A calibration method for industrial robots based on DH model and neural network

Abstract

The invention discloses an industrial robot calibration method based on a DH model and a neural network, comprising: controlling the industrial robot to run to a specified posture, measuring the actual posture reached by the industrial robot through a laser tracker, obtaining the theoretical posture that the industrial robot should reach through the DH model, and calculating the error before calibration; establishing a kinematic error model through the DH model, obtaining a model parameter error value, compensating the obtained error value back into the model parameter, and calculating the error between the posture after calibrating the DH model and the actual posture; using the six-axis angle of the industrial robot as the input of the neural network, the actual posture as the target value, and the posture after calibrating the DH model as a priori, obtaining a neural network model combining DH model calibration with neural network calibration, and using the neural network model to predict the posture of the industrial robot. The invention can comprehensively compensate for model errors and non-model errors, thereby greatly improving the working accuracy of the industrial robot.

Description

Technical Field

[0001] The present invention belongs to the field of computers and advanced manufacturing and automation, and specifically relates to an industrial robot calibration method based on a DH (Denavit-Hartenberg) model and a neural network, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Industrial robot calibration is an important research topic in the fields of computer and advanced manufacturing and automation. As a high-precision method that takes into account both accuracy and cost, it is currently a relatively reliable control method for improving the accuracy of industrial robots, with the advantages of low cost, large accuracy improvement, and convenience. Calibrated industrial robots have been widely used in various fields, such as aerospace equipment manufacturing, automobile manufacturing and parts processing, medical equipment manufacturing, intelligent assembly and transportation, etc.

[0003] In industrial robot systems, since industrial robots are usually composed of thousands of parts, and due to the differences in the manufacturing development levels of each part, the manufacturing accuracy and assembly process of important parts in the system have become the main factors restricting the improvement of the accuracy of industrial robot systems. The calibration technology of industrial robots can greatly affect its absolute positioning accuracy, making the position and posture of the calibrated industrial robot closer to the specified position and posture, thereby improving the working accuracy. Because there may be errors in the manufacturing process of the robot itself, which is inconsistent with the given theoretical value model, resulting in the difference between the measured actual value and the nominal value, the model error of the industrial robot can be compensated by model calibration, but the method of establishing a theoretical model of the robot and calculating the error with the actual work has limited improvement in accuracy. It may happen that the accuracy cannot be further improved after reaching a certain value. This is because in addition to the above-mentioned situation that restricts the accuracy of industrial robots, there are also non-model errors. Non-model errors are usually caused by external factors such as environment and load. The existing technology cannot compensate for the robot positioning error caused by these external factors. Summary of the invention

[0004] The purpose of the present invention is to provide an industrial robot calibration method, computer equipment, computer-readable storage medium and computer program product based on the DH model and neural network. By integrating the DH model calibration with the neural network calibration, the model error and non-model error can be comprehensively compensated, thereby greatly improving the working accuracy of the industrial robot.

[0005] One aspect of the present invention provides an industrial robot calibration method based on a DH model and a neural network, comprising:

[0006] Step S1, controlling the industrial robot to run to a specified posture, and measuring the actual posture reached by the industrial robot through a laser tracker, using the six-axis angle after the industrial robot reaches the specified posture and the actual posture measured by the laser tracker, establishing a DH model for the industrial robot, obtaining the theoretical posture that the industrial robot should reach through the DH model, and calculating the posture error between the theoretical posture and the actual posture in the same coordinate system as the error before calibration of the industrial robot;

[0007] Step S2, according to the posture error before calibration, establish the kinematic error model of the industrial robot through the DH model, obtain the error value of the kinematic error model parameter, compensate the obtained parameter error value back to the kinematic error model parameter, and calculate the error between the posture after calibrating the DH model and the actual posture measured by the laser tracker;

[0008] In step S3, the six-axis angle of the industrial robot is used as the input of the neural network, the actual posture measured by the laser tracker is used as the target value, and the posture after DH model calibration is used as the prior. Through iterative learning of the neural network, a neural network model combining DH model calibration and neural network calibration is obtained, and the neural network model is used to predict the posture of the industrial robot.

[0009] Preferably, in step S1, calculating the posture error between the theoretical posture and the actual posture in the same coordinate system includes: converting the actual posture in the virtual base coordinate system where the laser measuring instrument is located to the base coordinate system of the robot, obtaining the conversion relationship from the base coordinate system to the terminal coordinate system through the six-axis angle of the industrial robot after reaching the specified posture, and obtaining the theoretical posture that the industrial robot should reach, thereby obtaining the posture error in the same coordinate system.

[0010] Preferably, the transformation relationship T from the virtual base coordinate system to the terminal coordinate system is:

[0011]

[0012] Where Tr is the transformation matrix from the imaginary base coordinate system to the base coordinate system, is the transformation matrix from the base coordinate system to the one-axis coordinate system, is the conversion between one-axis coordinate system and two-axis coordinate system. is the transformation matrix between the two-axis and adjacent three-axis coordinate systems, is the transformation matrix between three axes and four axes, is the transformation matrix between the four-axis and the five-axis, is the transformation matrix between the five-axis and the six-axis, and T7 is the transformation matrix of the sixth-axis terminal coordinate system.

[0013] Preferably, in step S2, obtaining the kinematic error model parameter error value includes: using the actual robot posture measured by a laser tracker and the robot model posture obtained by a DH model, constructing an AX=B matrix, and using a singular value decomposition method to solve the matrix to obtain a parameter error value, wherein A is a partial derivative of the model parameter calculated by a Jacobian matrix, X is a parameter error, and B is an error between the model posture and the actual posture.

[0014] Preferably, the model parameters include 45 parameters related to the connecting rods, end flanges and virtual base coordinate systems of the industrial robot. The Jacobian matrix and the corresponding error of each posture are substituted into the matrix AX=B. Assuming that m measurements are performed, A is a (m×3, 45) matrix, X is a (45, 1) column vector, and B is a (m×3, 1) column vector. The matrix equation is solved by the singular value decomposition method to obtain the parameter error value.

[0015] Preferably, in step S3, the input data set is divided into a training set and a test set, the training set data is input into the neural network, the neural network learns the calibrated posture of the DH model, the predicted posture is calculated by forward propagation, and then the gradient is calculated and the weight is updated by back propagation. After thousands of iterative learning, the learned neural network model parameters are saved, and the test set data is imported into the neural network model to test the model effect, wherein the L2 norm is used as the loss function, and the Adam optimizer is used to update the neural network model weight parameters.

[0016] Preferably, in step S1, the end measurement points of the industrial robot are set as five target seats on the flange, the position relationship of each target seat on the end flange relative to the base coordinate system is obtained, the randomly generated posture is used as the specified posture, and the robot is controlled to run to the specified posture.

[0017] Another aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0018] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0019] Another aspect of the present invention provides a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0020] According to the industrial robot calibration method based on the DH model and neural network, computer equipment, computer-readable storage medium and computer program product of the above aspects of the present invention, by integrating the DH model calibration with the neural network calibration, it is possible to comprehensively compensate for the model error and the non-model error, thereby greatly improving the working accuracy of the industrial robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings used in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work:

[0022] Figure 1 is a flow chart of an industrial robot calibration method based on a DH model and a neural network according to an embodiment of the present invention;

[0023] Figure 2 is a calibration principle diagram of an industrial robot calibration method based on a DH model and a neural network according to an embodiment of the present invention;

[0024] Figure 3 4 is a structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The embodiment of the present invention provides an industrial robot calibration method based on the DH model and the neural network. Figure 1 As shown, the industrial robot calibration method based on the DH model and the neural network of the embodiment of the present invention includes steps S1 to S3. Figure 2 The schematic diagram of the embodiment of the present invention is used to describe each step in detail.

[0027] In step S1, a set of postures is randomly generated, the robot is controlled to run to the specified posture, the actual posture reached by the industrial robot is measured by the laser tracker, the actual posture is converted to the base coordinate system of the robot, and the theoretical position reached by the industrial robot is obtained by the DH model, and the position error of the industrial robot before calibration is calculated. Among them, the six-axis angle after the industrial robot reaches the specified posture and the actual posture measured by the laser tracker are used to establish the DH model for the industrial robot.

[0028] In this step, a set of poses is generated within the measurable range of the laser tracker. The poses are dispersed as much as possible within the measurement range to ensure that various poses that may occur in the working environment are measured. When collecting data, the measurement points are set to the five target seats on the end flange of the industrial robot. The base coordinate system of the industrial robot is established, and the positional relationship of the end coordinate system where each target seat on the end flange is located relative to the base coordinate system, as well as the conversion relationship from the virtual base coordinate system where the laser tracker is located to the base coordinate system, is obtained.

[0029] The actual position of the industrial robot measured by the laser tracker is converted to the position in the robot base coordinate system through rigid transformation (i.e. the process of converting the origin of the virtual base coordinate system where the laser measuring instrument is located to the origin of the industrial robot base coordinate system through translation and rotation operations), and the DH model is established for the robot through the six-axis angle returned by the industrial robot controller. i-1 That is, along the motion axis of joint i, X i It is along Z i-1 and Z i The common perpendicular line of the DH model only needs 4 parameters a i (Peripheral line a i It's Z i-1 and Z i The minimum distance between the two axes), α i (Z i-1 Axis and Z i The angle between the axes), d i (Two perpendicular lines a i-1 and a i The distance between i (X i-1 Axis and X i The angle between the axes) can determine the coordinate transformation relationship between two adjacent joints. The transformation matrix is ​​shown in the following formula (1):

[0030]

[0031] By multiplying all six axes according to the transformation relationship between adjacent links, we can obtain the transformation relationship T from the virtual base coordinate system to the terminal coordinate system, as shown in the following formula (2).

[0032]

[0033] Where Tr is the transformation matrix from the imaginary base coordinate system to the base coordinate system, is the transformation matrix from the base coordinate system to the one-axis coordinate system, is the conversion between one-axis coordinate system and two-axis coordinate system. is the transformation matrix between the two-axis and adjacent three-axis coordinate systems, is the transformation matrix between three axes and four axes, is the transformation matrix between the four-axis and the five-axis, is the transformation matrix between the five-axis and the six-axis, and T7 is the transformation matrix of the sixth-axis terminal coordinate system.

[0034] Usually, the robot terminal coordinate system is established on the target seat of the robot terminal flange. Because it is difficult to accurately establish the robot base coordinate system on the robot base, a virtual base coordinate system is established. The theoretical position and posture that the industrial robot should reach is obtained through the transformation relationship from the base coordinate system to the terminal coordinate system. The difference between the two positions is the error before the industrial robot calibration.

[0035] In step S2, after the DH model is established for the industrial robot, the Jacobian matrix of the model parameters and the model error based on the absolute position accuracy are obtained through the differential kinematic model of the DH model. Combined with the conversion error of the robot's virtual base coordinate system and the error of the end posture, the SVD (Singular Value Decomposition) algorithm is used to solve the model parameter error and compensate it to the model parameters to improve the absolute positioning accuracy of the industrial robot.

[0036] In this step, the DH model is preliminarily calibrated and the SVD method is used to solve the parameter error value of each position coordinate. The actual position and posture of the robot are obtained by the laser tracker, and the robot model position and posture are obtained by the DH model. The AX=B matrix is ​​constructed, and the SVD method is used to solve the above formula to obtain the parameter error value of each position coordinate, and then the error value is compensated to the model parameter, and the error between the calibrated position coordinate and the laser tracker measured coordinate is calculated.

[0037] For an industrial robot with N degrees of freedom, if there are errors in the structural parameters of each joint, the homogeneous coordinate transformations of adjacent joints all contain differential transformation terms. Then the actual posture of the end coordinate system relative to the robot base coordinate system is the transformed posture plus the error generated in the transformed posture.

[0038] The error at each pose can be expressed in matrix form:

[0039] Δ=M a Δa+M d Δd+M a Δα+M θ Δθ+M t Δt+M1Δl+M p Δp (3)

[0040] M a , M d , M α , Mθ , M t , M l , M p It is the Jacobian matrix corresponding to each link parameter, end flange, and imaginary base coordinate system, which describes the influence of each parameter change on the end position of the robot.

[0041] The above formula can also be written in the form of AX=B:

[0042] Where A(3*45) is the partial derivative of the model parameters calculated using the Jacobian matrix (the 45 parameters include M α1 ~M α6 is the partial derivative of the six torsion angle model parameters calculated by the Jacobian matrix, M d1 ~M d6 is the partial derivative of the model parameters of the six link distances calculated by the Jacobian matrix, M a1 ~M a6 is the partial derivative of the six connecting rod length model parameters, M θ1 ~M θ6 is the model parameter error of the six connecting rod angles, M tx , M ty , M tz are the partial derivatives of the three model parameters corresponding to the five target seats on the end flange, M l (M &x , M &y , M &z ) is the position partial derivative of the reference imaginary base coordinate system, M p (M dx , M dy , M dz ) are the attitude partial derivatives of the three reference imaginary base coordinate systems), X is the column vector with the unknown parameter (45*1) (where α1~α6 are the torsion angle errors, that is, the errors caused by the angles between two adjacent motion axes, d1~d6 are the distance errors between two adjacent connecting rods, a1~a6 are the connecting rod length parameter errors, θ1~θ6 are the connecting rod angle parameter errors, tx, ty, tz are the errors of T7, which are the errors caused by the five target seats on the end flange during the manufacturing and assembly process, &x, &y, &z are the position errors of the reference imaginary base coordinate system, dx, dy, dz are the attitude errors of the reference imaginary base coordinate system), and B is the errors Δx, Δy, Δz between the model and the actual posture.

[0043] Substitute the calculated Jacobian matrix of each posture and the error corresponding to the point into AX=B. Assuming that m measurement experiments are performed during the calibration process, A is a (m×3, 45) matrix, X is a (45, 1) column vector, and B is a (m×3, 1) column vector.

[0044] The system of equations is solved by the singular value decomposition (SVD) method to obtain the actual error X of the robot.

[0045] Let A∈R (m×3)×45 , then the singular value decomposition of A is:

[0046] A=US·V T (4)

[0047] where U∈C m×45 、VEC 45×45 AA T , A T A is a matrix of eigenvectors, and UU T =I m×m , W T =I 45×45 , S m×45 =diag(σ1,σ2,σ3,…,σ r ) are singular values ​​of A, and σ1>σ2>…>σ r , σ i >0 (i=1, 2,...r), r=rank(A).

[0048] Singular value decomposition is widely used in solving linear matrix equations such as Ax=B, especially when A is ill-conditioned. The singular value decomposition method is used to solve the matrix equation. The solution method is as follows:

[0049] X=(V·S + ·U T )Y (5)

[0050] in

[0051] The kinematic parameter error X is calculated using the above method. Based on the parameter calibration results, X can be compensated to the theoretical value of the kinematic model parameters, and the positioning error of each measured position of the robot can be recalculated, that is, the calibrated error.

[0052] In step S3, the DH model calibration and neural network fusion algorithm are used, the six-axis angle is used as the input of the neural network, the predicted posture is used as the output, and the posture after DH model calibration is used as the prior of the neural network. The multi-layer structure of the deep neural network (DNN) is used to learn the multi-level abstract features of the data, and the ReLU activation function is used to predict the complex nonlinear data. The two calibration methods are combined to achieve the purpose of improving the overall working accuracy and comprehensively compensate for the static error and dynamic error of the industrial robot.

[0053] In this step, the six-axis angle of the industrial robot is used as the input of the neural network, and the actual posture measured by the laser tracker is used as the target value. The multi-layer structure of the DNN allows it to learn the multi-level abstract features of the input data. The lower layer can capture the simple features of the original data, while the higher layer can combine and understand these simple features to form a more complex and abstract representation. In the process of forward propagation, the neural network integrates the prior with the physical model into the DNN by learning the posture calibrated by the DH model, further compensating for static errors and reducing the influence of non-model parameter errors. The deep neural network has multiple feature transformations of nonlinear mappings and can fit highly complex functions. The DNN neural network uses several weight coefficient matrices W, bias vectors b and input value vectors x to perform a series of linear operations and activation operations. The input layer consists of a group of perceptron neurons, which are used to connect variables from the external world with the internal neural network. The number of initial neurons is the size of the input data, and the number of output neurons is 64. It will directly transmit the data of the test set to the hidden layer neurons. The ReLU activation function is used between the output of the upper-layer nodes and the input of the lower-layer nodes to learn the complex nonlinear relationship between the input data and the prediction target. The features of the activated neurons are retained and mapped through the function, which is responsible for mapping the input of the neurons to the output.

[0054] Before performing the DNN back-propagation algorithm, you need to select a loss function to measure the loss between the output calculated by the training sample and the actual training sample output. In order to focus on the algorithm, the L2 norm is used as the loss function to measure the loss. That is, for each sample, the following formula is expected to be minimized, y i represents the target value, f(x i ) represents the predicted value:

[0055]

[0056] The gradient descent method is used to iteratively solve w and b of each layer. Gradient descent is mainly used to update weights in the neural network model, that is, to update and adjust the parameters of the model in one direction to minimize the loss function. The embodiment of the present invention uses the Adam algorithm, that is, the adaptive moment estimation method, which can calculate the adaptive learning rate of each parameter according to the first and second order moments of the gradient. This helps to speed up the convergence speed and improve the generalization ability of the model, and maintains the exponential decay average of the previous gradient.

[0057] The data set is divided into a training set and a test set. The model is trained using the data from the training set, and then the test set data is used to verify the model effect. The learning goal of the training set is the posture measured by the laser tracker. The posture is predicted through forward propagation calculation, and then the gradient is calculated and the weight is updated through back propagation. After five thousand iterative learning, the learned neural network model parameters are saved, and the test set data is imported into the model to test the model effect. The hyperparameters are continuously adjusted, the learning rate is initialized, and the model performance is optimized.

[0058] The trained model parameters are saved and their performance is verified in the test set. In the prediction process, the DH model compensation is used as a priori through the learning of the neural network and the data iteration of the two hidden layers. Finally, a set of compensated poses are predicted and the predicted poses are used as outputs. Finally, the pose error after neural network compensation is calculated.

[0059] In summary, according to the industrial robot calibration method based on the DH model and neural network in the above embodiment of the present invention, a DH model is established for the industrial robot using two sets of data: the six-axis angle after the industrial robot reaches the specified posture and the actual arrival posture measured by the laser tracker; the partial derivatives of each model parameter are calculated through the Jacobian matrix, the partial derivatives are substituted into the matrix equation to solve the model parameter error, and then the error is compensated back to the DH model parameter to calculate the calibrated posture error; then the six-axis angle is used as the input of the DNN neural network, the laser tracker data is used as the target value, and the posture after DH model calibration is used as the prior, and the neural network is iterated through thousands of times to learn, so as to obtain a neural network model combining the two calibration methods. Through this model, a more accurate posture can be predicted to meet the high-precision working requirements of the industrial robot.

[0060] The industrial robot calibration method based on the DH model and neural network in the above-mentioned embodiment of the present invention is a method for comprehensive calibration of different error sources of dynamic error and static error. It can not only realize the accurate compensation of the dynamic error generated in the manufacturing and assembly process of the industrial robot itself, but also in the case of unknown non-model error sources, it can predict the robot error only through the input value and target value and the data after DH model calibration, thereby calibrating the static error under the influence of environment, load, etc., and can provide effective calibration technical support for the complex tasks and high-precision work requirements of industrial robots in intelligent manufacturing.

[0061] An embodiment of the present invention further provides a computer device, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 3As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store operating parameter data of each framework. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps of the method of the embodiment of the present invention are implemented.

[0062] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0063] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method of the embodiment of the present invention are implemented.

[0064] An embodiment of the present invention further provides a computer program product, including a computer program, which implements the steps of the method of the embodiment of the present invention when executed by a processor.

[0065] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An industrial robot calibration method based on DH model and neural network, characterized in that: include: Step S1, controlling the industrial robot to run to a specified posture, and measuring the actual posture reached by the industrial robot through a laser tracker, using the six-axis angle after the industrial robot reaches the specified posture and the actual posture measured by the laser tracker, establishing a DH model for the industrial robot, obtaining the theoretical posture that the industrial robot should reach through the DH model, and calculating the posture error between the theoretical posture and the actual posture in the same coordinate system as the error before calibration of the industrial robot; Step S2, according to the posture error before calibration, establish the kinematic error model of the industrial robot through the DH model, obtain the error value of the kinematic error model parameter, compensate the obtained parameter error value back to the kinematic error model parameter, and calculate the error between the posture after calibrating the DH model and the actual posture measured by the laser tracker; Step S3, taking the six-axis angle of the industrial robot as the input of the neural network, the actual posture measured by the laser tracker as the target value, and the posture after DH model calibration as the prior, through iterative learning of the neural network, a neural network model combining DH model calibration and neural network calibration is obtained, and the posture of the industrial robot is predicted using the neural network model; Wherein, in step S2, obtaining the kinematic error model parameter error value includes: constructing an AX=B matrix using the actual robot posture measured by a laser tracker and the robot model posture obtained by a DH model, and using a singular value decomposition method to solve the matrix to obtain a parameter error value, wherein A is a partial derivative of a model parameter calculated by a Jacobian matrix, X is a parameter error, and B is an error between a model posture and an actual posture; The model parameters include 45 parameters related to the links, end flanges, and virtual base coordinate systems of the industrial robot. The Jacobian matrix and the corresponding error of each posture are substituted into the matrix AX=B. Assuming that m measurements are performed, A is a (m×3, 45) matrix, X is a (45, 1) column vector, and B is a (m×3, 1) column vector. The matrix equation is solved by the singular value decomposition method to obtain the parameter error value.

2. The method according to claim 1, characterized in that In step S1, the posture error between the theoretical posture and the actual posture in the same coordinate system is calculated, including: converting the actual posture in the virtual base coordinate system where the laser measuring instrument is located to the base coordinate system of the robot, obtaining the conversion relationship from the base coordinate system to the terminal coordinate system through the six-axis angle of the industrial robot after reaching the specified posture, and obtaining the theoretical posture that the industrial robot should reach, thereby obtaining the posture error in the same coordinate system.

3. The method according to claim 2, characterized in that The transformation relationship T from the virtual base coordinate system to the terminal coordinate system is: Where Tr is the transformation matrix from the imaginary base coordinate system to the base coordinate system, is the transformation matrix from the base coordinate system to the one-axis coordinate system, is the conversion between one-axis coordinate system and two-axis coordinate system. is the transformation matrix between the two-axis and adjacent three-axis coordinate systems, is the transformation matrix between three axes and four axes, is the transformation matrix between the four-axis and the five-axis, is the transformation matrix between the five-axis and the six-axis, and T7 is the transformation matrix of the sixth-axis terminal coordinate system.

4. The method according to claim 1 or 2, characterized in that: In step S3, the input data set is divided into a training set and a test set. The training set data is input into the neural network. The neural network learns the calibrated posture of the DH model, calculates the predicted posture through forward propagation, and then calculates the gradient and updates the weight through back propagation. After thousands of iterative learning, the learned neural network model parameters are saved, and the test set data is imported into the neural network model to test the model effect. The L2 norm is used as the loss function, and the Adam optimizer is used to update the neural network model weight parameters.

5. The method according to claim 1 or 2, characterized in that: In step S1, the end measurement points of the industrial robot are set as five target seats on the flange, the position relationship of each target seat on the end flange relative to the base coordinate system is obtained, the randomly generated posture is used as the specified posture, and the robot is controlled to run to the specified posture.

6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images