Automatic visual detection device and detection method for surface defect detection
By combining high-definition microscopes, micro-curved lenses, residual phase compensation devices, high-frequency ultrasonic testing and OCT probes with adaptive rotary fixtures and FPGA high-speed synchronous trigger modules, the limitations of surface defect detection and the problems of internal defect detection in existing technologies are solved, achieving efficient, comprehensive and accurate defect detection.
Patent Information
- Application Number
- CN202510765895.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-10
- Publication Date
- 2025-09-26
AI Technical Summary
Existing surface defect detection technologies have limitations such as single spectral imaging, insufficient multi-sensor synchronization accuracy, inability to detect internal defects, and low precision of rotating fixtures, resulting in high missed detection rates, insufficient synchronization accuracy, and poor dynamic adaptability.
A high-definition microscope combined with a micro-curved lens and a residual phase compensation device, combined with high-frequency ultrasonic testing and an OCT probe, an adaptive rotating fixture and an FPGA high-speed synchronous trigger module are used to achieve multi-angle scanning and real-time data processing, and the deep learning model YOLO is used for defect classification and positioning.
It achieves high-precision detection of product surface and internal defects, improves detection efficiency and comprehensiveness, enhances detection adaptability and data consistency, and can accurately identify complex defects such as tiny dents.
Smart Images
Figure CN120703092A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of surface defect detection, and in particular to an automated visual detection device and a detection method for surface defect detection. Background Art
[0002] With the continuous improvement of industrial automation, surface defect detection technology plays an increasingly important role in manufacturing quality control. Surface defect detection technology primarily uses optical imaging, image processing, and artificial intelligence to automatically identify and classify surface defects such as scratches, dents, and bubbles, thereby improving product quality and production efficiency.
[0003] The existing surface defect detection technology has the following main problems: First, the existing surface defect detection technology only covers a single band, which makes it difficult to capture the complementary characteristics of defects in different materials, resulting in a high missed detection rate; second, the trigger jitter (microsecond level) between multiple lenses or sensors leads to inconsistent data in time and space, affecting the fusion effect and insufficient synchronization accuracy; third, optical detection cannot penetrate the surface layer, and the ability to identify internal defects of the workpiece (such as bubbles and delamination) is insufficient, resulting in the lack of internal defect detection; fourth, the traditional rotary fixture has low angle adjustment accuracy (>0.1°), and cannot adapt to the linear speed changes of workpieces of different diameters, and has poor dynamic adaptability. These problems seriously restrict the application of surface defect detection technology in high-precision and high-efficiency production environments. Summary of the Invention
[0004] In order to solve the problems of single spectral imaging limitations, insufficient multi-sensor synchronization accuracy, inability to detect internal defects, and low precision of rotating fixtures in existing surface defect detection technologies, and to achieve technical effects such as improved detection accuracy, optimized efficiency, enhanced adaptability, improved internal defect detection capabilities, and improved data consistency, the present invention provides an automated visual inspection device and inspection method for surface defect detection.
[0005] The technical solution adopted by the present invention to solve its technical problems is: an automated visual inspection device and inspection method for surface defect detection, the inspection device includes an adaptive rotating fixture 2 and a positioning fixture 3, and the adaptive rotating fixture 2 is equipped with a detection lens 21 to scan the object on the positioning fixture 3.
[0006] The detection lens 21 includes a high-definition microscope 22 and a synchronous trigger and data collector 23. The high-definition microscope 22 includes a residual phase compensation device 22-1. The synchronous trigger and data collector 23 includes a time-sharing strobe control module 23-1, an FPGA high-speed synchronous trigger module 23-2 and a data collector module 23-3. The high-definition microscope 22 uses a micro-curved lens, which is a spherical glass bead. The residual phase compensation device 22-1 integrates a multi-stage TDI sensor. The FPGA synchronous trigger and high-speed data acquisition are distributed to the high-definition microscope 22 through the LVDS interface to ensure that the line frequency and exposure time of each band image are strictly synchronized. The FPGA high-speed synchronous trigger module 23-2 has built-in cache logic, supports parallel reception of multi-channel data, and buffers to reduce transmission delay, thereby ensuring the real-time and accuracy of data acquisition.
[0007] Furthermore, the inspection lens 21 uses pulse-echo 5-500MHz high-frequency ultrasound to receive reflected and transmitted signals during scanning. Wavelet transforms are used to de-noise the ultrasonic echo signals, and gain compensation is employed during imaging to enhance attenuated signals, ensuring clear imaging of deep structures. This approach not only detects surface defects but also internal defects, expanding the detection range.
[0008] The detection lens 21 is integrated with an OCT probe. When the OCT probe is rotated and scanned, it can step forward to obtain a three-dimensional data volume. This process can be decomposed into a triple integral of spatial position: V(x, y, z) = ∫∫∫ Ω I(ρ,φ,z)δ(ρ-xcosθ-ysinθ,z)ρdρdφ, where ρ and φ are polar coordinate parameters and θ is the probe rotation angle. Reconstructing the C-scan data matrix from 3D volume data allows for 2D slice reconstruction and 3D visualization in any orientation within the processing system, effectively observing the complex internal structure of a product.
[0009] The high-definition microscope 22 and residual phase compensation device 22-1 scan the target object from multiple angles, collecting ultrasonic echo data from different angles. The data acquisition module 23-3 performs synthetic focusing on the ultrasonic echo data to construct multiple virtual transducer models with larger apertures than actual physical transducers. The data acquisition module 23-3 transmits this data to the processing system via a data transmission interface, where it uses the deep learning model YOLO for defect classification and location.
[0010] During the synthetic focusing process, the synchronous trigger and data collector 23 delays the signal according to the difference in arrival time of the echo signals received by different virtual transducers, and then superimposes the delayed signals, so that the signals from the same position of the target object are enhanced after superposition, thereby optimizing the focusing depth and improving the resolution and clarity of the imaging.
[0011] Preferably, the adaptive rotary fixture 2 is driven by a precision stepping motor with an encoder and cooperates with the encoder closed-loop feedback to dynamically adjust the trigger frequency to match the linear speed.
[0012] Preferably, a precision stepping motor with an encoder under the positioning fixture 3 drives the object to accurately control the rotation angle and speed, thereby ensuring the comprehensiveness and accuracy of the detection.
[0013] Preferably, the positioning fixture 3 also includes a clamping fixture to assist in fixing the target object.
[0014] Preferably, the adaptive rotating fixture 2 is mounted on the bracket 1 , and the bracket 1 is provided with a guide rail and a control switch to control the movement of the adaptive rotating fixture 2 .
[0015] The present invention also provides an automated visual inspection method for surface defect detection, which includes global synchronous control, real-time data processing, data fusion and defect identification steps.
[0016] The global synchronization control and real-time data processing include an FPGA global synchronization trigger mechanism and time-sharing strobe control, a high-speed data transmission and dynamic feedback mechanism, and a real-time data processing mechanism. The FPGA global synchronization trigger mechanism generates high-precision trigger signals based on the FPGA, synchronously controlling the line frequency timing of the high-definition microscope via the LVDS interface to ensure alignment of the optical and electromagnetic signal acquisition timelines. The real-time data processing mechanism prioritizes high-priority data streams by deploying an embedded real-time operating system in the FPGA's real-time task scheduling module, employing the CUDA parallel architecture for image preprocessing, and simultaneously calculating defect depth information in real time using a sparse iterative algorithm.
[0017] The data fusion and defect recognition process includes a digital imaging analysis mechanism, a defect detection and classification learning mechanism, and a performance optimization and verification mechanism. This defect detection and classification learning mechanism uses an improved YOLOv8s network, which takes the fused image and electromagnetic signature as input. It optimizes defect location through an attention mechanism, calculates defect depth using phase difference, and constructs a three-dimensional surface topography map, improving the ability to detect tiny dents.
[0018] The beneficial effects of the present invention are:
[0019] 1 The automated visual inspection device of the present invention can effectively improve the quality of image acquisition and achieve high-precision detection of tiny defects by using a high-definition microscope combined with a special micro-curved lens and a residual phase compensation device.
[0020] 2. Using high-frequency ultrasonic testing technology and synthetic focusing processing methods, not only can surface defects of products be detected, but also internal defects of products can be detected, which expands the detection range and improves the comprehensiveness of detection.
[0021] The application of 3FPGA high-speed synchronous trigger module and real-time data processing mechanism ensures the real-time and accuracy of data acquisition and processing, and improves detection efficiency.
[0022] 4. The deep learning model YOLO is used for defect classification and positioning, and it is improved. Combined with multiple data fusion and analysis methods, it can accurately identify and classify various types of defects, and improve the detection ability of complex defects such as tiny dents.
[0023] 5. The adaptive rotary fixture is driven by a precision stepper motor, which can accurately control the rotation of the object and cooperate with the detection lens to scan from multiple angles to ensure comprehensive detection of the product surface. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a structural schematic diagram of the present invention.
[0025] Figure 2 Schematic diagram of the scanning of the present invention.
[0026] Figure 3 Schematic diagram of the control module of the present invention. DETAILED DESCRIPTION
[0027] The technical solutions of the present invention will be described clearly and completely below with reference to the accompanying drawings and embodiments. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.
[0028] like Figure 1-3 As shown, the inspection device includes a bracket 1, an adaptive rotating fixture 2, and a positioning fixture 3. The bracket is equipped with a guide rail and a control switch to control the movement of the adaptive rotating fixture 2. The adaptive rotating fixture 2 is equipped with an inspection lens 21, which scans the object on the positioning fixture 3. The positioning fixture 3 has a clamping fixture to help secure the object to be inspected.
[0029] The inspection lens 21 includes a high-definition microscope 22 and a synchronous trigger and data acquisition device 23. The high-definition microscope 22 includes a residual phase compensation device 22-1, which is supported by multiple auxiliary inspection lenses. The high-definition microscope 22 and the auxiliary inspection lenses utilize micro-curved lenses, which are spherical glass beads. The residual phase compensation device 22-1 integrates a multi-stage TDI sensor. This sensor compensates for residual phase errors (such as phase variations caused by optical system aberrations and surface roughness of the object being measured) in real time by integrating optical signals from different paths in a time-sharing manner.
[0030] The detection lens 21 is integrated with the OCT probe. When the OCT probe is rotated and scanned, it can step forward to obtain a three-dimensional data volume. This process can be decomposed into a triple integral of spatial position: V(x, y, z) = ∫∫∫ Ω I(ρ,φ,z)δ(ρ-x cosθ-y sinθ,z)ρdρdφ, where ρ and φ are polar coordinate parameters and θ is the probe rotation angle. Reconstructing the C-scan data matrix from 3D volume data enables 2D slice reconstruction and 3D visualization within the processing system, enabling effective observation of complex internal product structures. The operator can observe the product's internal structure and defect distribution from different angles through a human-computer interface, helping to more accurately determine the nature and severity of defects.
[0031] The synchronous trigger and data acquisition unit 23 comprises a time-sharing strobe control module 23-1, an FPGA high-speed synchronous trigger module 23-2, and a data acquisition module 23-3. The FPGA synchronous trigger and high-speed data acquisition are distributed to the high-definition microscope 22 and the auxiliary inspection lens via an LVDS interface, ensuring strict synchronization of the line rate and exposure time of each band image. The LVDS interface utilizes differential signal transmission, offering strong anti-interference capabilities, a transmission rate of up to 700Mbps, a signal delay of less than 5ns, and a jitter of less than 100ps.
[0032] The synchronous trigger and data acquisition unit 23 uses an FPGA-based high-speed synchronous trigger module 23-2 and a data acquisition module 23-3 for high-speed triggering. The FPGA also features built-in DDR4 cache logic, supporting parallel data reception from multiple channels and reducing transmission latency. The DDR4 memory capacity is 8GB, with a data transfer rate of 3200 MT / s. This supports parallel data processing on eight channels, with a single-channel bandwidth of 25.6GB / s.
[0033] Data collector module 23-3 transmits data to the processing system via a data transmission interface, where it uses the YOLO deep learning model for defect classification and location. The data transmission interface supports high-speed data transmission at 12.5 Gbps, with a latency of less than 1 μs and high reliability. The processing system utilizes an embedded AI computer with 10,752 CUDA cores, 48 GB of video memory, and 38.7 TFLOPS of FP32 computing power, enabling efficient execution of deep learning models. The YOLO model utilizes the YOLOv8s architecture.
[0034] The detection method includes global synchronization control and real-time data processing, data fusion, and defect identification steps. The global synchronization control and real-time data processing include an FPGA global synchronization trigger mechanism and time-sharing strobe control, a high-speed data transmission and dynamic feedback mechanism, and a real-time data processing mechanism. The FPGA global synchronization trigger mechanism generates high-precision trigger signals based on the FPGA, synchronously controlling the line frequency timing of the high-definition microscope via the LVDS interface to ensure alignment of the acquisition time axes of the optical and electromagnetic signals. The real-time data processing mechanism prioritizes high-priority data streams by deploying an embedded real-time operating system in the FPGA's real-time task scheduling module, employing the CUDA parallel architecture for image preprocessing, and simultaneously calculating defect depth information in real time using a sparse iterative algorithm.
[0035] Data fusion and defect recognition include a digital imaging analysis mechanism, a defect detection and classification learning mechanism, and a performance optimization and verification mechanism. This mechanism uses an improved YOLOv8s network as input, combining the fused image and electromagnetic signature map. It optimizes defect location through an attention mechanism, calculates defect depth using phase difference, and constructs a 3D surface topography map, improving the ability to detect even tiny dents.
[0036] Example 1 Implementation of automated visual inspection device
[0037] 1. Device Assembly: Install and debug the adaptive rotating fixture and positioning fixture according to design requirements, ensuring their accurate relative positioning. Mount the inspection lens on the adaptive rotating fixture and adjust the distance and angle between the inspection lens and the object on the positioning fixture to ensure a clear scan of the object's surface. Also, complete the integrated installation and debugging of the OCT probe and inspection lens to ensure proper operation.
[0038] 2 Operation of the HD microscope: Start the HD microscope, and its micro-curved lens (spherical glass beads) performs optical imaging of the target object. The multi-stage TDI sensor integrated in the residual phase compensation device starts working to compensate for the phase error generated during the imaging process, thereby improving the clarity and quality of the image.
[0039] 3. Synchronous Triggering and Data Acquisition: The FPGA high-speed synchronous trigger module generates high-precision trigger signals, which are used to synchronously control the line frequency timing of the HD microscope via the LVDS interface. Simultaneously, the time-sharing strobe control module controls the flashing frequency of the strobe light source based on inspection requirements. The data acquisition module collects image data output by the HD microscope, ultrasonic echo data collected by the ultrasonic testing device, and 3D data acquired by the OCT probe in real time.
[0040] Ultrasonic testing: The detection lens uses 5-500MHz high-frequency ultrasonic waves to scan the target object and receive reflected and transmitted signals. Wavelet transform is used to denoise the ultrasonic echo signal, and gain compensation is used during the imaging process to enhance the attenuated signal, ensuring clear acquisition of the target object's deep structural information.
[0041] 5OCT data acquisition: The OCT probe advances along the axial direction (z-axis) according to the set step distance while rotating and scanning. During the advancement process, the three-dimensional data volume is collected according to the formula. The data acquisition module transmits the collected three-dimensional data volume to the processing system.
[0042] 6. Data Processing and Analysis: The data acquisition module transmits the collected image data, ultrasonic echo data, and 3D data volumes to the processing system. The processing system performs synthetic focusing on the ultrasonic echo data, constructs a virtual transducer model, and delays and superimposes the signals based on the arrival time differences of the echo signals received by different virtual transducers, optimizing the focus depth and improving the resolution and clarity of the imaging. The deep learning model YOLO is used to classify and locate defects in the processed images, enabling accurate identification of product surface defects. Simultaneously, the processing system performs 2D slice reconstruction and 3D visualization rendering of the 3D data volume based on the C-scan data matrix, generating intuitive images of the product's internal structure and defect distribution.
[0043] 7. Operation of the adaptive rotary fixture: A precision stepper motor with an encoder drives the adaptive rotary fixture to rotate according to the set program, driving the detection lens to scan the target object from multiple angles to ensure comprehensive detection of the target object's surface.
[0044] Example 2 Implementation of the automated visual inspection method
[0045] 1Global synchronization control and real-time data processing
[0046] 1.1 FPGA Global Synchronous Trigger Mechanism and Time-Sharing Strobe Control: A high-precision trigger signal is generated based on the FPGA and transmitted to the HD microscope via the LVDS interface. This precisely controls the microscope's line frequency timing, ensuring strict time alignment between optical and electromagnetic signal acquisition. Simultaneously, the time-sharing strobe control module flexibly adjusts the flashing frequency of the strobe light source based on different inspection scenarios and requirements, providing optimal lighting conditions for image acquisition. Furthermore, the trigger signal is used to synchronously control the rotation and axial propulsion rhythm of the OCT probe, ensuring accurate 3D data acquisition.
[0047] 1.2 High-Speed Data Transmission and Dynamic Feedback Mechanism: The built-in cache logic of the FPGA high-speed synchronous trigger module supports parallel reception of multi-channel data, effectively reducing data transmission latency. During the data transmission process, a dynamic feedback mechanism is established to monitor the transmission status of image data, ultrasonic echo data, and 3D data volumes in real time. If any data transmission anomalies are detected, the transmission parameters are adjusted promptly to ensure stable data transmission.
[0048] 1.3 Real-time Data Processing Mechanism: An embedded real-time operating system is deployed in the FPGA's real-time task scheduling module. Tasks are scheduled based on data flow priorities, prioritizing high-priority data flows. The CUDA parallel architecture is used to pre-process captured images, including filtering and enhancement, to improve image quality. Furthermore, a sparse iterative algorithm is used to calculate defect depth information in real time, providing accurate data support for subsequent defect analysis. For the 3D data volumes acquired by OCT, computing resources are rationally allocated in real-time task scheduling to ensure initial data processing and storage.
[0049] 2. Data fusion and defect identification
[0050] 2.1 Digital imaging analysis mechanism: Comprehensively analyze the image data collected by high-definition microscopes, the imaging data obtained by ultrasonic testing, and the images reconstructed from OCT three-dimensional data to extract key feature information in the images, such as edges, textures, and internal structure contours.
[0051] 2.2 Defect Detection and Classification Learning Mechanism: The fused image data and electromagnetic signature map are fed into an improved YOLOv8s network. The network's attention mechanism optimizes defect location, improving the accuracy of detecting small defects. Phase difference is used to calculate defect depth, and a 3D surface topography map is constructed based on image features. The 3D OCT visualization results are also referenced to accurately classify and locate different types of defects (both surface and internal).
[0052] 2.3 3D Data Processing and Visualization: After receiving the 3D data volume acquired by the OCT probe, the processing system performs 2D slice reconstruction based on the C-scan data matrix. Operators can select different slice orientations (such as cross-section, longitudinal section, and oblique section) to observe the internal structure of the product as needed. Simultaneously, 3D visualization rendering is performed to generate a 3D stereo image. Operators can use the human-computer interaction interface (such as mouse dragging, zooming, and rotating) to observe the internal structure of the product and the distribution of defects from different angles, more intuitively determining the location, shape, size, and severity of defects, providing a strong basis for product quality assessment and subsequent processing.
[0053] Finally, it should be noted that the above embodiments and implementation methods are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An automated visual inspection device and method for detecting surface defects, comprising an adaptive rotating fixture 2 and a positioning fixture 3. The adaptive rotating fixture 2 is equipped with an inspection lens 21 for scanning an object on the positioning fixture 3. The device is characterized by: The detection lens 21 includes a high-definition microscope 22 and a synchronous trigger and data collector 23. The high-definition microscope 22 includes a residual phase compensation device 22-1. The synchronous trigger and data collector 23 includes a time-sharing strobe control module 23-1, an FPGA high-speed synchronous trigger module 23-2 and a data collector module 23-3. The high-definition microscope 22 uses a micro-curved lens, which is a spherical glass bead. The residual phase compensation device 22-1 integrates a multi-stage TDI sensor. The FPGA synchronous trigger and high-speed data acquisition are distributed to the high-definition microscope 22 through the LVDS interface to ensure that the line frequency and exposure time of each band image are strictly synchronized. The FPGA high-speed synchronous trigger module 23-2 has built-in cache logic to support parallel reception of multi-channel data, and buffering reduces transmission delay.
2. The automated visual inspection device for surface defect detection according to claim 1, characterized in that: The detection lens 21 is provided with an OTC probe to obtain the three-dimensional structure of the object. When scanning, the detection lens 21 uses pulse echo 5-500MHz high-frequency ultrasound to receive reflected and transmitted signals; wavelet transform is used to denoise the ultrasonic echo signal, and gain compensation method is used in the imaging process to enhance the attenuated signal to ensure that the deep structure can be clearly imaged.
3. The automated visual inspection device for surface defect detection according to claim 1, characterized in that: The high-definition microscope 22 and the residual phase compensation device 22-1 scan the target object from multiple angles and collect ultrasonic echo data at different angles.
4. The automated visual inspection device for surface defect detection according to claim 1, characterized in that: The data collector module 23 - 3 performs synthetic focusing processing on the ultrasonic echo data to construct multiple virtual transducer models with larger apertures than actual physical transducers.
5. The automated visual inspection device for surface defect detection according to claim 1, characterized in that: During the synthetic focusing process, the synchronous trigger and data collector 23 delays the signal according to the difference in arrival time of the echo signals received by different virtual transducers, and then superimposes the delayed signals, so that the signals from the same position of the target object are enhanced after superposition, thereby optimizing the focusing depth and improving the resolution and clarity of the imaging.
6. The automated visual inspection device for surface defect detection according to claim 1, characterized in that: The data collector module 23 - 3 transmits the data to the processing system via the data transmission interface, and uses the deep learning model YOLO to classify and locate defects.
7. The automated visual inspection device for surface defect detection according to claim 1, characterized in that: The self-adaptive rotating fixture 2 is driven by a precision stepping motor with an encoder, and the positioning fixture 3 is provided with a clamping fixture.
8. An automated visual inspection device and method for surface defect detection, characterized by: The detection method includes global synchronous control and real-time data processing and data fusion and defect identification steps. The global synchronous control and real-time data processing include the steps of FPGA global synchronous trigger mechanism and time-sharing strobe control, high-speed data transmission and dynamic feedback mechanism and real-time data processing mechanism; the data fusion and defect identification include the steps of digital imaging analysis mechanism, defect detection and classification learning mechanism and performance optimization and verification mechanism.
9. The automated visual inspection method for surface defect detection according to claim 8, characterized in that: The FPGA global synchronization trigger mechanism generates high-precision trigger signals based on the FPGA, and synchronously controls the line frequency timing of the high-definition microscope through the LVDS interface to ensure that the acquisition time axes of the optical and electromagnetic signals are aligned; The real-time data processing mechanism deploys an embedded real-time operating system in the FPGA's real-time task scheduling module to prioritize high-priority data streams, uses the CUDA parallel architecture to pre-process images, and simultaneously calculates defect depth information in real time through a sparse iterative algorithm. The defect detection and classification learning mechanism uses an improved YOLOv8s network, inputs a fused image and an electromagnetic feature map, optimizes defect positioning through an attention mechanism, calculates defect depth using phase difference, constructs a three-dimensional surface topography map, and improves the ability to detect tiny depressions.