Control method and system of assembly equipment

Abstract

The invention relates to the technical field of equipment control, and discloses a control method and system for assembly equipment. The method comprises the following steps: carrying out initialization scanning and digital twinborn model construction on assembly equipment to obtain an initial digital twinborn model; performing control parameter analysis on the initial digital twin model to obtain initial equipment control parameters; assembling task operation data are obtained through the initial digital twin model, feature deviation prediction is carried out, and feature deviation prediction data are obtained; establishing a derivation gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain target equipment control parameters; performing model updating on the initial digital twin model according to the target equipment control parameters to obtain a target digital twin model; and performing operation data fault diagnosis through the target digital twinning model to obtain an equipment fault type and an equipment fault probability, and realizing intelligent control of the assembly equipment by adopting a digital twinning technology.

Description

Technical Field

[0001] The present application relates to the technical field of equipment control, and in particular to a control method and system for assembly equipment. Background Art

[0002] Precise control and efficient operation of assembly equipment are the key to improving production efficiency and ensuring product quality. With the rapid development of industrial automation technology, the control methods of assembly equipment are constantly improving. However, there are still many challenges in actual operation. Traditional control methods often rely on preset parameters and static control models, which are inadequate when dealing with complex or changing production conditions. For example, when the equipment is in different working environments or facing different material properties, it may be necessary to continuously adjust the control strategy, which requires the assembly system to have higher adaptability and flexibility.

[0003] In addition, the existing control methods need to be strengthened in terms of fault diagnosis and prevention. Assembly equipment may experience various expected and unexpected faults during long-term operation. If these faults are not discovered and handled in time, they will seriously affect production safety and efficiency. However, traditional fault diagnosis technology usually relies on post-processing analysis, which makes it difficult to achieve real-time monitoring and prediction. Summary of the invention

[0004] The present application provides a control method and system for assembly equipment, which are used to realize intelligent control of assembly equipment by adopting digital twin technology.

[0005] In a first aspect, the present application provides a control method for an assembly device, the control method for the assembly device comprising: Perform initial scanning and digital twin model construction on the assembly equipment to obtain an initial digital twin model; Performing control parameter analysis on the initial digital twin model to obtain initial equipment control parameters; Acquiring assembly task operation data through the initial digital twin model and performing feature deviation prediction to obtain feature deviation prediction data; Establishing a derived gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain target device control parameters; Updating the initial digital twin model according to the target device control parameters to obtain a target digital twin model; The target digital twin model is used to perform operation data fault diagnosis to obtain the equipment fault type and equipment failure probability.

[0006] In a second aspect, the present application provides a control system for an assembly device, the control system for the assembly device comprising: A construction module is used to perform initial scanning on the assembly equipment and build a digital twin model to obtain an initial digital twin model; An analysis module, used to perform control parameter analysis on the initial digital twin model to obtain initial device control parameters; A prediction module, used to obtain assembly task operation data through the initial digital twin model and perform feature deviation prediction to obtain feature deviation prediction data; An optimization module, used for establishing a derived gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain target device control parameters; An updating module, used to update the initial digital twin model according to the target device control parameters to obtain a target digital twin model; The diagnostic module is used to perform operation data fault diagnosis through the target digital twin model to obtain the equipment fault type and equipment fault probability.

[0007] The third aspect of the present application provides a control device for an assembly device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the control device of the assembly device executes the above-mentioned control method of the assembly device.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned control method for the assembly equipment.

[0009] In the technical solution provided by the present application, by utilizing the digital twin model, the state of the assembly equipment and the changes in the operating environment can be reflected in real time. The dynamic update function of the digital twin model allows the assembly equipment to automatically adjust the control parameters according to the actual working conditions, thereby enhancing the adaptability and flexibility of the equipment to complex working environments. This combination of real-time simulation and prediction significantly improves the response speed and efficiency of the control system. By accurately analyzing and optimizing the control parameters of the initial digital twin model, the operation of the equipment can be more accurately adjusted to meet specific assembly requirements. The use of reinforcement learning and particle swarm optimization algorithms to analyze and adjust the control parameters not only optimizes the efficiency of equipment operation, but also improves the accuracy and quality of the production process. By comprehensively analyzing the operating data and feature deviations, potential equipment failures can be effectively predicted and diagnosed. This prediction-based fault diagnosis system can promptly discover and solve problems, reduce equipment downtime, reduce maintenance costs, and improve production continuity and equipment life. By real-time monitoring and maintaining the operating status of assembly equipment, losses caused by failures or improper operations can be effectively reduced. In addition, the optimized control parameters can ensure that the equipment operates in the best state, thereby saving energy consumption and reducing unnecessary material waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 A schematic diagram of an embodiment of a control method for assembling equipment in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a control system for assembling equipment in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application embodiment provides a control method and system for assembly equipment. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the control method of the assembly equipment in the embodiment of the present application includes: Step S101, performing initialization scanning and digital twin model construction on the assembly equipment to obtain an initial digital twin model; It is understandable that the execution subject of the present application may be a control system of an assembly device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0014] Specifically, the original image data of the assembly equipment is collected through the image acquisition system. These data usually contain various noises and need to be denoised to improve the image quality. The denoised image data is clearer and easier to process. High-dimensional feature recognition technology is applied to the denoised image data, and computer vision algorithms such as convolutional neural networks are used to detect and extract key visual features of the assembly equipment. The high-dimensional image features are encoded to obtain feature encoding vectors, and the image features are converted into a series of digital codes that can more effectively represent the image content. The three-dimensional model of the assembly equipment is reconstructed according to the migration feature vector to obtain the three-dimensional model of the equipment, and the image features of the equipment are converted into a form suitable for model reconstruction. After style transfer, the obtained migration feature vector is used in the three-dimensional modeling software to reconstruct the accurate three-dimensional model of the assembly equipment. The three-dimensional model provides detailed geometric and structural information of the assembly equipment and is the basis for the construction of the digital twin model. In order to ensure the accuracy and applicability of the established three-dimensional model, the three-dimensional model of the equipment is verified from multiple angles and multiple environments. The verification process ensures that the model can effectively reflect the performance and operating characteristics of the physical assembly equipment in practical applications, and obtain the initial digital twin model of the assembly equipment.

[0015] Step S102: Analyze the control parameters of the initial digital twin model to obtain initial equipment control parameters; Specifically, in order to optimize the performance of assembly equipment, a reward function based on reinforcement learning is set, and the performance of equipment control parameters is evaluated and optimized through the reward function. At the same time, a knowledge dissemination mechanism supported by meme theory is introduced, which uses the concept of meme, i.e., cultural gene, to promote the effective dissemination of knowledge and information in the assembly system, imitating the natural selection mechanism in biological evolution, thereby optimizing the learning process. The control parameters of the initial digital twin model are initialized by applying the particle swarm optimization algorithm. The algorithm finds the optimal solution by simulating the group behavior of a group of birds. Each particle represents a set of potential equipment control parameter configurations, improves the efficiency of parameter search, and explores a variety of potential effective parameter combinations. Using the set reinforcement learning reward function, the reward feedback calculation is performed on the first equipment control parameters, and each set of parameters obtains a corresponding reward value according to its positive impact on the equipment performance. According to the obtained reward feedback value, the first equipment control parameters are adjusted to form multiple sets of second equipment control parameters. Using the knowledge dissemination mechanism in meme theory, the most promising search direction and range are determined, which helps to guide the system to further search and optimize the second equipment control parameters. According to the search direction and search range, multiple second device control parameters are searched to obtain multiple third device control parameters. Fitness calculation is performed on multiple third device control parameters to evaluate the actual contribution of each set of parameter configurations to the performance of the assembly equipment. Through mathematical models and algorithm support, the parameters are finally optimized and solved to ensure that each set of control parameters can provide the best performance in actual applications, and finally the initial device control parameters are obtained.

[0016] Step S103, obtaining assembly task operation data through the initial digital twin model and performing feature deviation prediction to obtain feature deviation prediction data; Specifically, the assembly task operation data, including dynamic parameters, operation sequence and environmental interaction information in the assembly process, are collected through the initial digital twin model. Node and edge feature information is extracted from the assembly task operation data. Nodes may represent key components or operation steps in the assembly process, while edges represent the interactions and dependencies between these components or steps. Multi-layer attention hierarchy analysis is performed on the feature information of nodes and edges, and the attention mechanism is used to determine the importance of dependencies between different nodes and edges. Through analysis, the mutual influence between nodes is identified, and the understanding of edge connection characteristics is deepened to improve the dimension and quality of feature information. Feature encoding is performed on the feature information of nodes and edges according to the node and edge dependencies, and the complex relationship is converted into a usable numerical representation to obtain the first operation feature. Time series analysis is performed on the same data set to identify and extract time dependencies and periodic changes in the operation data to obtain the second operation feature. Feature deviation prediction is performed on the first operation feature and the second operation feature. Feature deviation prediction is performed through a machine learning model, such as a deep learning network. The prediction model automatically adjusts its parameters based on historical operation data and real-time feedback to improve the accuracy and adaptability of the prediction, and finally generates feature deviation prediction data.

[0017] Step S104, establishing a derived gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain the target device control parameters; Specifically, the influence of each target parameter in the initial equipment control parameters on the performance of the assembly equipment is calculated based on the characteristic deviation prediction data, and the sensitivity of each parameter change to the output result is determined to obtain the influence gradient. The corresponding derivation gradient matrix is ​​constructed according to the influence gradient of each target parameter. The derivation gradient matrix is ​​an efficient tool that integrates all key control parameters and their contribution to equipment performance, providing a clear framework for parameter optimization. Through the matrix, the potential impact of different parameter adjustments on assembly efficiency and accuracy can be intuitively seen. Based on the derivation gradient matrix, each target parameter in the initial equipment control parameters is optimized. Algorithms such as gradient descent or other optimization techniques are used to ensure that each parameter can be adjusted to the optimal state while ensuring the stability and efficiency of equipment operation. The optimized target parameters are comprehensively combined to form the target equipment control parameters. The combination process takes into account the interaction and synergy between different parameters to ensure the overall optimization of the control strategy.

[0018] Step S105: updating the initial digital twin model according to the target device control parameters to obtain the target digital twin model; Specifically, the initial digital twin model is controlled and simulated based on the target device control parameters to obtain a simulated control data set, which simulates the expected operation of the device under the new control parameters. At the same time, in order to conduct effective comparative analysis, a real control data set is collected from the assembly equipment in actual operation, which reflects the actual performance of the device under the influence of the new control parameters. The Wasserstein distance algorithm is used to perform a difference comparison analysis on the simulated control data set and the real control data set. Wasserstein distance, as an efficient measurement method, is suitable for the measurement of distribution differences and can quantify the "movement" cost between two sets of data distributions, that is, the minimum effort required to transform one set of distributions into another. Through this algorithm, the difference between the simulated data and the real data is evaluated to provide a quantitative basis for the adjustment of the model. Based on the results of the difference comparison analysis, the initial digital twin model is updated, and specific parameters and structures in the model are adjusted to reduce the gap between the simulated data and the real data, and to improve the prediction accuracy and reliability of the model. The update may include adjusting the internal simulation logic, optimizing parameter settings, or retraining the algorithm components in the model to ensure that the model can more accurately reflect the working status and response of the actual equipment, and finally obtain the target digital twin model. The updated model is not only more consistent with the actual operating conditions of the equipment in theory, but can also provide more accurate predictions and decision support in practical applications, thereby helping to optimize the operating efficiency and performance of assembly equipment and ensure the efficiency and accuracy of the assembly process.

[0019] Step S106: Perform operating data fault diagnosis through the target digital twin model to obtain the equipment fault type and equipment failure probability.

[0020] Specifically, the target digital twin model is used to simulate the assembly task and collect the target assembly operation data without affecting the actual production line. These data reflect the performance of the equipment in the simulated working environment and provide real-time information for fault diagnosis. Cluster analysis is performed on the target assembly operation data to group the data into multiple assembly operation feature centers, each of which represents a group of data points with similar characteristics. Cluster analysis helps the system understand the normal and abnormal modes in equipment operation, so as to more effectively identify potential abnormalities or signs of failure. Multiple assembly operation feature centers are input into the preset multi-layer neural network model for operation data fault diagnosis to obtain the equipment fault type. According to the equipment fault type identified by the multi-layer neural network model, the fault probability is calculated. Combining historical fault data and current operating status, probabilistic statistical methods are applied to estimate the probability of various faults. Help the system predict and prepare for possible maintenance work, thereby optimizing maintenance plans and reducing unexpected downtime.

[0021] In the embodiment of the present application, by using the digital twin model, the state of the assembly equipment and the changes in the operating environment can be reflected in real time. The dynamic update function of the digital twin model allows the assembly equipment to automatically adjust the control parameters according to the actual working conditions, thereby enhancing the adaptability and flexibility of the equipment to complex working environments. This combination of real-time simulation and prediction significantly improves the response speed and efficiency of the control system. By accurately analyzing and optimizing the control parameters of the initial digital twin model, the operation of the equipment can be more accurately adjusted to meet specific assembly requirements. The use of reinforcement learning and particle swarm optimization algorithms to analyze and adjust the control parameters not only optimizes the efficiency of equipment operation, but also improves the accuracy and quality of the production process. By comprehensively analyzing the operating data and feature deviations, potential equipment failures can be effectively predicted and diagnosed. This prediction-based fault diagnosis system can promptly discover and solve problems, reduce equipment downtime, reduce maintenance costs, and improve production continuity and equipment life. By real-time monitoring and maintaining the operating status of assembly equipment, losses caused by failures or improper operations can be effectively reduced. In addition, the optimized control parameters can ensure that the equipment operates in the best state, thereby saving energy consumption and reducing unnecessary material waste.

[0022] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Collecting raw image data of the assembly equipment and denoising the raw image data to obtain denoised image data; (2) Perform high-dimensional feature recognition on the denoised image data to obtain high-dimensional image features, and encode the high-dimensional image features to obtain a feature encoding vector; (3) Perform style migration on the feature encoding vector to obtain the migration feature vector; (4) Reconstructing the three-dimensional model of the assembly equipment according to the migration feature vector to obtain the three-dimensional model of the equipment; (5) Verify the equipment three-dimensional model from multiple angles and environments to obtain the initial digital twin model.

[0023] Specifically, the original image data of the assembly equipment is collected, and the various components of the equipment and their operating status are captured by a high-resolution camera. The original image data is denoised. The original image is often affected by various interferences, such as changes in ambient light, camera shake, and other factors. The denoising process usually uses filtering techniques, such as Gaussian blur or median filtering, to effectively remove random noise in the image and retain important visual information. High-dimensional feature recognition is performed on the denoised image data. Computer vision algorithms, such as convolutional neural networks, are used to identify complex patterns and features from the image, such as the shape, size, and relative position of specific components of the equipment. High-dimensional image features are encoded to obtain feature encoding vectors. High-dimensional data is compressed into a more efficient numerical representation, making the features more economical in calculation and storage, and facilitating subsequent processing and analysis. The feature encoding vector is style-transferred to merge the style of the equipment features with the pre-defined model style. By changing the style of the feature encoding vector, it can adapt to different model requirements and visual effects. After style transfer, the three-dimensional model of the assembly equipment is reconstructed according to the transferred feature vector. The 3D modeling process usually relies on professional CAD software or 3D modeling tools. By inputting feature vectors, the software can generate a detailed 3D representation of the equipment. In order to ensure that the generated 3D model can accurately reflect the physical and operational characteristics of the actual equipment, the 3D model of the equipment is verified from multiple angles and in multiple environments. This includes testing the model under different lighting, temperature and operating conditions to evaluate its performance and response. For example, by installing multiple high-resolution cameras in the assembly workshop, every movement and component state of the robotic arm during operation is captured. After denoising and feature extraction, these images use deep learning technology to identify the specific position and motion state of each joint and operating tool of the robotic arm. These data are converted into feature vectors and style transfer is performed to adapt to the pre-defined high-performance model style. Use this data to reconstruct an accurate 3D digital model of the robotic arm in the 3D modeling software, and test the model in simulated different operating environments to ensure that it can maintain efficient and accurate operation in any given situation.

[0024] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Setting the reward function for reinforcement learning and introducing the knowledge dissemination mechanism through meme theory; (2) Initializing the control parameters of the initial digital twin model through the particle swarm algorithm to obtain multiple first device control parameters; (3) performing reward feedback calculation on the plurality of first device control parameters based on the reward function to obtain a reward feedback value of each first device control parameter; (4) performing parameter feedback adjustment on the plurality of first device control parameters according to the reward feedback value to obtain a plurality of second device control parameters; (5) determining a corresponding search direction and a search range based on the knowledge propagation mechanism, and performing parameter search on the plurality of second device control parameters according to the search direction and the search range to obtain a plurality of third device control parameters; (6) Perform fitness calculation and parameter optimization on multiple third device control parameters to obtain initial device control parameters.

[0025] Furthermore, a reinforcement learning environment is constructed, with the initial digital twin model as the environment, the state space defined as the current control parameter configuration, and the action space as the possible parameter adjustment schemes; a reward function is designed to quantify the performance change after each parameter adjustment based on key performance indicators such as assembly efficiency, product quality, and energy consumption; a Q-value table or a deep Q-network is initialized to estimate the long-term reward of taking an action under a specific state; an ε-greedy strategy is implemented to strike a balance between exploring new parameter combinations and using known optimal parameters; an experience replay buffer is set to store the experience tuples (state, action, reward, next state) of the agent's interaction with the environment; a virtual assembly task is performed in the initial digital twin model, and the performance indicators in the initial state are recorded; the agent selects an action according to the ε-greedy strategy, that is, a set of parameter adjustment schemes is selected; the selected parameter adjustment is applied in the digital twin model, the virtual assembly task is performed, and the new state and the reward obtained are observed; the interaction experience is recorded in the buffer. The experience (state, action, reward, new state) is stored in the experience replay buffer; a batch of experience is randomly sampled from the experience replay buffer to update the Q value table or train the deep Q network; the Q value estimate or the neural network weight is updated using the temporal difference learning or back propagation algorithm; until the preset number of training rounds is reached or the performance index meets the requirements; the performance of the trained agent in different initial states is evaluated to verify its adaptability under various parameter configurations; the optimal strategy after training is completed is extracted, which guides how to select the optimal parameter adjustment action according to the current state; the optimal strategy is applied to different operating scenarios of the initial digital twin model to collect a series of high-performance parameter configurations; statistical analysis is performed on the collected high-performance parameter configurations to calculate the mean, variance and other statistical characteristics of each control parameter; based on the statistical characteristics and combined with the knowledge of domain experts, the reasonable range and initial value of each control parameter are determined; the determined parameter range and initial value are combined to form an initial device control parameter set.

[0026] Specifically, a reward function for reinforcement learning is set, which evaluates the effect of each control parameter adjustment to guide future parameter settings. The design of the reward function depends on the specific goals of the equipment operation, such as improving efficiency, reducing energy consumption, or improving product quality. For example, in an automated robotic assembly line, the reward function may be set based on the speed and accuracy of completing the task, where operations with fast speed and small errors receive higher rewards. The knowledge dissemination mechanism of meme theory is introduced to enhance the learning process. Meme theory promotes the dissemination of excellent operating habits and strategies in manufacturing systems by simulating cultural dissemination. By simulating this "cultural inheritance", the optimization process of control parameters is more in line with actual production needs. For example, if a certain parameter configuration is found in a robotic assembly line to significantly improve operating efficiency, this configuration can be regarded as a "good meme" and quickly spread to other parts of the entire production line. The control parameters of the initial digital twin model are initialized using a particle swarm algorithm. The particle swarm optimization algorithm is a swarm intelligence algorithm that finds the optimal solution to a problem by simulating the group behavior of a flock of birds or a school of fish. In the control parameter optimization of assembly equipment, each particle represents a set of possible parameter settings. The particles search in the parameter space and update their own positions and speeds by following the currently found optimal solution (i.e., the global optimal particle) and the individual historical optimal solution (i.e., the individual optimal particle). In this way, they quickly converge to a set of better control parameters and obtain the first device control parameters. Based on the reward function, reward feedback calculations are performed on multiple first device control parameters to obtain the reward feedback value of each first device control parameter. Each set of parameters obtains a corresponding reward value according to the operation results it produces (such as the speed and accuracy of completing the task). According to the reward feedback value, the parameters are adjusted to generate the second device control parameters. The knowledge propagation mechanism based on meme theory determines the direction and scope of the search, ensuring that the parameter search is not only based on mathematical optimization, but also reflects the best practices in actual operation. According to the determined direction and scope, the second device control parameters are searched, and finally the third device control parameters are obtained. These parameters are further adjusted through fitness calculation and parameter optimization solution, and finally the initial device control parameters are obtained. This process makes full use of the mechanism of simulating biological evolution, ensuring that the parameter setting is not only scientific and reasonable, but also feasible and efficient in actual operation. For example, in a high-precision electronic component assembly system, the initial control parameters may include the robot's movement speed, precision setting, and work sequence. Through the above steps, these parameters are systematically adjusted so that the robot can achieve the fastest assembly speed and the highest energy efficiency while ensuring assembly quality, thereby achieving dual optimization of production efficiency and quality.

[0027] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Collect assembly task operation data through the initial digital twin model; (2) Extract node and edge feature information from assembly task operation data to obtain node and edge feature information; (3) Perform multi-layer attention hierarchy analysis on the feature information of nodes and edges to obtain the node and edge dependency relationship, and perform feature encoding on the feature information of nodes and edges according to the node and edge dependency relationship to obtain the first operation feature; (4) Performing time series analysis on the assembly task operation data to obtain the second operation feature; (5) Perform feature deviation prediction on the first operation feature and the second operation feature to obtain feature deviation prediction data.

[0028] Specifically, the initial digital twin model is used to collect assembly task operation data, including the movement trajectory of the robot arm, the speed, accuracy, force of component assembly, and other related operating parameters. The digital twin model can provide virtual data that is fully synchronized with the physical assembly line, allowing the system to simulate and analyze various possible operation scenarios and configurations without interfering with actual production. For example, assuming that the system is analyzing an automated production line, the model will record every detail from component picking to precise placement, including each joint movement of the robot arm and the readings of the pressure sensor. The collected data contains a large amount of node and edge information. Nodes may represent specific assembly points, such as screw insertion points or component contact points, while edges represent the action sequence in the task and the dependencies between components. The feature information of nodes and edges is extracted, and data processing techniques such as image processing and sensor data analysis are used to obtain accurate operation characteristics and interaction information between assembly components. The information of these nodes and edges is processed using multi-layer attention hierarchy analysis. By applying the attention mechanism, the features that have the greatest impact on the assembly task are prioritized, the efficiency of data processing is improved, and the accuracy of analysis is improved. For example, if a screw is often driven at the wrong angle in a certain assembly step, the attention model will highlight this step to ensure that it is analyzed in detail. With the help of multi-layer attention analysis, the dependencies of nodes and edges are clearly mapped, and feature information is encoded according to the dependencies to obtain the first set of operational features. Time series analysis of assembly task operation data reveals the time-dependent characteristics and potential periodic problems in the assembly process. By using techniques such as autoregressive models or Fourier transforms, key time series features such as periodic fluctuations, mutation points or trend changes are extracted from the operation data to obtain the second set of operational features. The first and second operational features are combined and machine learning models such as random forests or neural networks are applied to predict feature deviations. The prediction process involves comparing the expected behavior generated by the model with the actual behavior to determine possible causes of deviations and solutions. For example, if the pressure data during the assembly of a component is found to have large deviations, it may indicate that the assembly pressure needs to be adjusted or the stability of the robot arm needs to be improved.

[0029] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Calculate the influence gradient of each target parameter in the initial equipment control parameters based on the characteristic deviation prediction data; (2) Construct the corresponding derived gradient matrix according to the influence gradient of each target parameter; (3) Based on the derived gradient matrix, each target parameter in the initial device control parameters is optimized to obtain the optimization result of each target parameter; (4) Combine the optimization results of each target parameter to obtain the target device control parameters.

[0030] Specifically, the influence gradient of each target parameter in the initial equipment control parameters is calculated based on the characteristic deviation prediction data. The prediction data usually comes from the comparative analysis of the historical operation data of the system and the current operating state. Through data analysis techniques, such as machine learning models, the potential impact on assembly quality or efficiency when the parameters change is predicted. The purpose of calculating the influence gradient is to determine the expected impact on the assembly task results (such as assembly accuracy or speed) when each control parameter changes by one unit. The mathematical expression of the influence gradient usually involves partial derivatives, that is, the calculation of local derivatives around the current set point, which can be achieved through automatic differentiation techniques or numerical approximation methods. A derived gradient matrix is ​​constructed based on the influence gradient of each target parameter. The matrix is ​​a multidimensional array in which each element represents the degree of influence of a specific control parameter on a certain performance indicator. Constructing the matrix helps to systematically understand the interactions and independent effects between different control parameters. Based on the derived gradient matrix, the initial equipment control parameters are optimized. Optimization algorithms, such as gradient descent, genetic algorithms, or other heuristic search techniques are used to ensure that the optimal parameter combination is found to maximize production efficiency and quality while minimizing cost and resource consumption. In this process, each target parameter is adjusted according to its corresponding gradient value, with the goal of reducing the predicted feature deviation and improving the overall performance of the assembly equipment. The optimization results of each target parameter are combined to form the final target equipment control parameters. The adjustment results of each parameter are evaluated and integrated to ensure that all parameter changes work together to achieve the optimal state of the entire assembly system. For example, suppose a robot system for part assembly needs to accurately control the temperature, pressure and speed of the welding point. By analyzing historical data, it is found that when the welding temperature and pressure exceed a certain range, the strength of the weld joint will decrease significantly. Based on this data, the influence gradients of temperature and pressure are calculated, and the corresponding derived gradient matrix is ​​constructed. The temperature and pressure parameters are adjusted using an optimization algorithm to ensure the maximum strength of the weld joint and the optimal production rate. After multiple optimization iterations, the final control parameters can ensure the welding quality while improving the efficiency of the entire assembly line.

[0031] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Perform control simulation on the initial digital twin model based on the target equipment control parameters to obtain a simulated control data set, and obtain the real control data set of the assembly equipment based on the target equipment control parameters; (2) Based on the Wasserstein distance algorithm, a difference comparison analysis is performed on the simulated control data set and the real control data set to obtain the difference comparison analysis results; (3) Based on the difference comparison analysis results, the initial digital twin model is updated to obtain the target digital twin model.

[0032] Specifically, the initial digital twin model is controlled and simulated based on the target device control parameters. The initial digital twin model is driven by the device control parameters to simulate its performance in real operation, and a simulated control data set is obtained, which contains information such as the possible operating status, performance indicators, and possible faults of the device. For example, suppose the system is processing a digital twin model of a robot assembly arm, which uses the latest optimized speed and force parameters to simulate the operation of the robot when assembling parts. At the same time, a real control data set is obtained from the actual assembly equipment. These data reflect the actual performance of the equipment under the same control parameters in the actual production environment. By comparing the simulated data with the actual data, the accuracy and reliability of the digital twin model are evaluated. For example, the real control data may include the actual assembly time, the energy consumed, the completed quality inspection results, etc. The Wasserstein distance algorithm is used to compare and analyze the differences between the simulated control data set and the real control data set. Wasserstein distance, also known as the earth moving distance, is a method to measure the difference between two probability distributions. Calculating the "minimum amount of work" required to transform one distribution into another is suitable for evaluating the difference between data sets with actual physical meaning. For example, the Wasserstein distance can help the system quantify the difference in performance between simulated assembly operation data and actual assembly operation data, and the quantitative results can help identify inaccuracies in the model. The initial digital twin model is updated based on the results of the difference comparison analysis. Adjust certain parameters of the model, such as changing the dynamic model of robot motion, or updating certain assumptions in the simulated assembly process, such as friction coefficient or material fatigue properties. In this way, the digital twin model is constantly approaching an ideal model that can accurately predict and reflect the actual equipment state, thereby improving the efficiency and reliability of the entire manufacturing system. For example, if it is found that the assembly speed of the robot in actual operation is generally lower than the speed predicted by the simulation, it may be necessary to adjust the algorithm or parameters of the robot speed in the model to more accurately simulate the dynamic limitations and actual resistance encountered in actual operation. Through the iterative process, the model is continuously optimized until the Wasserstein distance between the simulated data and the actual data is minimized, that is, the two are very close, indicating that the model can very accurately predict and simulate the behavior of the actual equipment.

[0033] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Perform assembly task operations through the target digital twin model to obtain target assembly operation data; (2) Performing cluster analysis on the target assembly operation data to obtain multiple assembly operation feature centers; (3) Input multiple assembly operation feature centers into a preset multi-layer neural network model to perform operation data fault diagnosis and obtain the equipment fault type; (4) Calculate the failure probability of the assembly equipment according to the equipment failure type to obtain the equipment failure probability.

[0034] Specifically, the assembly task operation is performed through the target digital twin model. The target digital twin model is used to simulate the operation of physical assembly equipment and generate operation data of the assembly task. These data simulate the behavior of the actual equipment when performing a specific task, such as the action of a robot arm installing parts on an assembly line. By setting different operation parameters, such as speed, force, accuracy, etc., the model can output various operation data in the assembly process, such as the movement trajectory, torque output, and response time of the robot arm. Cluster analysis is performed on the target assembly operation data. Cluster analysis is a statistical tool used to group observations in a data set so that data points in the same group are more similar than data points in other groups. Different feature centers in the assembly operation data are identified, each center representing a typical operation mode in the assembly process. For example, when performing cluster analysis, it may be found that some data points show repeated pauses of the robot arm during the assembly process, which may indicate mechanical failure or programming errors.

[0035] Furthermore, in this embodiment, the target digital twin model is used to perform assembly task operations to obtain target assembly operation data; cluster analysis is performed on the target assembly operation data to obtain multiple assembly operation feature centers; the multiple assembly operation feature centers are input into a preset multi-layer neural network model to perform operation data fault diagnosis to obtain the equipment fault type, wherein the step of inputting the multiple assembly operation feature centers into the preset multi-layer neural network model to perform operation data fault diagnosis to obtain the equipment fault type includes: inputting the multiple assembly operation feature centers into the input layer of the multi-layer neural network model, the input layer including the number of neurons matching the feature dimension; performing batch normalization processing on the output data of the input layer to obtain normalized feature data; inputting the normalized feature data into the first hidden layer, the first hidden layer adopts a fully connected structure, including 1024 neurons, and uses the ReLU activation function for nonlinear transformation to obtain the first layer feature representation; applying the Dropout operation to the first layer feature representation, with the dropout rate set to 0.5, to obtain the first layer sparse feature representation; inputting the first layer sparse feature representation into the second hidden layer, the second hidden layer adopts a fully connected structure, including 512 neurons, and uses the Leaky The ReLU activation function is used for nonlinear transformation to obtain the second-layer feature representation; L2 regularization is applied to the second-layer feature representation, and the regularization coefficient is set to 0.01 to obtain the second-layer regularized feature representation; the second-layer regularized feature representation is input into the attention mechanism layer, and the attention mechanism layer uses the self-attention mechanism to calculate the feature weights to obtain the weighted feature representation; the weighted feature representation is input into the third hidden layer, and the third hidden layer adopts a fully connected structure, including 256 neurons, and the ELU activation function is used for nonlinear transformation to obtain the third-layer feature representation; the third-layer feature representation is input into the output layer, and the output layer adopts the Softmax activation function, and the number of neurons is the same as the number of fault types, to obtain the probability distribution of each fault type; based on the probability distribution, the fault type corresponding to the highest probability is determined, and it is used as the diagnosis result of the equipment fault type; the fault probability of the assembly equipment is calculated according to the equipment fault type to obtain the equipment fault probability.

[0036] The control method of the assembly equipment in the embodiment of the present application is described above. The control system of the assembly equipment in the embodiment of the present application is described below. Figure 2 In one embodiment of the present application, a control system for an assembly device includes: A construction module 201 is used to perform initialization scanning on the assembly equipment and construct a digital twin model to obtain an initial digital twin model; An analysis module 202 is used to analyze the control parameters of the initial digital twin model to obtain initial equipment control parameters; The prediction module 203 is used to obtain the assembly task operation data through the initial digital twin model and perform feature deviation prediction to obtain feature deviation prediction data; An optimization module 204 is used to establish a derived gradient matrix and perform parameter optimization according to the characteristic deviation prediction data to obtain target device control parameters; An updating module 205 is used to update the initial digital twin model according to the target device control parameters to obtain a target digital twin model; The diagnosis module 206 is used to perform operation data fault diagnosis through the target digital twin model to obtain the equipment fault type and equipment fault probability.

[0037] Through the collaborative cooperation of the above components, the digital twin model can reflect the changes in the status of assembly equipment and the operating environment in real time. The dynamic update function of the digital twin model allows the assembly equipment to automatically adjust the control parameters according to the actual working conditions, thereby enhancing the adaptability and flexibility of the equipment to complex working environments. This combination of real-time simulation and prediction significantly improves the response speed and efficiency of the control system. By accurately analyzing and optimizing the control parameters of the initial digital twin model, the operation of the equipment can be adjusted more accurately to adapt to specific assembly requirements. The use of reinforcement learning and particle swarm optimization algorithms to analyze and adjust the control parameters not only optimizes the efficiency of equipment operation, but also improves the accuracy and quality of the production process. Through comprehensive analysis of operating data and feature deviations, potential equipment failures can be effectively predicted and diagnosed. This prediction-based fault diagnosis system can detect and solve problems in a timely manner, reduce equipment downtime, reduce maintenance costs, and improve production continuity and equipment life. By monitoring and maintaining the operating status of assembly equipment in real time, losses caused by failures or improper operations can be effectively reduced. In addition, the optimized control parameters can ensure that the equipment operates in the best condition, thereby saving energy consumption and reducing unnecessary material waste.

[0038] The present application also provides a control device for an assembly device, the control device for the assembly device comprising a memory and a processor, the memory storing computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the control method for the assembly device in the above-mentioned embodiments.

[0039] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the control method of the assembly equipment.

[0040] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0041] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0042] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 embodiments of the present application.

Claims

1. A control method for assembly equipment, characterized in that: The control method of the assembly equipment comprises: Perform initial scanning and digital twin model construction on the assembly equipment to obtain an initial digital twin model; Performing control parameter analysis on the initial digital twin model to obtain initial equipment control parameters; Acquiring assembly task operation data through the initial digital twin model and performing feature deviation prediction to obtain feature deviation prediction data; Establishing a derived gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain target device control parameters; Updating the initial digital twin model according to the target device control parameters to obtain a target digital twin model; The target digital twin model is used to perform operation data fault diagnosis to obtain the equipment fault type and equipment failure probability.

2. The control method of assembly equipment according to claim 1, characterized in that: The initialization scanning of the assembly equipment and the construction of the digital twin model to obtain the initial digital twin model include: Collecting original image data of the assembly equipment, and performing denoising processing on the original image data to obtain denoised image data; Performing high-dimensional feature recognition on the denoised image data to obtain high-dimensional image features, and encoding the high-dimensional image features to obtain a feature encoding vector; Performing style migration on the feature encoding vector to obtain a migration feature vector; Reconstruct the three-dimensional model of the assembly equipment according to the migration feature vector to obtain a three-dimensional model of the equipment; The three-dimensional model of the equipment is verified from multiple angles and multiple environments to obtain an initial digital twin model.

3. The control method of assembly equipment according to claim 1, characterized in that: The control parameter analysis of the initial digital twin model to obtain initial equipment control parameters includes: Set the reward function of reinforcement learning and introduce the knowledge dissemination mechanism through meme theory; Initializing control parameters of the initial digital twin model by using a particle swarm algorithm to obtain a plurality of first device control parameters; Performing reward feedback calculation on the plurality of first device control parameters based on the reward function to obtain a reward feedback value of each first device control parameter; Performing parameter feedback adjustment on the plurality of first device control parameters according to the reward feedback value to obtain a plurality of second device control parameters; Determine a corresponding search direction and a search range based on the knowledge propagation mechanism, and perform parameter search on the plurality of second device control parameters according to the search direction and the search range to obtain a plurality of third device control parameters; Fitness calculation and parameter optimization are performed on the plurality of third device control parameters to obtain initial device control parameters.

4. The control method of assembly equipment according to claim 1, characterized in that: The obtaining of assembly task operation data through the initial digital twin model and performing feature deviation prediction to obtain feature deviation prediction data includes: Collecting assembly task operation data through the initial digital twin model; Extracting node and edge feature information from the assembly task operation data to obtain node and edge feature information; Performing a multi-layer attention hierarchy analysis on the feature information of the nodes and edges to obtain node and edge dependencies, and performing feature encoding on the feature information of the nodes and edges according to the node and edge dependencies to obtain a first operation feature; Performing time series analysis on the assembly task operation data to obtain a second operation feature; Perform feature deviation prediction on the first operation feature and the second operation feature to obtain feature deviation prediction data.

5. The control method of assembly equipment according to claim 1, characterized in that: The step of establishing a derived gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain target device control parameters includes: Calculating the influence gradient of each target parameter in the initial device control parameters according to the characteristic deviation prediction data; Construct the corresponding derived gradient matrix according to the influence gradient of each target parameter; Based on the derived gradient matrix, each target parameter in the initial device control parameters is optimized to obtain an optimization result of each target parameter; The optimization results of each target parameter are combined to obtain the target device control parameters.

6. The control method of assembly equipment according to claim 1, characterized in that: The updating of the initial digital twin model according to the target device control parameter to obtain the target digital twin model includes: Performing control simulation on the initial digital twin model based on the target device control parameters to obtain a simulated control data set, and obtaining a real control data set of the assembly device based on the target device control parameters; Performing a difference comparison analysis on the simulated control data set and the real control data set based on a Wasserstein distance algorithm to obtain a difference comparison analysis result; The initial digital twin model is updated according to the difference comparison analysis results to obtain a target digital twin model.

7. The control method of assembly equipment according to claim 1, characterized in that: The operating data fault diagnosis is performed through the target digital twin model to obtain the equipment fault type and equipment fault probability, including: The assembly task operation is performed through the target digital twin model to obtain the target assembly operation data; the specific process is: initialize the virtual environment of the target digital twin model, set parameters that match the actual assembly environment, and the parameters include environmental factors such as temperature, humidity, and vibration; convert the operation instructions of the actual assembly task into an instruction set that can be recognized by the digital twin model, and the instruction set includes action sequence and operation parameters; load the virtual parts and tool models required for the assembly task in the target digital twin model to ensure that their geometric features and physical properties are consistent with the actual ones; activate the virtual sensor system in the target digital twin model to simulate various sensors on the actual assembly equipment, including position sensors, force sensors, and visual sensors; start the virtual actuator in the target digital twin model to simulate the actions of the mechanical arm and fixture actuator of the actual assembly equipment; according to the assembly task instructions, in the target digital twin model Execute the virtual assembly process step by step, and record the motion trajectory, speed, and acceleration dynamic parameters of each virtual actuator in real time. During the virtual assembly process, use the physical engine of the target digital twin model to simulate the contact, friction, and collision physical interactions between parts, and record relevant data. Collect various data in the virtual assembly process in real time through the virtual sensor system, including position data, force data, and temperature data. Simulate and execute quality inspection procedures in the target digital twin model to evaluate the accuracy and stability indicators of the virtual assembly results. Record abnormal events in the virtual assembly process, such as deformation of virtual parts and over-limit motion of virtual actuators. Calculate and record various performance indicators of the virtual assembly process, including assembly time, energy consumption, and resource utilization, to form a complete target assembly operation data set, which includes time series data, event logs, and performance indicators. Performing cluster analysis on the target assembly operation data to obtain a plurality of assembly operation feature centers; Inputting the plurality of assembly operation feature centers into a preset multi-layer neural network model to perform operation data fault diagnosis to obtain equipment fault types; The failure probability of the assembly equipment is calculated according to the equipment failure type to obtain the equipment failure probability.

8. A control system for assembly equipment, characterized in that: The control system of the assembly equipment comprises: A construction module is used to perform initial scanning on the assembly equipment and build a digital twin model to obtain an initial digital twin model; An analysis module, used to perform control parameter analysis on the initial digital twin model to obtain initial device control parameters; A prediction module, used to obtain assembly task operation data through the initial digital twin model and perform feature deviation prediction to obtain feature deviation prediction data; An optimization module, used for establishing a derived gradient matrix and performing parameter optimization according to the characteristic deviation prediction data to obtain target device control parameters; An updating module, used to update the initial digital twin model according to the target device control parameters to obtain a target digital twin model; The diagnostic module is used to perform operation data fault diagnosis through the target digital twin model to obtain the equipment fault type and equipment fault probability.

9. A control device for an assembly device, characterized in that: The control device of the assembly device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the control device of the assembly device to execute the control method of the assembly device according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, a control method for an assembly device as described in any one of claims 1 to 7 is implemented.

Images