Robot automatic welding equipment control system and method
Through deep learning technology, the drum motion feature vector is extracted and the welding path trajectory is generated, which solves the problems of low welding efficiency and difficult quality of traditional drums, and efficient and accurate robot welding control is achieved, improving the stability of the production line and product quality.
Patent Information
- Application Number
- CN202411052081.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Relying on manual operations during traditional roller welding results in inefficient efficiency, difficult to control welding quality, and more difficult to weld large or heavy rollers, making welding problems prone to welding defects and unevenness.
The camera collects drum motion posture images and sensors to collect motion information, and uses deep learning technology to extract feature vectors to generate welding path trajectories of robot welding equipment, reduce manual dependence, and improve welding stability and consistency.
Efficient and precise welding control is achieved, the risk of human error is reduced, the stability of the production line and product quality are improved, and the aesthetics and overall quality of the welding results are ensured.
Smart Images

Figure CN118977236A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of welding equipment control, and more specifically, to a robot automatic welding equipment control system and method. Background Art
[0002] As a common component in mechanical equipment, the drum is usually composed of pipe fittings and end caps at both ends. In the traditional drum welding process, the end caps are usually spot welded on both ends of the pipe fittings first, and then the end caps are fully welded. In this process, the operator needs to constantly turn the drum to ensure that all parts of the end caps can be fully welded. This manual welding method not only consumes a lot of manpower, but also has low efficiency.
[0003] Especially for large or heavy rollers, manual welding is more difficult and time-consuming. Since the operation depends on manual experience and skills, the welding quality is often difficult to fully control, which may lead to uneven weld joints or unstable weld quality. In this case, the appearance of the welded roller is usually not beautiful, and there may be welding defects or unevenness, which is not conducive to the aesthetics and overall quality of the product.
[0004] Therefore, a robot automatic welding equipment control system and method are desired. Summary of the invention
[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a robot automatic welding equipment control system and method, which first obtains the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor, and then uses deep learning technology to perform feature extraction and correlation analysis on the two, and finally generates a welding path trajectory of the robot welding equipment through a generator to reduce dependence on operators, reduce the risk of human errors, and improve the stability and consistency of the production line.
[0006] According to one aspect of the present application, a robot automatic welding equipment control system is provided, which includes:
[0007] A robot roller welding data acquisition module, used to acquire roller motion posture images at multiple predetermined time points acquired by a camera and roller motion information at multiple predetermined time points acquired by a sensor;
[0008] A robot roller welding data extraction module, used to extract a roller motion posture global feature vector and optimize the roller motion information associated feature vector from the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor;
[0009] The welding path trajectory generation module is used to generate a welding path trajectory of a robot welding device based on the global feature vector of the roller motion posture and the associated feature vector of the optimized roller motion information.
[0010] According to another aspect of the present application, a method for controlling a robot automatic welding device is provided, comprising:
[0011] Acquire the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor;
[0012] Extracting a global feature vector of the roller motion posture and optimizing a feature vector associated with the roller motion information from the roller motion posture images at the plurality of predetermined time points collected by the camera and the roller motion information at the plurality of predetermined time points collected by the sensor;
[0013] Based on the global feature vector of the roller motion posture and the optimized roller motion information associated feature vector, a welding path trajectory of the robot welding equipment is generated.
[0014] Compared with the prior art, the present application provides a control system and method for robotic automatic welding equipment, which first obtains the roller motion posture images at multiple predetermined time points collected by a camera and the roller motion information at multiple predetermined time points collected by a sensor, and then uses deep learning technology to perform feature extraction and correlation analysis on the two, and finally uses a generator to generate a welding path trajectory of the robotic welding equipment to reduce dependence on operators, reduce the risk of human errors, and improve the stability and consistency of the production line. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.
[0016] Figure 1 It is a block diagram of a control system of a robot automatic welding equipment according to an embodiment of the present application.
[0017] Figure 2 It is a block diagram of a robot roller welding data extraction module in a robot automatic welding equipment control system according to an embodiment of the present application.
[0018] Figure 3 It is a block diagram of a drum motion posture image feature extraction unit in a robot automatic welding equipment control system according to an embodiment of the present application.
[0019] Figure 4 The block diagram is a welding path trajectory generation module in a robot automatic welding equipment control system according to an embodiment of the present application.
[0020] Figure 5 The present invention is a flowchart of a method for controlling robot automatic welding equipment according to an embodiment of the present application.
[0021] Figure 6 is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.
[0023] Figure 1 FIG. 1 is a block diagram of a control system for a robot automatic welding device according to an embodiment of the present application. Figure 1 As shown, according to the robot automatic welding equipment control system 100 of the embodiment of the present application, it includes: a robot roller welding data acquisition module 110, which is used to obtain the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor; a robot roller welding data extraction module 120, which is used to extract the roller motion posture global feature vector and the optimized roller motion information associated feature vector from the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor; a welding path trajectory generation module 130, which is used to generate the welding path trajectory of the robot welding equipment based on the roller motion posture global feature vector and the optimized roller motion information associated feature vector.
[0024] In the above-mentioned robot automatic welding equipment control system 100, the robot roller welding data acquisition module 110 is used to obtain the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor. It should be understood that in the traditional roller welding process, the end caps are usually spot welded first, the end caps are fixed to both ends of the pipe fitting, and then comprehensive welding is performed. This process requires the operator to repeatedly turn the roller to ensure that every part of the end cap can be fully welded. This manual operation is not only time-consuming and labor-intensive, but also inefficient. Especially when dealing with large or heavy rollers, manual welding becomes more difficult. Since the operation depends on manual experience and skills, the welding quality is often difficult to guarantee. This may result in uneven welding joints or unstable weld quality, making the appearance of the final welded roller not beautiful enough, and there may be welding defects or uneven surfaces, which is not conducive to the appearance and overall quality of the product. Therefore, in the technical solution of the present application, by acquiring the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor, and combining deep learning technology, by generating the welding path trajectory of the robotic welding equipment, it is possible to reduce dependence on operators, reduce the risk of human errors, and improve the stability and uniformity of the production line.
[0025] Specifically, the posture image captured by the camera can provide visual information, including the position, angle and motion trajectory of the roller in space. The motion information collected by the sensor includes dynamic parameters such as the speed and acceleration of the roller, which can quantify the motion characteristics of the roller and further ensure the stability and accuracy of the welding process. By combining visual data and dynamic parameters, the system can accurately analyze and identify the various motion states of the roller, provide precise control and path planning for the robotic welding equipment, thereby improving welding efficiency and quality, and ensuring that the final welding results meet the expected requirements.
[0026] In the above-mentioned robot automatic welding equipment control system 100, the robot roller welding data extraction module 120 is used to extract the roller motion posture global feature vector and optimize the roller motion information associated feature vector from the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor. It should be understood that the purpose of extracting the roller motion posture global feature vector and optimizing the roller motion information associated feature vector from the roller motion posture images and motion information at multiple predetermined time points collected by the camera and the sensor is to comprehensively understand and describe the motion state and characteristics of the roller through comprehensive analysis and extraction of key features. Combining the feature vectors of these two types of data, it is possible to achieve a global grasp and accurate description of the roller motion state. The global feature vector includes a variety of state characteristics from static to dynamic, such as posture stability, frequency of posture changes, speed range of motion, etc.; while the associated feature vector focuses more on linking various features of posture images and motion information to provide a more comprehensive motion characteristic analysis. Therefore, extracting feature vectors from data obtained by cameras and sensors is a key step in achieving efficient and accurate welding control, which can significantly improve production efficiency and product quality while ensuring safety and stability during the welding process.
[0027] Figure 2 FIG. 1 is a block diagram of a robot roller welding data extraction module in a robot automatic welding equipment control system according to an embodiment of the present application. Figure 2 As shown, in a specific embodiment of the present application, the robot roller welding data extraction module 120 includes: a roller motion posture image feature extraction unit 121, used to perform feature extraction on the roller motion posture images at multiple predetermined time points collected by the camera to obtain the global feature vector of the roller motion posture; a roller motion information semantic encoding unit 122, used to perform semantic encoding on the roller motion information at multiple predetermined time points collected by the sensor to obtain multiple roller motion information feature vectors; a roller motion information feature encoding unit 123, used to perform feature encoding on the multiple roller motion information feature vectors to obtain the optimized roller motion information associated feature vector.
[0028] It should be understood that through algorithms and models, key visual features that affect the movement of the drum are extracted from the image at each time point. The feature extraction process usually includes analyzing the geometric features such as the position, posture angle, contour shape, and visual features such as color and texture from the image. These features can be extracted and quantified through various mathematical methods and image processing techniques, such as edge detection, region segmentation, feature descriptor extraction, etc. Through these processes, the image at each time point can be converted into a feature vector, which contains the key information of the drum posture and movement state expressed by the image.
[0029] Furthermore, the purpose of semantically encoding the drum motion information collected by the sensor at multiple predetermined time points is to convert the original dynamic data into a high-level feature representation with semantics and practical significance, so that the system can better understand and utilize these data. The motion information collected by the sensor includes physical quantities such as the speed, acceleration, and angular velocity of the drum, which directly reflect the real-time motion state and dynamic characteristics of the drum in space. Among them, the semantic encoding process usually involves technologies such as signal processing, feature extraction, and pattern recognition. First, through technologies such as data preprocessing and filtering, the noise and interference in the sensor data are eliminated or reduced to ensure the accuracy and reliability of the data. Next, an appropriate feature extraction algorithm is used, such as time domain feature extraction, frequency domain analysis, or feature learning methods based on machine learning, to extract representative and discriminative features from the original time series data.
[0030] Furthermore, feature encoding of multiple drum motion information feature vectors can further refine and integrate data, thereby revealing deeper associations and patterns of drum motion. Multiple drum motion information feature vectors are high-level feature representations extracted from time series data collected by sensors, and each vector contains an abstract description of the motion state of the drum at different time points. The feature encoding process involves data dimension reduction, feature selection, pattern recognition and other technologies. First, through dimension reduction techniques such as principal component analysis (PCA) or feature selection algorithms, the high-dimensional feature space is compressed to a lower-dimensional representation that retains the most important information. This helps to reduce the complexity of the data while retaining the key features in the data. Secondly, using pattern recognition and cluster analysis methods, potential associations and patterns between drum motion information feature vectors are identified, such as the periodicity of motion, the transition between different motion modes, etc.
[0031] Figure 3 FIG. 1 is a block diagram of a drum motion posture image feature extraction unit in a robot automatic welding equipment control system according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the drum motion posture image feature extraction unit 121 includes: a drum motion posture image embedding coding subunit 1211, which is used to embed the drum motion posture images of multiple predetermined time points collected by the camera through drum motion posture to obtain a sequence of drum motion posture image embedding vectors; a drum motion posture feature coding subunit 1212, which is used to feature encode the sequence of drum motion posture image embedding vectors to obtain the drum motion posture global feature vector.
[0032] It should be understood that the drum motion posture image embedding coding usually uses convolutional neural networks (CNN) or other deep learning models to extract multi-level feature expressions from the original image data through multi-level convolution and pooling operations. These features can be geometric shapes, texture information, color distribution, etc. in the image. Through learning and training, the model can effectively map these features into a high-dimensional embedding vector space. The sequence of drum motion posture image embedding vectors can not only express the drum posture characteristics at a single time point, but also capture the dynamic changes and trends during the drum motion process. This enables the system to continuously monitor and analyze the overall characteristics of the drum motion, thereby achieving refined control and prediction of the motion state. For example, in an automated welding system, these embedded vectors can be used to detect and correct posture deviations, optimize welding path planning, and support the implementation of adaptive control strategies. The embedded vector sequence obtained by the drum motion posture image embedding coding provides a new way of image understanding and data processing for intelligent systems. Through efficient feature extraction and encoding, the system's perception of motion posture changes is effectively enhanced, thereby improving the accuracy and reliability of equipment operation, while optimizing efficiency and quality management in the production process.
[0033] Furthermore, the purpose of feature encoding the sequence of drum motion posture image embedding vectors is to convert the continuous, high-dimensional image embedding vector sequence into a more compact and representative global feature representation so that the system can more effectively describe and analyze the overall motion posture of the drum. This process involves integrating and summarizing complex sequence data to extract the most discriminative and information-rich features to achieve a comprehensive evaluation and understanding of the drum motion state. Preliminary feature extraction of the embedding vector sequence is performed through statistical methods such as mean, variance and other descriptive statistics to capture the overall trend and changes in the sequence. Secondly, clustering analysis methods such as k-means clustering are used to group the embedding vector sequence and extract representative feature vectors for each group, thereby reducing the complexity of the data while retaining important feature information. On the other hand, deep learning techniques such as long short-term memory networks (LSTM) or attention mechanisms can more finely model the time dependency and dynamic changes in the sequence, and extract more representative global feature vectors by learning the intrinsic structure and pattern in the sequence. These global feature vectors not only contain the local posture information at each time point, but also reflect important features such as motion trends, periodicity and mutation points in the entire motion sequence.
[0034] In a specific embodiment of the present application, the roller motion posture feature encoding subunit 1212 includes: passing the sequence of roller motion posture image embedding vectors through a converter-based roller motion posture ViT model to obtain a roller motion posture global feature matrix; passing the roller motion posture global feature matrix through a roller motion posture feature encoder to obtain the roller motion posture global feature vector.
[0035] It should be understood that the core of the ViT model is to divide the image into small blocks, and then learn the relationship and global features between the blocks through a multi-layer transformer structure. For the drum motion posture image sequence, it is first necessary to reorganize the embedding vector sequence of each time point into a form suitable for the ViT model input. This may involve the appropriate encoding and integration of the time dimension and the spatial dimension to ensure that the model can effectively capture the dynamic features and time dependencies in the image sequence. When processing the drum motion posture image sequence, the ViT model can use its self-attention mechanism to simultaneously focus on the information at different time points in the image sequence, and effectively capture the long-term dependencies and local patterns in the sequence through the multi-head attention mechanism. This ability enables the ViT model to more comprehensively understand and extract the global dynamic features in the image sequence, such as the periodicity of the motion, the trend of change, and possible abnormal patterns. The drum motion posture global feature matrix obtained by the transformer-based drum motion posture ViT model will contain the global feature representations of each time point, which not only retains the local information of each time point, but also integrates the context and global structure of the entire sequence. This global feature matrix can be directly used in more advanced motion prediction, dynamic posture classification, and anomaly detection tasks, providing more accurate and reliable motion analysis and control capabilities for intelligent systems.
[0036] Furthermore, the global feature matrix of the drum motion posture is usually a high-dimensional data structure, and each row or column may represent information at different time points or different feature dimensions. Through the drum motion posture feature encoder, these complex matrix data can be mapped to a vector space with lower dimensions and retaining key features through various mathematical and statistical methods, such as linear transformation, nonlinear mapping or deep neural network. Among them, the selection of feature encoder depends on the characteristics of the data and the needs of the application. For example, for time series data, recursive neural network (RNN) or long short-term memory network (LSTM) can be used to process time dependency and dynamic features, thereby generating a global feature vector with sequence properties. For static image data, convolutional neural network (CNN) or self-attention mechanism and other methods can be used to effectively integrate spatial information into the global feature vector. Specifically, each layer of the drum motion posture feature encoder is used to perform convolution processing, mean pooling processing based on the local feature matrix and nonlinear activation processing on the input data in the forward transfer of the layer to output the drum motion posture global feature vector by the last layer of the drum motion posture feature encoder, wherein the input of the drum motion posture feature encoder is the drum motion posture global feature matrix.
[0037] In a specific embodiment of the present application, the drum motion information semantic encoding unit 122 includes: performing data preprocessing on the drum motion information at multiple predetermined time points collected by the sensor to obtain a drum motion information input sequence; passing the drum motion information input sequence through an LSTM-based drum motion information semantic encoder to obtain the multiple drum motion information feature vectors.
[0038] It should be understood that data preprocessing includes quality control and outlier detection of the drum motion information collected by the sensor. This means checking whether the data has missing values, duplicate values or abnormal values, such as extreme values outside the predetermined range, and handling these abnormal situations to ensure the integrity and reliability of the data. Preprocessing also includes feature extraction and data reconstruction, which converts the raw data collected by the sensor into a more informative and analyzable form. For drum motion information, it may be necessary to extract key features such as motion speed, acceleration, position, angle, etc. from the raw data, and integrate and organize these features according to the predetermined time points and data formats. Normalization and standardization are also important steps in data preprocessing, which are used to eliminate differences in measurement units between different sensors or differences in scale between data, and ensure that each feature is within the same numerical range. Doing so helps to improve the stability and efficiency of data analysis and model training, and avoid errors or deviations caused by different data scales. Data preprocessing may also include data smoothing and noise reduction, which are used to reduce random noise or oscillations in the sensor collected data, making the data more stable and reliable. These techniques can be implemented based on rolling averages, filters or other signal processing methods to ensure that the extracted motion information has good continuity and consistency.
[0039] Furthermore, the drum motion information input sequence is converted into multiple drum motion information feature vectors through a drum motion information semantic encoder based on a long short-term memory network (LSTM) in order to better capture and understand the time series characteristics of the drum motion, thereby providing a more accurate and comprehensive data representation. As a recursive neural network structure, LSTM is particularly suitable for processing time series data and can effectively model long-term dependencies and dynamic changes in data, so it has significant advantages in processing drum motion information. The LSTM encoder can effectively capture the time dependency in the drum motion information sequence through its memory unit and gating mechanism. This means that when processing continuous time point data collected by sensors, LSTM can learn and memorize the complex relationship between historical data points, including dynamic features such as changes in motion speed, fluctuations in acceleration, and rotation of angles. Through semantic encoding by LSTM, the drum motion information sequence can be converted into a highly abstract feature vector representation. These feature vectors not only contain the local feature information of each time point, but also comprehensively consider the global dynamic characteristics of the entire sequence, such as periodic changes, long-term trends, and emergencies. This global feature vector is very critical for subsequent data analysis and model application, and can provide a more comprehensive and comprehensive data expression.
[0040] In a specific embodiment of the present application, the drum motion information feature encoding unit 123 includes: passing the multiple drum motion information feature vectors through a drum motion information feature encoder based on an attention mechanism to obtain multiple drum motion information attention feature vectors; integrating the multiple drum motion information attention feature vectors to obtain a drum motion information association feature vector; based on the drum motion posture global feature vector, performing inverse error perception based on feature multi-viewpoint quantization on the drum motion information association feature vector to obtain an optimized drum motion information association feature vector.
[0041] It should be understood that the drum motion information feature encoder based on the attention mechanism can dynamically adjust the importance of each feature vector in the entire sequence. This dynamic adjustment is achieved by assigning different attention weights to each feature vector, so that the model can focus on those features that are more critical and have a decisive impact on the current task or problem. The attention feature encoder can help reduce the impact of irrelevant or noisy information in the feature vector on the final result, thereby improving the data representation accuracy and the generalization ability of the model. It strengthens the representation of important features while suppressing the influence of irrelevant features, making the final feature vector more compact and effective. The drum motion information feature encoder based on the attention mechanism can handle the relationship between features at different levels and granularities through a combination of multi-head attention, self-attention or other attention mechanisms. For example, for time series data, the attention weight of the feature vector at each time point can be adjusted according to the specific context and task requirements, so as to better explain and utilize time-dependent information during model training and prediction. Specifically, the convolutional encoding part of the drum motion information feature encoder based on the attention mechanism is used to perform deep convolutional encoding on the multiple drum motion information feature vectors to obtain an initial convolutional feature map; the initial convolutional feature map is input into the spatial attention part of the drum motion information feature encoder based on the attention mechanism to obtain a spatial attention map; the spatial attention map is activated by the Softmax function to obtain a spatial attention feature map; the spatial attention feature map and the initial convolutional feature map are multiplied by the position points to obtain the multiple drum motion information attention feature vectors. More specifically, the initial convolutional feature map is input into the spatial attention part of the drum motion information feature encoder based on the attention mechanism to obtain the spatial attention map, including: performing average pooling and maximum pooling along the channel dimension on the initial convolutional feature map to obtain an average feature matrix and a maximum feature matrix; cascading and channel-adjusting the average feature matrix and the maximum feature matrix to obtain a channel feature matrix; using the convolutional layer of the spatial attention feature map to perform convolutional encoding on the channel feature matrix to obtain a spatial attention map.
[0042] Furthermore, integrating multiple attention feature vectors of drum motion information to obtain drum motion information association feature vectors is to synthesize and summarize the key information captured under different attention mechanisms, and further enhance the comprehensive representation ability of data and the comprehensive decision-making ability of the model. In this process, feature vectors from different sources and dimensions are effectively combined to reflect the diversity and complexity of data. The integration process utilizes the complementarity and interaction of feature vectors captured under multiple attention mechanisms. Different attention mechanisms may focus on different aspects or time points in the data. By integrating these feature vectors, important information from all aspects can be comprehensively considered, thereby improving the comprehensive understanding and representation of the overall characteristics of the data. Integrating multiple attention feature vectors helps to reduce the possible bias or shortcomings of a single feature vector. Each attention mechanism may perform better under specific conditions. By integrating these advantages, the limitations of a single model in terms of expression ability and application scope can be compensated, thereby obtaining a more comprehensive and multi-dimensional data feature representation. Among them, integrating multiple attention feature vectors of drum motion information can also enhance the robustness and generalization ability of the model. By fusing feature vectors from different sources and types, the errors or uncertainties that the model may encounter in processing new data or complex environments can be reduced, so that the model can show better performance and reliability in various application scenarios.
[0043] Furthermore, in the technical solution of the present application, it is considered that the global feature vector of the drum motion posture is extracted based on visual image data, while the drum motion information association feature vector is based on motion data collected by the sensor. These two data sources may capture different aspects of the drum motion, resulting in differences in feature expression. The drum motion posture feature is extracted by the converter-based ViT model, focusing on the visual features in the image; while the drum motion information feature is extracted by the encoder based on LSTM and attention mechanism, focusing on the dynamic features of time series data. These two different feature extraction methods may lead to semantic differences in the feature vectors. The drum motion information collected by the sensor is time series data, reflecting the motion state of the drum over time; while the drum motion posture image collected by the camera is static and may not fully reflect the dynamic changes in the time series. When fusing the global feature vector of the drum motion posture and the drum motion information association feature vector, if the fusion strategy does not fully consider the temporal and spatial characteristics of the two, it may cause the drum motion information association feature vector to have a posterior-prior causal association missing relative to the global feature vector of the drum motion posture. In order to solve this problem, in the technical solution of the present application, based on the global feature vector of the drum motion posture, the feature vector associated with the drum motion information is subjected to inverse error perception based on feature multi-viewpoint quantization.
[0044] Specifically, based on the global feature vector of the drum motion posture, the drum motion information associated feature vector is subjected to inverse error perception based on feature multi-viewpoint quantization to obtain an optimized drum motion information associated feature vector, including: calculating the position difference between the drum motion information associated feature vector and the global feature vector of the drum motion posture to obtain a first viewpoint measurement representation vector; calculating the position point multiplication between the drum motion information associated feature vector and the global feature vector of the drum motion posture to obtain a second viewpoint measurement representation vector; calculating the position sum between the drum motion information associated feature vector and the global feature vector of the drum motion posture to obtain a third viewpoint measurement Representation vector; calculate the gradient amplitude of each position in the first viewpoint measurement representation vector, the second viewpoint measurement representation vector and the third viewpoint measurement representation vector to obtain the first viewpoint measurement perception vector, the second viewpoint measurement perception vector and the third viewpoint measurement perception vector; input the first viewpoint measurement perception vector, the second viewpoint measurement perception vector and the third viewpoint measurement perception vector into the backward machine learning reverse multi-viewpoint perception optimization module to obtain the reverse multi-viewpoint perception optimization representation vector; calculate the position point multiplication between the reverse multi-viewpoint perception optimization representation vector and the drum motion information associated feature vector to obtain the optimized drum motion information associated feature vector.
[0045] The backward machine learning reverse multi-view perception optimization module is expressed as follows:
[0046]
[0047] in, and represents the eigenvalues of the i-th position of the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector, respectively, and μ represents the three eigenvalues of the corresponding positions in the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector , σ is the standard deviation of the three eigenvalues at corresponding positions in the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector, represents the maximum value of the three eigenvalues of the corresponding positions in the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector, ReLU represents a linear rectification function, softmax represents a normalized exponential function, and v ri Represents the eigenvalue of the i-th position of the inverse multi-view perceptual optimization representation vector.
[0048] Specifically, in the technical solution of the present application, based on the global feature vector of the drum motion posture, the drum motion information associated feature vector is optimized by reverse error perception based on feature multi-viewpoint quantization, and the process includes first calculating the position difference, dot multiplication and sum between the drum motion information associated feature vector and the global feature vector of the drum motion posture through multi-view analysis (Multi-View Analysis) to obtain three viewpoint measurement representation vectors, and quantify the relationship between features from different angles. Then, the gradient amplitude of each position in these viewpoint measurement representation vectors is calculated through reverse error perception (Backward Error Perception) to identify key change points to provide a basis for feature optimization.
[0049] Then, the first viewpoint measurement perception vector, the second viewpoint measurement perception vector and the third viewpoint measurement perception vector are input into the backward machine learning reverse multi-viewpoint perception optimization module, wherein the backward machine learning reverse multi-viewpoint perception optimization module captures the significant comprehensive ability factor of the feature importance perception (Feature Importance Perception) of the feature sub-dimension distribution of the corresponding positions in the first to third viewpoint measurement perception vectors by adopting a squeeze-incentive mechanism composed of -functions. Then, the position point multiplication between the reverse multi-viewpoint perception optimization representation vector and the drum motion information associated feature vector is calculated to perform backward optimization perception (Backward Optimization Perception) on the drum motion information associated feature vector to adjust the drum motion information associated feature vector so as to better associate with the drum motion posture global feature vector, thereby enhancing the causal relationship between the drum motion information associated feature vector and the drum motion posture global feature vector.
[0050] In the above-mentioned robot automatic welding equipment control system 100, the welding path trajectory generation module 130 is used to generate the welding path trajectory of the robot welding equipment based on the global feature vector of the roller motion posture and the optimized roller motion information associated feature vector. It should be understood that the global feature vector of the roller motion posture contains the position, angle and posture information of the robot in space, which is crucial to ensure the accuracy and stability of the welding process. These global feature vectors can guide the position adjustment and posture control of the robot throughout the welding process to ensure the correctness and consistency of the welding path in three-dimensional space. The optimized roller motion information associated feature vector provides a comprehensive representation of various motion information in the welding process by integrating different time points and key dynamic features. These feature vectors include not only basic information such as motion speed and acceleration, but may also cover advanced features such as curve optimization of motion trajectory and dynamic adjustment of path planning, thereby providing a basis for fine control and optimization of welding paths. In this way, the energy-saving efficiency and time cost of the welding path can be optimized, and the welding efficiency and quality can be maximized by accurately controlling the motion speed and posture adjustment, and the adaptive ability of the welding process is enhanced. The welding path is dynamically adjusted according to real-time sensor data and environmental feedback to cope with changes in complex environments and workpiece morphology.
[0051] Figure 4 FIG. 1 is a block diagram of a welding path trajectory generation module in a robot automatic welding equipment control system according to an embodiment of the present application. Figure 4 As shown, in a specific embodiment of the present application, the welding path trajectory generation module 130 includes: a roller motion data feature fusion unit 131, which is used to fuse the roller motion posture global feature vector and the optimized roller motion information associated feature vector to obtain a welding path trajectory generation feature vector; a robot welding trajectory generation unit 132, which is used to pass the welding path trajectory generation feature vector through a generator to generate a welding path trajectory for a robot welding device.
[0052] It should be understood that by fusing these two types of feature vectors, the following key goals can be achieved: First, ensure the continuity and smoothness of the welding path, and avoid sharp angle turns or excessive movements in path planning by comprehensively considering the robot's posture and dynamic motion characteristics; second, optimize the efficiency and quality of the welding path, and maximize the speed and accuracy of the welding process through precise posture adjustment and dynamic path planning; third, enhance the adaptability and intelligence level of welding operations, and dynamically adjust the generated path trajectory according to real-time feedback and environmental changes to adapt to different workpiece forms and production requirements. Fusion of feature vectors to generate welding path trajectories can also support autonomous decision-making and automatic control of the robot system. This method not only improves the automation level of welding equipment and the accuracy of operation, but also maintains consistency and repeatability in complex production environments, reduces human intervention and error risks, and significantly improves the efficiency and cost-effectiveness of the welding process.
[0053] Furthermore, the welding path trajectory is generated by the generator from the characteristic vector of the welding path trajectory, so as to realize the automatic and precisely controlled welding path planning by effectively utilizing the key information and parameters contained in the characteristic vector. As a key computing module, the generator can generate a welding path trajectory that meets the requirements according to the input characteristic vector and the preset rules and algorithms. In this process, the generator dynamically adjusts the generated path by parsing the spatial posture, dynamic motion information and other relevant parameters in the characteristic vector to ensure the continuity and smoothness of the path in three-dimensional space. The key goal of welding path trajectory generation is to optimize the efficiency and quality of the welding process. The generator can intelligently adjust the speed, processing sequence and welding angle of the welding path according to specific production requirements and workpiece requirements to maximize the accuracy and consistency of welding. This precise control capability is particularly important in large-scale production, which can significantly reduce production costs and manpower input. The generation of welding path trajectories by the generator can also realize the automation and intelligence of welding operations. The generator can autonomously adjust and optimize the welding path according to the preset optimization strategy and real-time feedback to adapt to the changes in the morphology of different workpieces and complex environments, thereby improving the autonomous decision-making ability of the robot system and its ability to cope with complex tasks.
[0054] In summary, the purpose of generating the welding path trajectory of the robot welding equipment by generating the welding path trajectory of the welding path trajectory through the generator is to achieve efficient, accurate and intelligent welding operation. This method not only improves production efficiency and product quality, but also promotes the transformation and progress of the manufacturing industry towards intelligent manufacturing and automated production.
[0055] In summary, the embodiment of the present application first obtains the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor, and then uses deep learning technology to perform feature extraction and correlation analysis on the two. Finally, a generator is used to generate the welding path trajectory of the robot welding equipment to reduce dependence on the operator, reduce the risk of human error, and improve the stability and consistency of the production line.
[0056] As described above, the robot automatic welding equipment control system 100 according to the embodiment of the present application can be implemented in various terminal devices. In one example, the robot automatic welding equipment control system 100 can be integrated into the terminal device as a software module and / or a hardware module. For example, the robot automatic welding equipment control system 100 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the robot automatic welding equipment control system 100 can also be one of the many hardware modules of the terminal device.
[0057] Alternatively, in another example, the robot automatic welding equipment control system 100 and the terminal device may also be separate devices, and the robot automatic welding equipment control system 100 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.
[0058] Figure 5 FIG. 1 is a flow chart of a method for controlling a robot automatic welding device according to an embodiment of the present application. Figure 5 As shown, according to the robot automatic welding equipment control method of the embodiment of the present application, it includes: S110, acquiring the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor; S120, extracting the roller motion posture global feature vector and the optimized roller motion information associated feature vector from the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor; S130, generating the welding path trajectory of the robot welding equipment based on the roller motion posture global feature vector and the optimized roller motion information associated feature vector.
[0059] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned robot automatic welding equipment control method have been referred to above. Figures 1 to 4 The description of the robot automatic welding equipment control system has been described in detail, and therefore, its repeated description will be omitted.
[0060] Below, reference Figure 6 To describe an electronic device according to an embodiment of the present application.
[0061] Figure 6 A block diagram of an electronic device according to an embodiment of the present application is illustrated.
[0062] like Figure 6 As shown, the electronic device 10 includes an input device 11, an input interface 12, a central processing unit 13, a memory 14, an output interface 15, an output device 16, and a bus 17. The input interface 12, the central processing unit 13, the memory 14, and the output interface 15 are connected to each other through the bus 17, and the input device 11 and the output device 16 are connected to the bus 17 through the input interface 14 and the output interface 15, respectively, and then connected to other components of the electronic device 10.
[0063] Specifically, the input device 11 receives input information from the outside and transmits the input information to the central processing unit 13 through the input interface 12; the central processing unit 13 processes the input information based on the computer executable instructions stored in the memory 14 to generate output information, temporarily or permanently stores the output information in the memory 14, and then transmits the output information to the output device 16 through the output interface 15; the output device 16 outputs the output information to the outside of the electronic device 10 for user use.
[0064] In one embodiment, Figure 6 The electronic device 10 shown can be implemented as a network device, which may include: a memory configured to store programs; a processor configured to run the programs stored in the memory to execute any one of the robot automatic welding equipment control methods described in the above embodiments.
[0065] According to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network, and / or installed from a removable storage medium.
[0066] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.
Claims
1. A robot automatic welding equipment control system, characterized in that: include: A robot roller welding data acquisition module, used to acquire roller motion posture images at multiple predetermined time points acquired by a camera and roller motion information at multiple predetermined time points acquired by a sensor; A robot roller welding data extraction module, used to extract a roller motion posture global feature vector and optimize the roller motion information associated feature vector from the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor; The welding path trajectory generation module is used to generate a welding path trajectory of a robot welding device based on the global feature vector of the roller motion posture and the associated feature vector of the optimized roller motion information.
2. The robot automatic welding equipment control system according to claim 1, characterized in that: The robot roller welding data extraction module comprises: A drum motion posture image feature extraction unit, used for extracting features from the drum motion posture images at a plurality of predetermined time points collected by the camera to obtain a global feature vector of the drum motion posture; A drum motion information semantic encoding unit, used for semantically encoding the drum motion information at a plurality of predetermined time points collected by the sensor to obtain a plurality of drum motion information feature vectors; The drum motion information feature encoding unit is used to perform feature encoding on the multiple drum motion information feature vectors to obtain the optimized drum motion information associated feature vector.
3. The robot automatic welding equipment control system according to claim 2 is characterized in that: The drum motion posture image feature extraction unit comprises: A drum motion posture image embedding coding subunit, used for performing drum motion posture embedding coding on the drum motion posture images at a plurality of predetermined time points acquired by the camera to obtain a sequence of drum motion posture image embedding vectors; The drum motion posture feature encoding subunit is used to perform feature encoding on the sequence of the drum motion posture image embedding vectors to obtain the drum motion posture global feature vector.
4. The robot automatic welding equipment control system according to claim 3 is characterized in that: The roller motion posture feature encoding subunit comprises: The sequence of the roller motion posture image embedding vectors is passed through a converter-based roller motion posture ViT model to obtain a roller motion posture global feature matrix; The global feature matrix of the roller motion posture is passed through a roller motion posture feature encoder to obtain the global feature vector of the roller motion posture.
5. The robot automatic welding equipment control system according to claim 4, characterized in that: The drum motion information semantic encoding unit comprises: Performing data preprocessing on the roller motion information collected by the sensor at a plurality of predetermined time points to obtain a roller motion information input sequence; The drum motion information input sequence is passed through a drum motion information semantic encoder based on LSTM to obtain the multiple drum motion information feature vectors.
6. The robot automatic welding equipment control system according to claim 5, characterized in that: The drum motion information feature encoding unit comprises: Passing the plurality of roller motion information feature vectors through a roller motion information feature encoder based on an attention mechanism to obtain a plurality of roller motion information attention feature vectors; integrating the plurality of roller motion information attention feature vectors to obtain a roller motion information association feature vector; Based on the global feature vector of the drum motion posture, the drum motion information associated feature vector is subjected to inverse error perception based on feature multi-viewpoint quantization to obtain the optimized drum motion information associated feature vector.
7. The robot automatic welding equipment control system according to claim 6, characterized in that: Based on the global feature vector of the drum motion posture, the drum motion information associated feature vector is subjected to inverse error perception based on feature multi-viewpoint quantization to obtain the optimized drum motion information associated feature vector, including: Calculating the position difference between the drum motion information associated feature vector and the drum motion posture global feature vector to obtain a first viewpoint metric representation vector; Calculate the position point multiplication between the drum motion information associated feature vector and the drum motion posture global feature vector to obtain a second viewpoint metric representation vector; Calculating the positional sum of the drum motion information associated feature vector and the drum motion posture global feature vector to obtain a third viewpoint metric representation vector; Calculating the gradient magnitude of each position in the first viewpoint metric representation vector, the second viewpoint metric representation vector, and the third viewpoint metric representation vector to obtain a first viewpoint metric perception vector, a second viewpoint metric perception vector, and a third viewpoint metric perception vector; Inputting the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector into a backward machine learning inverse multi-view perception optimization module to obtain a reverse multi-view perception optimization representation vector; The optimized drum motion information associated feature vector is obtained by calculating the position point multiplication between the reverse multi-view perception optimization representation vector and the drum motion information associated feature vector.
8. The robot automatic welding equipment control system according to claim 7, characterized in that: The welding path trajectory generating module comprises: A drum motion data feature fusion unit, used for fusing the drum motion posture global feature vector and the optimized drum motion information associated feature vector to obtain a welding path trajectory generation feature vector; The robot welding trajectory generating unit is used to generate the welding path trajectory of the robot welding device by passing the welding path trajectory generating feature vector through a generator.
9. A robot automatic welding equipment control method, characterized in that: include: Acquire the roller motion posture images at multiple predetermined time points collected by the camera and the roller motion information at multiple predetermined time points collected by the sensor; Extracting a global feature vector of the roller motion posture and optimizing a feature vector associated with the roller motion information from the roller motion posture images at the plurality of predetermined time points collected by the camera and the roller motion information at the plurality of predetermined time points collected by the sensor; Based on the global feature vector of the roller motion posture and the optimized roller motion information associated feature vector, a welding path trajectory of the robot welding equipment is generated.
10. The robot automatic welding equipment control method according to claim 9, characterized in that: Extracting a global feature vector of the roller motion posture and optimizing a feature vector associated with the roller motion information from the roller motion posture images at the plurality of predetermined time points collected by the camera and the roller motion information at the plurality of predetermined time points collected by the sensor, including: Performing feature extraction on the roller motion posture images at a plurality of predetermined time points collected by the camera to obtain a global feature vector of the roller motion posture; semantically encoding the roller motion information at a plurality of predetermined time points collected by the sensor to obtain a plurality of roller motion information feature vectors; The plurality of drum motion information feature vectors are feature encoded to obtain the optimized drum motion information associated feature vector.
Citation Information
Cited By
Welding path real-time planning system based on deep learning
CN120422220A
