A method and apparatus for compensating for absolute positioning accuracy of an industrial robot

Abstract

The application provides an industrial robot absolute positioning accuracy compensation method and device. The method provided by the application comprises the following steps: determining multiple targets for training of an accuracy compensation model, wherein the multiple targets at least comprise a sub-target of a training dimension and a sub-target of a compensation accuracy dimension; determining a to-be-trained original parameter set of a deep belief network; calculating an influence degree of each parameter in the original parameter set on the multiple targets; screening parameters with an influence degree higher than a preset threshold as candidate hyperparameters; obtaining current pose data of multiple sampling points of the robot based on a laser tracker; generating a position error based on pre-determined expected pose data and the current pose data of the robot; training the deep belief network to obtain a trained target deep belief network; calculating a contribution weight of each feature vector based on evidence theory; and generating a compensation result. The industrial robot absolute positioning accuracy compensation method provided by the application is used to improve the positioning accuracy during robot operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot manufacturing, and in particular to an industrial robot absolute positioning accuracy compensation method and device. BACKGROUND

[0002] With the rapid development of manufacturing industry, industrial robots play an increasingly important role in the field of automatic production and assembly due to their high efficiency, stability and flexibility. However, due to the complexity of industrial robot system, its absolute positioning accuracy is often affected by many factors, making it difficult to meet the needs of high-precision tasks in practical applications.

[0003] In order to improve the absolute positioning accuracy of industrial robots, a common method is the traditional manual calibration method which relies on the experience and skills of professional technicians, and adjusts the industrial robot manually to correct the errors of its position and attitude. However, the accuracy and consistency of manual calibration are difficult to guarantee, which affects the stability and reliability of the robot. Another method based on neural network is to fit the experimental semi-variogram function with the joint angle of the robot and the corresponding position error, and to modify the target position using semi-variogram parameters and RBF neural network. However, this method has low precision and large error in model processing. Therefore, there is an urgent need for a more efficient and accurate industrial robot absolute positioning accuracy compensation technology to overcome the shortcomings of existing technologies. SUMMARY

[0004] Therefore, the present application provides an industrial robot absolute positioning accuracy compensation method and device to improve the positioning accuracy of the robot during operation.

[0005] Specifically, the present application is realized by the following technical solutions:

[0006] The first aspect of the present application provides an industrial robot absolute positioning accuracy compensation method, which comprises:

[0007] determining multiple targets for training the accuracy compensation model, the multiple targets including at least a sub-target of training dimension and a sub-target of compensation accuracy dimension;

[0008] determining a set of original parameters to be trained by the deep belief network;

[0009] calculating the influence degree of each parameter in the set of original parameters on the multiple targets;

[0010] selecting parameters with an influence degree higher than a preset threshold as candidate hyperparameters;

[0011] obtaining current pose data of multiple sampling points of the robot based on a laser tracker;

[0012] generate a position error based on the predetermined expected pose data of the robot and the current pose data, input the expected pose data into a deep belief network, use the position error as the output of the deep belief network, and train the deep belief network;

[0013] optimize the candidate hyperparameters of the deep belief network based on a genetic algorithm while training the deep belief network, to obtain a trained target deep belief network;

[0014] obtain to-be-compensated data during robot operation;

[0015] extract feature vectors of the to-be-compensated data using the target deep belief network, and calculate contribution weights of each feature vector based on evidence theory;

[0016] weight process the feature vectors based on the contribution weights, to generate a compensation result.

[0017] The second aspect of the application provides an industrial robot absolute positioning accuracy compensation device, the device comprises an acquisition module, a training module, a calculation module and a processing module; wherein,

[0018] The processing module is configured to determine multiple targets for training of the accuracy compensation model, the multiple targets including at least a sub-target of a training dimension and a sub-target of a compensation accuracy dimension;

[0019] The processing module is further configured to determine a set of original parameters to be trained for the deep belief network;

[0020] The calculation module is configured to calculate the influence degree of each parameter in the set of original parameters on the multiple targets;

[0021] The processing module is further configured to select parameters with an influence degree higher than a preset threshold as candidate hyperparameters;

[0022] The acquisition module is configured to obtain current pose data of multiple sampling points of a robot based on a laser tracker;

[0023] The training module is configured to generate a position error based on predetermined expected pose data of the robot and the current pose data, input the expected pose data into a deep belief network, use the position error as the output of the deep belief network, and train the deep belief network;

[0024] The training module is further configured to optimize the candidate hyperparameters of the deep belief network based on a genetic algorithm while training the deep belief network, to obtain a trained target deep belief network;

[0025] The acquisition module is further configured to obtain to-be-compensated data during robot operation;

[0026] The computing module is further configured to extract feature vectors of the to-be-compensated data using the target deep belief network, and calculate contribution weights of each feature vector based on evidence theory;

[0027] The processing module is further configured to process the feature vectors based on the contribution weights to generate a compensation result.

[0028] The industrial robot absolute positioning accuracy compensation method and device provided in this application start from the robot's control results, that is, from the coordinate position after movement, ignoring the fluctuations of internal forces generated in the intermediate process. This effectively reduces the computational complexity and ensures the accuracy of the calculation. Firstly, by combining deep belief networks (DBNs) with genetic algorithms and applying evidence theory, the DBN uses a multi-layer restricted Boltzmann machine to extract deep features from the input data. The genetic algorithm further optimizes the hyperparameters of the DBN, ensuring optimal network model performance. Combined with evidence theory to assess prediction uncertainty, the organic integration of these three elements in different directions improves the accuracy and stability of the network output. Secondly, when optimizing the DBN, the genetic algorithm not only globally searches for the optimal hyperparameter combination but also uses a fitness function to screen individuals, ensuring that only high-performing individuals enter the next generation. By screening out the key factors that have the greatest impact on model performance, the genetic algorithm effectively avoids local optima, enabling the DBN to better adapt to complex industrial environments and improve the accuracy of prediction and compensation. This optimization strategy significantly reduces the complexity of model training while improving the efficiency and accuracy of compensation control. Thirdly, the application of evidence theory in deep belief networks goes beyond simply correcting the final prediction result; it penetrates the intermediate processes of the network. By precisely calculating the contribution weights of each feature vector, the network can more accurately capture key information and ignore irrelevant noise during feature extraction and pattern recognition. This intervention method actually improves the accuracy and robustness of the network itself, enabling deep belief networks to maintain stable prediction performance even in complex and ever-changing task environments. Thus, by training deep belief networks with expected pose data and position errors, the deep belief network learns the complex mapping relationship between robot pose and position error from the data, enabling accurate prediction. The robot's position error under arbitrary poses provides a reliable predictive model for subsequent compensation. By optimizing the hyperparameters of the deep belief network using a genetic algorithm, the mechanisms of natural selection and genetics are simulated. The optimal parameter combination is searched globally, effectively avoiding the problem of local optima and enabling more accurate prediction of the robot's position error under different poses. Based on evidence theory, the contribution weight of each feature vector is calculated. Evidence theory can handle uncertain information and assign reasonable weights to each feature vector, reflecting their importance in the compensation process. This ensures that each feature vector can reasonably influence the compensation result, guaranteeing the accuracy of positioning. In this way, through the organic combination of deep belief network, genetic algorithm, and evidence theory, accurate compensation results are obtained, achieving an improvement in the absolute positioning accuracy of industrial robots. Attached Figure Description

[0029] Figure 1Flow chart of the absolute positioning accuracy compensation method for industrial robots according to an embodiment of the present application;

[0030] Figure 2 Schematic diagram of the absolute positioning accuracy compensation process for industrial robots according to an exemplary embodiment of the present application;

[0031] Figure 3 Schematic diagram of the communication control mode according to an exemplary embodiment of the present application;

[0032] Figure 4 Schematic diagram of the data set division according to an exemplary embodiment of the present application;

[0033] Figure 5 Schematic diagram of the data set division when training the deep belief network alone according to an exemplary embodiment of the present application;

[0034] Figure 6 Schematic diagram of the fitness function curve according to an exemplary embodiment of the present application;

[0035] Figure 7 Error prediction result diagram according to an exemplary embodiment of the present application;

[0036] Figure 8 Error prediction result diagram according to another exemplary embodiment of the present application;

[0037] Figure 9 Error prediction result diagram according to still another exemplary embodiment of the present application;

[0038] Figure 10 Schematic diagram of the x-direction comparison according to an exemplary embodiment of the present application;

[0039] Figure 11 Schematic diagram of the y-direction comparison according to another exemplary embodiment of the present application;

[0040] Figure 12 Schematic diagram of the z-direction comparison according to still another exemplary embodiment of the present application;

[0041] Figure 13 Schematic diagram of the compensation effect according to an exemplary embodiment of the present application;

[0042] Figure 14 Schematic diagram of the absolute positioning accuracy compensation device for industrial robots according to an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0043] The exemplary embodiments will be described in detail herein with reference to several drawings. The following description is presented in connection with the described embodiments, it is to be understood that various changes in form and detail can be made. The following detailed description is exemplary in nature and is intended to provide a thorough description of the natural embodiments.

[0044] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0045] It is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0046] The following detailed description is exemplary in nature and is intended to provide a thorough description of the natural embodiments.

[0047] Figure 1 The flow chart of the industrial robot absolute positioning accuracy compensation method provided by the present application is shown in Figure 1. Please refer to Figure 1 The method provided by the present application can include:

[0048] S101, determine the multi-objective of precision compensation model training, the multi-objective at least includes the sub-objective of training dimension and the sub-objective of compensation accuracy dimension.

[0049] Specifically, the precision compensation model is a mathematical model for improving the positioning accuracy of the industrial robot. The actual positioning error of the robot is analyzed and compensated by the precision compensation model to reduce the positioning error and improve the positioning accuracy.

[0050] Further, the sub-objective of training dimension focuses on the precision compensation model training process itself. Through the sub-objective of training dimension, the improvement of model learning ability and the efficiency of optimization process can be represented.

[0051] Further, the sub-objective of compensation accuracy dimension is directly related to the compensation effect of the precision compensation model on the positioning error of the industrial robot.

[0052] In practical implementation, the specific application scenarios and processing accuracy requirements of the industrial robot are clearly defined, and a first set of candidate sub-objectives for the compensation accuracy dimension is determined accordingly. For example, in one embodiment, if the robot is used for precision assembly, it may be necessary to include "maximum positioning error not exceeding 0.1mm" as a sub-objective. Then, considering factors such as model training time and computational resource consumption, a second set of candidate sub-objectives for the training dimension is determined. For example, in one embodiment, sub-objectives such as "model training time not exceeding 24 hours" or "using no more than 10GB of GPU memory" can be set. Then, for each sub-objective in the first and second objective configurations, potential conflicts between them are analyzed, and based on the degree of repulsion between sub-objectives, multi-objective optimization methods (such as weighted summation, Pareto front, etc.) are used to select the optimal combination of sub-objectives, so that the model can reach an acceptable level in multiple dimensions such as compensation accuracy and training efficiency. Furthermore, based on actual needs and technical feasibility, the final multi-objective set is further expanded. Finally, based on the selected optimal combination of sub-objectives, the accuracy compensation model is trained and optimized to achieve the predetermined multi-objective requirements.

[0053] Furthermore, the accuracy compensation model can be defined as optimizing the training dimensions of 6 dimensions and the compensation accuracy dimensions of 9 parameters in these 6 dimensions, to ensure the optimization effect.

[0054] S102. Determine the original parameter set to be trained in the deep belief network.

[0055] Specifically, deep belief networks are composed of multiple restricted Boltzmann machines stacked together to extract deep features from data.

[0056] In a specific implementation, for example, in one embodiment, each layer of the deep belief network has its corresponding parameters, so the original parameter set to be trained may include 20 dimensions, and each dimension has 8 corresponding features.

[0057] S103. Calculate the degree of influence of each parameter in the original parameter set on the multi-objective.

[0058] In a specific implementation, the parameters in the original parameter set can be clustered according to their network layer, role type (such as weight or bias), and other attributes. For example, in an embodiment, all weights belonging to the connection between the first layer of hidden layer and the input layer can be clustered into one category, and all biases belonging to the first layer of hidden layer can be clustered into another category, and so on. Then for each dimension of parameters, one or more representative parameters can be selected to evaluate the influence of the parameters in this category on the multi-objective. For example, in an embodiment, other parameters can be fixed, and only the size of the weights in a certain category can be adjusted, and then the changes in compensation accuracy and training efficiency of the model can be evaluated. Then it can be determined which parameter categories have a high degree of influence on the multi-objective, and the parameters in these dimensions should be paid more attention to in the subsequent optimization process. Then for the parameter categories with high degree of influence, the specific influence of each parameter in this category on the multi-objective can be analyzed in detail. For example, in an embodiment, the value of each parameter can be adjusted one by one, and the change in model performance can be observed. Finally, according to the degree of influence of each parameter on the multi-objective, the parameters with the greatest influence can be selected for optimization. For example, in an embodiment, during the optimization process, optimization methods such as genetic algorithm, gradient descent, and the like can be used to continuously adjust the values of these parameters to improve the overall performance of the model.

[0059] S104, screening parameters with a degree of influence higher than a preset threshold as candidate hyperparameters.

[0060] Specifically, the specific value of the preset threshold is set according to actual needs, which is not limited in the embodiment.

[0061] In a specific implementation, the collected parameter influence degree data can be compared with the preset threshold. If the degree of influence of a certain parameter is higher than the threshold, it is considered as a candidate hyperparameter; otherwise, it is excluded, and the final candidate hyperparameters are as shown in Table 1, which is a candidate hyperparameter table provided by an exemplary embodiment of the present application:

[0062] Table 1

[0063]

[0064] S105, obtaining current pose data of multiple sampling points of the robot based on the laser tracker.

[0065] Specifically, the specific positions of the sampling points are determined according to actual needs, which are not limited in the embodiment.

[0066] In a specific implementation, the sampling points can be uniformly distributed in the workspace of the robot to fully reflect the performance of the robot in different positions and postures. For example, in an embodiment, the mechanical arm and the joint points of the robot can be determined as sampling points.

[0067] Further, the laser tracker can be selected as AT901-B of Leica, with an error of ±15 μm + 6 μm / m.

[0068] It should be noted that the error of the laser tracker increases with the increase of the distance. For example, in an embodiment, in combination with the above embodiment, the AT901-B uses an angular 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 range finder. The two lasers can work independently. The laser beams emitted by the lasers pass through the gimbal mirror and are directed to the target. The interferometer laser beam simultaneously serves as the collimation axis of the tracker. The reflected laser is measured by the double-axis position detector built in the tracker. The pulses generated by the position detector are processed by the processor of the tracker and fed back to the servo motor of the tracker to drive the motor to track the target mirror of the tracker in real time. Ultimately, the actual pose of the end effector of the robot is measured.

[0069] As an optional embodiment, before the step S101, the method further includes: determining the plurality of sampling points comprises: establishing a space environment model of the robot; determining a space working trajectory of the robot in the space environment model; determining a change curve of a change rate of the space working trajectory; and determining a plurality of sampling points according to a change in amplitude of the change curve, wherein the number of sampling points at a position with a large amplitude is less than the number of sampling points at a position with a small amplitude. Specifically, during the machining of the robot, if the machining parameters frequently change, the calculation frequency of the position compensation of the robot is relatively dense. If the machining of the robot does not change, the original calculated value of the position compensation can be used to remain unchanged. The variable-period sampling method can reduce unnecessary resampling and improve the efficiency of sampling. At the same time, the number of sampling points is increased in the area with frequent changes, thereby improving the accuracy of sampling. Specifically, in an interval where the amplitude change is less than a preset threshold, a target position in the change curve is determined at a fixed period, and a target sampling position on the space working trajectory is determined according to the value of the change rate of the target position. In an interval where the amplitude change is greater than the preset threshold, a product of the amplitude change rate and the fixed period is used as a corrected adaptive variable period value, and a target sampling position on the space working trajectory is determined according to the adaptive variable period value.

[0070] S106, generating a position error based on the pre-determined expected pose data of the robot and the current pose data, inputting the expected pose data into a deep confidence network, using the position error as an output of the deep confidence network, and training the deep confidence network.

[0071] Specifically, the desired pose data is pre-set by a control system or a programming environment of the industrial robot, and represents an ideal position and orientation that the robot should reach at each sampling point.

[0072] In a specific implementation, the desired pose data can include the same types of data as the current pose data, and can include X, Y, Z coordinate values of the robot in the workspace and rotation angles (e.g., pitch angle, yaw angle, and roll angle) around the three axes.

[0073] Further, the input layer of the deep belief network has 9 channels, which are the theoretical position coordinates of the robot and the angles of the corresponding six joints. The output layer of the deep belief network has 3 channels, which are the position errors of the robot. .

[0074] Further, the number of hidden layers of the deep belief network nodes is determined to be a maximum of 4 layers, the number of nodes in the hidden layer ranges from 10 to 101, the initial learning rate is 0.01, the momentum factor is 0.8, the activation function is Sigmoid, and the mean square error MSE is used as the loss function; the hyperparameters that need to be optimized by the network are the number of hidden layers of the DBN, the number of hidden nodes, the learning rate, the momentum factor, the number of RBM iterations, and the number of DBN fine-tuning iterations.

[0075] Further, the desired pose data is compared with the current pose data to calculate the position error of the robot at each sampling point.

[0076] In a specific implementation, for example, in an embodiment, the position error can be represented by the vector difference between the desired position and the actual position: that is, the position error includes three translation errors (ΔX, ΔY, ΔZ) and three rotation errors (ΔRoll, ΔPitch, ΔYaw).

[0077] Further, the calculated position error is used as the output target of the deep belief network, and the desired pose data is used as the input feature of the network, and the training of the deep belief network is completed through these two data.

[0078] A specific embodiment is given below to introduce the training process of the deep belief network in detail:

[0079] (1) Adjust the parameters of the deep belief network based on a layer-by-layer greedy algorithm to obtain initialized model parameters; wherein the training process of the layer-by-layer greedy algorithm is unsupervised pre-training.

[0080] Specifically, the greedy algorithm is a training method for gradually constructing a model by obtaining the current optimal parameters at each step, so as to achieve global optimization through local optimal selection step by step.

[0081] In a specific implementation, input data can be used as visible layer input, and weights and biases can be adjusted by using unsupervised learning algorithms such as contrastive divergence, so that the deep belief network can learn an effective representation of the data.

[0082] A specific embodiment is given below to describe the process of initializing model parameters in detail.

[0083] Step one: processing the expected pose data based on the visible layer of the first layer restricted Boltzmann machine, and outputting target corresponding data of the expected pose data through the hidden layer of the restricted Boltzmann machine.

[0084] Specifically, in the deep belief network, the visible layer of the restricted Boltzmann machine is used to receive external input data, and the restricted Boltzmann machine can have multiple layers. In a specific implementation, the expected pose data is input to the visible layer of the first layer restricted Boltzmann machine.

[0085] Further, the restricted Boltzmann machine is composed of a visible layer and a hidden layer The visible layer is used to receive training data, and in this paper, it is used to receive the end expected position in the robot coordinate system and the current robot joint angle; the input of the hidden layer is the output of the visible layer, which is used to extract features. There is a relationship of "no connection within the layer and full connection between the layers" between the neurons of the two layers.

[0086] Step two: adjusting the weights and biases of the visible layer based on the feedback of the target corresponding data, and obtaining the trained first layer restricted Boltzmann machine.

[0087] Specifically, each layer of the restricted Boltzmann machine includes a visible layer and a hidden layer, which processes the input data through the visible layer and completes the output through the hidden layer.

[0088] In a specific implementation, the expected pose data is processed by the restricted Boltzmann machine, the reconstructed data is output based on the activation state of the hidden layer, the reconstruction error is calculated by comparing the difference between the reconstructed data and the original input data.

[0089] Further, according to the reconstruction error, the weights and biases between the visible layer and the hidden layer are updated using algorithms such as contrastive divergence.

[0090] In a specific implementation, the target corresponding data is transmitted to the hidden layer, the state of the hidden layer neurons is sampled according to the activation probability of the hidden layer neurons, then the state of the visible layer is reconstructed using the sampled hidden layer state through weights and biases, for each visible layer neuron, the reconstruction probability is calculated, then the difference between the original visible layer input and the reconstructed visible layer output is compared, and the weights and biases are updated according to the reconstruction error using the gradient descent method or other optimization algorithms.

[0091] It can be understood that by minimizing the reconstruction error, the restricted Boltzmann machine can better capture the internal structure of the data, and determine accurate weights and biases.

[0092] Further, step two can be repeated multiple times until the training of the restricted Boltzmann machine converges or reaches a predetermined number of iterations, to ensure that the weights and biases are gradually adjusted to an optimal state.

[0093] Step three: processing the expected pose data based on the trained first layer of the restricted Boltzmann machine, taking the output data as the input of the second layer of the restricted Boltzmann machine, and repeating the step of obtaining the target corresponding data.

[0094] In specific implementation, the hidden layer activation state of the trained first layer of the restricted Boltzmann machine is taken as the visible layer input of the second layer of the restricted Boltzmann machine, to realize layer-by-layer data transmission.

[0095] Further, the training process in steps one and two is repeated for the second layer of the restricted Boltzmann machine. The hidden layer output of the second layer of the restricted Boltzmann machine is taken as the visible layer input of the next layer of the restricted Boltzmann machine, and the training process is continued.

[0096] Step four: after the number of layers of the trained restricted Boltzmann machine reaches a preset number of layers, it is determined that the model parameter initialization is completed.

[0097] Specifically, the specific number of the preset number of layers is set according to actual needs, which is not limited in the embodiment. In specific implementation, the number of layers can be determined according to the complexity of robot positioning.

[0098] The industrial robot absolute positioning accuracy compensation method provided in the embodiment can iteratively adjust the weights and biases between the visible layer and the hidden layer according to the reconstruction error from the hidden layer to the visible layer through the contrast divergence algorithm and the like. This unsupervised training process can minimize the reconstruction error, so that the restricted Boltzmann machine can better capture the internal structure of the data. At the same time, by stacking the restricted Boltzmann machines layer by layer, a deep neural network is constructed, so as to process more complex data relationships and patterns, and ensure the reliability and accuracy of the deep belief network.

[0099] (2) Fine-tuning the parameters of the deep belief network based on the BP algorithm to complete the parameter adjustment of the deep belief network; wherein the fine-tuning process of the BP algorithm is supervised fine-tuning.

[0100] Specifically, the BP algorithm is a backpropagation algorithm (BP algorithm for short), which is used to update the weights in the network by calculating the difference between the network output and the expected output and propagating back to the network, so that the output of the network is closer and closer to the expected output.

[0101] Further, before fine-tuning the BP algorithm, a labeled data set needs to be prepared, which can include expected pose data of an industrial robot and corresponding actual position error (where the actual position error is used as a label).

[0102] The following gives a specific embodiment to introduce the process of fine-tuning the BP algorithm in detail:

[0103] Step one: input the expected pose data into the deep belief network to obtain the predicted result of the output.

[0104] In specific implementation, the expected pose data is input into the deep belief network, and the deep belief network is processed layer by layer to obtain the possible pose data of the robot as the predicted result.

[0105] Step two: calculate the loss function based on the predicted result and the position error, and obtain the gradient parameters of each gradient of the loss function; each gradient corresponds to a layer of the restricted Boltzmann machine.

[0106] In specific implementation, the position error corresponding to the expected pose data is determined according to the label, and the position error and the predicted result are compared to calculate the loss function. For example, in an embodiment, the average of the sum of squares of the difference between the position error and the predicted result can be calculated as the loss function.

[0107] Further, each layer of the restricted Boltzmann machine corresponds to a layer of gradient, which can be calculated by the chain rule from the output layer to calculate the gradient of the loss function with respect to the weight of each layer of the restricted Boltzmann machine, and the direction and amplitude of the weight adjustment of each layer of the restricted Boltzmann machine are represented by these gradient parameters to update the network parameters.

[0108] It can be understood that since each layer of the restricted Boltzmann machine corresponds to a gradient, each gradient corresponds to a layer of the restricted Boltzmann machine.

[0109] Step three: backpropagate the gradient parameters in the deep belief network to fine-tune the parameters of the deep belief network.

[0110] In specific implementation, the calculated gradient parameters are backpropagated from the output layer to the network. In the backpropagation process, each layer of the restricted Boltzmann machine adjusts its weight according to the received gradient parameters to reduce the prediction error.

[0111] The absolute positioning accuracy compensation method for the industrial robot provided by the embodiment compares the prediction result of the model with the actual position error, then derives the loss function by using the chain rule, since each gradient corresponds to a layer of restricted Boltzmann machine in the deep belief network, so that the restricted Boltzmann machine that produces the error can be accurately located, providing accurate guidance for subsequent parameter fine-tuning, and ensuring the accuracy of model parameter adjustment.

[0112] S107, based on the genetic algorithm, the hyperparameters of the deep belief network are optimized to obtain a trained target deep belief network.

[0113] Specifically, the genetic algorithm searches for the optimal solution of the problem by simulating selection, inheritance, mutation, and crossover.

[0114] Further, the performance of the deep belief network is measured by the genetic algorithm to obtain a corresponding fitness value, and the adjustment of the hyperparameters of the deep belief network is realized by the fitness value.

[0115] A specific embodiment is given below to introduce in detail the process of optimizing the hyperparameters of the deep belief network based on the genetic algorithm:

[0116] (1) For the hyperparameters of the deep belief network, an optimization catalog and an initialization configuration are generated.

[0117] In specific implementation, the optimization catalog is generated according to the range of the hyperparameters, and the optimization catalog is the basis of the search space of the genetic algorithm.

[0118] Further, the initialization configuration contains multiple individuals, and each individual represents a combination of a group of hyperparameters. In specific implementation, the optimization catalog and the initialization configuration can be randomly generated.

[0119] (2) The mean square error of the output deviation of the deep belief network and the actual actual deviation is calculated to obtain the fitness function of the genetic algorithm.

[0120] Specifically, the deep belief network is used for prediction based on the optimization catalog and the initialization configuration to obtain the prediction result of the robot position, and the mean square error of the prediction result and the actual deviation of the actual position is calculated to obtain the mean square error as the fitness function.

[0121] A specific embodiment is given below to introduce in detail the process of obtaining the mean square error:

[0122] Step 1: input the expected pose data into the deep belief network to obtain the predicted output result.

[0123] In specific implementation, the expected pose data is input into the deep belief network, and the corresponding predicted output result is output by the deep belief network.

[0124] Step two: calculate the square of the difference between the predicted output result and the position error to obtain a target intermediate value.

[0125] Further, the predicted output result is subtracted from the position error, and the square of the difference is obtained to obtain the target intermediate value.

[0126] In specific implementation, the target intermediate value can be calculated based on the following formula:

[0127] E= ;

[0128] Wherein, E is the target intermediate value, is the predicted output result, is the position error.

[0129] Step three: based on the square sum of the target intermediate value, the output dimension of the deep belief network and the number of data contained in the expected pose data, calculate the mean square error.

[0130] In specific implementation, the mean square error can be calculated based on the following formula.

[0131] ;

[0132] Wherein, N is the number of data, m is the output dimension of the deep belief network, is the predicted output result, is the position error, is the mean square error.

[0133] (3) Based on the fitness function, sort the initial configuration and select part of it as the first target configuration.

[0134] Specifically, the initial configuration is sorted according to the fitness function value, and the part with higher fitness is selected as the first target configuration.

[0135] It can be understood that the individual corresponding to the first target configuration performs well in the genetic process and has high genetic value.

[0136] (4) Perform cross operation and compilation operation on the first target configuration to obtain a group of second target configurations, update the initial configuration based on the second target configuration to obtain an updated configuration.

[0137] Specifically, the cross operation is to randomly select two first target configurations and generate new individuals by exchanging part of their hyperparameter values.

[0138] Further, the offspring individuals obtained after the cross operation are subjected to mutation operation, that is, the value of one or more hyperparameters of the offspring individuals is randomly changed with a certain probability.

[0139] Further, the second target configuration obtained after the cross operation and the mutation operation is replaced with part of the individuals in the initial configuration to form an updated configuration, and the updated configuration is the new hyperparameter value.

[0140] (5) The initial configuration is replaced with the updated configuration, and the step of obtaining the updated configuration is re-executed for a preset number of times, and the hyperparameters of the deep belief network are optimized based on the obtained updated configuration.

[0141] In a specific implementation, the hyperparameters of the deep belief network are updated to the updated configuration, and the process of using the new deep belief network to update the hyperparameters is iterated multiple times until the hyperparameter value is stable.

[0142] The industrial robot absolute positioning accuracy compensation method provided in the embodiment can continuously update individuals in the population, continuously maintain excellent genes of the population, and gradually eliminate individuals with poor performance, so that the overall performance of the population is continuously improved in the iteration process, and the robustness and accuracy of the deep belief network are ensured.

[0143] S108, obtain the data to be compensated when the robot works.

[0144] In a specific implementation, the data to be compensated can be obtained according to the training data used when the deep belief network is trained.

[0145] Further, the specific data type of the data to be compensated is determined according to actual needs, which is not limited in the embodiment. For example, in an embodiment, the data to be compensated can include the position coordinates of each mechanical arm of the robot and the joint angles of each joint of the robot.

[0146] S109, using the target deep belief network to extract features of the data to be compensated to obtain a feature vector, and calculating a contribution weight of each feature vector based on evidence theory.

[0147] In a specific implementation, the data to be compensated is input into the target deep belief network, and the feature vector is extracted based on the restricted Boltzmann machine.

[0148] Further, the evidence theory is a mathematical tool for handling uncertainty, which can effectively fuse information from multiple sources and evaluate its uncertainty.

[0149] It should be noted that the evidence theory is to obtain the feature vector calculated by each layer of the restricted Boltzmann machine, and calculate the contribution weight thereof,

[0150] A specific embodiment is given below to introduce the generation process of the contribution weight in detail.

[0151] (1) Based on the evidence theory, the feature vector is processed to obtain multiple basic probability assignment functions corresponding to each feature vector.

[0152] Specifically, the basic probability assignment function reflects the degree to which the feature vector supports the prediction result.

[0153] In practice, evidence theory is used to process feature vectors, treating each feature vector as a source of evidence. Based on the value of the feature vector and its performance in the deep belief network, one or more basic probability assignment functions are assigned to it.

[0154] (2) For a given feature vector, fuse all the corresponding basic probability assignment functions.

[0155] In practice, the Dempster synthesis rule can be applied to consider the conflict and consistency between different basic probability assignment functions to obtain the fusion result corresponding to the feature vector.

[0156] (3) Generate the contribution weight based on the trust interval and the fused basic probability assignment function.

[0157] Specifically, the confidence interval is used to quantify the range of proposition confidence. The confidence interval represents the minimum and maximum confidence estimates for a correctly calculated proposition.

[0158] Furthermore, the trust interval can be a closed interval composed of the trust function (Bel) and the likelihood function (Pl), used to represent the likelihood of an event. level of support Bel(A) is the sum of the basic probability assignments of all subsets of A, representing the degree of trust in A, and satisfies the formula:

[0159] ;

[0160] Where Pl(A) is the sum of the basic probability assignments of all subsets intersecting with A, representing the degree of non-denial of A, and satisfies the formula:

[0161] ;

[0162] Where Pl(A) is the sum of the basic probability assignments of all subsets intersecting with A, representing the degree to which A is not denied.

[0163] Furthermore, let there be a finite nonempty set. To identify the frame, the function To identify the frame The basic probability assignment function.

[0164] It should be noted that Bel and Pl are used to represent the upper and lower bounds of the reliability of the positioning accuracy of industrial robots.

[0165] In a specific implementation, after the basic probability assignment function of all feature vectors is fused, the contribution weight of each feature vector is generated based on the trust interval and the fused basic probability assignment function.

[0166] Further, the contribution weight can be calculated based on the distribution of the fused basic probability assignment function within the trust interval, and the greater the weight value, the greater the contribution of the feature vector to the prediction result.

[0167] Further, assuming is the n feature vectors of the input sample, in the recognition framework , each feature can be regarded as an evidence corresponding to a subset in the recognition framework, and for each class , the evidence is considered to support or support , and the specific support depends on the weight coefficient of the evidence :

[0168] ;

[0169] wherein and are two weight parameters.

[0170] Further, let the evidence weight of and be positive part and negative part respectively, then for each and each class , there are two basic probability assignment functions:

[0171] ;

[0172] ;

[0173] wherein represents the support degree of to , and represents the support degree of to .

[0174] The industrial robot absolute positioning accuracy compensation method provided by the embodiment introduces evidence theory to process feature vectors and generate contribution weights, so that the entire error compensation system has higher robustness and adaptability when facing complex and variable environments. Meanwhile, the error compensation method based on the contribution weights can concentrate more computing resources and attention on feature vectors that have greater influence on the final prediction result, thereby improving the overall efficiency of the system and ensuring the accuracy of the deep belief network.

[0175] S110, based on the contribution weight, the feature vector is weighted and processed, and a compensation result is generated by combining the to-be-compensated data.

[0176] In a specific implementation, the contribution weight can be directly multiplied by the feature vector to obtain position information, and the difference between the position information and the to-be-compensated data is used to determine how to compensate the position, thereby obtaining the compensation result.

[0177] Figure 2 FIG. 1 shows an industrial robot absolute positioning accuracy compensation process according to an example embodiment of the present application. Figure 2 First, the multi-objective and original parameter set for training the accuracy compensation model are determined, and then candidate hyperparameters with an influence degree higher than a preset threshold are obtained through screening. The training of the deep belief network is realized by combining the candidate hyperparameters, and a trained target deep belief network is obtained. Further, the contribution weight of each feature vector is calculated based on the evidence theory, and the compensation result is generated based on the contribution weight.

[0178] A specific embodiment is given below to introduce the calculation process of the compensation result in detail.

[0179] (1) The position coordinates and joint angles in the to-be-compensated data are obtained, and the position coordinates and joint angles are input into the visible layer of the target deep belief network. Feature extraction is performed through the hidden layer of the target deep belief network, and the feature vector is obtained.

[0180] In a specific implementation, the position coordinates and joint angles of the robot are determined from the to-be-compensated data obtained by the industrial robot, and these data are input into the optimized target deep belief network. Deep feature extraction is performed through the visible layer and hidden layer of the target deep belief network, and the feature vector is obtained.

[0181] (2) The feature vector is multiplied by the corresponding contribution weight to obtain a weighted vector, and the compensation result of the robot in three directions is calculated. The compensation result is output by the target deep belief network.

[0182] In a specific implementation, after the feature vector is extracted, the contribution weight calculated by the evidence theory for each feature vector is multiplied by the corresponding feature vector to obtain a weighted vector.

[0183] It can be understood that, by multiplying the feature vector with the corresponding contribution weight, features that have greater impact on positioning accuracy can be emphasized, while features that have less impact can be weakened.

[0184] Further, after obtaining the weighted vectors, the weighted vectors are used to calculate the compensation results of the robot in three directions (X-axis, Y-axis, and Z-axis directions).

[0185] In specific implementation, the deep belief network calculates the predicted error of the robot in each direction according to the weighted vectors and pre-trained model parameters, and takes the error value as the amount of compensation.

[0186] The industrial robot absolute positioning accuracy compensation method provided in the embodiment has stronger robustness, because the deep learning model can automatically learn and adapt to data features in different environments, and the uncertainty processing method can reduce uncertainty in prediction.

[0187] A specific embodiment is given below to describe the process of robot position correction in detail.

[0188] (1) Determine a robot joint angle adjustment vector and a robot position coordinate adjustment vector based on the compensation result.

[0189] Specifically, the joint angle adjustment vector is used to indicate the angle size and direction that each joint needs to adjust, and the position coordinate adjustment vector directly indicates the distance and direction that the position of the robot in the three-dimensional space needs to adjust.

[0190] In specific implementation, the compensation result is analyzed in depth to determine the robot joint angle adjustment vector and the robot position coordinate adjustment vector of the robot in the X, Y, and Z directions.

[0191] (2) Correct the joint angle of the robot based on the robot joint angle adjustment vector, and correct the end effector position of the robot in the Cartesian coordinate system based on the robot joint angle adjustment vector.

[0192] In specific implementation, an adjustment instruction can be sent to the corresponding joint driver to make it adjust the joint angle according to the requirements of the adjustment vector.

[0193] Further, the end effector position of the robot can be corrected in the Cartesian coordinate system, and the end effector position of the robot is accurately controlled to ensure that the robot can accurately reach the predetermined work point.

[0194] The industrial robot absolute positioning accuracy compensation method provided by the embodiment reduces robot failure and downtime caused by positioning errors through high-precision positioning, reduces additional costs and time losses caused by failure maintenance, and ensures the accuracy of robot adjustment in combination with a robot joint angle adjustment vector and a robot position coordinate adjustment vector.

[0195] Corresponding to the foregoing embodiment of the industrial robot absolute positioning accuracy compensation method, the application also provides an experimental verification process of an industrial robot absolute positioning accuracy compensation method.

[0196] The experimental platform for industrial robot absolute positioning accuracy compensation is a KUKA KR6_R700 sixx_CR industrial robot with a 6 kg load capacity and a 700 mm radius work range, a 1.36 m3 work space volume, a position repeatability of ±0.03 mm, and a position absolute accuracy of ±0.6 mm. The laser tracker used to measure the position error is a Leica AT901-B, with an error of ±15 μm + 6 μm / m, and the error of the laser tracker increases with the increase of the distance.

[0197] The AT901-B measures the angle using an angle encoder and measures the distance using an absolute interferometer. The absolute interferometer in the AT901 integrates a helium-neon laser interferometer and an absolute range finder internally. The two lasers can work independently. The laser beams emitted by the lasers pass through the gimbal mirror and are shot at the target. The interferometer laser beam simultaneously serves as the collimation axis of the tracker. The reflected laser is measured by the double-axis position detector built into the tracker. The pulses generated by the position detector are processed by the tracker processor and fed back to the servo motor of the tracker to drive the motor to track the target mirror of the tracker in real time. Finally, the tracking and ranging are realized to measure the actual pose of the end effector of the industrial robot.

[0198] The control and communication schematic diagram of the error compensation platform is shown in Figure 3 . Figure 3 The communication control mode schematic diagram shown in the example embodiment of the application. The computer is used as the TwinCAT host, that is, the main controller of the control system; the TwinCAT host communicates with the industrial robot using the industrial Ethernet EtherCAT, and the control period is 12 ms; the laser tracker communicates with the TwinCAT host through the Ethernet (TCP / IP protocol), and the sampling period of the laser tracker can reach 1 ms.

[0199] The actual position of the mechanical arm can be obtained using the laser tracker, the error is calculated, and finally the error compensation function is realized using the deep belief network based on the genetic algorithm.

[0200] Under the premise of meeting the measurement requirements of the laser tracker, about 8000 groups of data were measured in the 240mmx240mmx200mm workspace of the target ball installed on the fixed tooling of the industrial robot. In order to make the experimental data have universality and randomness, the random number module drand in TwinCAT3 was used to randomly generate specific sampling data in the predetermined sampling space. In order to obtain the real pose of the robot and the laser tracker in the stable state, each sampling was divided into three steps. First, the robot arrived at the sampling point and stayed for 2 seconds, then the laser tracker recorded the data for 1 second; finally, in order to reset, it was delayed for 1 second. The theoretical position coordinates and joint angles in the robot were the input of the model, and the absolute position error of the robot end was the output of the model.

[0201] The data set was divided into a training set and a test set, and the division ratio was 0.3. The 8000 groups of collected data were divided into a 5600-group training set and a 2400-group test set. As shown in Figure 4 , Figure 4 The data set division diagram shown in an example embodiment of the application. Blue dots represent the training set, and red dots represent the test set.

[0202] Further, the input layer of the deep belief network has 9 channels, which are the theoretical position coordinates and joint angles of the robot ; the output layer of the deep belief network has 3 channels, which are the position errors of the robot , Figure 5 The data set division diagram when the deep belief network is trained alone, as shown in an example embodiment of the application. The number of hidden layers, the number of hidden layer nodes, the learning rate, the momentum factor, the RBM iteration number and the deep belief network fine-tuning iteration number of the deep belief network are determined by the genetic algorithm Figure 6 The fitness function curve diagram, as shown in an example embodiment of the application. The fitness of the DE optimized deep belief network decreases with the increase of the iteration number, from 0.387 to 0.152, a decrease of 60.7%; after 150 iterations of training, the optimal fitness iteration number can be found at the end of the 92nd time, and the optimal parameters of the deep belief network are output.

[0203] The number of hidden layers, the number of hidden layer nodes, the learning rate, the momentum factor, the RBM iteration number and the deep belief network fine-tuning iteration number of the deep belief network are determined by the genetic algorithm (GA). The fitness of each particle is calculated according to the fitness calculation condition, and the genetic (GA) algorithm iteration is terminated when the training error reaches the allowable value or the iteration number reaches the maximum value.

[0204] Finally, the deep belief network hyperparameters determined by the genetic algorithm are shown in Table 2.

[0205] Table 2

[0206]

[0207] Supervised learning in machine learning, essentially given a series of training samples, to establish a mapping relationship, so that the fitting result is as close as possible to the true output. Loss function is an important indicator to analyze the pros and cons of the training results, and there are many ways to define the loss function. This paper adopts five indicators of Mean squared error (MSE), Root mean squared error (RMSE), Mean absolute percentage error (MAPE), Mean absolute error (MAE) and R2 Score to analyze the deep belief network. The following gives the definition and explanation of the five indicators.

[0208] ;

[0209] ;

[0210] ;

[0211] ;

[0212] ;

[0213] In the above formula, represents the true value of the data set, represents the predicted value of the data set, represents the average value of the predicted value of the data set, represents the number of data sets, represents the variance of the data.

[0214] Mean squared error is the mean of the sum of squares of the corresponding point errors between predicted data and original data; Root mean squared error is the square root of mean squared error, also called fitting standard deviation of regression system; Mean absolute percentage error is often used to measure the accuracy of prediction, and mean absolute error refers to the average value of the absolute value of each measurement deviation, which can accurately reflect the size of the actual prediction error; The closer the four indicators are to 0, the closer the predicted value is to the true value, and the better the prediction effect.

[0215] R2 represents the determination coefficient of the model, the best score is 1.0, which means that the model perfectly predicts the true value; it can also be negative because the model can be arbitrarily worse, that is, there is no mapping fitting relationship between the predicted data and the true data. Table 3 shows the values of MSE, RMSE, MAPE, MAR and R2 of the deep belief network fitted for the robot end error.

[0216] Table 3

[0217]

[0218] Using the proposed industrial robot position error prediction model, the MSE, RMSE, MAPE and MAE of the predicted value of the position error of the robot end effector are close to 0. The determination coefficient R2 of the predicted value of the position error is shown in Table 3. Figures 7-11 Figure 7 The error prediction result graph shown in the present application is shown in the present application. Figure 8 The error prediction result graph shown in the present application is shown in the present application. Figure 9 The error prediction result graph shown in the present application is shown in the present application. It can be seen that the predicted value (blue dot) is closely distributed around the true value (red line), and R2 is close to 1, indicating that the predicted value has high correlation with the actual value and high fitting accuracy. Therefore, the proposed machine learning model has good adaptability and robustness for predicting the position error of the industrial robot.

[0219] The R2 of the robot end precision compensation error is about 0.87 to 0.95, and the overall effect is good. Since the three dimensions are trained and iterated together during the training of the deep belief network, the features of the three dimensions will affect and couple with each other; it is also disturbed by the non-linear factors such as the accuracy of data acquisition and environmental conditions, so that the robot end precision error compensation effect has certain differences.

[0220] In order to verify the effectiveness and improvement effect of the genetic algorithm optimized deep belief network proposed in this paper, 150 random verification points are selected in the robot motion space range of 240mmx240mmx200mm, Figure 10 The x-direction comparison schematic diagram shown in the present application is shown in the present application. Figure 11 The y-direction comparison schematic diagram shown in the present application is shown in the present application. Figure 12 ​A schematic diagram of z-direction contrast is shown for another exemplary embodiment of the present application. As can be seen from the results of compensation, before compensation, the x-direction error is substantially uniformly distributed above and below 0; the y-direction error is also distributed around 0, but is more biased towards negative values; the z-direction error is substantially negative; using the error compensation technology of the genetic algorithm optimized deep belief network proposed in this paper, the errors in the three directions are substantially distributed around 0, and fluctuate around ±0.2mm, ±0.1mm and ±0.05mm, the fluctuation range is very small, which shows that the accuracy after compensation has high stability, and can improve the accuracy of robot operation.

[0221] The static statistical analysis results before and after the robot end position error compensation are shown in Table 4, and Table 4 is a compensation effect table. The x, y and z directions are improved by 65.56%, 55.22% and 49.12% respectively.

[0222] Table 4

[0223]

[0224] The experimental platform for data collection and verification of the robot is a light industrial robot KR6_R700 sixx_CR, and the error interval is much smaller than that of traditional heavy industrial robots, so it is more difficult to use deep belief network for feature extraction and model training and optimization. This paper proposes to optimize the deep belief network using genetic algorithm, and establishes an industrial robot position error mapping model combined with evidence theory, Figure 13 A schematic diagram of the compensation effect is shown for an exemplary embodiment of the present application, please refer to Figure 13 is the compensation effect of the robot end accuracy analyzed by combining the three directions.

[0225] In order to verify the test results, the method in this paper and the previous method are compared, and the results are shown in Table 5. Table 5 is a comparison result table of different methods. After offline compensation, the minimum value is improved from 0.097mm to 0.006mm, and the average value is improved from 0.110mm to 0.083mm, so the minimum value and average value after robot end error compensation are improved by the GA-deep belief network method proposed in this paper.

[0226] Table 5

[0227]

[0228] Corresponding to the foregoing embodiment of the industrial robot absolute positioning accuracy compensation method, the present application also provides an embodiment of an industrial robot absolute positioning accuracy compensation device.

[0229] The embodiment of the industrial robot absolute positioning accuracy compensation device provided by the application can be applied to an industrial robot absolute positioning accuracy compensation device. The device embodiment can be realized by software, hardware or a combination of software and hardware. Taking software implementation as an example, as a logical device, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory and running by the processor of the industrial robot absolute positioning accuracy compensation device where the device is located. From the hardware level, the industrial robot absolute positioning accuracy compensation device where the device is located in the embodiment can also include other hardware according to the actual function of the industrial robot absolute positioning accuracy compensation device, which will not be described here.

[0230] Figure 14 The industrial robot absolute positioning accuracy compensation device provided by the example embodiment of the application is shown in the figure. Referring to Figure 14 The device provided by the embodiment includes an acquisition module 1410, a training module 1420, a calculation module 1430 and a processing module 1440. Wherein,

[0231] The acquisition module 1410 is configured to acquire current pose data of multiple sampling points of a robot based on a laser tracker.

[0232] The training module 1420 is configured to generate position error based on pre-determined expected pose data and current pose data of the robot, input the expected pose data into a deep confidence network, use the position error as the output of the deep confidence network, and train the deep confidence network.

[0233] The training module 1420 is also configured to optimize hyperparameters of the deep confidence network based on a genetic algorithm to obtain a trained target deep confidence network.

[0234] The acquisition module 1410 is also configured to acquire compensation data when the robot is working.

[0235] The calculation module 1430 is configured to extract features of the compensation data using the target deep confidence network to obtain feature vectors, and calculate contribution weights of each feature vector based on evidence theory.

[0236] The processing module 1440 is configured to weight process the feature vectors based on the contribution weights to generate compensation results.

[0237] The device of the embodiment can be used to execute Figure 1 The steps of the method embodiment are similar to the specific implementation principle and implementation process, which will not be described here.

[0238] This application also provides an absolute positioning accuracy compensation device for an industrial robot, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the methods provided in the first aspect of this application.

[0239] 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.

[0240] 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.

[0241] 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.

[0242] 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 compensating the absolute positioning accuracy of an industrial robot, characterized in that, The method includes: Determine the multiple objectives for training the accuracy compensation model, wherein the multiple objectives include at least sub-objectives of the training dimension and sub-objectives of the accuracy compensation dimension; Determine the original set of parameters to be trained in the deep belief network; Calculate the degree of influence of each parameter in the original parameter set on the multi-objective; Parameters whose influence exceeds a preset threshold are selected as candidate hyperparameters; The current pose data of the robot at multiple sampling points is obtained using a laser tracker; Based on the predetermined desired pose data and the current pose data of the robot, a position error is generated. The desired pose data is input into the deep belief network, and the position error is used as the output of the deep belief network to train the deep belief network. While training the deep belief network, the candidate hyperparameters of the deep belief network are optimized using a genetic algorithm to obtain a trained target deep belief network. Specifically, for the candidate hyperparameters of the deep belief network, an optimized catalog and initial configuration are generated. The mean square error between the output bias and the actual deviation of the deep belief network is calculated to obtain the fitness function of the genetic algorithm. The initial configurations are sorted based on the fitness function, and those with fitness function values ​​higher than a preset threshold are selected as first target configurations. Crossover and compilation operations are performed on the first target configurations to obtain a set of second target configurations, and the target deep belief network is updated based on the second target configurations. Initialize the configuration to obtain an updated configuration; replace the initial configuration with the updated configuration, and repeat the step of obtaining the updated configuration a preset number of times; optimize the candidate hyperparameters of the deep belief network based on the obtained updated configuration; the calculation of the mean square error between the output bias of the deep belief network and the actual bias includes: inputting the desired pose data into the deep belief network to obtain a predicted output result; calculating the square of the difference between the predicted output result and the position error to obtain a target median; and calculating the mean square error based on the sum of squares of the target median, the output dimension of the deep belief network, and the number of data contained in the desired pose data. Acquire the data to be compensated during robot operation; The target deep belief network is used to extract features from the data to be compensated to obtain feature vectors, and the contribution weight of each feature vector is calculated based on evidence theory. The feature vector is weighted based on the contribution weights, and a compensation result is generated by combining the data to be compensated.

2. The method according to claim 1, characterized in that, The step of inputting the desired pose data into a depth belief network and using the position error as the output of the depth belief network to train the depth belief network includes: The parameters of the deep belief network are adjusted based on a layer-by-layer greedy algorithm to obtain the initialized model parameters; wherein, the training process of the layer-by-layer greedy algorithm is unsupervised pre-training. The parameters of the deep belief network are fine-tuned based on the BP algorithm to complete the parameter adjustment of the deep belief network; wherein, the fine-tuning process of the BP algorithm is supervised fine-tuning.

3. The method according to claim 2, characterized in that, The process of adjusting the parameters of the deep belief network based on a layer-by-layer greedy algorithm to obtain the initialized model parameters includes: The desired pose data is processed based on the visible layer of the first restricted Boltzmann machine, and the target corresponding data of the desired pose data is output through the hidden layer of the restricted Boltzmann machine. Based on the data feedback corresponding to the target, the weights and biases of the visible layer are adjusted to obtain the trained first layer of the restricted Boltzmann machine; The desired pose data is processed based on the trained first-layer restricted Boltzmann machine, and the output data is used as the input of the second-layer restricted Boltzmann machine. The steps of obtaining the target corresponding data are repeated. Once the trained restricted Boltzmann machine has reached the preset number of layers, the model parameter initialization is confirmed to be complete.

4. The method according to claim 2, characterized in that, The fine-tuning of the parameters of the deep belief network based on the BP algorithm, to complete the parameter adjustment of the deep belief network, includes: The desired pose data is input into the deep belief network to obtain the output prediction result; Based on the prediction results and the position error, a loss function is calculated, and the gradient parameters of each gradient of the loss function are obtained; each gradient corresponds to a restricted Boltzmann machine layer. The gradient parameters are backpropagated in the deep belief network to fine-tune the parameters of the deep belief network.

5. The method according to claim 1, characterized in that, The calculation of the contribution weight of each feature vector based on evidence theory includes: Based on the evidence theory, the performance of the feature vector is calculated to obtain multiple basic probability assignment functions corresponding to each feature vector; For a given feature vector, fuse all the corresponding basic probability assignment functions; The contribution weights are generated based on the trust interval and the fused basic probability assignment function.

6. The method according to claim 1, characterized in that, The step of weighting the feature vector based on the contribution weights to generate a compensation result includes: The position coordinates and joint angles in the data to be compensated are obtained, and the position coordinates and joint angles are input into the visible layer of the target depth belief network. Feature extraction is performed through the hidden layer of the target depth belief network to obtain the feature vector. The feature vector is multiplied by the corresponding contribution weight to obtain a weighted vector. The weighted vector is then input into the restricted Boltzmann machine of the target depth belief network to obtain the compensation results for the robot in three directions. The compensation results are then output through the target depth belief network.

7. The method according to claim 1, characterized in that, After the method generates a compensation result by weighting the feature vector based on the contribution weight, it further includes: Based on the compensation results, determine the robot joint angle adjustment vector and the robot position coordinate adjustment vector; The robot's joint angles are corrected based on the robot joint angle adjustment vector, and the robot's end effector position is corrected in Cartesian coordinates based on the robot joint angle adjustment vector.

8. An offline compensation device for absolute positioning accuracy of an industrial robot, characterized in that, The device includes an acquisition module, a training module, a calculation module, and a processing module; wherein, The processing module is used to determine the multiple objectives of training the accuracy compensation model, wherein the multiple objectives include at least the sub-objectives of the training dimension and the sub-objectives of the accuracy compensation dimension. The processing module is also used to determine the original parameter set of the deep belief network to be trained; The calculation module is used to calculate the degree of influence of each parameter in the original parameter set on the multi-objective; The processing module is also used to filter parameters whose influence is higher than a preset threshold as candidate hyperparameters; The acquisition module is used to acquire the current pose data of multiple sampling points of the robot based on the laser tracker; The training module is used to generate a position error based on the predetermined desired pose data and the current pose data of the robot, input the desired pose data into the depth belief network, and use the position error as the output of the depth belief network to train the depth belief network. The training module is further configured to optimize the candidate hyperparameters of the deep belief network using a genetic algorithm while training the deep belief network, thereby obtaining a trained target deep belief network. Specifically, for the candidate hyperparameters of the deep belief network, an optimized catalog and initialization configuration are generated; the mean square error between the output bias of the deep belief network and the actual deviation is calculated to obtain the fitness function of the genetic algorithm; the initialization configurations are sorted based on the fitness function, and the portion with fitness function values ​​higher than a preset threshold is selected as the first target configuration; cross-operations and compilation operations are performed on the first target configurations to obtain a set of second target configurations; and based on the second target configurations... The initialization configuration is updated to obtain an updated configuration; the initialization configuration is replaced with the updated configuration, and the step of obtaining the updated configuration is repeated a preset number of times; the candidate hyperparameters of the deep belief network are optimized based on the obtained updated configuration; the calculation of the mean square error between the output bias of the deep belief network and the actual bias includes: inputting the desired pose data into the deep belief network to obtain a predicted output result; calculating the square of the difference between the predicted output result and the position error to obtain a target median; and calculating the mean square error based on the sum of squares of the target median, the output dimension of the deep belief network, and the number of data contained in the desired pose data. The acquisition module is also used to acquire the data to be compensated during robot operation; The calculation module is also used to extract feature vectors of the data to be compensated using the target deep belief network, and to calculate the contribution weight of each feature vector based on evidence theory. The processing module is further configured to weight the feature vector based on the contribution weight to generate a compensation result.

Images