Robot polishing real-time defect detection and self-adaptive correction system based on deep learning

Abstract

The invention discloses a robot grinding real-time defect detection and self-adaptive correction system based on deep learning. The problem that in the prior art, real-time detection, defect correction and accurate path optimization are difficult to achieve in the grinding process is solved. The image, the three-dimensional point cloud and the moment data are fused through a multi-modal sensing system, and data fusion processing is performed by adopting a time synchronization mechanism, so that high-precision defect detection is realized. And the self-adaptive correction module dynamically adjusts the pressure, the speed and the angle of the grinding tool according to real-time feedback by using a reinforcement learning algorithm, so that accurate grinding for different workpiece surfaces is ensured. The system is further combined with closed-loop feedback control, the grinding path is continuously optimized, and the consistency and stability of the grinding quality are improved. Compared with the prior art, efficient and automatic defect detection and self-adaptive correction can be achieved on the complex and irregular workpiece surface, and the production efficiency and the surface treatment quality are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automated surface treatment controlled by industrial robots, and in particular to a real-time defect detection and adaptive correction system for robot polishing based on deep learning. Background Art

[0002] Industrial robot grinding systems in the prior art are widely used in aerospace, automobile manufacturing, precision instrument manufacturing and other fields, with the purpose of replacing manual labor with automated equipment for high-precision surface treatment. However, traditional grinding systems mainly rely on preset grinding paths and fixed parameters to complete the work. These systems usually use torque sensors or simple visual sensors to detect the surface state of the workpiece to achieve grinding control. The feedback during the grinding process is usually a single data source, relying on preset programs for limited adjustments, making it difficult to achieve real-time defect detection and adaptive correction.

[0003] In order to improve the grinding quality, the following methods are usually used for optimization in the prior art: 1. Vision or torque sensor: The surface of the workpiece is detected and fed back through a camera or torque sensor to adjust the pressure and path of the grinding tool. Although this method can automate the grinding process to a certain extent, it is not effective for complex geometric surfaces or irregular workpieces, and can only handle relatively simple defects or planes.

[0004] 2. Fixed grinding paths and parameters: Traditional grinding systems usually rely on pre-set paths and parameters and cannot flexibly respond to real-time changes in the grinding process. This method cannot respond quickly to real-time changes in defect location and severity, which may lead to repeated rework or inconsistent surface quality.

[0005] However, the existing technologies have the following shortcomings, such as insufficient real-time performance: traditional grinding methods are difficult to achieve real-time detection of surface defects, especially when dealing with complex shapes or multiple materials, and surface defects are often unable to be discovered and corrected in a timely manner. Poor adaptability: existing systems mostly rely on preset grinding paths and fixed parameters, and cannot be adaptively adjusted according to the specific needs of different workpieces. For complex geometric shapes or irregular surfaces, existing technologies cannot guarantee accurate grinding effects. Low detection accuracy: a single data source (such as image or torque data) is usually unable to fully capture tiny defects on the workpiece surface, resulting in defects being easily missed or over-grinding during the grinding process, affecting the quality of the final product.

[0006] In order to solve the above problems, the industry has tried to optimize by enhancing sensor accuracy or increasing the complexity of the grinding algorithm, but these methods still cannot completely solve the problem of real-time detection and adaptive correction. For example, although increasing the number and accuracy of sensors can improve the detection effect in certain specific situations, the multimodal information on the surface of complex workpieces (such as images, three-dimensional point clouds, torque, etc.) cannot be effectively integrated and synchronized, resulting in insufficient system response and limited grinding effect. In addition, the existing fixed paths and parameter settings cannot be adjusted in real time through feedback mechanisms, which often leads to over-reliance on manual adjustments or reduced production efficiency.

[0007] Therefore, how to achieve multimodal data fusion and synchronous processing, and adjust the working parameters of the grinding tool through real-time feedback to ensure high-precision detection and adaptive correction of complex surface defects has become a technical problem to be solved by the present invention. Summary of the invention

[0008] The technical problem solved by the present invention is to provide a real-time defect detection and adaptive correction system based on deep learning robot polishing in response to the defects existing in the above-mentioned prior art, so as to solve the problems of poor real-time performance, insufficient adaptability and low detection accuracy of the traditional polishing system proposed in the above-mentioned background technology.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: A real-time defect detection and adaptive correction system for deep learning robot polishing, the system comprising: A multimodal sensing system, used to collect multimodal data in the polishing process in real time, wherein the multimodal data includes image data, three-dimensional point cloud data and torque data; A data synchronization and fusion mechanism for synchronously processing the multimodal data through a time synchronization mechanism to ensure the temporal and spatial consistency of different data sources, and to fuse image data, three-dimensional point cloud data and moment data to generate comprehensive defect detection information; The defect detection module, based on a deep learning convolutional neural network, processes real-time data from the multimodal perception system to detect defects on the polished surface and identify the type, location and severity of the defects; An adaptive correction module generates an adaptive grinding path and control parameters of the grinding tool using a reinforcement learning algorithm based on the defect information output by the defect detection module. The path automatically adjusts the working state of the grinding tool, including pressure, speed and angle, based on the accurate description of the complex geometric shape of the workpiece by the three-dimensional point cloud data; The feedback control module uses a closed-loop control system to monitor the feedback data during the grinding process in real time. The feedback data includes torque data, state information of the grinding surface, and defect repair rate. The grinding parameters are dynamically optimized through a reinforcement learning algorithm to further improve the accuracy of defect detection and correction capabilities. The human-computer interaction module is used to provide the operator with a grinding status monitoring interface, displaying real-time defect detection results and grinding parameter adjustment options.

[0010] As a further solution of the present invention, the data synchronization and fusion mechanism includes: Add timestamps to different types of data from the multimodal perception system for time synchronization, ensuring that image data, 3D point cloud data, and torque data are comprehensively analyzed at the same time point; The image data and 3D point cloud data are combined using a fusion algorithm to generate accurate grinding path and defect location information.

[0011] As a further solution of the present invention, the three-dimensional point cloud data of the multimodal perception system is acquired in real time by a laser scanner and is used to generate a high-precision three-dimensional contour of the workpiece surface, and the accuracy and computational efficiency of the point cloud data are improved by downsampling and denoising.

[0012] As a further solution of the present invention, the defect detection module further includes: Preprocess the image data, including denoising, contrast enhancement, and image resizing, to ensure high-quality data when the image is input into the convolutional neural network to improve the accuracy of defect detection.

[0013] As a further solution of the present invention, the adaptive correction module uses a reinforcement learning algorithm to dynamically adjust the working parameters of the grinding tool, where the working parameters include the pressure, speed, angle and path of the grinding tool, and is optimized and adjusted specifically according to the severity and location of the surface defects of the workpiece.

[0014] As a further solution of the present invention, the reward mechanism of the reinforcement learning algorithm includes: based on the defect repair rate, grinding time and surface smoothness as reward indicators, real-time evaluation of the effect of the grinding process, and maximizing the above indicators by adjusting the grinding parameters to optimize the working path of the grinding tool.

[0015] As a further solution of the present invention, the feedback control module adopts closed-loop control, and the feedback control module continuously monitors the torque data and surface defect repair status during the grinding process, and adjusts the grinding parameters of the adaptive correction module in real time to improve the grinding quality.

[0016] As a further solution of the present invention, the human-computer interaction module includes: Real-time status monitoring interface, used to display defect detection results, tool working parameters and surface status during grinding process; The parameter adjustment interface allows the operator to adjust the working parameters of the grinding tool in real time through the graphical user interface and manually confirm the defect detection results.

[0017] As a further solution of the present invention, the system can automatically adjust the grinding strategy according to the surface of the workpiece of different materials to meet the grinding needs of various materials and ensure that the surface treatment of each material meets the expected standards. As a further solution of the present invention, the multimodal perception system and the adaptive correction module can automatically adjust the grinding path according to the complex geometric shape of the workpiece surface to ensure uniform grinding on irregular surfaces and avoid insufficient or excessive grinding caused by changes in surface height.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. Through the real-time synchronization and closed-loop feedback mechanism of multimodal data fusion, the real-time acquisition, synchronization and comprehensive processing of images, three-dimensional point clouds and torque data during the grinding process are realized. Unlike the existing systems that rely on a single data source, the present invention can provide more accurate defect identification and processing decisions on complex and irregular workpiece surfaces through the synchronous fusion of multi-sensor data, significantly improving the overall accuracy and automation of the grinding process. This multimodal data processing greatly reduces the misjudgment caused by the failure or insufficiency of a single data and enhances the robustness of the system.

[0019] 2. The point cloud-based three-dimensional grinding path optimization realizes the generation of accurate grinding paths for workpieces with complex geometric shapes. The present invention uses highly accurate descriptions of three-dimensional point cloud data to enable grinding tools to perform accurate grinding and defect repair according to the actual three-dimensional surface conditions of the workpiece. Compared with traditional grinding systems that only rely on two-dimensional images, this system can handle irregular surfaces and make adaptive adjustments, reducing the problem of under-grinding or over-grinding and improving the surface quality of the product.

[0020] 3. Use reinforcement learning-driven adaptive correction strategies to optimize the working parameters of grinding tools through real-time feedback. The reinforcement learning algorithm can dynamically adjust the pressure, speed and angle of the grinding tool based on the feedback data during defect detection and grinding, and achieve continuous self-optimization of the grinding process. This feedback-based reinforcement learning not only improves grinding efficiency, but also significantly reduces the need for human intervention, solving the problem of insufficient adaptive ability of existing grinding systems when facing different workpieces, materials and complex surfaces.

[0021] 4. The combination of closed-loop control and adaptive correction enables the system to achieve real-time adjustment and optimization during the grinding process. Through the closed-loop feedback mechanism, the system can immediately correct the grinding path and tool parameters while detecting defects to ensure the quality consistency of each workpiece surface treatment. Compared with the traditional static grinding path, the dynamic optimization strategy of the present invention can cope with complex working conditions and significantly improve the consistency and stability of the grinding process.

[0022] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1 This is a structural block diagram of the deep learning-based robot polishing system of the present invention. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] See also Figure 1 In an embodiment of the present invention, a real-time defect detection and adaptive correction system based on deep learning robot polishing is provided, the system comprising: A multimodal sensing system, used to collect multimodal data in the polishing process in real time, wherein the multimodal data includes image data, three-dimensional point cloud data and torque data; A data synchronization and fusion mechanism for synchronously processing the multimodal data through a time synchronization mechanism to ensure the temporal and spatial consistency of different data sources, and to fuse image data, three-dimensional point cloud data and moment data to generate comprehensive defect detection information; The defect detection module, based on a deep learning convolutional neural network, processes real-time data from the multimodal perception system to detect defects on the polished surface and identify the type, location and severity of the defects; An adaptive correction module generates an adaptive grinding path and control parameters of the grinding tool using a reinforcement learning algorithm based on the defect information output by the defect detection module. The path automatically adjusts the working state of the grinding tool, including pressure, speed and angle, based on the accurate description of the complex geometric shape of the workpiece by the three-dimensional point cloud data; The feedback control module uses a closed-loop control system to monitor the feedback data during the grinding process in real time. The feedback data includes torque data, state information of the grinding surface, and defect repair rate. The grinding parameters are dynamically optimized through a reinforcement learning algorithm to further improve the accuracy of defect detection and correction capabilities. The human-computer interaction module is used to provide the operator with a grinding status monitoring interface, displaying real-time defect detection results and grinding parameter adjustment options.

[0027] As a further specific implementation of the present invention, the following is a detailed description of each technical solution, which explains how these functions are specifically implemented in combination with actual application scenarios.

[0028] 1. Time synchronization mechanism: During the polishing process, different types of data (image data, 3D point cloud data, and torque data) from the multimodal perception system are collected at different frequencies and time sequences. In order to ensure that these data can accurately reflect the state of the workpiece surface at the same time, the system adds a timestamp to each data stream. The time synchronization mechanism corrects and aligns all data based on these timestamps to ensure that the image, 3D contour, and torque data can be comprehensively analyzed at the same time. For example, when the robot simultaneously collects visual information and 3D topography of the workpiece surface, the time synchronization mechanism ensures that this information is consistent with each other, thereby avoiding a decrease in polishing accuracy due to time differences.

[0029] 2. Data fusion algorithm: In order to generate more accurate grinding paths and defect location information, the system uses a data fusion algorithm to combine image data and 3D point cloud data. Specifically, image data provides high-definition visual information of the workpiece surface, and 3D point cloud data captures the 3D structure and height changes of the workpiece surface. The fusion algorithm superimposes and processes data from these different sources through a matching algorithm to generate more comprehensive surface information. In practical applications, such as processing workpieces with complex geometric shapes (such as aircraft engine blades or car bodies), the fusion algorithm can accurately locate the position and shape of surface defects, helping the system to better plan the grinding path and ensure that every surface defect can be accurately repaired.

[0030] 3. Acquisition and processing of three-dimensional point cloud data of multimodal perception system: In actual applications, three-dimensional point cloud data is acquired in real time by a laser scanner installed on the robot. The laser scanner emits a laser beam to the surface of the workpiece and generates the three-dimensional contour of the workpiece through the reflected signal. This process can accurately capture the height changes and shape details of the workpiece surface, and is especially suitable for surface modeling of workpieces with complex geometric shapes. To improve processing efficiency, the system downsamples the collected point cloud data to reduce the amount of data and increase processing speed. At the same time, the system uses a denoising algorithm to remove noise and abnormal data in the point cloud to ensure that the final input point cloud data is highly accurate and has higher computational efficiency. Taking aviation blade grinding as an example, the laser scanner can generate a fine three-dimensional model of the blade and identify tiny depressions and protrusions on the surface to help subsequent grinding tools accurately adjust the path.

[0031] 4. Preprocessing of image data: To ensure that the image data can be input into the convolutional neural network (CNN) for defect detection with high quality, the system performs multi-step preprocessing on the image data collected by the camera. First, Gaussian filtering or median filtering is used to remove image noise to avoid interference with subsequent image analysis. Secondly, the system uses techniques such as histogram equalization to enhance image contrast, making surface details of the workpiece (such as scratches or uneven grinding areas) clearer. Finally, the system adjusts the image size and scales the image to a fixed size suitable for CNN model input (such as 224x224 pixels). Through these preprocessing steps, the image data input to CNN is ensured to have sufficient quality, thereby improving the accuracy of defect detection. For example, when grinding complex-shaped body panels, the system can clearly capture tiny scratches on the surface, avoiding situations that traditional vision systems cannot recognize.

[0032] 5. Dynamic adjustment of the adaptive correction module: The adaptive correction module uses a reinforcement learning algorithm to dynamically adjust the working parameters of the grinding tool, including pressure, speed, angle, and path. Specifically, the system generates a corresponding grinding strategy based on the defect location and severity fed back by the defect detection module. For example, for a deep scratch on a complex surface, the system will automatically increase the grinding pressure and reduce the speed to ensure that the grinding tool can effectively repair the scratched area; for minor surface depressions, the system may adjust the angle and path of the grinding tool to complete the repair with lighter pressure to avoid excessive grinding. This dynamic adjustment process is driven by real-time feedback and defect detection results to ensure that the grinding tool is always in the best working condition.

[0033] 6. Reward mechanism of reinforcement learning algorithm: The reinforcement learning algorithm in the system evaluates the effect of the polishing process through a reward mechanism and adjusts the polishing parameters according to the reward indicators. The mechanism includes three main indicators: defect repair rate, polishing time and surface smoothness. Reinforcement learning dynamically adjusts the polishing path and working parameters by optimizing these indicators. For example, in automobile manufacturing, the system rewards the most effective polishing strategy and minimizes the polishing time based on the detected scratch repair and surface smoothness score. This mechanism ensures that the polishing tool can repair defects with the optimal strategy, improving efficiency and polishing quality.

[0034] 7. Closed-loop control of the feedback control module: The feedback control module uses closed-loop control technology to continuously monitor the torque data and surface defect repair during the grinding process. Through the torque sensor, the system monitors the pressure applied by the grinding tool on the workpiece surface in real time to ensure stability during the grinding process. At the same time, the system compares the surface smoothness and defect repair effect after grinding in real time. If insufficient grinding or over-grinding is detected, the system will automatically adjust the pressure, speed and path to improve the grinding effect. For example, when processing large aluminum plates, if insufficient edge grinding is detected, the system will immediately re-plan the grinding path and increase the pressure appropriately to achieve the best repair effect.

[0035] 8. Real-time monitoring and parameter adjustment of the human-computer interaction module: The human-computer interaction module provides an intuitive graphical user interface (GUI). The operator can view the defect detection results, tool working parameters (such as pressure, speed, angle) and surface repair status of the current grinding process through the real-time status monitoring interface. If the system detects complex defects, the operator can manually fine-tune the working parameters of the grinding tool through the parameter adjustment interface. Through the graphical interface, the operator can easily confirm or correct the grinding strategy automatically generated by the system, so that automation and manual operation can be flexibly combined. For example, the operator can fine-tune the parameters of a specific material to ensure that the surface treatment of the workpiece meets strict quality requirements.

[0036] 9. Automatically adjust the grinding strategy to suit different materials: The system can automatically adjust the grinding strategy according to the surface of workpieces of different materials (such as metal, plastic or composite materials). For example, for the grinding of metal materials, the system will use higher pressure and lower speed to ensure uniform surface treatment; while for plastic surfaces, the system will reduce the grinding pressure and tool speed to avoid damaging the surface. By automatically identifying material properties and adjusting the grinding strategy, the system can adapt to a variety of grinding needs to ensure that the surface treatment of each material meets the expected standards.

[0037] 10. Adjust the grinding path according to complex geometric shapes: The multimodal perception system and adaptive correction module in the system can automatically adjust the grinding path according to the complex geometric shapes of the workpiece (such as curved surfaces or uneven surfaces). Taking the grinding of engine blades as an example, the system obtains the three-dimensional contour of the blade surface in real time, and automatically adjusts the angle and path of the grinding tool according to the height change to ensure uniform grinding on irregular surfaces. Through this adjustment mechanism, the system effectively avoids the problem of insufficient or excessive grinding caused by changes in surface height, and improves the overall quality of grinding.

[0038] Embodiment 1: This embodiment provides a real-time defect detection and adaptive correction system for robot polishing based on deep learning, which is applied to the precision polishing of the body surface in automobile manufacturing. In this application scenario, the polishing quality of the body surface is crucial to the appearance and performance of the final product, especially in high-end automobile manufacturing, where the requirements for body smoothness and flawlessness are extremely high. Traditional polishing methods are difficult to ensure efficient and accurate polishing on complex body geometries and surfaces of various materials, which often leads to problems such as uneven surface treatment and untimely detection of defects. Therefore, the introduction of the system of the present invention significantly improves the automation level of the production line and the polishing effect.

[0039] In actual application, the robot first uses a multimodal perception system to perform real-time detection of the body surface. The perception system includes a high-definition camera, a laser scanner, and a torque sensor, which are used to simultaneously collect images, three-dimensional point cloud data, and torque data applied by the grinding tool. During the grinding process of the body surface, the data of these sensors are synchronized and fused through a data synchronization and fusion mechanism to ensure that the visual information, shape information, and grinding force information of the body surface can be coordinated at any time, thereby providing high-precision basic data for subsequent defect detection and adaptive correction.

[0040] For example, when polishing complex body contours (such as the transition area between the door and the body), the system uses a laser scanner to obtain real-time 3D point cloud data of the surface, and combines it with high-definition images captured by the camera to generate an accurate body surface model. These data are processed by a deep learning convolutional neural network (CNN) to identify subtle scratches, irregular bumps or edge defects on the surface. Traditional polishing systems may not be able to detect these minor defects at the first time, especially when dealing with areas with poor lighting or complex surfaces, but the present invention can detect these problems in a timely and accurate manner with the help of high-precision fusion data from a multimodal perception system.

[0041] When defect detection is completed, the system will automatically generate an optimized grinding path and parameter adjustment strategy based on the reinforcement learning algorithm in the adaptive correction module. Taking the presence of slight scratches on the body surface as an example, the system will adjust the angle and pressure of the grinding tool by analyzing the three-dimensional profile data of the area. For areas with deeper scratches, the system will reduce the grinding speed and increase the pressure to ensure that the scratches can be completely repaired. For areas where the body is more fragile or on the edge, the system will automatically reduce the grinding force and speed to avoid damage to the body surface caused by excessive grinding. The reinforcement learning algorithm plays a key role in this process. The system adjusts the parameters of the grinding tool through real-time feedback to ensure the grinding effect of each area.

[0042] In addition, the feedback control module continuously monitors the torque feedback data of the grinding tool and the real-time status of the grinding surface to ensure that each action is performed within the expected range. If the system detects that there are still defects or insufficient processing on the surface after grinding, it will immediately readjust the grinding path or parameters and continue to perform correction operations until the detected defects are completely repaired. For example, at the edge of the car body or in areas with complex curves, the traditional method is prone to edge wear due to fixed grinding paths or excessive force, while the closed-loop control mechanism of the present invention ensures real-time response to every tiny change, greatly reducing the defective rate.

[0043] The human-computer interaction module of the system provides a real-time monitoring interface for the operator, who can view the defect detection results and tool parameter adjustment during the grinding process at any time through the graphical interface. If necessary, the operator can also manually adjust certain parameters or paths for fine-tuning operations. This greatly enhances the operability of the system, achieving fully automated grinding while retaining the flexibility of manual intervention, ensuring that the system can adapt to different production needs.

[0044] Through the technical solution of this embodiment, the system realizes efficient polishing of complex car body surfaces in the field of automobile manufacturing, which is a significant technological advancement compared to traditional technologies. Its multimodal data fusion processing capability ensures higher detection accuracy, and the adaptive correction strategy based on reinforcement learning ensures the processing quality of each workpiece surface, whether it is a flat area or a complex curved area, it can achieve a uniform and precise polishing effect. Especially when processing high-gloss car body surfaces in high-end automobile manufacturing, this system can significantly improve the polishing quality, reduce manual intervention, greatly improve production efficiency, reduce rework rate, and ensure product consistency and stability.

[0045] Embodiment 2: This embodiment demonstrates the application of the present invention in the surface treatment of complex parts in the aerospace field. In aviation manufacturing, the surface quality of parts such as aircraft engine blades and fuselage structural parts is crucial to the overall performance. Especially in high-speed flight, whether the surface of the parts is smooth and flawless directly affects the aerodynamic performance and fuel efficiency. Traditional manual grinding or fixed path grinding methods are often unable to cope with complex blade curves or material surface irregularities, have low grinding efficiency and poor surface treatment consistency, and easily lead to uneven grinding or local defects.

[0046] In this application scenario, the deep learning-based robot grinding real-time defect detection and adaptive correction system of the present invention provides significant technological progress. First, the system's multimodal perception system uses high-definition cameras, laser scanners, and torque sensors to simultaneously collect multimodal data on the surface of complex blades, including three-dimensional contour data of surface morphology, torque information during grinding, and high-definition image data. These data are processed in real time through data synchronization and fusion mechanisms to ensure that the complex geometric structure of the blade can be accurately captured, providing accurate basic information for subsequent grinding defect detection and adaptive correction.

[0047] In specific operations, when the robot is grinding the blade surface, the system uses a laser scanner to obtain real-time 3D point cloud data of the blade surface, combined with the surface details captured by the camera, to detect surface defects such as tiny pits, wear or uneven surfaces generated during the grinding process. This defect detection module is trained based on a deep learning convolutional neural network (CNN) and can accurately identify different types of surface defects, even tiny defects that are difficult to detect with traditional visual systems.

[0048] For example, when the system detects a slight dent in a curved area of ​​the blade, the adaptive correction module immediately generates a correction strategy to adjust the angle, pressure, and speed of the grinding tool. Through the reinforcement learning algorithm, the system dynamically optimizes the grinding parameters to ensure that the depressed area is accurately repaired without causing damage to the surrounding surface. For the processing of complex curved surfaces of blades, traditional fixed-path grinding can easily lead to insufficient or excessive grinding in certain areas. This system ensures the uniformity of the blade surface grinding through real-time feedback control and path adjustment.

[0049] During the grinding process, the feedback control module monitors the changes in the blade surface and the operating status of the grinding tool in real time, such as real-time torque data and surface repair rate. If the system detects that the smoothness of the blade surface or the defect repair rate does not meet the preset standards, the system will immediately adjust the parameters of the grinding tool and re-plan the grinding path. Especially on the blade edge or sharp curved surface area, traditional grinding methods often lead to over-grinding or under-grinding due to operational errors, and the closed-loop feedback mechanism of the present invention can avoid these problems and ensure that the entire surface treatment meets the high standards required for aviation manufacturing.

[0050] In addition, the human-computer interaction module of this system provides an intuitive operation interface for operators. Operators can use the interface to monitor the grinding status in real time, view defect detection results, and adjust system parameters as needed. For example, when processing certain specific materials or particularly complex surfaces, operators can manually fine-tune the grinding parameters, so that the system's automation and manual operation can be flexibly combined to meet individual special needs that may arise during the production process.

[0051] Through this embodiment, the present invention realizes efficient grinding of complex blade surfaces and irregular parts in the aviation manufacturing industry, and solves the problems of uneven grinding and untimely defect detection in traditional grinding systems when dealing with complex surfaces. The system can detect and correct defects in the grinding process in real time, significantly improving the consistency and quality of surface treatment. Compared with traditional technologies, the present invention not only reduces manual intervention, but also greatly reduces the rework rate, improves production efficiency, and provides an intelligent solution for high-end manufacturing in the aerospace field.

[0052] Embodiment 3: This embodiment introduces the specific implementation process of the deep learning-based robot polishing real-time defect detection and adaptive correction system in data collection and defect detection.

[0053] The system first collects data through a multimodal perception system, including a high-definition camera, laser scanner, and torque sensor. The high-definition camera is responsible for capturing details on the workpiece surface, such as scratches or dents; the laser scanner is used to generate a three-dimensional contour map of the workpiece, which can accurately detect surface unevenness or insufficient grinding areas; the torque sensor is installed between the grinding tool and the robot arm to monitor the pressure applied to the workpiece surface in real time to ensure the stability of the force.

[0054] During the data collection process, the collection system is first started, and each sensor is initialized and self-checked to ensure its normal operation. During the grinding process, the high-definition camera collects image data at a fixed frame rate, and reduces the impact of reflections and shadows through appropriate lighting conditions (such as LED lights) to ensure clear images; the laser scanner scans the surface of the workpiece to obtain high-precision three-dimensional point cloud data, which will be used for three-dimensional modeling and analysis later. The torque sensor records the real-time pressure and torque applied by the grinding tool, and stores this data in a time series for subsequent analysis.

[0055] In order to ensure that the data from different sensors correspond in time, the system uses a timestamp marking mechanism to synchronize the data streams, ensuring that the image, 3D profile and torque data are analyzed at the same time. All collected data is transmitted to the control system in real time through a high-speed data transmission protocol and stored in a local database for use by subsequent defect detection modules.

[0056] During the defect detection phase, the system receives and processes the collected image data, 3D point cloud data, and moment data. The image data is first preprocessed, including denoising (such as Gaussian filtering), contrast enhancement (such as histogram equalization), and image resizing to ensure high quality data input to the convolutional neural network (CNN), thereby improving the accuracy of defect detection. The point cloud data is downsampled and outlier removed to reduce the amount of calculation and remove noise, retaining valid data.

[0057] The defect detection module analyzes preprocessed images and point cloud data. The system uses a trained convolutional neural network (CNN) to identify different types of defects (such as scratches and dents) and mark the location and severity of the defects. Taking the polishing of a car body as an example, the system can detect tiny scratches or dents on the surface of the car body and output the category and location of the defect based on the confidence of the model.

[0058] After the inspection is completed, the system passes the defect information to the adaptive correction module, which automatically generates a repair strategy and adjusts the working parameters of the grinding tool (such as pressure, speed and angle) according to the defect type, location and severity. If a scratch is detected, the system may increase the grinding pressure and reduce the speed; if it is a recessed area, the system may adjust the angle of the grinding tool to cover the defective area.

[0059] The entire process is continuously optimized through a closed-loop feedback control mechanism, ensuring that the system can dynamically adjust parameters during the polishing process and detect and repair surface defects in real time.

[0060] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.

Claims

1. A real-time defect detection and adaptive correction system based on deep learning robot polishing, characterized in that: The system comprises: A multimodal sensing system, used to collect multimodal data in the polishing process in real time, wherein the multimodal data includes image data, three-dimensional point cloud data and torque data; A data synchronization and fusion mechanism for synchronously processing the multimodal data through a time synchronization mechanism to ensure the temporal and spatial consistency of different data sources, and to fuse image data, three-dimensional point cloud data and moment data to generate comprehensive defect detection information; The defect detection module, based on a deep learning convolutional neural network, processes real-time data from the multimodal perception system to detect defects on the polished surface and identify the type, location and severity of the defects; An adaptive correction module generates an adaptive grinding path and control parameters of the grinding tool using a reinforcement learning algorithm based on the defect information output by the defect detection module. The path automatically adjusts the working state of the grinding tool, including pressure, speed and angle, based on the accurate description of the complex geometric shape of the workpiece by the three-dimensional point cloud data; The feedback control module uses a closed-loop control system to monitor the feedback data during the grinding process in real time. The feedback data includes torque data, state information of the grinding surface, and defect repair rate. The grinding parameters are dynamically optimized through a reinforcement learning algorithm to further improve the accuracy of defect detection and correction capabilities. The human-computer interaction module is used to provide the operator with a grinding status monitoring interface, displaying real-time defect detection results and grinding parameter adjustment options.

2. According to claim 1, a deep learning robot polishing real-time defect detection and adaptive correction system is characterized in that: Data synchronization and fusion mechanisms include: Add timestamps to different types of data from the multimodal perception system for time synchronization, ensuring that image data, 3D point cloud data, and torque data are comprehensively analyzed at the same time point; The image data and 3D point cloud data are combined using a fusion algorithm to generate accurate grinding path and defect location information.

3. According to claim 1, a real-time defect detection and adaptive correction system based on deep learning robot polishing is characterized in that: The three-dimensional point cloud data of the multimodal perception system is acquired in real time by a laser scanner and is used to generate a high-precision three-dimensional contour of the workpiece surface. The accuracy and computational efficiency of the point cloud data are improved by downsampling and denoising.

4. According to claim 1, a real-time defect detection and adaptive correction system based on deep learning robot polishing is characterized in that: The defect detection module further includes: Preprocess the image data, including denoising, contrast enhancement, and image resizing, to ensure high-quality data when the image is input into the convolutional neural network to improve the accuracy of defect detection.

5. According to the deep learning robot polishing real-time defect detection and adaptive correction system of claim 1, it is characterized in that: The adaptive correction module uses a reinforcement learning algorithm to dynamically adjust the working parameters of the grinding tool, wherein the working parameters include the pressure, speed, angle and path of the grinding tool.

6. The real-time defect detection and adaptive correction system based on deep learning robot polishing according to claim 1 is characterized in that: The reward mechanism of the reinforcement learning algorithm includes: based on the defect repair rate, grinding time and surface smoothness as reward indicators, real-time evaluation of the effect of the grinding process, and maximizing the above indicators by adjusting the grinding parameters to optimize the working path of the grinding tool.

7. The real-time defect detection and adaptive correction system based on deep learning robot polishing according to claim 1 is characterized in that: The feedback control module adopts closed-loop control, which continuously monitors the torque data and surface defect repair status during the grinding process, and adjusts the grinding parameters of the adaptive correction module in real time to improve the grinding quality.

8. The real-time defect detection and adaptive correction system based on deep learning robot polishing according to claim 1 is characterized in that: The human-computer interaction module includes: Real-time status monitoring interface, used to display defect detection results, tool working parameters and surface status during grinding process; The parameter adjustment interface allows the operator to adjust the working parameters of the grinding tool in real time through the graphical user interface and manually confirm the defect detection results.

9. The real-time defect detection and adaptive correction system based on deep learning robot polishing according to claim 1 is characterized in that: The system can automatically adjust the grinding strategy according to the surface of workpieces of different materials to meet the grinding needs of various materials and ensure that the surface treatment of each material meets the expected standards.

10. The real-time defect detection and adaptive correction system based on deep learning robot polishing according to claim 1 is characterized in that: The multimodal perception system and the adaptive correction module can automatically adjust the grinding path according to the complex geometry of the workpiece surface to ensure uniform grinding on irregular surfaces and avoid under-grinding or over-grinding problems caused by surface height changes.

Images