficient adaptability. In traditional gripper design methods, most rely on experience and trial-and-error methods, lacking systematic optimization of design parameters, which limits the performance improvement of grippers. Although some design methods introduce intelligent optimization algorithms, they have the disadvantages of being prone to falling into local optima and having a slow convergence speed, and it is difficult to efficiently solve multi-variable and multi-objective optimization problems. The lack of effective performance prediction and verification means in the design process leads to a high dependence on subsequent experimental verification for design parameter adjustment, increasing the development cost and cycle. Therefore, there is an urgent need in this field for an efficient and intelligent optimization design method to meet the diverse needs of industrial production by improving the comprehensive performance of robot grippers in terms of gripping force, stability, structural strength, and quality. Summary of the Invention

[0003] In view of this, aiming at the technical problems existing in this field, the present invention provides an intelligent optimization design method for a robot constant-force gripper, which specifically includes the following steps:

[0004] Step 1: Determine design variables and establish corresponding geometric models and force models according to the structure, size, and gripping force design objectives of the robot gripper; establish a corresponding objective function with the change of gripping force as the optimization objective, and define geometric constraints and value range constraint conditions for each design variable;

[0005] Step 2: Use the solved design variable combinations to execute the good point set algorithm to generate an initial population, and set the non-linear convergence factor and the population mutation strategy as the t-distribution, thereby completing the improvement and initialization of the Black Kite Algorithm (BKA);

[0006] Step 3: Execute the improved Black Kite Algorithm to iteratively calculate the optimal design variable combination that minimizes the objective function under the set constraint conditions;

[0007] Step 4: Based on the optimal overall machine design scheme of the gripper obtained in Step 3, use the central composite method to design the length, angle, and cross-section design parameters of the key components of the gripper, so as to obtain multiple groups of design schemes for the key component design parameters;

[0008] Step 5: Perform finite element analysis on the stress and deformation of multiple groups of design schemes, load actual working condition data and material data in the simulation environment, and simulate and extract the performance parameters of key components, including total mass, maximum total deformation, and maximum equivalent stress data;

[0009] Step 6: Establish a performance prediction model of the gripper based on the BP neural network. Use the design parameters of the key components determined in Step 4 as the model input, and the performance parameters determined in Step 5 as the model output. Establish training sets and test sets respectively to train the model and verify the training effect. Then establish a response surface model of the design parameters of the key components of the gripper, and thus determine the optimal combination of design parameters;

[0010] Step 7: Determine the optimal overall design scheme of the gripper based on the optimal design variable combination obtained in Step 3, and obtain the optimal design scheme of the key components based on the optimal design parameter combination obtained in Step 6. On this basis, build a virtual prototype of the gripper in the simulation environment and verify the rationality of its assembly and movement. Finally, select the corresponding materials to make the physical entity of the robot gripper, and verify its actual performance through experiments.

[0011] Furthermore, in Step 1, specifically for the parallelogram clamping structure and its motion mode, define parameters including the lengths of each connecting rod, the angles between the connecting rods, the clamping force related to the driving force of the motor screw, and the displacement of the connecting rod as the design variables to be optimized.

[0012] Furthermore, in Step 2, specifically execute the following good point set algorithm to generate the initial population:

[0013] For an i-population and a j-dimensional problem, calculate the optimal point set through the following formula:

[0014]

[0015] where, P j i is the optimal point set of the i-th population in the j-th dimension, where lb j and ub j are the lower and upper bounds of the j-th variable respectively, and mod is the remainder function;

[0016] Set the following adaptive selection probability p for adjusting the step size of the black-winged kite algorithm as the non-linear convergence factor:

[0017]

[0018] where, T is the maximum number of iterations, t is the current number of iterations, and k is used to adjust the exploration scale.

[0019] Further, the specific process of obtaining each performance parameter in step 5 includes:

[0020] 1) According to the multiple groups of design schemes and the corresponding key component design parameters generated in step 4, geometric models of the key components are respectively established in the simulation environment;

[0021] 2) Using triangular elements and selecting a suitable mesh density to perform mesh division processing on the geometric model; setting the material of the model;

[0022] 3) Applying the maximum clamping force that conforms to the actual working conditions to the model, and setting fixed supports and free movement boundaries for it to simulate the constraint conditions during actual operation;

[0023] 4) Extracting the performance parameters including total mass, maximum total deformation, and maximum equivalent stress data for each group of design schemes through finite element analysis;

[0024] 5) Using the data in the foregoing steps to establish a basic data set corresponding to the design parameters and performance parameters for subsequent analysis and BP neural network training.

[0025] Further, in step 6, a BP neural network with a 3-input 3-output structure including 5 hidden layers is specifically established, and its learning rate is specifically set to 0.01; the output of each layer is calculated using the following formula:

[0026]

[0027] where, b ij and w ij respectively represent the bias and weight of the neuron, and both i and j are layer numbers;

[0028] Then calculate the output and the total error E between the true value y i is:

[0029]

[0030] After that, the gradient descent method is used to minimize the weight w ij :

[0031]

[0032] where, η is the learning rate;

[0033] The model with the minimum error between the calculated value and the true value is obtained by adjusting the bias and weight.

[0034] Further, in step 7, a response surface in the following second-order polynomial form is specifically designed:

[0035]

[0036] Among them, Y is the response variable, X i is the independent variable, β0 is the constant term, β i is the coefficient of the first-order term, β ii is the coefficient of the quadratic term of the independent variable X i is the coefficient of the quadratic term of the independent variable X ij is the coefficient of the cross-term of the independent variables X i and X j and ε is the error term;

[0037] Based on the above formula, the relationships between the design parameters of the key components including length, angle, and cross-section and the performance parameters including total mass, maximum total deformation, and maximum equivalent stress data are plotted respectively, and the optimal design parameter combination that optimizes the performance is selected from them.

[0038] Furthermore, in step 7, a 3D printing method is specifically used to fabricate the physical entity of the robot gripper and perform corresponding verification experiments.

[0039] The intelligent optimization design method of the robot constant-force gripper provided by the present invention above, through improving the black-winged kite algorithm and combining it with the BP neural network algorithm, performs multi-dimensional optimization on the parameters of the overall gripper and key components, and can significantly improve the comprehensive performance of the gripper. Based on the optimization results, finite element analysis can quickly generate the optimal design scheme that meets industrial requirements in a virtual environment. Combining with the optimization effect experiment verification of the physical prototype, it can provide a theoretical basis and practical guidance for the production of similar products with lightweight and high-strength indicators, not only greatly reducing the design time and cost, but also avoiding the repeated experimental process in traditional design. Description of the Drawings

[0040] Figure 1 is a schematic flow chart of the method provided by the present invention;

[0041] Figure 2 is an optimization process diagram of the improved black-winged kite algorithm in the example of the present invention;

[0042] Figure 3 is a design parameter optimization process diagram based on finite element analysis and BP neural network in the example of the present invention;

[0043] Figure 4 is a structural diagram of the virtual prototype of the gripper built by simulation. Detailed Embodiments

[0044] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] The intelligent optimization design method of the robot constant force gripper provided by the present invention is as follows Figure 1 shown, and specifically includes the following steps:

[0046] Step 1: Determine the design variables and establish the corresponding geometric model and force model according to the structure, size and clamping force design objectives of the robot gripper; establish the corresponding objective function with the change of clamping force as the optimization objective, and define the geometric constraints and value range constraints of each design variable;

[0047] Step 2: Use the solved design variable combination to execute the good point set algorithm to generate the initial population, and set the non-linear convergence factor and the population mutation strategy as the t-distribution, thereby improving and initializing the black-winged kite algorithm;

[0048] Step 3: Execute the improved black-winged kite algorithm, and iteratively calculate the optimal design variable combination that minimizes the objective function under the set constraints;

[0049] Step 4: Based on the optimal overall design scheme of the gripper obtained in Step 3, design the length, angle and cross-section design parameters of the key components of the gripper by the central composite method, so as to obtain multiple groups of design schemes for the key component design parameters;

[0050] Step 5: Conduct finite element analysis on the stress and deformation of multiple groups of design schemes, load the actual working condition data and material data in the simulation environment, and simulate and extract the performance parameters of the key components including the total mass, maximum total deformation and maximum equivalent stress data;

[0051] Step 6: Establish a performance prediction model of the gripper based on the BP neural network. Use the design parameters of each key component determined in Step 4 as the model input, and use the performance parameters determined in Step 5 as the model output. Establish the training set and test set respectively to train the model and verify the training effect, and then establish the response surface model of the key component design parameters of the gripper, thereby determining the optimal design parameter combination;

[0052] Step 7: Determine the optimal overall design scheme of the gripper based on the optimal design variable combination obtained in Step 3, and obtain the optimal key component design scheme based on the optimal design parameter combination obtained in Step 6. On this basis, build a virtual prototype of the gripper in the simulation environment and verify the rationality of its assembly and movement. Finally, select the corresponding materials to make the physical entity of the robot gripper, and verify its actual performance through experiments.

[0053] In a preferred embodiment of the present invention, specifically for the parallelogram clamping structure and its motion mode in step 1, parameters including the lengths of the connecting rods, the angles between the connecting rods, the clamping force related to the driving force of the motor screw, and the displacements of the connecting rods are defined as the design variables to be optimized. As Figure 2 shown, Fk represents the clamping force when the gripper is working; a, b, c, δ represent the link structure parameters of the gripper; e, f, l represent the position structure parameters of the gripper, and these parameters are all design variables to be optimized for the gripper; z, α, β represent the motion state at a certain moment. At any moment, the length from point A to point C is denoted as g, and the included angle ∠CAD is Each parameter can be calculated and expressed as:

[0054]

[0055] When the driving force of the motor screw is P, by calculating the static balance between the force passing through the connecting rod a and the clamping force Fk, Fk can be calculated:

[0056]

[0057] Considering the displacement y of the clamp, this displacement is related to the opening and closing angle of the connecting rod c and the position parameters e, f. y can be calculated as follows:

[0058]

[0059] To achieve the parallel grasping ability of the robot gripper, a fixed point C' is added to the designed parallelogram mechanism. The distance between point C and point C' is L1, and the horizontal angle of CC' is A1. Δy is the opening and closing distance of the grasping, and Δx is the height difference during opening and closing.

[0060] With the goal of minimizing the change in the clamping force, the key parameters are solved. These key parameters include structural variables such as the lengths and positions of the connecting rods, and performance indicators such as the clamping force, stability, and efficiency are established to evaluate the effectiveness of the parameters. The established design variables, objective function, and constraint conditions:

[0061] X = (x1, x2, x3, x4, x5, x6, x7) = (a, b, c, e, f, l, δ)

[0062]

[0063] The value ranges of each variable are limited: Ymin = 50, Ymax = 100, YG = 150, Zmax = 100. The boundaries of the design variables are: 20 ≤ a ≤ 100, 30 ≤ b ≤ 100, 100 ≤ c ≤ 200, 10 ≤ e ≤ 100, 10 ≤ f ≤ 30, 50 ≤ l ≤ 200, π / 2 ≤ δ ≤ π.

[0064] In step 2, the following good point set algorithm is specifically executed to generate the initial population:

[0065] For the problem of i populations and j dimensions, the optimal point set is calculated through the following formula:

[0066]

[0067] where P j i is the good point set of the i-th population in the j-th dimension, where lb j and ub j are the lower bound and upper bound of the j-th variable respectively, and mod is the remainder function;

[0068] Set the following adaptive selection probability p for adjusting the step size of the black-winged kite algorithm as the non-linear convergence factor:

[0069]

[0070] where T is the maximum number of iterations, t is the current number of iterations, and k is used to adjust the exploration scale. The non-linear convergence factor can better control the step size or perturbation size of the algorithm during the search process, so as to achieve a balanced relationship between exploration and exploitation. In this embodiment, k = 0.8 is taken, so that the algorithm can conduct large-scale exploration in the early stage and concentrate on developing near the excellent solutions in the later stage, thereby increasing the probability of finding the global optimal solution.

[0071] To verify the improvement effect of the algorithm, good results have been obtained by attempting to calculate 23 test functions of CEC2017 with the improved black-winged kite algorithm, which also reflects the effectiveness of the improved algorithm of the present invention.

[0072] In step 3, the improved black-winged kite algorithm is used to iteratively calculate and obtain the theoretical optimal design parameters of the whole machine structure. To ensure the accuracy of the optimization calculation and the effectiveness of the improved BKA algorithm, a variety of heuristic optimization algorithms are adopted to solve the optimization problem of the gripper model. Other algorithms include the dung beetle optimization (DBO) algorithm, the Harris hawk optimization (HHO) algorithm, and the grey wolf optimization (GWO) algorithm. For the continuous design of the robot fixture, the obtained design variables are normalized to X = (a, b, c, e, f, l, d) = (100, 85, 120, 10, 25, 100, 105).

[0073] In step 4, the central composite design method is used to design the structural scheme of the key components. According to the actual design requirements of the gripper, geometric parameters such as the link length, cross-sectional width, and structural angle of the key component parallelogram mechanism are selected as design variables, and the maximum equivalent stress, maximum deformation, and total mass are used as optimization objectives. In this embodiment, 15 groups of design samples are generated. The schematic diagrams of this step and subsequent steps are as Figure 3 shown.

[0074] The specific process of obtaining each performance parameter in step 5 includes:

[0075] 1) According to the multiple groups of design schemes and the corresponding key component design parameters generated in step 4, geometric models of the key components are established in the simulation environment respectively;

[0076] 2) Use triangular elements and select an appropriate mesh density to perform mesh division processing on the geometric model; set the material of the model;

[0077] 3) Apply the maximum clamping force that conforms to the actual working conditions to the model, and set fixed supports and free movement boundaries for it to simulate the constraint conditions during actual operation;

[0078] 4) Through finite element analysis, performance parameters including total mass, maximum total deformation, and maximum equivalent stress data of each group of design schemes are extracted;

[0079] 5) Use the data in the previous steps to establish a basic data set corresponding to design parameters and performance parameters for subsequent analysis and BP neural network training.

[0080] According to the design samples generated by the central composite design method, geometric models of the key components of the gripper are established, including main structure parameters such as the length of the connecting rod, cross-sectional width, and angle. Mesh division is performed on the geometric model, and triangular elements and mesh density are selected. Set the ABS material for the model, and apply the maximum clamping force of 50 N consistent with the actual working conditions. Set fixed supports and free movement boundaries to simulate the constraint conditions under the actual working state. Through finite element analysis, the total mass, maximum deformation, and maximum equivalent stress results of each group of samples are extracted. Organize the analysis results into a data table, including detailed data of each group of design parameters and corresponding performance indicators, providing a basic data set for the subsequent steps.

[0081] In step 6, a BP neural network with a 3-input 3-output structure containing 5 hidden layers is specifically established, and its learning rate is specifically set to 0.01; the output of each layer is calculated using the following formula:

[0082]

[0083] Among them, b ij and w ij respectively represent the bias and weight of the neuron, and both i and j are layer numbers;

[0084] Then calculate the output and the total error E between the true value y i is:

[0085]

[0086] Subsequently, the gradient descent method is used to minimize the weight w ij :

[0087]

[0088] where η is the learning rate.

[0089] By adjusting the bias and weight, a model with the minimum error between the calculated value and the true value is obtained.

[0090] In this embodiment, the root mean square error of the training result of the BP neural network is established for the dataset involved. The results show that the root mean square error of the training set is 0.001615, and the root mean square error of the test set is 0.0029328, indicating that the accuracy of the prediction model reaches a good level.

[0091] In step 6, the following response surface in the form of a second-order polynomial is specifically designed:

[0092]

[0093] where Y is the response variable, X i is the independent variable, β0 is the constant term, β i is the coefficient of the first-order term, β ii is the coefficient of the second-order term of the independent variable X i is the coefficient of the cross-term of the independent variables X ij and X i and X j , and ε is the error term;

[0094] Based on the above formula, the second-order response surface relationship between the design parameters of the key components including length, angle, and cross-section and the performance parameters including total mass, maximum total deformation, and maximum equivalent stress data is plotted respectively, and the optimal design parameter combination that optimizes the performance is selected from them. In this embodiment, the optimal parameter combination of length L1 = 31.535 mm, angle A1 = 74.783°, and cross-section radius R2 = 5.0026 mm is finally determined.

[0095] Step 7: Establish a virtual prototype of the robot gripper according to the optimal design scheme to verify the rationality of assembly and motion. Establish a virtual prototype as shown in Figure 4 according to the overall machine structure parameters and the optimal design scheme parameters of the key components obtained in steps 3 and 6.

[0096] In step 7, a physical entity of the robot gripper is specifically fabricated by 3D printing and corresponding verification experiments are performed. The physical gripper is driven by a motor to work, and the gripping force and deformation of the gripper under different loads and the gripping force during the opening and closing process are measured. Finally, the rationality and practicality of the present invention are effectively verified.

[0097] It should be understood that the sequence numbers of the steps in the embodiments of the present invention do not imply the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0098] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent optimization design method for a robot constant force gripper, characterized in that: Specifically, it includes the following steps: Step 1: According to the structure, size, and clamping force design objectives of the robot gripper, determine the design variables and establish the corresponding geometric model and force model; establish the corresponding objective function with the change of clamping force as the optimization objective, and define the geometric constraints and value range constraints of each design variable; Step 2: Use the solved design variable combination to execute the good point set algorithm to generate the initial population, and set the nonlinear convergence factor and the population mutation strategy as the t-distribution, thereby completing the improvement and initialization of the Black-winged Kite Algorithm (BKA); Step 3: Execute the improved Black-winged Kite Algorithm, and iteratively calculate the optimal design variable combination that minimizes the objective function under the set constraints; Step 4: Based on the optimal overall design scheme of the gripper obtained in Step 3, use the central composite method to design the length, angle, and cross-section design parameters of the key components of the gripper, so as to obtain multiple groups of design schemes for the key component design parameters; Step 5: Conduct finite element analysis on the stress and deformation of multiple groups of design schemes, load the actual working condition data and material data in the simulation environment, and simulate and extract the performance parameters of the key components including the total mass, maximum total deformation, and maximum equivalent stress data; Step 6: Establish a gripper performance prediction model based on the BP neural network. Take the design parameters of each key component determined in Step 4 as the model input, and take the performance parameters determined in Step 5 as the model output. Establish the training set and test set respectively to train the model and verify the training effect, and then establish the response surface model of the key component design parameters of the gripper, thereby determining the optimal design parameter combination; Step 7: Determine the optimal overall design scheme of the gripper based on the optimal design variable combination obtained in Step 3, and obtain the optimal key component design scheme based on the optimal design parameter combination obtained in Step 6. On this basis, build a virtual prototype of the gripper in the simulation environment and verify the rationality of its assembly and movement. Finally, select the corresponding materials to make the physical entity of the robot gripper, and verify its actual performance through experiments.

2. The method according to claim 1, wherein: In Step 1, specifically for the parallelogram clamping structure and its movement mode, define the parameters including the length of each connecting rod, the angle between the connecting rods, the clamping force related to the driving force of the motor lead screw, and the displacement of the connecting rod as the design variables to be optimized.

3. The method according to claim 1, characterized in that: In Step 2, specifically execute the following good point set algorithm to generate the initial population: For an i-population and j-dimensional problem, calculate the optimal point set through the following formula: Among them, P j i is the optimal point set of the $i$-th population in the $j$-th dimension, where lb j and ub j are the lower bound and upper bound of the $j$-th variable respectively, and mod is the remainder function; Set the following adaptive selection probability p for adjusting the step size of the Black-winged Kite Algorithm as the nonlinear convergence factor: where T is the maximum number of iterations, t is the current number of iterations, and k is used to adjust the exploration scale.

4. The method according to claim 1, wherein: The specific process of obtaining each performance parameter in Step 5 includes: 1) According to the multiple groups of design schemes generated in Step 4 and the corresponding key component design parameters, establish the geometric models of the key components in the simulation environment respectively; 2) Use triangular elements and select a suitable mesh density to perform mesh division processing on the geometric model; set the material of the model; 3) Apply the maximum clamping force that conforms to the actual working condition to the model, set its fixed support and free movement boundary to simulate the constraint conditions during actual work; 4) Extract the performance parameters of each group of design schemes, including the total mass, the maximum total deformation, and the maximum equivalent stress data, through finite element analysis; 5) Use the data in the foregoing steps to establish a basic data set corresponding to the design parameters and performance parameters for subsequent analysis and BP neural network training.

5. The method according to claim 1, characterized in that: In step 6, specifically establish a 3-input 3-output structure BP neural network with 5 hidden layers, and set its learning rate to 0.01; the output of each layer is calculated using the following formula: where b ij and w ij represent the bias and weight of the neuron respectively, and both i and j are layer numbers; Calculate the output again and the true value y i The total error E between them is as follows: After that, the gradient descent method is used to minimize the weight w ij : where η is the learning rate; Obtain a model with the smallest error between the calculated value and the true value by adjusting the bias and weights.

6. The method according to claim 1, characterized in that: In step 6, specifically design the following response surface in the form of a second-order polynomial: Among them, Y is the response variable, X i is the independent variable, β0 is the constant term, β i is the coefficient of the first-order term, β ii is the coefficient of the quadratic term of the independent variable X i is the coefficient of the quadratic term, β ij is the coefficient of the cross-term of the independent variables X i and X j ; ε is the error term Based on the above formula, respectively plot the relationships between the design parameters of the key components, including length, angle, and cross-section, and the performance parameters, including the total mass, the maximum total deformation, and the maximum equivalent stress data, and select the optimal design parameter combination that makes the performance reach the best.

7. The method according to claim 1, wherein: In step 7, specifically use 3D printing to fabricate the physical entity of the robot gripper and perform the corresponding verification experiment.

Images