Multi-type industrial workpiece AI recognition intelligent detection system and method

Abstract

The application discloses a multi-type industrial workpiece AI recognition intelligent detection system and method, and belongs to the technical field of industrial quality detection. The system comprises a workpiece standard library module, a defect feature standard library module, a workpiece identity recognition module, a two-dimensional acquisition module, a three-dimensional scanning module, a data processing module, a data fusion module, a detection model training module, an online defect detection module, an artificial review feedback module and a report tracing module. The method comprises the following steps: collecting multi-view data of a workpiece; completing image and point cloud preprocessing and coordinate alignment; constructing a defect analysis object; training a detection model; carrying out online defect detection; performing artificial review feedback, generating a report and archiving and tracing. The application completes two-dimensional image and three-dimensional point cloud acquisition of a workpiece through a set of conveying and detection line, cooperatively judges through two-dimensional, three-dimensional or fusion models matched with defect types, considers surface texture defect recognition and three-dimensional geometric defect measurement, and effectively improves the intelligent level of workpiece defect detection.

Description

Technical Field

[0001] This invention relates to the field of industrial quality inspection technology, and in particular to an artificial intelligence defect identification and intelligent inspection system for various types of industrial workpieces. Background Technology

[0002] Industrial workpieces typically require quality checks during procurement, manufacturing, rework, and service maintenance, including inspections for appearance defects, dimensional deviations, localized deformation, surface damage, and material deficiencies. Intelligent inspection of industrial workpieces is a core component of quality control in intelligent manufacturing. Furthermore, the accuracy of workpiece defect identification, dimensional measurement precision, multi-category adaptability, and end-to-end traceability directly determine product pass rates and the efficiency of refined production line management. Multimodal data fusion capabilities, standardized inspection configuration systems, dynamic algorithm scheduling mechanisms, and continuous model iteration and learning capabilities are key factors influencing the overall performance of intelligent inspection of industrial workpieces.

[0003] Traditional methods primarily rely on manual visual inspection, paper records, partial photography, single-point measurement, or independent 2D vision inspection equipment. Some automated inspection solutions use line scan cameras, area scan cameras, or 3D scanners for data acquisition, but their inspection processes are often designed around a single workpiece type or a single defect type, lacking a unified correlation between data acquisition, model recognition, dimensional measurement, and report generation. Currently, deep learning object detection models, 3D point cloud processing algorithms, and multi-sensor fusion algorithms are gradually being used in industrial inspection; however, existing technologies still suffer from the following core technical shortcomings: 1. Insufficient adaptability to single workpieces: Existing inspection processes typically rely on fixed fixtures, fixed shooting angles, and fixed models. After a change of model, a large amount of manual reconfiguration is required, making it difficult to adapt to industrial workpieces of various shapes, sizes, and inspection standards.

[0004] 2. Incomplete single-modal detection information and lack of 2D / 3D fusion: Low detection accuracy; 2D detection lacks depth information and cannot identify spatial deformation defects, easily leading to false positives and false negatives; 3D detection lacks texture details and is difficult to identify surface defects such as scratches, color differences, and contamination. Furthermore, the two types of data are processed independently, lacking coordinate unification, region mapping, depth compensation, and consistency verification.

[0005] 3. Lack of data traceability and intelligent iteration closed loop: Existing detection data, defect records, review information and reports are independent and cannot be uniformly bound for traceability; the model is fixed and cannot be iterated, lacking continuous training and version update mechanisms; manual review results cannot be structured and recycled, and automatic detection and manual review are disconnected from each other. Summary of the Invention

[0006] The purpose of this invention is to provide an AI-based intelligent detection system and method for multi-type industrial workpieces. To address the aforementioned problems in the prior art, this invention takes workpiece identification and detection tasks as the entry point, integrating the two-dimensional image, three-dimensional point cloud, standard model, detection area, defect rules, and manual review results of the industrial workpiece into a unified detection data link. Multi-view data of the workpiece is acquired through a two-dimensional acquisition subsystem and a three-dimensional scanning subsystem. Image preprocessing, point cloud preprocessing, and coordinate alignment are completed through a data processing module. Two-dimensional texture information, three-dimensional geometric information, and standard model area information are fused into a unified defect analysis object through a data fusion module. A two-dimensional detection model, a three-dimensional detection model, and a fused detection model are trained through a detection model training module. An online defect detection module outputs the defect type, defect location, length, width, and depth dimensions, area or volume, severity, and judgment criteria. A closed loop is formed through manual review, evaluation optimization, and report traceability modules.

[0007] To achieve the above objectives, this invention provides an AI-based intelligent detection system and method for recognizing multiple types of industrial workpieces, comprising the following steps: A multi-type industrial workpiece AI recognition and intelligent inspection system, including: The system includes a workpiece standard library module, a defect feature standard library module, a workpiece identification module, a 2D acquisition module, a 3D scanning module, a data processing module, a data fusion module, a detection model training module, an online defect detection module, a manual review and feedback module, and a report traceability module. The workpiece standard library module is communicatively connected to the workpiece identification module, the two-dimensional acquisition module, the three-dimensional scanning module, and the online defect detection module. It is used to store standard models, detection areas, acquisition paths, and judgment rules for various types of industrial workpieces, providing a workpiece-dimensional configuration basis for the entire detection process. The defect feature standard library module is connected to the detection model training module and the online defect detection module, respectively. It is used to store defect categories, two-dimensional features, three-dimensional features, size calculation methods and severity rules, providing standard basis for defect dimensions for model training and defect judgment. The workpiece identification module is communicatively connected to the workpiece standard library module, the two-dimensional acquisition module, and the three-dimensional scanning module. It is used to obtain the model and unique serial number of the workpiece to be inspected and create an inspection task. After retrieving the configuration parameters of the corresponding workpiece, it sends an acquisition command to the downstream acquisition module. The two-dimensional acquisition module is communicatively connected to the workpiece identification module, the workpiece standard library module, and the data processing module, respectively. It is used to acquire two-dimensional images of workpieces under the same detection task according to a preset acquisition path and transmit the original two-dimensional images to the data processing module. The 3D scanning module is communicatively connected to the workpiece identification module, the workpiece standard library module, and the data processing module, respectively. It is used to collect the 3D point cloud of the workpiece under the same detection task according to the preset acquisition path, and transmit the original 3D point cloud to the data processing module. The data processing module is communicatively connected to the two-dimensional acquisition module, the three-dimensional scanning module, and the data fusion module, respectively, and is used to complete the two-dimensional image preprocessing, the three-dimensional point cloud preprocessing and coordinate alignment, and synchronously transmit the preprocessed two-dimensional image and the three-dimensional point cloud to the data fusion module. The data fusion module is communicatively connected to the data processing module and the online defect detection module, respectively, and is used to register the preprocessed two-dimensional image and three-dimensional point cloud to a unified coordinate system, generate fused defect features and transmit them to the online defect detection module. The detection model training module is connected to the defect feature standard library module, the manual review and feedback module, and the online defect detection module, respectively. It is used to train and update the two-dimensional detection model, the three-dimensional detection model, and the fusion detection model based on defect feature rules and return samples, and to deploy the updated detection model to the online defect detection module. The online defect detection module is connected to the workpiece standard library module, the defect feature standard library module, the data fusion module, the detection model training module, the manual review and feedback module, and the report traceability module. It is used to dynamically call the corresponding detection model according to the workpiece type, defect category, and fused defect features, output the defect location, defect type, defect size, and severity, and transmit the detection results to the manual review and feedback module and the report traceability module respectively. The manual review and feedback module is connected to the online defect detection module, the detection model training module, and the report traceability module, respectively. It is used to receive the detection results and output the manual confirmation or correction results, transmit the final judgment result after review to the report traceability module, and at the same time, return the labeled and corrected sample data to the detection model training module. The report traceability module is connected to the online defect detection module, the manual review and feedback module, and the detection model training module. It is used to associate and store the detection results, review results, model version, and detection report with the unique serial number of the workpiece, and supports multi-dimensional historical data traceability and query.

[0008] Preferably, the workpiece identification module communicates with both the 2D acquisition module and the 3D scanning module to collect workpiece data; the workpiece standard library module includes a standard 3D model of the workpiece, detection attributes, acquisition parameters, size thresholds, and version number, which is used to abstract multiple types of workpieces into configurable detection objects and support the reuse of the same detection process on different workpieces; the defect feature standard library module presets multiple types of defect standards, covering workpiece surface, edge, shape, coating, and contact abnormality defects, and uses the defect feature library to drive model training, detection pipeline selection, and result judgment.

[0009] Preferably, the 2D acquisition module and the 3D scanning module communicate with the data processing module and the data fusion module respectively to complete data preprocessing and feature fusion; the 2D acquisition module is used to acquire multi-view 2D images of the workpiece; the 3D scanning module is used to acquire point clouds or mesh models of the workpiece; the data processing module performs 2D image preprocessing and 3D point cloud preprocessing; the 2D image preprocessing includes one or more of the following: noise removal, distortion correction, image stitching, ROI extraction, multi-view alignment, brightness and contrast optimization; the 3D point cloud preprocessing includes one or more of the following: noise filtering, point cloud simplification, resampling, normal calculation, mesh reconstruction, boundary clipping, coordinate unification, and multi-segment stitching.

[0010] Preferably, the data fusion module and the online defect detection module communicate to transmit defect feature data; the data fusion module is used to perform coordinate system registration, feature point matching, and texture mapping between the two-dimensional image and the three-dimensional point cloud, and to perform depth compensation and consistency verification between the two; the registration between the two-dimensional image and the three-dimensional point cloud includes coordinate transformation based on calibration parameters, matching based on feature points, registration optimization based on ICP, or a combination thereof.

[0011] Preferably, the detection model training module and the online defect detection module communicate to deploy and update the detection model. The detection model training module manages sample import, annotation, dataset partitioning, model training, hyperparameter tuning, model evaluation, version management, and deployment. The online defect detection module has a built-in dynamic detection pipeline. The dynamic detection pipeline can select a detection model based on defect category, regional attributes, data integrity, model confidence, or consistency results between 2D and 3D. When the 2D and 3D detection results are inconsistent, consistency verification or manual review is triggered. The online defect detection module communicates with the manual review feedback module and the report traceability module to transmit defect judgment results and result summaries. The manual review feedback module displays the defect location, 2D image, 3D morphology, and fusion judgment criteria, and supports defect editing operations. The report traceability module supports multi-dimensional queries based on the workpiece's unique serial number, detection task ID, model version, detection time, and defect category.

[0012] A preferred method for AI-based intelligent detection of various types of industrial workpieces includes the following steps: S1. Establish a workpiece standard library and a defect feature standard library; S2. Identify the model and unique serial number of the workpiece to be inspected and create an inspection task; S3. Determine the detection area and data acquisition path based on the workpiece standard library; S4. Acquire two-dimensional images and three-dimensional point clouds of the workpiece to be inspected; S5. Perform preprocessing on the two-dimensional image and the three-dimensional point cloud respectively; S6. Register the two-dimensional image and the three-dimensional point cloud to a unified coordinate system and merge them into a defect analysis object; S7. Dynamically call two-dimensional detection models, three-dimensional detection models, or fusion detection models based on the defect feature standard library and model confidence; compare the two-dimensional detection results with the three-dimensional detection results, and if the two conclusions are inconsistent, trigger the consistency verification or manual review process. S8. Output the location, type, size, and severity of the defect; S9. Receive the results of manual review and feed them back to the model training data; S10. Generate and store a test report associated with a unique serial number.

[0013] Preferably, a workpiece standard library and a defect feature standard library are established, including: Input the workpiece type, standard 3D model, external dimensions, key inspection areas, inspection surfaces, prohibited inspection areas, acquisition path, clamping posture, area attributes and quality judgment criteria of various industrial workpieces to construct the workpiece standard library; The two-dimensional features, three-dimensional features, size calculation methods, and judgment thresholds of various defects are entered to construct the defect feature standard library.

[0014] Preferably, the model and unique serial number of the workpiece to be inspected are identified through one or more methods such as machine vision, OCR, barcode, QR code, or RFID; the workpiece surface images are acquired by controlling an industrial camera, light source, turntable, robot, or conveyor mechanism in multiple postures, and the images are bound with task ID, viewpoint, timestamp, and spatial posture parameters to complete the acquisition of two-dimensional images; the workpiece point cloud or mesh data is acquired by controlling a three-dimensional scanning device, and the point cloud coordinates, scanning posture, scanning segment information, and task ID are bound to complete the acquisition of three-dimensional data; the two-dimensional images and three-dimensional point cloud data are transmitted to the data processing module for preprocessing, including noise removal, distortion correction, brightness and contrast optimization, image stitching, ROI extraction, and multi-viewpoint alignment for the two-dimensional images; and noise filtering, point cloud simplification, resampling, normal calculation, mesh reconstruction, boundary clipping, coordinate unification, and multi-segment stitching for the three-dimensional data.

[0015] Preferably, the specific steps for registering two-dimensional images and three-dimensional point clouds to a unified coordinate system and fusing them into a defect analysis object include: By using coordinate transformation matrices, calibration parameters, feature point matching, or ICP registration algorithms, two-dimensional images and three-dimensional point clouds can be unified into the same coordinate system. Two-dimensional textures or defect candidate regions are mapped to three-dimensional point clouds or mesh models, and three-dimensional depth information is used to supplement the spatial information missing in the two-dimensional image. The current workpiece data is matched with the standard models and regional partitioning information in the workpiece standard library, and different defect thresholds, model versions and judgment rules are called for different detection areas.

[0016] Preferably, the detection model is dynamically invoked, and the corresponding detection model is matched according to the defect category. For surface texture defects, a two-dimensional detection model is invoked to extract texture features, and for spatial deformation defects, a three-dimensional detection model is invoked to extract geometric features. A fusion detection model is used to combine two-dimensional texture features and three-dimensional geometric features for joint defect judgment and size calculation. The results are manually reviewed and fed back to the model training data. Confirmed defects are used as positive samples, false detection areas are used as negative samples, and corrected categories or boundaries are used as annotation correction samples to achieve iterative optimization of the detection model. The detection report is associated with and stores the workpiece identity, collected data, detection results, review opinions, model version, report file, and operation log to support multi-dimensional traceability and query.

[0017] Therefore, the beneficial effects of the above-mentioned AI-based intelligent detection system and method for multi-type industrial workpieces are as follows: By configuring workpiece types, inspection areas, and judgment rules through workpiece standard libraries and defect feature standard libraries, the manual reconstruction cost during model changeover can be reduced, and the adaptability of multiple workpiece types can be improved. Two-dimensional images provide surface texture information, while three-dimensional point clouds provide depth and spatial geometry information. The fusion of the two can compensate for the shortcomings of a single sensor and reduce detection blind spots. The system can simultaneously output the length, width, depth, area, volume, and spatial location of defects, making it particularly suitable for defects that require depth or thickness determination, thus improving the reliability of defect size measurement. By dynamically calling two-dimensional models, three-dimensional models, and fusion models, and combining multi-view consistency verification, the risk of false detection and false negative detection is reduced, and the recognition accuracy and robustness are improved. The system can automatically complete workpiece identification, data acquisition path recall, data processing, model reasoning, and report generation, reducing repetitive manual operations and improving inspection efficiency. The workpiece serial number, collected data, defect results, review comments, model version, and report file are uniformly bound together, which facilitates subsequent traceability and quality analysis and enhances the traceability of results. The results of manual review and new samples can be fed back into the training dataset to continuously expand the ability to identify new workpieces and new defects, supporting continuous learning; When adding new 2D acquisition devices, 3D scanning devices, algorithm models, or reporting systems, they can be connected through data interfaces and model version management modules, facilitating system expansion.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This invention provides an overall architecture block diagram of an AI-based intelligent detection system for multiple types of industrial workpieces. Figure 2 This invention provides an overall flowchart of an AI-based intelligent detection method for multiple types of industrial workpieces. Figure 3 This invention provides a block diagram illustrating the principle of fusing two-dimensional images and three-dimensional point cloud data in an AI-based intelligent detection method for multiple types of industrial workpieces. Figure 4 This invention provides a flowchart of the dynamic scheduling and result determination process for the detection model in an AI-based intelligent detection method for multiple types of industrial workpieces. Figure 5 This invention provides a flowchart of the detection model training and closed-loop iterative optimization process in an AI-based intelligent detection method for multiple types of industrial workpieces. Figure 6 This invention provides a block diagram illustrating the full-link data traceability association for the detection task in an AI-based intelligent detection method for multiple types of industrial workpieces. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0021] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0022] The following is in conjunction with the appendix Figure 1-6 The embodiments of the present invention will be described in detail below.

[0023] Example 1: Intelligent inspection of complex curved surface industrial workpieces throughout the entire process: This embodiment takes complex curved metal industrial workpieces with multiple detection surfaces, curved transition areas, and local groove structures as the detection object, and realizes intelligent defect identification and dimensional detection throughout the entire process.

[0024] Before the inspection begins, the system uses a workpiece identification module with machine vision to identify the model and unique serial number of the current workpiece, generating a unique inspection task ID. It then retrieves the corresponding standard 3D model of the workpiece, key inspection area division rules, standard acquisition paths, and quality judgment thresholds pre-stored in the workpiece standard library module. Subsequently, the system controls a conveyor mechanism to transport the workpiece to the inspection station, and automatically adjusts the workpiece's posture via a rotary table or robot to ensure that the acquisition view fully covers the key inspection areas of the workpiece.

[0025] Once in place, the 2D acquisition module, equipped with an industrial camera and a matching light source, acquires 2D images of the workpiece surface from multiple perspectives. The data processing module sequentially performs distortion correction, brightness and contrast optimization, image stitching, ROI defect region extraction, and multi-view image alignment on the acquired 2D images, preserving 2D features such as workpiece surface texture, color difference, and edge details. Simultaneously, the 3D scanning module acquires 3D point clouds of the entire workpiece and key areas such as grooves and curved surfaces. The data processing module performs noise reduction, resampling, normal calculation, mesh reconstruction, and multi-segment point cloud stitching on the original point cloud to restore the complete 3D geometric shape of the workpiece.

[0026] The data fusion module retrieves preset camera calibration parameters and point cloud coordinate system parameters to construct a registration matrix between the image and the point cloud. For defects that can be identified in the 2D image but lack depth, the system uses the 3D point cloud to supplement the depth. For abnormal shapes discovered in the 3D point cloud, the system checks the corresponding 2D image to confirm texture or color features. The candidate defect regions extracted from the 2D image are accurately mapped to the 3D mesh model, while the 3D depth and deformation information are back-projected to the pixel regions of the 2D image, completing the coordinate unification and feature complementarity of 2D texture features and 3D geometric features.

[0027] The online defect detection module outputs the defect type, location, length, width, depth, area, volume, and severity. Based on the currently identified defect category, the module dynamically matches and calls the corresponding detection model: for surface texture defects such as scratches, stains, coating abnormalities, and uneven textures, it prioritizes using a 2D detection model to complete texture feature recognition and preliminary judgment; for spatial deformation defects such as depressions, protrusions, local thinning, surface deformation, and boundary defects, it prioritizes using a 3D detection model to extract geometric deformation parameters; for defects exhibiting both texture and geometric deformation characteristics, it calls a fusion detection model to combine 2D texture features and 3D geometric features to complete defect type determination, defect location, and multi-dimensional dimension calculations such as length, width, depth, area, and volume. During the inspection process, if the two-dimensional and three-dimensional inspection results are consistent, the system automatically generates a preliminary conclusion. If the two types of inspection results conflict or the confidence level is below the threshold, the system marks the defect as pending manual review. Inspectors can simultaneously view the workpiece's two-dimensional real-shot image, three-dimensional point cloud morphology, multimodal fusion analysis basis, and the system's initial judgment results on the interactive interface, and confirm, correct, or delete the defect. After manual review, the review results are synchronously fed back to the model training sample library. The manual review results are not only used in the final report but are also structured and saved as sample update data, providing data support for subsequent model iteration and optimization.

[0028] After a single workpiece inspection and manual review process is completed, the report traceability module automatically binds the current workpiece's unique serial number to the corresponding inspection task ID. It integrates workpiece basic parameters, multi-view 2D images, 3D point cloud data, defect detection coordinates and multi-dimensional dimensional parameters, model inference version, system judgment results, and manual review opinions to automatically generate a standardized inspection report and complete its archiving. The system establishes associated archiving relationships for various types of inspection data, supporting subsequent tracing and querying of historical inspection records using conditions such as workpiece serial number, defect category, inspection time, and model version. It also retains inspection report versions before and after model iterations, enabling long-term traceability and quality review of workpiece quality inspection data.

[0029] Example 2: Rapid Adaptation and Deployment of New Workpiece Types: When a new type of industrial workpiece needs to be added for inspection, the administrator can import the standard 3D model or CAD conversion model of the workpiece through the system backend, define the inspection area, acquisition posture, inspection rules and allowed defect types of the workpiece in the workpiece standard library module, generate or recommend 2D photography viewpoint and 3D scanning path according to the workpiece geometry, and save it as the standard configuration for the workpiece type.

[0030] To address the initial shortage of samples for inspecting new workpieces, the inspection model training module uses pre-trained 2D and 3D inspection models of existing workpieces of the same type and material as the base model, and performs fine-tuning training using a small number of labeled samples from the new workpieces. Inspectors label defect samples of the new workpieces, and the model training module divides the samples into training, validation, and test sets to complete model training, evaluation, and deployment.

[0031] After deployment, the system records false detections, missed detections, and manual corrections in real time during the detection process, continuously supplementing the training dataset with valid samples, enabling the model to continuously iterate and optimize for new workpiece types, and gradually improving the defect detection accuracy and adaptation stability of newly added workpieces.

[0032] Example 3: Extended identification mechanism for newly added defect categories: When a new defect category needs to be added, the administrator enters the name, morphological characteristics, typical 2D texture representation, typical 3D geometric representation, size calculation method, minimum recording threshold, severity grading rules, and applicable workpiece area into the defect feature standard library module. Based on the core feature attributes of the new defect, the system recommends suitable detection models: defects with only texture, color difference, and surface anomalies are suitable for 2D detection models; defects with only spatial deformation and dimensional deviations are suitable for 3D detection models; and complex defects with both texture and geometric anomalies are suitable for multimodal fusion detection models.

[0033] Based on the configured defect feature rules and labeled samples, the model training module completes the training, evaluation, and deployment of a dedicated model for each newly added defect, and incorporates the defect category into the system's dynamic algorithm pipeline. During online detection, if a suspected defect is detected and the area where the defect is located matches a preset defect application area, the system calls the corresponding adapted model to complete the defect identification and outputs the defect size and severity judgment results according to the rules in the defect feature standard library. Valid data after manual review is continuously fed back into the sample library, gradually improving the identification capability of new defect categories and enabling the sustainable expansion of defect categories.

[0034] Example 4: Automatic generation of test reports and end-to-end data traceability: After a single workpiece inspection process is completed, the system report traceability module automatically binds the current inspection task ID, workpiece model, and unique serial number, and automatically generates an inspection report. The report fully includes the workpiece model, workpiece serial number, inspection time, inspection equipment, inspection model version, defect type, defect location, defect size, 2D screenshot, 3D partial view, fusion analysis basis, manual review comments, and final judgment.

[0035] All inspection reports, raw data collection, model operation logs, and manual review records are archived and stored with a unique workpiece serial number. Staff can quickly query historical inspection data using multiple criteria such as workpiece serial number, workpiece type, defect category, inspection time, model version, and review status. If the inspection model is subsequently upgraded or the workpiece re-inspection results change, the system retains both old and new version reports and establishes version relationships, enabling long-term traceability of industrial workpiece quality inspection data.

[0036] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-type industrial workpiece AI recognition and intelligent inspection system, characterized in that, include: The system includes a workpiece standard library module, a defect feature standard library module, a workpiece identification module, a 2D acquisition module, a 3D scanning module, a data processing module, a data fusion module, a detection model training module, an online defect detection module, a manual review and feedback module, and a report traceability module. The workpiece standard library module stores standard models, inspection areas, acquisition paths, and judgment rules for various types of industrial workpieces; the defect feature standard library module stores defect categories, two-dimensional features, three-dimensional features, size calculation methods, and severity rules; the workpiece identification module obtains the workpiece model and unique serial number and creates an inspection task; the two-dimensional acquisition module and the three-dimensional scanning module respectively acquire two-dimensional images and three-dimensional point clouds under the same inspection task; the data processing module performs image preprocessing, point cloud preprocessing, and coordinate alignment; and the data fusion module registers the two-dimensional image and three-dimensional point cloud to a unified coordinate system and generates fused defect features. The detection model training module is used to train and update the two-dimensional detection model, the three-dimensional detection model, and the fusion detection model; the online defect detection module dynamically calls the two-dimensional detection model, the three-dimensional detection model, or the fusion detection model according to the workpiece type, defect category, and fusion defect characteristics, and outputs the defect location, defect type, defect size, and severity; the manual review and feedback module is used to receive manual confirmation or correction results and return them to the training samples. The report traceability module is used to associate and store the test results, review results, model version, and test report with the unique serial number; The modules communicate with each other to complete data exchange.

2. The AI-based intelligent detection system for multi-type industrial workpieces according to claim 1, characterized in that, The workpiece identification module communicates with the two-dimensional acquisition module and the three-dimensional scanning module respectively to collect workpiece data; the workpiece standard library module includes a standard three-dimensional model of the workpiece, detection attributes, acquisition parameters, size thresholds and version numbers, which are used to abstract multiple types of workpieces into configurable detection objects and support the reuse of the same detection process on different workpieces. The defect feature standard library module presets multiple types of defect standards, covering defects on the workpiece surface, edges, shape, coating, and contact anomalies. The defect feature library drives model training, detection pipeline selection, and result determination.

3. The AI-based intelligent detection system for multi-type industrial workpieces according to claim 1, characterized in that, The 2D acquisition module and 3D scanning module communicate with the data processing module and data fusion module respectively to complete data preprocessing and feature fusion. The 2D acquisition module is used to acquire multi-view 2D images of the workpiece. The 3D scanning module is used to acquire point clouds or mesh models of the workpiece. The data processing module performs 2D image preprocessing and 3D point cloud preprocessing. The 2D image preprocessing includes one or more of the following: noise removal, distortion correction, image stitching, ROI extraction, multi-view alignment, brightness and contrast optimization. The 3D point cloud preprocessing includes one or more of the following: noise filtering, point cloud simplification, resampling, normal calculation, mesh reconstruction, boundary clipping, coordinate unification, and multi-segment stitching.

4. The AI-based intelligent detection system for multi-type industrial workpieces according to claim 1, characterized in that, The data fusion module communicates with the online defect detection module to transmit defect feature data; the data fusion module is used to perform coordinate system registration, feature point matching, and texture mapping between the two-dimensional image and the three-dimensional point cloud, and to perform depth compensation and consistency verification between the two; the registration between the two-dimensional image and the three-dimensional point cloud includes coordinate transformation based on calibration parameters, matching based on feature points, registration optimization based on ICP, or a combination thereof.

5. The AI-based intelligent detection system for multiple types of industrial workpieces according to claim 1, characterized in that, The detection model training module and the online defect detection module communicate to deploy and update the detection model. The detection model training module manages sample import, annotation, dataset partitioning, model training, hyperparameter tuning, model evaluation, version management, and deployment. The online defect detection module has a built-in dynamic detection pipeline. The dynamic detection pipeline can select a detection model based on defect category, regional attributes, data integrity, model confidence, or 2D and 3D consistency results. When the 2D and 3D detection results are inconsistent, consistency verification or manual review is triggered. The online defect detection module communicates with the manual review feedback module and the report traceability module to transmit defect judgment results and result summaries. The manual review feedback module displays the defect location, 2D image, 3D morphology, and fusion judgment criteria, and supports defect editing operations. The report traceability module supports multi-dimensional queries based on workpiece unique serial number, detection task ID, model version, detection time, and defect category.

6. A multi-type industrial workpiece AI recognition and intelligent detection method, used in the multi-type industrial workpiece AI recognition and intelligent detection system as described in any one of claims 1-5, characterized in that, Includes the following steps: S1. Establish a workpiece standard library and a defect feature standard library; S2. Identify the model and unique serial number of the workpiece to be inspected and create an inspection task; S3. Determine the detection area and data acquisition path based on the workpiece standard library; S4. Acquire two-dimensional images and three-dimensional point clouds of the workpiece to be inspected; S5. Perform preprocessing on the two-dimensional image and the three-dimensional point cloud respectively; S6. Register the two-dimensional image and the three-dimensional point cloud to a unified coordinate system and merge them into a defect analysis object; S7. Dynamically call two-dimensional detection models, three-dimensional detection models, or fusion detection models based on the defect feature standard library and model confidence; compare the two-dimensional detection results with the three-dimensional detection results, and if the two conclusions are inconsistent, trigger the consistency verification or manual review process. S8. Output the location, type, size, and severity of the defect; S9. Receive the results of manual review and feed them back to the model training data; S10. Generate and store a test report associated with a unique serial number.

7. The AI-based intelligent detection method for multiple types of industrial workpieces according to claim 6, characterized in that, The establishment of the workpiece standard library and the defect feature standard library includes: Input the workpiece type, standard 3D model, external dimensions, key inspection areas, inspection surfaces, prohibited inspection areas, acquisition path, clamping posture, area attributes and quality judgment criteria of various industrial workpieces to construct the workpiece standard library; The two-dimensional features, three-dimensional features, size calculation methods, and judgment thresholds of various defects are entered to construct the defect feature standard library.

8. The AI-based intelligent detection method for multiple types of industrial workpieces according to claim 6, characterized in that, The system identifies the model and unique serial number of the workpiece to be inspected through one or more methods such as machine vision, OCR, barcode, QR code, or RFID. It acquires 2D images by controlling industrial cameras, light sources, turntables, robots, or conveyor mechanisms to capture workpiece surface images in multiple postures, and binds these images with task IDs, viewpoints, timestamps, and spatial posture parameters. It also acquires 3D data by controlling a 3D scanning device to collect workpiece point cloud or mesh data, binding point cloud coordinates, scanning posture, scanning segment information, and task ID. The 2D images and 3D point cloud data are then transmitted to a data processing module for preprocessing, including noise removal, distortion correction, brightness and contrast optimization, image stitching, ROI extraction, and multi-viewpoint alignment of the 2D images. Perform noise filtering, point cloud simplification, resampling, normal calculation, mesh reconstruction, boundary clipping, coordinate unification, and multi-segment stitching on 3D data.

9. The AI-based intelligent detection method for multiple types of industrial workpieces according to claim 6, characterized in that, The specific steps for registering two-dimensional images and three-dimensional point clouds into a unified coordinate system and fusing them into a defect analysis object include: By using coordinate transformation matrices, calibration parameters, feature point matching, or ICP registration algorithms, two-dimensional images and three-dimensional point clouds can be unified into the same coordinate system. Two-dimensional textures or defect candidate regions are mapped to three-dimensional point clouds or mesh models, and three-dimensional depth information is used to supplement the spatial information missing in the two-dimensional image. The current workpiece data is matched with the standard models and regional partitioning information in the workpiece standard library, and different defect thresholds, model versions and judgment rules are called for different detection areas.

10. The AI-based intelligent detection method for multiple types of industrial workpieces according to claim 6, characterized in that, The detection model is dynamically invoked, and the corresponding detection model is matched according to the defect category. For surface texture defects, a two-dimensional detection model is invoked to extract texture features, and for spatial deformation defects, a three-dimensional detection model is invoked to extract geometric features. A fusion detection model is used to combine two-dimensional texture features and three-dimensional geometric features for joint defect judgment and size calculation. The results are manually reviewed and fed back to the model training data. Confirmed defects are used as positive samples, false detection areas are used as negative samples, and corrected categories or boundaries are used as labeled correction samples to achieve iterative optimization of the detection model. The inspection report is associated with and stores the workpiece identity, collected data, inspection results, review comments, model version, report file, and operation log to support multi-dimensional traceability queries.

Images