Intelligent processing system and control method based on artificial intelligence vision technology

Abstract

The intelligent processing system based on artificial intelligence vision technology of the present invention and the control method thereof are as follows: step 1, reading a local configuration file; step 2, obtaining user setting parameters; step 3, judging whether it is the first processing, jumping to step 4; judging not, jumping to step 9; step 4, initializing the configuration file; step 5, starting the target recognition module; step 6, judging whether the workpiece to be processed is consistent with the preset workpiece type; step 7, alarming; step 8, starting the stereoscopic vision detection module; step 9, calling the local configuration file, executing the processing module; step 10, executing the quality inspection module; step 11, judging whether the quality inspection is qualified, jumping to step 12; judging whether it is unqualified, jumping to step 13; step 12, saving the configuration information locally; step 13, executing the compensation module. The present invention ensures the processing quality and efficiency of small batch and multi-variety products.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial automation and intelligent manufacturing, and in particular to an intelligent processing system based on artificial intelligence vision technology and a control method thereof. Background Art

[0002] With technological advancements, the manufacturing industry is moving toward intelligent and automated processes. Traditional machining systems, which mostly rely on fixed processing procedures and preset parameter configurations, cannot meet the demands of processing high-variety, small-batch, and complex parts. Furthermore, automated machining systems based on traditional vision technology still have shortcomings in target recognition, positioning accuracy, and dynamic parameter adjustment, making it difficult to balance machining efficiency and quality.

[0003] With the rapid development of artificial intelligence (AI), especially deep learning algorithms, computer vision is gradually being applied to industrial processing. Technologies such as target recognition, 3D scanning and modeling, and path planning can significantly improve the level of intelligent processing. However, these systems still face the following challenges in practical applications:

[0004] (1) Insufficient processing complexity: When processing parts with complex surfaces or multiple degrees of freedom, the robot is prone to cumulative processing errors due to its own dynamic errors and insufficient path planning accuracy, making it difficult to meet the processing requirements of complex parts.

[0005] (2) Lack of flexibility in processing parameter configuration: usually requires manual preset or adjustment based on experience, which makes it difficult to meet the needs of dynamic environments and different materials.

[0006] (3) The existing processing system lacks dynamic monitoring and closed-loop feedback mechanisms for the entire processing process, which can easily lead to processing deviations or quality problems.

[0007] (4) The contradiction between cost and flexible production: Although CNC machine tools have high processing precision, their high equipment costs and long debugging time lead to high production costs in small-batch, high-variety production scenarios. Robots have an advantage in flexible production due to their lower cost and higher flexibility, but their lack of precision limits their widespread application.

[0008] To address these challenges, an intelligent machining system based on robotics, artificial intelligence, and vision technology is urgently needed. By combining the flexibility of robotics with the precise control of AI vision, this system enables accurate workpiece identification and positioning, dynamic parameter configuration, and closed-loop control of the machining process. This significantly improves machining accuracy and efficiency, meeting the modern manufacturing industry's demand for flexible production and intelligent machining. This system can reduce costs, improve production efficiency, and adapt to a wider range of production scenarios while ensuring machining quality. Summary of the Invention

[0009] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0010] To this end, the present invention proposes an intelligent processing system and its control method based on artificial intelligence vision technology to ensure the processing quality and efficiency of small batch and multi-variety products, and is suitable for flexible manufacturing scenarios of multi-variety, small batch and complex parts.

[0011] According to an embodiment of the present invention, a control method for intelligent processing based on artificial intelligence vision technology includes the following steps:

[0012] Step 1: Read the local configuration file, parse the basic parameters, and display the parsing results through a visual interface;

[0013] Step 2: Obtain the user's settings and adjustments to the parsed basic parameters, and update the corresponding configuration information according to the adjusted parameters;

[0014] Step 3: Determine whether it is the first processing:

[0015] If it is determined to be the first processing, the automatic processing parameters are obtained and the process goes to step 4;

[0016] If it is determined that this is not the first processing, jump to step 9;

[0017] Step 4: Initialize the configuration file;

[0018] Step 5: 3D model analysis of workpiece drawings and start the target recognition module;

[0019] Step 6: Determine whether the workpiece to be processed is consistent with the preset workpiece type:

[0020] If it is determined that the workpiece to be processed is the same as the preset workpiece type, jump to step 8;

[0021] If it is determined that the workpiece to be processed is inconsistent with the preset workpiece type, jump to step 7;

[0022] Step 7: Alarm prompt, manual intervention inspection;

[0023] Step 8: Start the stereo vision inspection module, scan and calculate the spatial posture of the workpiece in sequence according to the initial target point; calculate the workpiece parameters based on the 3D model and the spatial posture; automatically generate the processing plan according to the processing process logic; update the processing plan information in the configuration file of the workpiece and save it locally;

[0024] Step 9: Call the local configuration file, update the dynamic configuration information, and execute the processing module according to the configuration information;

[0025] Step 10: After processing is completed, execute the quality inspection module;

[0026] Step 11: Determine whether the quality inspection is qualified:

[0027] If the quality inspection is qualified, jump to step 12;

[0028] If the quality inspection is judged to be unqualified, jump to step 13;

[0029] Step 12: After passing the test, save the configuration information locally for subsequent use;

[0030] Step 13: Execute the compensation module and then jump to step 10.

[0031] The beneficial effect of the present invention is to solve the problems existing in the existing processing system in terms of complexity, flexibility, dynamic monitoring and cost control. By introducing artificial intelligence vision technology and multi-sensor fusion, high-precision target recognition, dynamic path planning and adaptive parameter adjustment can be achieved to ensure processing quality and efficiency. It is suitable for flexible manufacturing scenarios of multiple varieties, small batches and complex parts.

[0032] According to one embodiment of the present invention, in the first step, the local configuration file contains data information required during the processing; if the configuration file cannot be found locally, a prompt will be given and the non-first processing process will be closed.

[0033] According to one embodiment of the present invention, in the second step, in addition to the position parameter, each parameter value is indexed in the local configuration file, and the user is waited for to change its parameter information. After the change is completed, the user-modified value is obtained again to update the dynamic configuration information.

[0034] According to an embodiment of the present invention, in the fourth step, except for the parameters set by the user, all dynamic configuration information in the configuration file is cleared.

[0035] According to one embodiment of the present invention, in the fifth step, a 2D camera is used to collect image information of the object to be processed, and then deep learning is used to infer the material information of the workpiece to be processed and the preliminary target point of the object to be processed.

[0036] According to one embodiment of the present invention, in step 8, trajectory planning is performed according to the initial target point, and the robot carrying the 3D camera scans sequentially along the trajectory to accurately obtain the parameters within the field of view of the target point, which are used to calculate the machining allowance.

[0037] According to one embodiment of the present invention, in the twelfth step, the configuration information includes a processing trajectory, target recognition information, 3D image information, and quality inspection results.

[0038] An intelligent processing system that implements a control method for intelligent processing based on artificial intelligence vision technology includes: a target recognition module, which is used for model training and optimization testing before the system is installed, as well as for image acquisition, calling of an image processing algorithm library, and model reasoning after image processing during the working process. It combines 3D model analysis and processing drawings to determine whether the workpiece is a preset workpiece to be processed, and effectively divides the workpiece space area, and outputs the center point of each area as the initial target point; a stereoscopic vision detection module, which uses a 3D vision camera and a collaborative robot to scan the workpiece in space in sequence according to the trajectory generated by the initial target point, and calculates the processing parameters in combination with the raw material drawing and the processing drawing; a processing module, which is used to read processing parameters, match task strategies, automatically plan the processing robot trajectory, and supervise and control the robot to perform processing tasks; a quality inspection module, which uses a 3D vision camera to offset the initial target point according to the processing parameters and then scan it to generate a point cloud model of the workpiece in space after processing, and analyze whether its processing is accurate and whether there are defects such as burrs; and a compensation module, which is used to dynamically adjust and optimize the processing problems caused by errors during the processing process.

[0039] According to one embodiment of the present invention, the target recognition module, the visual positioning detection module and the quality inspection module all call the image acquisition service module and the image processing algorithm library function with image acquisition function; the visual positioning detection module and the quality inspection module both call the collaborative robot trajectory automatic planning function; the processing module and the compensation module both call the task strategy matching function; the target recognition module and the quality inspection module both call the reasoning service module with model reasoning function.

[0040] According to one embodiment of the present invention, the image acquisition service module provides a unified interface for the target recognition module, the visual positioning detection module and the quality inspection module to call; the reasoning service module provides a unified interface for the target recognition module and the quality inspection module to call.

[0041] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is a schematic diagram of the main functional components of the user-side software of the intelligent processing system of the present invention;

[0045] Figure 2 This is a flow chart of a control method for an intelligent processing system based on artificial intelligence vision technology of the present invention;

[0046] Figure 3 It is a schematic diagram of the calling method of dynamic configuration information and the composition of the configuration file of the intelligent processing system based on artificial intelligence vision technology of the present invention;

[0047] Figure 4 It is a schematic diagram of the module functional structure of the intelligent processing system of the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0049] The following describes in detail an intelligent processing system and a control method thereof based on artificial intelligence vision technology according to an embodiment of the present invention with reference to the accompanying drawings.

[0050] The user interface of the intelligent processing system consists of Figure 1 As shown in the figure, its design fully integrates practical application needs, providing users with an intuitive operating platform that covers the entire process from workpiece identification to processing quality inspection. The user interface integrates functional modules such as deep learning, stereo vision, robot control, and 3D model analysis. These functions work together to achieve full visualization and convenient operation of the processing process.

[0051] Through the user interface, the system provides target recognition capabilities, supports sample set collection, labeling, cleaning, and training, and also includes the ability to select appropriate training model algorithms. The recognition module infers the workpiece's material type and spatial pose (X, Y position, and Z angle of the ROI) based on user-uploaded data. Based on the recognition results, the system determines whether the material of the workpiece to be processed matches the preset material. If so, the system automatically loads the processing parameters (such as processing speed and cutting depth) for that material and displays them on the interface. If not, an alarm is issued to prevent processing errors.

[0052] The stereo vision function provides 3D camera parameter settings and ROI adjustment options through the user interface. During machining, the system uses the 3D camera to collect point cloud data of the target area and displays the workpiece's 3D model in real time through the user interface. Using this data, the system supports subsequent machining path planning and spatial dimensional analysis, allowing users to intuitively view the data and analysis results.

[0053] Robot control functions include parameter settings for both the machining robot and the collaborative robot, as well as intuitive display of parameters such as the robot's trajectory. Based on material information, point clouds, and the workpiece's raw and machining drawings, the system automatically generates machining paths and updates them in real time on the interface. Furthermore, the collaborative robot, equipped with a 3D camera, performs image acquisition tasks according to the planned path, allowing users to view the workpiece's trajectory in real time through the interface.

[0054] The system visualizes 3D model analysis through its user interface. The analysis results include 3D spatial data of the raw and processed workpieces, and the interface displays a comparison of theoretical and actual machining volume. When machining deviations are detected, the system generates a compensation strategy, and the compensation operations and their effects are displayed in real time on the interface for the user to review and confirm.

[0055] The user interface also displays dynamic data about the machining process in real time, including the acquisition and machining trajectory, quality inspection results, and compensation results. The quality inspection function displays the completed point cloud model to determine if there are defects such as machining deviations and burrs, and presents this information intuitively on the interface. The compensation function automatically plans a compensation path upon detection of a problem and prompts the user to confirm execution through the interface, ensuring that the final machining result meets the requirements.

[0056] The above user interface design is both practical and professional, and can effectively guide users to complete processing operations. At the same time, it provides real-time processing monitoring and feedback, greatly improving the intelligence and reliability of processing.

[0057] The flow chart of the control method of the intelligent processing system based on artificial intelligence vision technology is as follows Figure 2 As shown, the following steps are included:

[0058] Step 1: Read the local configuration file, parse the basic parameters, and display the parsing results through a visual interface. The local configuration file contains the data information needed during the processing. At this time, if the configuration file cannot be found locally, a prompt will be displayed and the non-first processing process will be closed. A blank configuration file will be generated according to the existing configuration file format.

[0059] Step 2: Obtain user-set parameters, update the configuration file, parse the parameters, and display them: Obtain the user's settings and adjustments to the parsed basic parameters and update the corresponding configuration information based on the adjusted parameters. Except for positional parameters, each parameter value is indexed from the local configuration file, and the user waits for the parameter information to be modified. After the change is completed, the user's modified value is retrieved and the dynamic configuration information is updated.

[0060] Step 3: Call the condition control module, select the condition based on the return value of whether it is the first processing, and judge whether it is the first processing: if it is the first processing, it is necessary to obtain the automatic processing parameters and jump to step 4; if it is not the first processing, it will be processed with the parameters in the configuration file (which is mostly used for small batch processing), and jump to step 9; It should be noted that the meaning of first processing is: due to the fact that new products often come in during small batch production, the first processing requires the formulation of a processing plan for the product. If such products are processed again, they can be processed directly according to the processing plan, reducing the time for formulating the processing plan.

[0061] Step 4: Initialize the configuration file. Except for the user-set parameters, clear all dynamic configuration information in the configuration file. Except for the user-set parameters, clear all dynamic configuration information in the configuration file.

[0062] Step 5: Analyze the workpiece drawings (raw material drawings and processing drawings) using the 3D model and start the target recognition module: Use a 2D camera to collect image information of the object to be processed, and then use deep learning to infer the material information of the workpiece to be processed and the preliminary target point of the object to be processed. Specifically, start the target recognition module, use a 2D camera to collect image information of the object to be processed, and then use deep learning to infer the material information of the workpiece to be processed. Combined with the results of the 3D model analysis of the workpiece drawings, the preliminary target point is determined. The preliminary target point is obtained based on the size, shape, and processing method of the workpiece. Its main function is to generate the spatial point position of the mapping trajectory carried by the collaborative robot with a 3D vision camera. The second function is to reference the mapping trajectory during quality inspection. The target point of quality inspection only needs to have a certain offset capability. It should be noted that target recognition not only recognizes, but also reasonably divides the area of ​​the workpiece based on the recognition results, and obtains the center position within the area as the target point. The collaborative robot will refer to the (x, y) position of these target points. Of course, the position accuracy of the target recognition output is relatively low.

[0063] Step 6: Call the condition control module and determine whether the workpiece to be processed is consistent with the preset workpiece type based on the returned result (to prevent the user from placing the wrong workpiece): If the workpiece to be processed is determined to be consistent with the preset workpiece type, jump to step 8; if the workpiece to be processed is determined to be inconsistent with the preset workpiece type, jump to step 7; specifically, determine whether the material and size of the inferred workpiece to be processed are consistent with the preset material, raw material size and other factors. If not, jump to step 7; if yes, jump to step 8.

[0064] Step 7: Alarm prompt and manual intervention inspection, that is, triggering an alarm and prompting manual intervention.

[0065] Step 8: Start the stereoscopic vision inspection module, scan and calculate the spatial posture of the workpiece in sequence according to the initial target point; calculate the workpiece parameters based on the 3D model and spatial posture; automatically generate a processing plan based on the processing process logic; update the processing plan information in the workpiece configuration file and store it locally; plan the trajectory according to the initial target point, and execute the robot carrying the 3D camera to scan along the trajectory in sequence, accurately obtaining the parameters within the target point's field of view, and use them to calculate the processing allowance. Specifically, start the stereoscopic vision inspection module, and the collaborative robot automatically generates a mapping path according to the initial target point. At a certain operating speed and waiting time, the 3D camera automatically scans the initial target point on the mapping trajectory to generate a point cloud model, accurately obtaining parameters such as height (thickness) within the target point's field of view, and calculates the processing allowance in combination with the processing drawing model of the workpiece. The processing plan is automatically generated according to the processing logic of this type of workpiece, and the processing plan information is dynamically configured to update the processing plan information. At the same time, the dynamic configuration information is named with the serial number of the workpiece to be processed and saved to a designated local folder.

[0066] Step 9: Call the local configuration file, update the dynamic configuration information, and execute the processing module according to the configuration file information: the processing module calls the configuration file of the corresponding workpiece, executes the processing module, and the processing robot performs processing according to the processing plan. At the same time, the system monitors the processing robot to execute the processing task.

[0067] Step 10: Execute the quality inspection module after processing is completed: Specifically, the quality inspection module is executed, and the collaborative robot carries a 3D vision camera to collect images along the quality inspection trajectory. The quality inspection trajectory is composed of quality inspection target points, and the quality inspection target points are obtained by offsetting the initial target points according to information such as the processing volume.

[0068] Step 11: Call the condition judgment module to judge whether the quality inspection is qualified based on the returned results: if the quality inspection is qualified, jump to step 12; if the quality inspection is unqualified, jump to step 13; specifically, the quality inspection will be based on unreasonable processing dimensions, burrs, etc. The detection result is given by the visual positioning detection module, which includes a target recognition module and a stereoscopic vision detection module, and the detection equipment used are 2D vision cameras and 3D vision cameras respectively.

[0069] Step 12: After passing the test, save the configuration information locally for subsequent use. Specifically, after passing the test, save the machining trajectory, quality inspection trajectory and other parameters to the workpiece configuration file. At the same time, save the target recognition information, 3D image information, quality inspection information and other parameters to a designated folder. The configuration information includes the machining trajectory, target recognition information, 3D image information and quality inspection results.

[0070] Step 13, execute the compensation module, then jump to step 10: Specifically, execute the compensation module, and after the compensation is completed, jump to step 10 and start the quality inspection module again until it is qualified. Specifically, the compensation module uses the robot to perform compensation according to the compensation trajectory and related strategy information. Specifically, the compensation trajectory is regenerated according to the compensation task strategy matching and modification based on the processing trajectory. The compensation trajectory of the processing robot only refers to a part of the processing trajectory. The compensation trajectory and the processing trajectory are similar in planning but are two completely different planning methods. Specifically, the processing trajectory is determined by the processing plan, and it matches the task strategy library of the processing plan. The compensation trajectory is determined by the compensation plan, and it matches the task strategy library of the compensation plan. Specifically, the compensation module does not necessarily compensate successfully. If the compensation module exceeds the working range, an alarm will be issued, and manual intervention will be prompted to find the problem. Specifically, exceeding the working range of the compensation module means that the actual size after processing exceeds the maximum allowable error range, or there is no compensation plan for this defect in the compensation task strategy library.

[0071] The calling method of the software's dynamic configuration information and the composition diagram of the configuration file are as follows: Figure 3 As shown in the figure, during system initialization, the system automatically generates dynamic configuration information based on the predefined format of the configuration file, which serves as the core basis for the operation of subsequent processing, quality inspection, and compensation modules. For the workpiece to be processed, the system will first query the configuration file records in the existing workpiece library. If a matching configuration file is found, the system will directly extract its content and dynamically load it as the dynamic configuration information for the current processing. If the corresponding record is not found in the workpiece library, the system will automatically create a new dynamic configuration information and gradually supplement and improve the relevant content during the processing. After the processing task is completed, the dynamic configuration information will be stored as a formal configuration file, providing a reference for subsequent workpiece processing and system optimization.

[0072] Specifically, the newly created configuration file includes basic information, target recognition results, stereo vision detection results, processing plan, quality inspection plan, compensation plan, etc.

[0073] Basic information includes the operator's identity, the workpiece's serial number, the information number, and the contract number. The operator's information and task number are entered by the user upon login, and the system then retrieves them through indexing. Information such as the date and time are automatically generated by the system. Key parameters such as the workpiece's material and processing options are entered directly by the user to ensure the accuracy of the initial parameters.

[0074] Specifically, the target recognition module includes the workpiece's initial parameters and target point information. Model inference determines the material, dimensions, and other key parameters of the workpiece to be processed. Combined with image segmentation results, the module outputs the coordinates of the workpiece's initial target point. All this information is automatically acquired and dynamically updated during system operation.

[0075] Specifically, the stereoscopic vision inspection results, jointly produced by the collaborative robot and a 3D vision camera, include three key elements: the collaborative robot's scanning trajectory, the current point cloud data of the workpiece after machining, and the ideal machining point cloud data based on model analysis. The collaborative robot's scanning trajectory is automatically planned based on parameters such as the initial target point; the current point cloud data of the workpiece is acquired through 3D camera scanning; and the ideal machining point cloud data, derived from 3D modeling analysis of the machining design drawings, serves as a benchmark for assessing machining quality.

[0076] Specifically, the machining plan encompasses core elements such as machining procedures, machining parameters, and machining trajectories. The machining procedures are flexibly configured by the user based on actual needs; machining parameters are automatically generated by calculating the error between the actual point cloud and the ideal point cloud; and the machining trajectory for each procedure is automatically generated by the system for the machining robot, combining the procedure definition and machining parameters to ensure high-precision trajectory control and machining operations.

[0077] Specifically, the quality inspection solution module includes the collaborative robot's inspection trajectory and inspection results. The inspection trajectory generates inspection target points by integrating parameters such as the initial target point and machining allowance. The collaborative robot then automatically generates an inspection path based on these target points. The inspection results are analyzed through model reasoning of the processed point cloud data to determine whether there are any dimensional errors, unevenness, burrs, or other machining defects.

[0078] The compensation plan consists of the compensation type, compensation parameters, and the robot's compensation trajectory. Compensation types include issues such as under-dimensioning, defects (such as burrs and unevenness), and dimensional deviations. If the system determines that compensation can address the problem, it replans the robot's trajectory based on the preset compensation strategy and executes the compensation. Cases where compensation is unavailable are marked and archived for later review and documentation.

[0079] Specifically, because the quality inspection and compensation process requires multiple iterations, configuration information needs to be stored in batches. For example, the first quality inspection and compensation, and the second quality inspection and compensation, must be recorded separately to provide complete data support and traceability for subsequent processing analysis.

[0080] Specifically, dynamic configuration information is highly scalable and can flexibly adapt to changing system functionality requirements. During later system development or functional expansion, only the structure or content of the configuration file needs to be adjusted and the corresponding new functional modules implemented, without the need for large-scale changes to the existing software architecture. Through this design, dynamic configuration information not only significantly reduces the complexity of functional upgrades and maintenance, but also provides strong compatibility and maintainability for continuous system optimization and functional expansion. This feature is particularly important in complex industrial application scenarios and can significantly improve the system's development efficiency and lifecycle management capabilities.

[0081] The module functional structure diagram of the intelligent processing system is as follows: Figure 4 As shown, the system adopts a modular design, dividing functions into independent modules and abstracting and integrating core functional units that are shared (such as image acquisition, image processing algorithm library, collaborative robot trajectory planning, task strategy matching, and model reasoning) to form a unified component. The modular design offers high independence and low coupling, significantly improving development efficiency and system stability while reducing redundant design and optimizing resource utilization. Furthermore, the modular design enhances the system's scalability and maintainability, enabling efficient collaboration through standardized interfaces. This eliminates the need for significant system architecture modifications when adding new features or adapting to different application scenarios, thus meeting the needs of flexible production and intelligent manufacturing.

[0082] Specifically, the modules of the intelligent processing system based on artificial intelligence vision technology include target recognition module, stereo vision detection module, processing module, quality inspection module and compensation module.

[0083] Specifically, the target recognition module is used for model training and optimization testing before the system is installed, as well as for image acquisition during the work process, calling the image processing algorithm library, and model reasoning after image processing. It combines the 3D model to analyze the processing drawings to determine whether the workpiece is the preset workpiece to be processed, and effectively divides the workpiece space area, outputting the center point of each area as the initial target point. It should be noted that the initial target point mentioned in the present invention is the position where the 3D camera in the stereo vision module should be located to scan the workpiece. There are multiple initial target points. After the 3D camera is located at multiple initial points and scans them separately, a point cloud model of the workpiece can be constructed. During quality inspection, the scan is also performed after offsetting a certain distance based on the initial target point.

[0084] The target recognition module specifically includes model training, optimization, and testing functions and must support the development of sample set-related operational tools, such as sample cleaning, sample labeling, and sample collection, to improve the model's training dataset management. This module must possess model training and inference capabilities, executing target recognition tasks by calling pre-trained or real-time trained models. Furthermore, the module's provided interface supports the rapid configuration of more technologically advanced training models during software updates to ensure that the system maintains a certain level of technical proficiency in target recognition performance and technical applicability. Furthermore, the target recognition module should support image acquisition and processing, invoking model inference to obtain the region of interest (ROI) of the workpiece to be processed, and further extracting the workpiece's length, width, and angle information. Furthermore, this module should be able to identify the material properties of the workpiece to be processed, providing the necessary basis for the task strategy matching function of the processing module, thereby achieving precise processing of workpieces of different materials and shapes.

[0085] Specifically, the stereoscopic vision inspection module uses a 3D vision camera in conjunction with a collaborative robot to sequentially scan the workpiece along a trajectory generated by the initial target point, completing a point cloud model in space. This model then calculates processing parameters based on the raw material and processing drawings. It should be noted that the collaborative robot, equipped with a 3D camera, scans the initial target point, and the resulting point cloud data can be assembled into a realistic model of the workpiece through methods such as splicing.

[0086] The stereo vision inspection module includes automated trajectory planning for the collaborative robot, image acquisition, and image algorithm processing. Leveraging 3D model analysis capabilities, the automated trajectory planning function analyzes raw material and process drawings. Using the 3D camera's field of view, it captures the model area within the field of view to generate corresponding point cloud data. A specific algorithm extracts target initial points from the point cloud, such as those based on the model's geometric boundary features, areas of curvature variation, or representative points selected through uniform spatial partitioning. These initial target points are then output as coordinates to support the collaborative robot's trajectory planning. Following the generated trajectory, the collaborative robot, carrying the 3D vision camera, moves to the designated target point, performing precise scanning to acquire point cloud data. The image algorithm processing function stitches and optimizes the captured point cloud data and, combined with the positional information of the initial target points, generates a realistic 3D point cloud model of the workpiece to be processed. Based on this model, the system calculates machining parameters such as cutting allowance and machining allowance, providing this reference data to the machining module for subsequent precision machining operations.

[0087] Specifically, the processing module is used to read processing parameters, match task strategies, automatically plan processing robot trajectories, and supervise and control the robot to perform processing tasks.

[0088] The processing module specifically includes processing parameter acquisition, processing plan task strategy matching, and automatic generation of processing robot trajectories. The processing parameter acquisition function is primarily responsible for receiving and processing key parameter information transmitted by other modules, such as the initial target point, the length, width, and angle of the workpiece, the cutting amount, and the cutting speed. These parameters provide data support for the generation and execution of processing plans. The processing plan task strategy matching function matches the processing task strategy based on the user's preset processing method library. The preset processing method library contains a variety of processing technologies such as drilling, milling, and composite grinding. Based on the task requirements, the system selects the appropriate processing technology from the processing method library and generates the corresponding processing plan. The processing robot automatically generates the processing trajectory by analyzing the above parameters and the processing plan. The generated trajectory not only ensures processing accuracy but also optimizes path planning to improve efficiency. After the processing trajectory is generated, the system monitors the processing robot and executes the processing task according to the planned trajectory, achieving high-precision processing operations on the workpiece.

[0089] The quality inspection module uses a 3D vision camera to offset the initial target point according to machining parameters before scanning. This generates a point cloud model of the finished workpiece in space, allowing analysis of machining accuracy and the presence of defects such as burrs. Specifically, the quality inspection module uses the 3D vision camera to offset the initial target point based on machining parameters before performing a scanning operation to generate a point cloud model of the finished workpiece. This model is used to analyze machining accuracy and detect defects such as burrs. The collaborative robot, equipped with a 3D camera, completes the scanning task according to the planned trajectory. In the event of a machining failure, for example, if the goal is to machine a 500mm cylinder to 395mm, but the actual machining result is still 500mm, the initial target point should be adjusted to 560mm, and the quality inspection target point should be 455mm. However, due to system anomalies, performing quality inspection based on the 455mm target point may result in an anomaly. To prevent such accidents, the collaborative robot's 3D camera housing has a protective mechanism during scanning. When the protective housing approaches the workpiece surface, the system triggers an alarm and an emergency stop, ensuring the safety of the equipment and workpiece and preventing machining accidents.

[0090] The quality inspection module specifically includes functions such as automated collaborative robot trajectory planning, image acquisition, image algorithm processing, and model inference. Unlike the automated collaborative robot trajectory planning in the stereo vision inspection module, the trajectory planning in the quality inspection module offsets the initial target point based on the machining volume to generate the inspection target point, and automatically generates the inspection trajectory based on this offset. Following the planned inspection trajectory, the collaborative robot, carrying a 3D vision camera, performs point cloud acquisition to obtain surface data of the finished workpiece. After image acquisition, the quality inspection module analyzes the dimensional and geometric characteristics of the workpiece surface through image preprocessing, feature extraction, and segmentation to determine whether it meets design requirements. Using model inference, the system further detects defects such as uneven machining, burrs, overcuts, or incomplete machining. Furthermore, the quality inspection module compares the overall point cloud data of the workpiece with the machining drawing or model to accurately assess machining errors. If the quality inspection results indicate defects, the system uses the compensation module to generate a repair solution, such as adjusting the machining trajectory to repair incomplete areas or implementing post-processing methods such as deburring to ensure that the workpiece meets quality standards. Through the closed-loop feedback mechanism, the quality inspection module can continuously optimize processing accuracy and improve the processing quality and stability of the entire system.

[0091] Specifically, the compensation module is used to dynamically adjust and optimize the processing problems caused by errors in the processing process. The system uses the post-processing point cloud model generated by scanning with a 3D vision camera, combined with the preset processing parameters, to perform real-time analysis of the processing accuracy. If an area that is not processed in place is detected, the system will automatically calculate the compensation path and compensation processing amount, and re-plan the trajectory of the processing tool, and perform secondary processing to ensure that the processing is in place. For the detected burr defects, the system will accurately locate the burr position based on the comparison between the point cloud model and the target model, and automatically plan the processing path for removing the burrs. After removing the burrs, the system scans again to verify the processing quality of the workpiece to ensure that it meets the accuracy requirements. The compensation module also supports multiple rounds of compensation processing, and continuously adjusts the processing trajectory and parameters through a closed-loop control mechanism until the processing is fully up to standard. For workpieces that are difficult to repair or exceed the processing tolerance range, the system will generate an alarm and mark the workpiece as unqualified to ensure the quality of the final product.

[0092] The compensation module often needs to be executed repeatedly until the processing is completed, depending on factors such as machining process allowances. The compensation module specifically includes functions such as task strategy matching and automatic compensation trajectory planning. The task strategy matching function compares the inspection results output by the quality inspection module with a preset compensation task strategy library. The compensation task strategy library contains specific compensation schemes for different processing defects (such as incomplete processing, burrs, overcuts, etc.). The system selects the corresponding compensation scheme based on the matching results and indexes the required parameters (such as compensation processing volume, path adjustment parameters, etc.) to generate a compensation trajectory. The automatic compensation trajectory planning function uses the defect location data provided by the quality inspection module and combines it with the processing parameters in the compensation scheme to generate a specific compensation path. The processing robot performs repair operations on the workpiece according to the planned compensation trajectory to ensure processing accuracy and quality. If no applicable compensation scheme can be matched in the task strategy library, or the system detects that the workpiece can no longer be processed, the system will trigger an alarm function, prompting the operator to manually intervene or terminate processing.

[0093] Specifically, the quality inspection and compensation modules are closely intertwined, often requiring repeated execution based on factors such as machining steps and allowances, forming a closed-loop control system. Through continuous quality inspection and compensation, the system can gradually correct machining errors, ultimately ensuring that the workpiece meets design requirements and completes the entire machining task.

[0094] The target recognition module, visual positioning detection module, and quality inspection module all call upon the image acquisition service module and image processing algorithm library functions, which have image acquisition capabilities. The visual positioning detection module and quality inspection module all call upon the collaborative robot's automatic trajectory planning function. The collaborative robot automatically generates a trajectory based on the initial target point, moving the 3D camera through each target point in sequence. The processing module and compensation module both call upon the task strategy matching function, which dynamically matches the corresponding processing strategy based on the preset processing method and process requirements, including tool selection, tool replacement, and operating parameter adjustment, to meet the execution requirements of different processing tasks. The target recognition module and quality inspection module both call upon the inference service module, which has model inference capabilities. The image acquisition service module provides a unified interface for the target recognition module, visual positioning detection module, and quality inspection module to call; the inference service module provides a unified interface for the target recognition module and quality inspection module to call. It should be noted that the target recognition module, visual positioning detection module, and quality inspection module all complete their corresponding functions by calling upon the image acquisition function unit and image processing algorithm library interfaces. The stereo vision detection module and quality inspection module both call upon the collaborative robot's automatic trajectory planning function to support their task execution.

[0095] The intelligent processing system and control method based on artificial intelligence vision technology of the present invention solve the problems of existing processing systems in complexity, flexibility, dynamic monitoring and cost control. By introducing artificial intelligence vision technology and multi-sensor fusion, high-precision target recognition, dynamic path planning and adaptive parameter adjustment are achieved, ensuring the processing quality and efficiency of small batch and multi-variety products. It is suitable for flexible manufacturing scenarios of multi-variety, small batch and complex parts.

[0096] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A control method for intelligent processing based on artificial intelligence vision technology, characterized in that: The following steps are involved: Step 1: Read the local configuration file, parse the basic parameters, and display the parsing results through a visual interface; The local configuration file contains the data information needed during the processing. If the configuration file cannot be found in the local search, a prompt will be displayed and the non-first processing process will be closed. A blank configuration file will be generated according to the existing configuration file format. During system initialization, the system automatically generates dynamic configuration information based on the predefined format of the configuration file, and serves as the core basis for the operation of subsequent processing modules, quality inspection modules, and compensation modules. For workpieces to be processed, the system will first query the configuration file records in the existing workpiece library. If a matching configuration file is found, the system will directly extract its content and dynamically load it as the dynamic configuration information for the current processing. If no corresponding record is found in the workpiece library, the system will automatically create a new dynamic configuration information and gradually supplement and improve the relevant content during the processing process. After the processing task is completed, the dynamic configuration information will be stored as a formal configuration file, providing a reference basis for subsequent workpiece processing and system optimization. Step 2: Obtain the user's settings and adjustments to the parsed basic parameters, and update the corresponding configuration information according to the adjusted parameters; In addition to positional parameters, each parameter value is indexed in the local configuration file, and the user waits for the parameter information to be changed. After the change is completed, the user's modified value is retrieved and the dynamic configuration information is updated; Step 3: Call the condition control module and make conditional selection based on the return value of whether it is the first processing to determine whether it is the first processing: If it is determined to be the first processing, the automatic processing parameters are obtained and the process goes to step 4; If it is determined that this is not the first processing, the processing will be performed according to the parameters in the configuration file and the process will jump to step 9. Step 4: Initialize the configuration file: except for the user-set parameters, leave the dynamic configuration information blank. Step 5: Analyze the workpiece drawing using the 3D model and start the target recognition module: Use a 2D camera to collect image information of the object to be processed, then use deep learning to infer the material information of the workpiece to be processed, and combine the results of the 3D model analysis of the workpiece drawing to determine the preliminary target point; The target recognition module supports the development of sample set-related operation tools to improve the management of model training datasets; The target recognition module performs the target recognition task by calling the pre-trained or real-time trained model; The interface provided by the target recognition module supports the rapid configuration of the most advanced training models during software updates, ensuring that the system always maintains a certain level of technical proficiency in target recognition performance and technical applicability. The target recognition module uses model inference to obtain the region of interest of the workpiece to be processed, and further extracts the length, width, and angle information of the workpiece. The target recognition module identifies the material properties of the workpiece to be processed, providing the necessary basis for the task strategy matching function of the processing module, thereby achieving precise processing of workpieces of different materials and shapes. Target recognition not only recognizes the workpiece, but also reasonably segments the workpiece area based on the recognition results, obtaining the center position within the area as the target point. The collaborative robot will refer to the location of these target points. Step 6: Call the condition control module and determine whether the workpiece to be processed is consistent with the preset workpiece type based on the returned result: If it is determined that the workpiece to be processed is the same as the preset workpiece type, jump to step 8; If it is determined that the workpiece to be processed is inconsistent with the preset workpiece type, jump to step 7; Step 7: Alarm prompt, manual intervention inspection; Step 8: Start the stereo vision detection module, scan and calculate the spatial posture of the workpiece in sequence according to the initial target point; calculate the processing part parameters according to the 3D model and spatial posture; automatically generate the processing plan according to the processing process logic; update the processing plan information in the configuration file of the workpiece and save it locally; perform trajectory planning according to the initial target point, execute the robot to carry the 3D camera to scan along the trajectory in sequence, accurately obtain the height parameters within the field of view of the target point, and calculate the processing allowance using the point cloud and the processing drawing model of the workpiece; dynamically configure the information of the processing plan, and at the same time, save the dynamic configuration information to the local designated folder after naming it with the serial number of the processed workpiece; the processing plan mainly covers the processing process, processing parameters and processing trajectory; the processing process is flexibly configured by the user according to actual needs; the processing parameters are automatically generated by calculating the error between the actual point cloud and the ideal point cloud; the processing trajectory of each process is automatically generated by the system for the processing robot in combination with the process definition and processing parameters to ensure high-precision trajectory control and processing operation; Step 9: Call the local configuration file, update the dynamic configuration information, and execute the processing module according to the configuration file information; Step 10: After processing is completed, execute the quality inspection module; Step 11: Call the condition judgment module and judge whether the quality inspection is qualified according to the returned results: If the quality inspection is qualified, jump to step 12; If the quality inspection is judged to be unqualified, jump to step 13; Step 12: After passing the test, the configuration information is saved locally for subsequent use; the configuration information includes processing trajectory, target recognition information, 3D image information and quality inspection results; Step 13: Execute the compensation module and then jump to step 10. The system uses the post-processing point cloud model generated by scanning with a 3D vision camera and combines it with the preset processing parameters to perform real-time analysis of the processing accuracy. If an area that is not processed is detected, the system will automatically calculate the compensation path and compensation processing amount, re-plan the trajectory of the processing tool, and perform secondary processing to ensure that the processing is in place. The compensation module also supports multiple rounds of compensation processing, and continuously adjusts the processing trajectory and parameters through a closed-loop control mechanism until the processing is fully qualified. For workpieces that are difficult to repair or exceed the processing tolerance range, the system will generate an alarm and mark the workpiece as unqualified to ensure the quality of the final product. An intelligent processing system that implements a control method for intelligent processing based on artificial intelligence vision technology adopts a modular design, divides functions into independent modules, and abstractly integrates the core functional units of shared calls to build a unified component. The core functional units include image acquisition, image processing algorithm library, collaborative robot trajectory planning, task strategy matching and model reasoning. The intelligent processing system includes: The target recognition module is used for model training and optimization testing before the system is installed, as well as for image acquisition during operation, calling the image processing algorithm library, and model reasoning after image processing. It combines 3D model analysis with processing drawings to determine whether the workpiece is the preset workpiece to be processed. At the same time, it effectively divides the workpiece space area and outputs the center point of each area as the initial target point. The stereo vision inspection module uses a 3D vision camera in conjunction with a collaborative robot to sequentially scan the workpiece's point cloud model in space along the trajectory generated by the initial target point, and calculates the processing parameters based on the raw material drawings and processing drawings. The processing module is used to read processing parameters, match task strategies, automatically plan processing robot trajectories, and supervise and control the robot to perform processing tasks; The quality inspection module is used to use a 3D vision camera to offset the initial target point according to the processing parameters and then scan it to generate a point cloud model of the workpiece in space after processing, and analyze whether its processing is accurate and whether there are any defects. That is, the quality inspection module uses a 3D vision camera in combination with the processing parameters to perform an offset adjustment on the initial target point and then performs a scanning operation to generate a spatial point cloud model of the workpiece after processing, which is used to analyze the processing accuracy and detect whether there are any defects. The collaborative robot carries a 3D camera and completes the scanning task according to the planned trajectory. When the collaborative robot is performing scanning, the 3D camera housing it carries has a protection mechanism. When the protective housing approaches the surface of the workpiece, the system triggers an alarm and stops the operation urgently. The compensation module is used to dynamically adjust and optimize the processing problems caused by errors during the processing process. The visual positioning detection module includes a target recognition module and a stereo vision detection module. Both the visual positioning detection module and the quality inspection module call the collaborative robot's automatic trajectory planning function; the collaborative robot automatically generates a trajectory based on the position of the target initial point, and moves the 3D camera through each target point in sequence; the processing module and the compensation module both call the task strategy matching function, which is used to dynamically match the corresponding processing strategy according to the preset processing method and process requirements to adapt to the execution requirements of different processing tasks; the target recognition module and the quality inspection module both call the reasoning service module with model reasoning function; The image acquisition service module provides a unified interface for the target recognition module, the visual positioning detection module and the quality inspection module to call; the reasoning service module provides a unified interface for the target recognition module and the quality inspection module to call; specifically, the target recognition module, the visual positioning detection module and the quality inspection module all complete the corresponding function execution by calling the image acquisition function unit and the image processing algorithm library interface; the stereo vision detection module and the quality inspection module both call the collaborative robot trajectory automatic planning function to support their task execution process.

Images