Spraying control method and system based on intelligent robot

Abstract

The embodiment of the invention provides a spraying control method and system based on an intelligent robot, and the method comprises the steps: carrying out the parameter learning of a target robot spraying control model through employing a sample robot spraying task data sequence, and generating model learning data; and then, based on the target robot spraying control model, learning data without a label to generate spraying control prediction data, thereby further performing parameter learning on the deep learning network model, and iteratively processing iteration sample data without a label to obtain spraying control prediction data. And learning and determining key sample data by using a deep learning network model, taking the key sample data as new sample data, and repeating the process until the model meets the convergence requirement. According to the method, the accuracy and robustness of the target robot spraying control model can be effectively improved, so that the finally obtained target robot spraying control model can perform spraying control operation prediction on the robot spraying task data more accurately, and the efficiency and quality of robot spraying operation are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a spraying control method and system based on an intelligent robot. Background Art

[0002] With the rapid development of intelligent robot technology, robots are increasingly used in various industrial fields, especially in spraying operations. Intelligent robots have gradually replaced traditional manual spraying methods with their high efficiency and precision. However, the control method of robot spraying operations still faces many challenges.

[0003] Traditional robot spraying control methods often rely on pre-set spraying parameters and rules, which are incapable of coping with complex and changeable spraying tasks. Since the quality of spraying operations is affected by many factors, such as the properties of the spraying material, the shape and material of the spraying surface, the temperature and humidity of the spraying environment, etc., fixed spraying parameters and rules are difficult to adapt to all situations, resulting in unstable spraying effects, and even problems such as uneven spraying, missed spraying or overspraying.

[0004] In order to solve this problem, robot spraying control methods based on machine learning have emerged in recent years. These methods use a large amount of spraying task data to train the model, so that the model can automatically learn the rules and characteristics of the spraying operation, thereby realizing intelligent control of the spraying task. However, the existing spraying control methods based on machine learning still have some shortcomings. For example, some methods only rely on limited labeled data to train the model, resulting in limited generalization ability of the model; although other methods use unlabeled data to expand the training set, they fail to effectively mine the valuable information in the unlabeled data, resulting in poor learning effect of the model. Summary of the invention

[0005] In view of this, the purpose of the present application is to provide a spraying control method and system based on an intelligent robot.

[0006] In conjunction with the first aspect of the present application, a spraying control method based on an intelligent robot is provided, which is applied to a spraying control system based on an intelligent robot, and the method comprises:

[0007] Acquire a sample robot spraying task data sequence, use the sample robot spraying task data sequence carrying spraying control label data to perform parameter learning on a target robot spraying control model, and generate model learning data of the sample robot spraying task data sequence;

[0008] Based on the target robot spraying control model that has completed parameter learning, the sample robot spraying task data sequence that does not carry spraying control label data is learned to generate spraying control prediction data of the sample robot spraying task data sequence;

[0009] Using the spray control prediction data of the sample robot spray task data sequence as input data and the model learning data of the sample robot spray task data sequence as output data, performing parameter learning on the deep learning network model;

[0010] Acquire an iterative sample robot spraying task data sequence, learn the iterative sample robot spraying task data sequence based on a deep learning network model that completes parameter learning, generate a learning result, and determine key sample robot spraying task data in the iterative sample robot spraying task data sequence based on the learning result to form a key sample robot spraying task data sequence; wherein the iterative sample robot spraying task data sequence is an iterative sample robot spraying task data sequence that does not carry spraying control label data;

[0011] The key sample robot spraying task data sequence carrying the spray control label data is used as the sample robot spraying task data sequence, and the target robot spraying control model that has completed parameter learning is used as the target robot spraying control model. Return to execute the above steps until the target robot spraying control model that has completed parameter learning meets the model convergence requirements, then end the parameter learning of the target robot spraying control model, and use the target robot spraying control model that has completed parameter learning as the final target robot spraying control model for predicting the spraying control operation of the robot spraying task data.

[0012] In a possible implementation of the first aspect, the spraying control prediction data of the sample robot spraying task data sequence includes a spraying control vector and a probability value corresponding to the sample robot spraying task data in the sample robot spraying task data sequence; the model learning data of the sample robot spraying task data sequence includes an error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence;

[0013] The method uses the spray control prediction data of the sample robot spray task data sequence as input data and the model learning data of the sample robot spray task data sequence as output data to perform parameter learning on the deep learning network model, including:

[0014] Using the spraying control vector and probability value corresponding to the sample robot spraying task data in the sample robot spraying task data sequence to perform parameter learning on the deep learning network model, and generating an estimated error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence;

[0015] Based on the estimated error floating parameters corresponding to the sample robot spraying task data in the sample robot spraying task data sequence and the error floating parameters corresponding to the corresponding sample robot spraying task data in the sample robot spraying task data sequence, the model parameter information of the deep learning network model is optimized to generate a deep learning network model that has completed parameter learning.

[0016] In a possible implementation manner of the first aspect, the learning result is an iterative estimation error floating parameter corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence;

[0017] The deep learning network model based on the completed parameter learning learns the iterative sample robot spraying task data sequence to generate a learning result, including:

[0018] Based on the target robot spraying control model that has completed parameter learning, the iterative sample robot spraying task data sequence is learned to generate a spraying control vector and a probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence;

[0019] Based on the deep learning network model that completes parameter learning, the spraying control vector and probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence are learned to generate iterative estimation error floating parameters corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence.

[0020] In a possible implementation manner of the first aspect, determining the key sample robot spraying task data in the iterative sample robot spraying task data sequence based on the learning result to form the key sample robot spraying task data sequence includes:

[0021] Determine that the sample robot spraying task data in which the iteration estimation error floating parameter in the iterative sample robot spraying task data sequence is greater than the set floating parameter is the key sample robot spraying task data;

[0022] A key sample robot spraying task data sequence is formed based on all the key sample robot spraying task data.

[0023] In a possible implementation of the first aspect, the method further includes:

[0024] After obtaining the target robot spraying control model that has completed parameter learning, calculating the prediction performance of the target robot spraying control model that has completed parameter learning;

[0025] Correspondingly, the target robot spraying control model that completes parameter learning meets the model convergence requirement, and then the parameter learning of the target robot spraying control model is terminated, including:

[0026] It is determined whether the predicted performance of the target robot spraying control model that has completed parameter learning is greater than a set performance index. If so, the parameter learning of the target robot spraying control model is terminated.

[0027] In a possible implementation of the first aspect, the performing parameter learning on the target robot spraying control model by using the sample robot spraying task data sequence carrying the spraying control label data further includes:

[0028] Calculate the first error floating mean parameter of the target robot spraying control model for each parameter learning;

[0029] It is determined whether the first error floating mean parameter is less than a first threshold value. If so, the parameter learning of the target robot spraying control model is terminated to generate a target robot spraying control model that has completed parameter learning.

[0030] In a possible implementation of the first aspect, the spray control prediction data of the sample robot spray task data sequence is used as input data, the model learning data of the sample robot spray task data sequence is used as output data, and parameter learning is performed on the deep learning network model, further comprising:

[0031] Calculate the second error floating mean parameter of the deep learning network model for each round of parameter learning;

[0032] Determine whether the second error floating mean parameter is less than a second threshold value. If so, end the parameter learning of the deep learning network model and generate a deep learning network model that has completed parameter learning.

[0033] In a possible implementation of the first aspect, the method further includes:

[0034] Obtain robot spraying task data of any input target robot;

[0035] Predicting the spraying control operation of the robot spraying task data based on the final target robot spraying control model to generate predicted spraying control data;

[0036] The target robot is spray-controlled based on the predicted spray-control data.

[0037] In combination with the second aspect of the present application, a spray control system based on an intelligent robot is provided, wherein the spray control system based on an intelligent robot includes a machine-readable storage medium and a processor, wherein the machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the spray control system based on an intelligent robot implements the aforementioned spray control method based on an intelligent robot.

[0038] In conjunction with the third aspect of the present application, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When the computer-executable instructions are executed, the aforementioned intelligent robot-based spray control method is implemented.

[0039] In combination with any of the above aspects, the embodiment of the present application performs parameter learning on the target robot spray control model by using the sample robot spray task data sequence, and generates model learning data. Then, based on the target robot spray control model, the data without labels is learned to generate spray control prediction data. With this prediction data as input and the model learning data as output, the deep learning network model is further parameter learned. By iteratively processing the iterative sample data without labels, the deep learning network model is used to learn and determine the key sample data to form a key sample sequence. This key sample sequence is used as the new sample data, and the above process is repeated until the model meets the convergence requirements. This method can effectively improve the accuracy and robustness of the target robot spray control model, so that the target robot spray control model finally obtained can more accurately predict the spray control operation of the robot spray task data, thereby improving the efficiency and quality of the robot spray operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained by combining these drawings without paying creative work.

[0041] Figure 1 A schematic flow chart of a spraying control method based on an intelligent robot provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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.

[0043] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0044] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0045] Figure 1 The flowchart of the spraying control method based on the intelligent robot provided in the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps in the spraying control method based on the intelligent robot of this embodiment can be shared with each other based on actual needs, or some steps can be omitted or maintained. The spraying control method based on the intelligent robot includes:

[0046] Step S110, obtaining a sample robot spraying task data sequence, using the sample robot spraying task data sequence carrying spraying control label data to perform parameter learning on a target robot spraying control model, and generating model learning data for the sample robot spraying task data sequence.

[0047] In this embodiment, it is assumed that a robot spraying task in an automobile production workshop is being processed. In this workshop, there are many different types of automobile parts that need to be sprayed, such as automobile body shells, doors, hoods, etc. The sample robot spraying task data sequence contains various data recorded when spraying tasks were performed on these parts in the past period of time. These data include the shape and size information of the parts (such as the length, width, and height of the body shell, the area and curvature of the door, etc.), material information (metal material or composite material, etc.), surface roughness information, and the desired spray color, thickness, etc.

[0048] The spray control label data is the correct spray control parameters corresponding to these sample tasks. For example, for the body shell of a specific model of car, its spray control label data may include setting the pressure of the spray gun to X Pascals, keeping the distance between the spray gun and the workpiece surface at Y centimeters, setting the spray speed to Z centimeters per second, etc. These parameters are the best spray control parameters obtained through experience or precise measurement.

[0049] The target robot spraying control model is a pre-built mathematical model, and its structure may include a neural network with multiple hidden layers, which is used to simulate the relationship between the input (sample robot spraying task data) and the output (spraying control parameters) in the robot spraying task. When the sample robot spraying task data sequence carrying spraying control label data is used to learn the parameters of the target robot spraying control model, it is like providing the model with a series of questions and answers for it to learn.

[0050] For example, the shape, size, material and other data of the body shell and the corresponding correct spraying control parameters (spray gun pressure, distance, speed, etc.) are input into the target robot spraying control model. The model will adjust the parameters inside the model (such as weights and biases in the neural network, etc.) based on these input and output data through specific algorithms (such as back propagation algorithms). In this process, the model constantly tries to find the parameter combination that best fits the relationship between the input data and the output data. After many such learning processes, the model will generate model learning data for the sample robot spraying task data sequence. These model learning data may reflect the error of the model in the learning process, such as the error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence. This error floating parameter can represent the degree of deviation between the spraying control parameters predicted by the model and the actual spraying control label data, which will play an important role in the subsequent steps.

[0051] Step S120, learning the sample robot spraying task data sequence that does not carry spraying control label data based on the target robot spraying control model that has completed parameter learning, and generating spraying control prediction data for the sample robot spraying task data sequence.

[0052] Let's continue with the example of the automobile production workshop. Now the parameter learning of the target robot spraying control model has been completed. Next, select another set of sample robot spraying task data sequences, which do not carry spraying control label data. For example, this set of data contains some relevant information about new automobile parts, such as the shape, size, material, surface roughness and other data of the bumper of a new model car.

[0053] Input these data into the target robot spray control model that has completed parameter learning. The model will process these input data according to the parameters and algorithms learned previously. Since the model has a certain understanding of the relationship between various types of automotive parts data and corresponding spray control parameters in the previous learning process, it can generate corresponding spray control prediction data based on the newly input parts data.

[0054] For example, for the bumper of this new model car, the model may predict that the pressure of the spray gun should be set to A Pascal, the distance between the spray gun and the workpiece surface should be kept at B cm, the spray speed should be set to C cm / s, etc. These prediction data are not just simple numerical values, but also include probability values ​​associated with these predictions. For example, the model predicts that the probability of the spray gun pressure being A Pascal is 80%, which means that the model has a high confidence in this prediction. At the same time, these prediction data also exist in the form of spray control vectors, which can represent the relationship between various spray control parameters, such as a certain quantitative correlation between spray gun pressure, distance and speed. These spray control prediction data will serve as input data for subsequent deep learning network model learning.

[0055] Step S130, using the spray control prediction data of the sample robot spray task data sequence as input data and the model learning data of the sample robot spray task data sequence as output data, to perform parameter learning on the deep learning network model.

[0056] In this scenario of an automobile production workshop, the deep learning network model is a model specifically used to further optimize the robot spray control. The spray control prediction data of the sample robot spray task data sequence, such as the spray gun pressure, distance, speed for the bumper mentioned above, as well as the corresponding probability value and spray control vector, are provided as input data to the deep learning network model.

[0057] The model learning data of the sample robot spraying task data sequence, that is, the error floating parameters corresponding to the sample robot spraying task data generated in the target robot spraying control model learning process, are used as output data. For example, for the previously input bumper data, there may be a certain error between the spraying control parameters predicted in the target robot spraying control model and the actual ones, and this error floating parameter reflects this difference.

[0058] The deep learning network model will use these input and output data for parameter learning. First, the spray control vector and probability value corresponding to the sample robot spray task data in the sample robot spray task data sequence are used to learn the parameters of the deep learning network model. For example, the neurons in the deep learning network model will perform weighted calculations based on the input spray control vector and probability value. This process is like building a more complex functional relationship in an attempt to find a model structure that can better generate corresponding error floating parameters based on the input prediction data.

[0059] In this process, the deep learning network model will generate an estimated error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence. This estimated error floating parameter is a value close to the actual error floating parameter inferred by the model based on the input data and its own learning algorithm. Then, the model parameter information of the deep learning network model is optimized based on the estimated error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence and the error floating parameter corresponding to the corresponding sample robot spraying task data in the sample robot spraying task data sequence.

[0060] For example, if the estimated error floating parameter is larger than the actual error floating parameter, it means that the model's prediction is too dispersed, and the model's parameters (such as the connection weights in the neural network, etc.) need to be adjusted to make the prediction more accurate. Through continuous comparison and adjustment, a deep learning network model with completed parameter learning is finally generated, which has better performance in predicting error floating parameters.

[0061] Step S140, obtaining an iterative sample robot spraying task data sequence, learning the iterative sample robot spraying task data sequence based on the deep learning network model that has completed parameter learning, generating a learning result, and determining the key sample robot spraying task data in the iterative sample robot spraying task data sequence based on the learning result to form a key sample robot spraying task data sequence. The iterative sample robot spraying task data sequence is an iterative sample robot spraying task data sequence that does not carry spraying control label data.

[0062] In the automobile production workshop scenario, a new set of iterative sample robot spraying task data sequences are obtained. This set of data sequences also does not carry spraying control label data. For example, it contains the shape, size, material, surface roughness and other data of some automobile interior parts (such as dashboards, seat armrests, etc.).

[0063] First, the target robot spraying control model based on the completed parameter learning learns the iterative sample robot spraying task data sequence. The target robot spraying control model processes the interior parts data based on the previously learned knowledge and generates the spraying control vector and probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence. For example, for the instrument panel, the model may predict a range of values ​​for the spray gun pressure, distance, and speed and the corresponding probability value, and give a spraying control vector representing the relationship between these parameters.

[0064] Then, based on the deep learning network model that has completed parameter learning, the spraying control vectors and probability values ​​corresponding to the sample robot spraying task data in these iterative sample robot spraying task data sequences are learned. The deep learning network model will perform complex calculations based on these input data to generate iterative estimated error floating parameters corresponding to the sample robot spraying task data in these iterative sample robot spraying task data sequences. This iterative estimated error floating parameter reflects the error estimation of the prediction of these interior parts spraying task data under the current model.

[0065] Next, based on this learning result, the key sample robot spraying task data in the iterative sample robot spraying task data sequence is determined. A set floating parameter is set. For example, if the iterative estimation error floating parameter of a certain interior part (such as a seat armrest) is greater than this set floating parameter, it means that the model has a large uncertainty or error in predicting the spraying control of this interior part. Then the relevant data of this seat armrest is determined as the key sample robot spraying task data. All such key sample robot spraying task data are collected to form a key sample robot spraying task data sequence. This key sample robot spraying task data sequence contains data whose model prediction effect is not ideal, and these data will be processed in the future.

[0066] Step S150, taking the key sample robot spraying task data sequence carrying the spraying control label data as the sample robot spraying task data sequence, taking the target robot spraying control model that has completed parameter learning as the target robot spraying control model, returning to execute the above steps until the target robot spraying control model that has completed parameter learning meets the model convergence requirements, then ending the parameter learning of the target robot spraying control model, and taking the target robot spraying control model that has completed parameter learning as the final target robot spraying control model for predicting the spraying control operation of the robot spraying task data.

[0067] In the example of the automobile production workshop, the key sample robot spraying task data sequence previously determined to carry spraying control label data (for example, data including correct spraying control parameters for interior parts that were previously determined to have less than ideal prediction effects) is used as a new sample robot spraying task data sequence, while keeping the target robot spraying control model that has completed parameter learning unchanged, and then returning to execute the previous steps S110-S140 again.

[0068] In each cycle of executing these steps, the target robot spray control model and the deep learning network model will continuously optimize their own parameters. For example, in step S110, when the target robot spray control model is trained with the new sample robot spray task data sequence (including key sample data), the model will pay more attention to the learning of the data that was previously predicted inaccurately, and adjust the internal parameters to better fit the relationship between these data and the corresponding spray control label data.

[0069] This process will be repeated until the target robot spray control model that has completed parameter learning meets the model convergence requirements. Model convergence requirements can be judged in a variety of ways. For example, the first error floating mean parameter of the target robot spray control model for each parameter learning can be calculated. This first error floating mean parameter reflects the average deviation between the predicted results and the actual results of the model in the current learning state. If this first error floating mean parameter is less than the first threshold value, it means that the model has reached a certain degree of convergence and the parameter learning of the target robot spray control model can be ended.

[0070] Alternatively, whether to terminate the learning can also be determined by judging whether the prediction performance of the target robot spray control model that has completed parameter learning is greater than the set performance index. The prediction performance can be measured in many ways, such as the prediction accuracy of the model for different types of automotive parts spray task data, the generalization ability for new unknown data, etc. If the prediction performance of the model reaches the set performance index, the parameter learning of the target robot spray control model is terminated.

[0071] When the target robot spray control model that has completed parameter learning meets the model convergence requirements, the target robot spray control model that has completed parameter learning is used as the final target robot spray control model. This final model can be used to predict the spray control operation of new robot spray task data. For example, when a workshop introduces a new car model or new parts need to be sprayed, these new robot spray task data (such as the shape, size, material and other information of the parts) can be input into the final target robot spray control model. The model will accurately predict the corresponding spray control parameters (spray gun pressure, distance, speed, etc.) based on the previously learned knowledge and parameters, thereby achieving precise control of the robot spray task.

[0072] Throughout the entire process in this automobile production workshop, through continuous cycle learning and optimization, the accuracy and reliability of the robot spraying control model can be gradually improved, thereby improving the quality and efficiency of automobile parts spraying.

[0073] Based on the above steps, the embodiment of the present application performs parameter learning on the target robot spray control model by using the sample robot spray task data sequence, and generates model learning data. Then, based on the target robot spray control model, the data without labels is learned to generate spray control prediction data. With this prediction data as input and the model learning data as output, the deep learning network model is further parameter learned. By iteratively processing the iterative sample data without labels, the deep learning network model is used to learn and determine the key sample data to form a key sample sequence. Using this key sample sequence as the new sample data, the above process is repeated until the model meets the convergence requirements. This method can effectively improve the accuracy and robustness of the target robot spray control model, so that the target robot spray control model finally obtained can more accurately predict the spray control operation of the robot spray task data, thereby improving the efficiency and quality of the robot spray operation.

[0074] In a possible implementation, the spray control prediction data of the sample robot spray task data sequence includes the spray control vector and probability value corresponding to the sample robot spray task data in the sample robot spray task data sequence. The model learning data of the sample robot spray task data sequence includes the error floating parameter corresponding to the sample robot spray task data in the sample robot spray task data sequence.

[0075] The method uses the spray control prediction data of the sample robot spray task data sequence as input data and the model learning data of the sample robot spray task data sequence as output data to perform parameter learning on the deep learning network model, including:

[0076] The spraying control vector and probability value corresponding to the sample robot spraying task data in the sample robot spraying task data sequence are used to perform parameter learning on the deep learning network model to generate estimated error floating parameters corresponding to the sample robot spraying task data in the sample robot spraying task data sequence.

[0077] Based on the estimated error floating parameters corresponding to the sample robot spraying task data in the sample robot spraying task data sequence and the error floating parameters corresponding to the corresponding sample robot spraying task data in the sample robot spraying task data sequence, the model parameter information of the deep learning network model is optimized to generate a deep learning network model that has completed parameter learning.

[0078] In a possible implementation, the learning result is an iterative estimation error floating parameter corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence.

[0079] The deep learning network model based on the completed parameter learning learns the iterative sample robot spraying task data sequence to generate a learning result, including:

[0080] The iterative sample robot spraying task data sequence is learned based on the target robot spraying control model that has completed parameter learning, and a spraying control vector and a probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence are generated.

[0081] Based on the deep learning network model that completes parameter learning, the spraying control vector and probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence are learned to generate iterative estimation error floating parameters corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence.

[0082] In a possible implementation, determining the key sample robot spraying task data in the iterative sample robot spraying task data sequence based on the learning result to form the key sample robot spraying task data sequence includes:

[0083] Determine that the sample robot spraying task data in the iterative sample robot spraying task data sequence whose iterative estimation error floating parameter is greater than the set floating parameter is the key sample robot spraying task data.

[0084] A key sample robot spraying task data sequence is formed based on all the key sample robot spraying task data.

[0085] In a possible implementation, the method further includes:

[0086] After obtaining the target robot spraying control model that has completed parameter learning, the prediction performance of the target robot spraying control model that has completed parameter learning is calculated.

[0087] Correspondingly, the target robot spraying control model that completes parameter learning meets the model convergence requirement, and then the parameter learning of the target robot spraying control model is terminated, including:

[0088] It is determined whether the predicted performance of the target robot spraying control model that has completed parameter learning is greater than a set performance index. If so, the parameter learning of the target robot spraying control model is terminated.

[0089] In a possible implementation manner, the performing parameter learning on the target robot spraying control model using the sample robot spraying task data sequence carrying the spraying control label data further includes:

[0090] Calculate the first error floating mean parameter of the target robot spray control model for each parameter learning.

[0091] It is determined whether the first error floating mean parameter is less than a first threshold value. If so, the parameter learning of the target robot spraying control model is terminated to generate a target robot spraying control model that has completed parameter learning.

[0092] In a possible implementation, the spray control prediction data of the sample robot spray task data sequence is used as input data, the model learning data of the sample robot spray task data sequence is used as output data, and the deep learning network model is subjected to parameter learning, further comprising:

[0093] Calculate the second error floating mean parameter of the deep learning network model for each round of parameter learning.

[0094] Determine whether the second error floating mean parameter is less than a second threshold value. If so, end the parameter learning of the deep learning network model and generate a deep learning network model that has completed parameter learning.

[0095] In this embodiment, in the scenario of an automobile production workshop, for a sample robot spraying task data sequence, the spraying control prediction data includes a spraying control vector and a probability value corresponding to the sample robot spraying task data, and the model learning data includes an error floating parameter corresponding to the sample robot spraying task data. When the spraying control prediction data of the sample robot spraying task data sequence is used as input data and the model learning data is used as output data to perform parameter learning on the deep learning network model, the following is a detailed operation process.

[0096] First, the spray control vector and probability value corresponding to the sample robot spray task data in the sample robot spray task data sequence are used to learn the parameters of the deep learning network model. For example, for the spray task data of the automobile body shell, its spray control vector may represent the combination relationship of the pressure, speed and other parameters of the spray gun in different directions, and the probability value represents the possibility of this combination relationship. The neuron structure in the deep learning network model performs complex calculations based on these input vectors and probability values. In this process, each neuron processes the input according to a preset activation function, which may be a linear rectifier function (ReLU) or other suitable functions. Through the calculation and transmission of multiple layers of neurons, the estimated error floating parameter corresponding to the sample robot spray task data in the sample robot spray task data sequence is finally generated. This estimated error floating parameter reflects the deep learning network model's estimation of the actual error floating parameter based on the input spray control vector and probability value.

[0097] Next, the model parameter information of the deep learning network model is optimized based on the estimated error floating parameters corresponding to the sample robot spraying task data in the sample robot spraying task data sequence and the error floating parameters corresponding to the corresponding sample robot spraying task data in the sample robot spraying task data sequence. For the spraying task data of the automobile body shell, the generated estimated error floating parameters are compared with the previously known actual error floating parameters. If the estimated error floating parameters are greater than the actual error floating parameters, it means that the model is too loose in predicting this part of the data, and the parameters of the model need to be adjusted. For example, in the neural network structure of the deep learning network model, the weight values ​​of the connections between neurons are adjusted. The adjustment of the weight value may be achieved through a gradient descent algorithm, and the direction and amplitude of the weight adjustment are calculated according to the difference in the error, so that the model can more accurately predict the error floating parameters in subsequent calculations. Through such processing of a large number of sample robot spraying task data, the model parameters are continuously optimized, and finally a deep learning network model with completed parameter learning is generated.

[0098] In the processing of the iterative sample robot spraying task data sequence, the learning result is the iterative estimated error floating parameter corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence. The process of learning the iterative sample robot spraying task data sequence based on the deep learning network model that completes parameter learning is as follows.

[0099] First, the target robot spraying control model that has completed parameter learning is used to learn the sequence of iterative sample robot spraying task data. For example, for iterative sample robot spraying task data of automotive interior parts such as dashboards, the target robot spraying control model processes the iterative sample robot spraying task data of dashboards based on the previously learned knowledge about automotive parts, including the relationship between shape, size, material, etc. and spraying control parameters, and generates the spraying control vector and probability value corresponding to the iterative sample robot spraying task data of dashboards. This spraying control vector contains the relationship between the pressure, speed and other parameters of the spray gun for the dashboard in different directions and positions, and the probability value indicates the credibility of this relationship.

[0100] Then, based on the deep learning network model that has completed parameter learning, the spray control vector and probability value corresponding to the iterative sample robot spray task data of the dashboard are learned. The deep learning network model performs multi-layer neuron calculations again, processes the input spray control vector and probability value according to the previously optimized model parameters, and finally generates the iterative estimated error floating parameter corresponding to the iterative sample robot spray task data of the dashboard. This iterative estimated error floating parameter reflects the error estimation of the dashboard spray task data prediction under the current model state.

[0101] Based on the above learning results, the key sample robot spraying task data in the iterative sample robot spraying task data sequence is determined, and the process of forming the key sample robot spraying task data sequence is as follows. For the iterative sample robot spraying task data sequence of automobile interior parts, a set floating parameter is set. For example, for the iterative sample robot spraying task data of the automobile seat armrest, if its iterative estimation error floating parameter is greater than the set floating parameter, this indicates that the model has a large uncertainty or error in the spraying control prediction of the automobile seat armrest. Then the relevant iterative sample robot spraying task data of the automobile seat armrest is determined as the key sample robot spraying task data. Through such a judgment of the entire iterative sample robot spraying task data sequence, all key sample robot spraying task data that meet the conditions are collected to form a key sample robot spraying task data sequence.

[0102] After obtaining the target robot spray control model that has completed parameter learning, the prediction performance of the target robot spray control model that has completed parameter learning is calculated. In the scenario of an automobile production workshop, the calculation of prediction performance involves multiple aspects. For example, for the spraying task data of different types of automobile parts (body shell, door, interior parts, etc.), it is input into the target robot spray control model that has completed parameter learning to obtain the predicted spray control parameters. Then it is compared with the actual optimal spray control parameters (from spray control label data or empirical data). The comparison indicators may include the mean value, standard deviation, etc. of the difference between the predicted spray control parameters (such as spray gun pressure, distance, speed, etc.) and the actual parameters. These indicators combined reflect the prediction performance of the model.

[0103] Accordingly, it is determined whether the predicted performance of the target robot spray control model that has completed parameter learning is greater than the set performance index. If so, the parameter learning of the target robot spray control model is terminated. The set performance index is pre-set according to the requirements of the automobile production workshop for robot spraying quality and efficiency. For example, the set performance index requires that the average value of the difference between the predicted spray control parameters and the actual parameters is within a certain range, and the standard deviation is also within a certain range. If the predicted performance of the target robot spray control model that has completed parameter learning meets these requirements, that is, it is greater than the set performance index, then the parameter learning of the target robot spray control model is terminated.

[0104] When the target robot spray control model is parameter-learned using the sample robot spray task data sequence carrying the spray control label data, the first error floating mean parameter of the target robot spray control model for each parameter learning is also calculated. For example, after the sample robot spray task data sequence (including shape, size, material, spray control label data, etc.) of automobile parts (body shell, door, etc.) is input into the target robot spray control model for parameter learning each time, the error between the spray control parameters predicted by the model and the actual spray control label data is calculated. For multiple sample robot spray task data, the mean of these errors is calculated to obtain the first error floating mean parameter. Then, it is determined whether the first error floating mean parameter is less than the first threshold value. If so, the parameter learning of the target robot spray control model is terminated, and the target robot spray control model with completed parameter learning is generated. The first threshold value is pre-set according to the convergence requirements of the model and the requirements of the automobile production workshop for spray accuracy. For example, the first threshold value is set to a certain smaller error value. When the first error floating mean parameter is less than this value, it means that the model has reached a certain degree of convergence and learning can be stopped.

[0105] When the spray control prediction data of the sample robot spray task data sequence is used as input data and the model learning data of the sample robot spray task data sequence is used as output data for parameter learning of the deep learning network model, it also includes calculating the second error floating mean parameter of the deep learning network model for each round of parameter learning. After the sample robot spray task data sequence (including spray control prediction data and model learning data) is input into the deep learning network model for parameter learning in each round, the error between the estimated error floating parameter generated by the model and the actual error floating parameter is calculated. For multiple sample robot spray task data, the mean of these errors is calculated to obtain the second error floating mean parameter. Then, it is determined whether the second error floating mean parameter is less than the second threshold value. If so, the parameter learning of the deep learning network model is terminated, and a deep learning network model with completed parameter learning is generated. The second threshold value is also pre-set according to the convergence requirements of the model and the requirements of the automobile production workshop for spraying accuracy. When the second error floating mean parameter is less than this value, it means that the deep learning network model has reached a certain degree of convergence, and learning can be stopped and the optimized model can be obtained for subsequent processing.

[0106] In a possible implementation, the method further includes:

[0107] Get the robot spraying task data of any input target robot.

[0108] The spraying control operation of the robot spraying task data is predicted based on the final target robot spraying control model to generate predicted spraying control data.

[0109] The target robot is spray-controlled based on the predicted spray-control data.

[0110] In the above embodiments, the intelligent robot-based spray control system for executing the above method embodiments has at least one processor, a control module (chip set) coupled to at least one of the (at least one) processors, a memory coupled to the control module, a non-volatile memory (NVM) / storage device coupled to the control module, at least one load / output device coupled to the control module, and a network interface coupled to the control module.

[0111] The processor may include at least one single-core or multi-core processor, and the processor may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.). For some alternative implementations, the spray control system based on the intelligent robot can be used as the gateway or other electronic device described in the embodiments of the present application.

[0112] For some alternative embodiments, the intelligent robot-based spray control system may include at least one computer-readable medium (e.g., a memory or NVM / storage device) having instructions and at least one processor integrated with the at least one computer-readable medium and configured to execute the instructions to implement a module to perform the actions described in the present disclosure.

[0113] For one embodiment, the control module may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) and / or any suitable device or component in communication with the control module.

[0114] The control module may include a memory controller module to provide an interface to the memory. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0115] The memory may be used, for example, to load and store data and / or instructions for a spray control system based on an intelligent robot. For one embodiment, the memory may include any suitable volatile memory, for example, a suitable DRAM.

[0116] For one embodiment, the control module may include at least one load-in / load-out controller to provide an interface to the NVM / storage device and the (at least one) load-in / load-out device.

[0117] For example, the NVM / storage device may be used to store data and / or instructions. The NVM / storage device may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (at least one) non-volatile storage device (e.g., at least one hard disk drive (HDD), at least one compact disk (CD) drive, and / or at least one digital versatile disk (DVD) drive).

[0118] The NVM / storage device may include storage resources that are physically part of the device on which the intelligent robot-based spray control system is installed, or it may be accessible to the device without being part of the device. For example, the NVM / storage device may be accessed via (at least one) load / output device based on a network.

[0119] (At least one) load-in / output device may provide an interface for the intelligent robot-based spray control system to communicate with any other appropriate device, and the load-in / output device may include a communication component, a phonetic component, a sensor component, etc. The network interface may provide an interface for the intelligent robot-based spray control system to communicate based on at least one network, and the intelligent robot-based spray control system may wirelessly communicate with at least one component of the wireless network based on any of the priors and / or protocols of at least one wireless network, such as accessing a wireless network based on a communication prior.

[0120] For one embodiment, at least one of the (at least one) processor may be loaded together with the logic of at least one controller of the control module (e.g., a memory controller module). For one embodiment, at least one of the (at least one) processor may be loaded together with the logic of at least one controller of the control module to form a system level load. For one embodiment, at least one of the (at least one) processor may be fused on the same die with the logic of at least one controller of the control module. For one embodiment, at least one of the (at least one) processor may be fused on the same die with the logic of at least one controller of the control module to form a system on chip (SoC).

[0121] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, based on the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0122] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the spraying control method based on an intelligent robot described in the aforementioned embodiment.

[0123] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps in the intelligent robot-based spray control method described in the aforementioned embodiment.

[0124] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed over multiple modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Those of ordinary skill in the art may understand and implement the present invention without creative effort.

[0125] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, a magnetic disk storage, a magnetic tape storage, or any other medium that can be used to be computer-readable with or store data.

[0126] Finally, it should be noted that what is disclosed above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A spraying control method based on an intelligent robot, characterized in that: The method comprises: Acquire a sample robot spraying task data sequence, use the sample robot spraying task data sequence carrying spraying control label data to perform parameter learning on a target robot spraying control model, and generate model learning data of the sample robot spraying task data sequence; Based on the target robot spraying control model that has completed parameter learning, the sample robot spraying task data sequence that does not carry spraying control label data is learned to generate spraying control prediction data of the sample robot spraying task data sequence; Using the spray control prediction data of the sample robot spray task data sequence as input data and the model learning data of the sample robot spray task data sequence as output data, performing parameter learning on the deep learning network model; Acquire an iterative sample robot spraying task data sequence, learn the iterative sample robot spraying task data sequence based on a deep learning network model that completes parameter learning, generate a learning result, and determine key sample robot spraying task data in the iterative sample robot spraying task data sequence based on the learning result to form a key sample robot spraying task data sequence; wherein the iterative sample robot spraying task data sequence is an iterative sample robot spraying task data sequence that does not carry spraying control label data; The key sample robot spraying task data sequence carrying the spray control label data is used as the sample robot spraying task data sequence, and the target robot spraying control model that has completed parameter learning is used as the target robot spraying control model. Return to execute the above steps until the target robot spraying control model that has completed parameter learning meets the model convergence requirements, then end the parameter learning of the target robot spraying control model, and use the target robot spraying control model that has completed parameter learning as the final target robot spraying control model for predicting the spraying control operation of the robot spraying task data.

2. The spraying control method based on an intelligent robot according to claim 1 is characterized in that: The spraying control prediction data of the sample robot spraying task data sequence includes the spraying control vector and probability value corresponding to the sample robot spraying task data in the sample robot spraying task data sequence; the model learning data of the sample robot spraying task data sequence includes the error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence; The method uses the spray control prediction data of the sample robot spray task data sequence as input data and the model learning data of the sample robot spray task data sequence as output data to perform parameter learning on the deep learning network model, including: Using the spraying control vector and probability value corresponding to the sample robot spraying task data in the sample robot spraying task data sequence to perform parameter learning on the deep learning network model, and generating an estimated error floating parameter corresponding to the sample robot spraying task data in the sample robot spraying task data sequence; Based on the estimated error floating parameters corresponding to the sample robot spraying task data in the sample robot spraying task data sequence and the error floating parameters corresponding to the corresponding sample robot spraying task data in the sample robot spraying task data sequence, the model parameter information of the deep learning network model is optimized to generate a deep learning network model that has completed parameter learning.

3. The spraying control method based on an intelligent robot according to claim 1, characterized in that: The learning result is an iterative estimation error floating parameter corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence; The deep learning network model based on the completed parameter learning learns the iterative sample robot spraying task data sequence to generate a learning result, including: Based on the target robot spraying control model that has completed parameter learning, the iterative sample robot spraying task data sequence is learned to generate a spraying control vector and a probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence; Based on the deep learning network model that completes parameter learning, the spraying control vector and probability value corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence are learned to generate iterative estimation error floating parameters corresponding to the sample robot spraying task data in the iterative sample robot spraying task data sequence.

4. The spraying control method based on an intelligent robot according to claim 1, characterized in that: The determining of the key sample robot spraying task data in the iterative sample robot spraying task data sequence based on the learning result to form a key sample robot spraying task data sequence includes: Determine that the sample robot spraying task data in which the iteration estimation error floating parameter in the iterative sample robot spraying task data sequence is greater than the set floating parameter is the key sample robot spraying task data; A key sample robot spraying task data sequence is formed based on all the key sample robot spraying task data.

5. The spraying control method based on an intelligent robot according to any one of claims 1 to 4, characterized in that: The method further comprises: After obtaining the target robot spraying control model that has completed parameter learning, calculating the prediction performance of the target robot spraying control model that has completed parameter learning; Correspondingly, the target robot spraying control model that completes parameter learning meets the model convergence requirement, and then the parameter learning of the target robot spraying control model is terminated, including: It is determined whether the predicted performance of the target robot spraying control model that has completed parameter learning is greater than a set performance index. If so, the parameter learning of the target robot spraying control model is terminated.

6. The spraying control method based on an intelligent robot according to claim 1, characterized in that: The method of using the sample robot spraying task data sequence carrying the spraying control label data to perform parameter learning on the target robot spraying control model also includes: Calculate the first error floating mean parameter of the target robot spraying control model for each parameter learning; It is determined whether the first error floating mean parameter is less than a first threshold value. If so, the parameter learning of the target robot spraying control model is terminated to generate a target robot spraying control model that has completed parameter learning.

7. The spraying control method based on an intelligent robot according to claim 1, characterized in that: The method uses the spray control prediction data of the sample robot spray task data sequence as input data, the model learning data of the sample robot spray task data sequence as output data, and performs parameter learning on the deep learning network model, further comprising: Calculate the second error floating mean parameter of the deep learning network model for each round of parameter learning; Determine whether the second error floating mean parameter is less than a second threshold value. If so, end the parameter learning of the deep learning network model and generate a deep learning network model that has completed parameter learning.

8. The spraying control method based on an intelligent robot according to claim 1, characterized in that: The method further comprises: Obtain robot spraying task data of any input target robot; Predicting the spraying control operation of the robot spraying task data based on the final target robot spraying control model to generate predicted spraying control data; The target robot is spray-controlled based on the predicted spray-control data.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the spraying control method based on an intelligent robot described in any one of claims 1 to 8 is implemented.

10. A spray control system based on an intelligent robot, characterized in that: It includes a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the intelligent robot-based spraying control method described in any one of claims 1 to 8 is implemented.

Images