A steel strand surface defect online identification system based on multispectral imaging

Abstract

The application discloses a kind of steel strand surface defect online identification systems based on multispectral imaging, comprising: motion perception and artifact correction module gathers the axial traction speed of steel strand and radial torsion angle, and the double-dimensional correction of axial motion drag shadow and radial phase distortion is carried out to multispectral image;Optical imaging module simultaneously collects the multispectral image of steel strand circumferential;Algorithm processing module carries out layered difference operation according to standardization twisted texture spectrum model to corrected image, adaptively shields twisted texture spectrum signal, and outputs the pixel level position information and multispectral feature of defect;Multi-sensing fusion module triggers laser force sensing unit to scan defect area according to pixel level position information to obtain physical parameters, and establishes the bidirectional mapping and check relationship of spectral feature and physical parameters;System main control module controls the working time sequence and data interaction of each module.The application realizes the high-precision online identification and quantization of micro-defects under the high-speed motion state of steel strand.

Description

Technical Field

[0001] This invention relates to the field of structural component testing technology, and in particular to an online identification system for surface defects in steel strands based on multispectral imaging. Background Technology

[0002] As a critical load-bearing component in bridge engineering, high-speed railways, building structures, and power facilities, the long-term service safety of steel strands is highly dependent on their surface quality. During manufacturing and subsequent use, steel strands may develop micro-scale defects such as microcracks, micro-pitting, and exposed inclusions. Although these defects are small, they can easily become fatigue sources under high stress environments, thus affecting the overall structural reliability. Therefore, achieving high-speed, stable, and accurate identification of micro-defects on the surface of steel strands during the production stage has become a core technical requirement for quality control.

[0003] Under continuous industrial production conditions, steel strands are typically transported at high speeds. Due to their stranded structure, their surfaces exhibit periodic spiral textures accompanied by a degree of spatial orientation variation. This complex interplay of motion and structure results in significant dynamic instability and texture interference in the surface imaging environment, placing high demands on the imaging quality and recognition accuracy of online detection systems.

[0004] Multispectral imaging technology has been applied in the field of metal surface inspection due to its advantages such as non-contact operation and the ability to acquire multidimensional spectral information. However, under high-speed dynamic conditions, the imaging results are easily affected by changes in motion and the inherent texture of the structure, resulting in image distortion and enhanced interference. Especially in microscale defect identification scenarios, weak anomalous signals are often submerged in structural texture and background response, making it difficult to guarantee the stability of the detection results.

[0005] On the other hand, existing detection systems typically rely primarily on image recognition results for judgment, lacking further verification and quantitative analysis of the physical properties of defects, making it difficult to establish an objective assessment mechanism for the degree of defects. Furthermore, insufficient correlation between multi-source detection data means that detection information fails to form an effective cross-verification relationship, limiting the reliability of detection conclusions under complex operating conditions.

[0006] In summary, the following technical problems still urgently need to be solved in the high-speed online inspection of steel strands: 1. How to obtain stable, high-quality imaging data that can be used for fine identification under complex dynamic working conditions; 2. How to effectively distinguish between inherent structural texture signals and real defect signals to improve the accuracy of microscale defect identification; 3. How to achieve objective quantification of defect physical parameters while ensuring detection efficiency; 4. How to establish a correlation mechanism between multidimensional detection data to improve the reliability and consistency of detection results.

[0007] Therefore, it is necessary to construct an online identification system suitable for high-speed steel strand production environments, and to systematically optimize aspects such as imaging quality control, texture interference suppression, accurate defect identification, parameter quantification, and data consistency verification, so as to meet the application requirements of high-precision and high-reliability detection in industrial sites. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide an online identification system for surface defects of steel strands based on multispectral imaging, which can be used to realize online identification and quantification of physical parameters of micro-defects on the surface of steel strands under high-speed dynamic conditions.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an online identification system for surface defects of steel strand based on multispectral imaging, comprising: An optical imaging module, connected to a motion sensing and artifact correction module, is used to synchronously acquire multispectral images of the steel strand in the circumferential direction and send the multispectral images to the motion sensing and artifact correction module. The motion sensing and artifact correction module is used to acquire the axial traction speed and radial torsion angle of the steel strand in real time under high-speed conveying conditions, and to receive the multispectral image acquired by the optical imaging module. Based on the axial traction speed and the radial torsion angle, the module performs two-dimensional correction of axial motion trailing and radial phase distortion on the multispectral image, and outputs the corrected image. The algorithm processing module, connected to the motion sensing and artifact correction module, is used to receive the corrected image, perform layered differential operations on the corrected image according to the pre-established standardized stranded texture spectral model, adaptively shield the stranded texture spectral signal on the surface of the steel strand, and perform micro-defect enhancement and localization on the shielded image, outputting the pixel-level location information of the defects and the corresponding multispectral features. The multi-sensor fusion module, connected to the algorithm processing module, is used to receive the pixel-level location information of the defect and the corresponding multispectral features, trigger a non-contact laser force sensing unit to scan the defect area based on the pixel-level location information, obtain the physical parameters of the defect, and establish a bidirectional mapping and mutual verification relationship between the multispectral features and the physical parameters based on the pre-trained fusion model, and output the defect quantification data. The system main control module is connected to the motion sensing and artifact correction module, the optical imaging module, the algorithm processing module, and the multi-sensor fusion module, respectively, and is used to control the system's working sequence and data interaction.

[0010] Furthermore, the optical imaging module includes: A ring-shaped multispectral camera array, comprising a predetermined number of miniature multispectral cameras arranged in a ring around the center of the steel strand transport center; A polarization light source compensation unit includes a polarization LED light source corresponding to the miniature multispectral camera. Each polarization LED light source is used to dynamically adjust the emitted light intensity to a preset light intensity threshold range and adjust the polarization angle to a preset angle threshold range according to the ambient light intensity. A high-speed image acquisition card, with its input end connected to each of the miniature multispectral cameras and its output end connected to the motion sensing and artifact correction module, is used to convert the multispectral image into a digital signal and send it to the motion sensing and artifact correction module at a speed not lower than a preset transmission rate threshold.

[0011] Furthermore, the motion sensing and artifact correction module includes: The motion parameter sensing unit is used to collect the axial traction speed and radial torsion angle of the steel strand in real time under high-speed conveying conditions; The wide-beam coherent combining unit is used to coherently combine the imaging beams of each band in the multispectral image acquired by the optical imaging module to form a wide-beam composite imaging beam, so as to improve the ability to capture high-speed moving targets. An adaptive phase compensation unit is connected to the motion parameter sensing unit and the wide-beam coherent synthesis unit, respectively. It is used to receive the axial traction speed and the radial torsion angle, and, in combination with the camera acquisition frame rate of the optical imaging module, establish a composite motion-phase deviation correlation model. It calculates the beam phase deviation caused by radial torsion in real time, and performs adaptive phase compensation on the wide-beam composite imaging beam through an electro-optic modulator to eliminate axial motion trailing and radial phase distortion, and outputs the corrected image. The motion parameter sensing unit transmits the collected axial traction speed and radial torsion angle to the adaptive phase compensation unit and synchronizes them to the system main control module.

[0012] Furthermore, the algorithm processing module includes: The texture modeling unit is used to extract the strand pitch, wire diameter, helix angle and texture spectral reflectance characteristics of steel strands, and to establish a standardized strand texture spectral model. An adaptive shielding unit, connected to the motion sensing and artifact correction module and the texture modeling unit, is used to receive the corrected image, perform layered difference operations on the corrected image and the standardized stranded texture spectral model, and shield the stranded texture spectral signal on the surface of the steel strand based on a preset adaptive threshold algorithm, and output the shielded image. The micro-defect enhancement unit, connected to the adaptive shielding unit, is used to perform edge enhancement processing on the shielded image using a Laplacian convolution kernel of a preset size to amplify the spectral features of the micro-defect, establish a micro-defect-specific spectral model, and output the enhanced micro-defect image. The defect localization unit, connected to the micro-defect enhancement unit, is used to perform pixel-level coordinate calibration of the micro-defect region in the enhanced micro-defect image and output the pixel-level location information of the defect and the corresponding multispectral features.

[0013] Furthermore, the micro-defect enhancement unit is specifically used to: perform feature magnification processing on the masked image using a 5×5 Laplacian convolution kernel edge enhancement algorithm to extract the spectral features of the micro-defects and establish a micro-defect-specific spectral model.

[0014] Furthermore, the multi-sensor fusion module includes: A laser force sensing unit is used to scan the defect area and collect the depth, area, and opening width of the defect as the physical parameters. The precise trigger linkage unit is connected to the algorithm processing module and the laser force sensing unit respectively, and includes a servo drive mechanism. The precise trigger linkage unit is used to receive the pixel-level position information of the defect, control the laser emitter to move to the defect area through the servo drive mechanism according to the pixel-level position information, and perform equidistant micro-step scanning on the defect area, and transmit the collected physical parameters to the data fusion modeling unit. The data fusion modeling unit, connected to the precise triggering linkage unit, is used to receive the physical parameters and corresponding multispectral features, establish a bidirectional mapping and mutual verification relationship between the multispectral features and the physical parameters based on the pre-trained fusion model, and output defect quantification data.

[0015] Furthermore, the data fusion modeling unit is specifically used to: based on a preset number of sets of experimental data of micro-defects in steel strands that have been corrected for artifacts, train and establish a nonlinear correlation model between the multispectral features of micro-defects in steel strands and the physical parameters of laser force sensing using a machine learning algorithm that combines random forest and gradient boosting, and the fitting degree of the nonlinear correlation model is not lower than a preset fitting degree threshold.

[0016] Furthermore, the system main control module includes: The anomaly handling unit is used to monitor the working status of each module and the motion parameters of the steel strand in real time. When at least one of the following abnormal events is detected: equipment failure, abnormal image acquisition, sensor data exceeding the range, or sudden change in the motion parameters of the steel strand, the unit immediately triggers the audible and visual alarm device to issue an alarm signal, suspends the detection process, and saves the detection data before the abnormal event occurred. An adaptive control unit is connected to the motion sensing and artifact correction module and the optical imaging module, respectively. It is used to automatically adjust the acquisition frame rate of the multispectral camera in the optical imaging module according to the axial traction speed of the steel strand, and to automatically adjust the phase compensation frequency of the adaptive phase compensation unit in the motion sensing and artifact correction module, so that the system operating parameters are adaptively matched with the steel strand conveying speed. A high-precision clock synchronization unit is used to provide a unified time reference for each module, ensuring the consistency of the working sequence of each module and adapting to the online detection requirements of steel strand in preset high-speed transmission scenarios.

[0017] Furthermore, it also includes a quantitative evaluation module, which is connected to the multi-sensor fusion module and the system main control module, and is used to determine the level of defect quantitative data and output visualization.

[0018] Furthermore, the quantitative evaluation module includes: The defect level determination unit is used to receive the defect quantification data output by the multi-sensor fusion module, divide the defect into multiple preset safety levels according to the preset defect depth threshold, defect area threshold and opening width threshold, and output the defect level determination result. The report output unit, connected to the defect level determination unit, is used to generate an integrated inspection report based on the defect quantification data and the defect level determination result; the integrated inspection report includes at least the defect location, defect type, physical parameters, safety level, multispectral characteristics, and motion parameters; The system linkage interface connects the report output unit and the system main control module respectively, and is used to transmit the defect level judgment result and the integrated inspection report to the PLC control system or operation and maintenance management platform of the steel strand production line in real time through a preset industrial communication protocol.

[0019] The beneficial effects of this invention are: 1. Improve the stability of high-speed dynamic imaging: By sensing the motion parameters of the steel strand in high-speed conveying state in real time and performing dual-dimensional correction processing on the imaging data, the image distortion caused by motion is effectively reduced, the imaging clarity under complex dynamic working conditions is improved, and a stable data foundation is provided for subsequent defect identification.

[0020] 2. Improve the accuracy of micro-defect identification: Based on the standardized texture spectrum model, the inherent twisted texture is adaptively masked, and the weak abnormal features are enhanced by combining layered differential operation, which effectively distinguishes the structural texture from the real defect signal, reduces the false judgment rate, and improves the accuracy of micro-scale defect identification.

[0021] 3. Achieve accurate quantification and verification of defects: Through a defect area-triggered multi-sensor collaborative detection mechanism, directional scanning and physical parameter acquisition of defect areas are achieved, and the correlation between spectral features and physical parameters is established, improving the objectivity of defect assessment and data consistency, and realizing the upgrade from qualitative identification to quantitative analysis.

[0022] 4. Enhance the reliability and consistency of test results: Construct a two-way mapping and mutual verification mechanism between multi-source data to improve the stability and credibility of test results under complex working conditions and reduce the error risk caused by a single test dimension.

[0023] 5. Suitable for high-speed online industrial scenarios: The system adopts a modular architecture design, which can adapt to different operating speeds and specifications of steel strands, realize continuous online detection, and meet the application requirements of industrial production that emphasize both detection efficiency and accuracy. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the online identification system for surface defects of steel strands based on multispectral imaging in this invention.

[0025] Figure reference numerals: 1. Optical imaging module; 11. Ring multispectral camera array; 12. Polarization light source compensation unit; 13. High-speed image acquisition card; 2. Motion sensing and artifact correction module; 21. Motion parameter sensing unit; 22. Wide-beam coherent synthesis unit; 23. Adaptive phase compensation unit; 3. Algorithm processing module; 31. Texture modeling unit; 32. Adaptive shielding unit; 33. Micro-defect enhancement unit; 34. Defect localization unit; 4. Multi-sensor fusion module; 41. Laser force sensing unit; 42. Precise triggering linkage unit; 43. Data fusion modeling unit; 5. System main control module; 51. Anomaly handling unit; 52. Adaptive control unit; 53. High-precision clock synchronization unit; 6. Quantitative evaluation module; 61. Defect level determination unit; 62. Report output unit; 63. System linkage interface. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0027] like Figure 1As shown in Example 1, which is the first embodiment of the present invention, this embodiment provides an online identification system for surface defects of steel strands based on multispectral imaging, applied to the online inspection scenario of a high-speed production line for φ15.2mm prestressed concrete steel strands. The axial traction speed of the steel strands is 8m / s, the radial torsion angle is ≤5° / s, the inspection environment is an industrial production workshop environment, the ambient light intensity fluctuates between 500-2000lx, and slight dust is present.

[0028] This system comprises: an optical imaging module 1, a motion sensing and artifact correction module 2, an algorithm processing module 3, a multi-sensor fusion module 4, and a system main control module 5. The modules are connected via high-speed industrial Ethernet and fiber optic communication links. The entire system adopts a closed, dustproof cabinet structure with an IP54 protection rating and an ambient temperature range of 0-45℃.

[0029] I. System Structure and Hardware Composition: (a) Motion sensing and artifact correction module 2; The motion sensing and artifact correction module 2 is used to acquire the axial traction speed and radial torsion angle of the steel strand in real time under high-speed conveying conditions, and to perform two-dimensional correction of axial motion trailing and radial phase distortion on the multispectral image based on the axial traction speed and the radial torsion angle.

[0030] Motion sensing and artifact correction module 2 includes: A laser velocity sensor with a velocity measurement accuracy of ±0.01m / s is used to collect the axial traction speed of steel strands in real time. The photoelectric torsion angle sensor has an angle measurement accuracy of ±0.1° and is used to collect the radial torsion angle of steel strands, ensuring high-precision output within the detection range of torsion angle ≤5° / s.

[0031] The wide-beam coherent combining unit 22 is used to coherently combine ultraviolet (300-400nm), visible (400-700nm) and near-infrared (700-1000nm) light beams to improve the consistency and anti-interference capability of multi-band illumination.

[0032] The adaptive phase compensation unit 23 is equipped with an electro-optic modulator with a phase compensation response time of ≤0.1ms. It performs phase inverse modulation based on the real-time acquired axial traction speed and radial torsion angle to achieve axial motion trail compensation and radial phase distortion correction.

[0033] In this embodiment, the dual-dimensional correction algorithm adopts a fusion of real-time interpolation reconstruction and phase rearrangement, so that the motion artifact elimination rate reaches 85% under the condition of high-speed transportation of 8m / s.

[0034] Technical effect: By synchronously acquiring motion parameters and performing dual-dimensional compensation, image blurring and phase distortion caused by high-speed motion are eliminated, improving the accuracy of subsequent defect identification.

[0035] (ii) Optical Imaging Module 1; Optical imaging module 1 is used to synchronously acquire multispectral images of the steel strand in the circumferential direction.

[0036] Optical imaging module 1 includes: Four miniature multispectral cameras are arranged in a ring, with an imaging angle of 90°, achieving 360° circumferential coverage. The parameters of the miniature multispectral cameras include a focal length of 20mm, an acquisition frame rate of 200fps, and a resolution of 2048×2048 pixels.

[0037] Four polarized LED light sources. The parameters of the polarized LED light sources include: light intensity adjustment range of 0-5000lx; polarization angle adjustable range of 0-90°.

[0038] The high-speed image acquisition card 13 has a transmission rate of 10Gbps and supports direct transmission of uncompressed raw data.

[0039] Under ambient light intensity fluctuations of 500-2000 lx, the optical imaging module 1 automatically adjusts the light intensity and polarization angle through a dynamic light source compensation mechanism to suppress reflection interference.

[0040] Technical benefits: Ensures high-contrast, multispectral consistent original image data in complex industrial environments, providing stable input for texture masking and micro-defect enhancement.

[0041] (III) Algorithm Processing Module 3; The algorithm processing module 3 is connected to the motion sensing and artifact correction module 2, and is used to receive the corrected image and perform defect identification.

[0042] This module is based on the FPGA chip XC7K325T to build a hardware parallel processing architecture, with a single frame image processing time of 0.7ms.

[0043] Establishment of a standardized twisted texture spectral model: First, the following structural parameters are extracted: steel strand diameter: φ15.2mm; stranding pitch: 120mm; wire diameter: 5.07mm; helix angle: 15°. A standardized stranding texture spectral model is then established.

[0044] Then, layered differential operation is performed: using a differential operation threshold of 0.02, layered differential calculation is performed on the corrected image to adaptively shield the stranded texture spectral signal on the surface of the steel strand.

[0045] Then, micro-defect enhancement and localization are performed: a 5×5 Laplacian convolution kernel is used for edge enhancement to strengthen micro-defects such as micro-cracks and indentations; the pixel-level location information of the defects and their corresponding multispectral features are output.

[0046] Technical effect: Effectively extracts minute defect features against a background of strong texture interference, achieving a micro-defect detection accuracy of 99.6%.

[0047] (iv) Multi-sensor fusion module 4; Multi-sensor fusion module 4 is used for physical quantitative verification of defects, including: The laser force sensing unit 41 scans the defect area and collects the depth, area and opening width of the defect as physical parameters. The parameters during the scanning process are configured as follows: laser phase reflection sensor accuracy 0.001mm; scanning step distance 0.005mm; servo drive mechanism positioning accuracy ≤0.001mm.

[0048] Data fusion modeling unit 43 trained a fusion model based on 6000 sets of artifact-corrected experimental data, with a model fit R. 2 =0.985, establishing a two-way mapping and mutual verification relationship between multispectral features and physical parameters.

[0049] After the algorithm processing module 3 outputs the pixel-level location information of the defect, the multi-sensor fusion module 4 accurately triggers the scan, performing a local scan only on the defect area.

[0050] Technical Results: The above module settings resulted in a defect depth measurement error of 0.007mm; scanning efficiency was improved by 88% compared to full-area scanning; and defect quantification data was output (e.g., depth 0.09mm, area 0.9mm²). 2 ).

[0051] (v) System main control module 5; The system main control module 5 is connected to the motion sensing and artifact correction module 2, the optical imaging module 1, the algorithm processing module 3, and the multi-sensor fusion module 4, respectively, and is used to control the system's working timing and data interaction.

[0052] In this embodiment, the system main control module 5 uses a Siemens S7-1500 series PLC with a clock synchronization accuracy of ±1ms. This module is also equipped with an audible and visual alarm device with an alarm sound pressure level of 85dB. The system linkage interface 63 uses the Modbus / TCP protocol to interface with the production line PLC. After the test report is generated, the production line automatically adjusts the axial traction speed from 8m / s to 7.5m / s and optimizes the twisting force parameters.

[0053] Technical effect: The above settings enable the response time between the test report and the production line to be ≤0.4s, achieving closed-loop control.

[0054] II. Working principle of Example 1: When the steel strand passes through the detection area with an axial pulling speed of 8 m / s and a radial torsion angle of ≤5° / s: Phase 1: Real-time acquisition of motion parameters and artifact correction: The motion sensing and artifact correction module 2 acquires axial traction speed and radial torsion angle data in real time, and the adaptive phase compensation unit 23 performs dynamic phase compensation according to the parameters to achieve dual-dimensional correction of axial motion trailing and radial phase distortion.

[0055] The second stage involves simultaneous multispectral imaging: The optical imaging module 1 uses a ring array to synchronously acquire 360° circumferential multispectral images, and the light source unit performs dynamic compensation to output images with high consistency correction.

[0056] The third stage, texture masking and micro-defect identification: Algorithm processing module 3 calls the standardized twisted texture spectral model, uses a differential operation threshold of 0.02 to mask the twisted texture signal, and enhances micro-defect features with a 5×5 Laplacian convolution kernel to achieve pixel-level localization.

[0057] The fourth stage is multi-sensor physical quantization: The multi-sensor fusion module 4 triggers the laser force sensing unit 41 to perform a 0.005mm step-size scan based on pixel-level position information, obtains the defect depth and area, and performs cross-validation through a bidirectional mapping model.

[0058] Phase 5: Defect Level Assessment and Production Line Integration The quantitative evaluation module 6 performs graded judgment based on depth and area thresholds, and the system main control module 5 generates a detection report and links with the production line PLC through the Modbus / TCP protocol to realize automatic optimization of process parameters.

[0059] III. Overall Technical Effects: This embodiment, under high-speed production conditions of 8m / s, brings the following parameter improvements: motion artifact elimination rate of 85%, micro-defect detection accuracy of 99.6%, defect depth measurement error of 0.007mm, area scanning efficiency improvement of 88%, linkage response time ≤0.4s, and single-frame image processing time of 0.7ms. Therefore, this embodiment achieves high-precision online identification and quantitative evaluation of surface defects in steel strands, meeting the needs of continuous and automated industrial production inspection.

[0060] Example 2 is the second embodiment of the present invention. Based on Example 1, this embodiment further limits and optimizes the structure and collaborative mechanism of the optical imaging module 1, the motion sensing and artifact correction module 2, and the algorithm processing module 3, in order to further improve the multispectral imaging stability and micro-defect identification accuracy of the steel strand under high-speed operation.

[0061] I. Structure and Working Mechanism of Optical Imaging Module 1: The optical imaging module 1 includes: a ring multispectral camera array 11, a polarization light source compensation unit 12, and a high-speed image acquisition card 13.

[0062] (a) Circular multispectral camera array 11; The ring-shaped multispectral camera array 11 includes a predetermined number of miniature multispectral cameras arranged in a ring around the center of the steel strand transport.

[0063] In this embodiment, the preset number is 4-6, preferably 4. Each miniature multispectral camera is arranged at equal angles along the circumference of the steel strand to achieve 360° circumferential coverage without blind spots.

[0064] Detection band coverage: ultraviolet band 200-400nm, visible light band 400-760nm, and near-infrared band 760-1100nm.

[0065] The camera captures frames at a rate of ≥200fps and matches them in real time with the composite motion-phase deviation correlation model in the motion perception and artifact correction module 2.

[0066] Technical effect: It enables the acquisition of different responses to defects such as microcracks, scratches, and oxidation points on the surface of steel strands in different wavelength bands, thereby improving the resolution capability of multispectral features.

[0067] (ii) Polarization light source compensation unit 12; The polarization light source compensation unit 12 includes polarization LED light sources that correspond one-to-one with each miniature multispectral camera.

[0068] Each polarized LED light source is used to: dynamically adjust the emitted light intensity to a preset light intensity threshold range based on ambient light intensity. The preset light intensity threshold range is 0-5000 lx. It also adjusts the polarization angle to a preset angle threshold range. The preset angle threshold range is 0-180°.

[0069] When the ambient light intensity changes, the polarization light source compensation unit 12 adjusts the light intensity and polarization angle in real time to keep the imaging light intensity on the steel strand surface stable within the set range, avoiding interference from specular reflection or diffuse reflection.

[0070] Technical effect: It can still ensure the stability of grayscale distribution of multispectral images under fluctuating ambient light conditions in industrial workshops, thereby improving the reliability of subsequent differential operations.

[0071] (iii) High-speed image acquisition card 13; The input end of the high-speed image acquisition card 13 is connected to each miniature multispectral camera, and the output end is connected to the motion sensing and artifact correction module 2.

[0072] The high-speed image acquisition card 13 is used to convert multispectral images into digital signals and send them to the motion sensing and artifact correction module 2 at a speed not lower than a preset transmission rate threshold.

[0073] In this embodiment, the preset transmission rate threshold is ≥10Gbps.

[0074] Technical effect: Ensures no data delay and no frame loss under a capture frame rate of ≥200fps, providing a continuous image stream for real-time artifact correction.

[0075] II. Structure and Working Principle of Motion Sensing and Artifact Correction Module 2: The motion sensing and artifact correction module 2 includes: a motion parameter sensing unit 21, a wide-beam coherent synthesis unit 22, and an adaptive phase compensation unit 23.

[0076] (a) Motion parameter sensing unit 21; The motion parameter sensing unit 21 is used to collect the axial traction speed and radial torsion angle of the steel strand in real time under high-speed conveying conditions. The axial traction speed detection range is 0-10m / s with an accuracy of ±0.01m / s; the radial torsion angle detection range is 0-360° with an accuracy of ±0.1°.

[0077] The collected axial traction speed and radial torsion angle are simultaneously transmitted to the adaptive phase compensation unit 23 and synchronized to the system main control module 5.

[0078] (ii) Wide-beam coherent combining unit 22; The wide-beam coherent combining unit 22 is used to coherently combine the imaging beams of each band in the multispectral image acquired by the optical imaging module 1 to form a wide-beam composite imaging beam.

[0079] Its working mechanism is as follows: by superimposing the phase consistency of the beams in each band, the equivalent illumination area of ​​the imaging beam is expanded, reducing the problem of insufficient exposure time caused by high-speed axial movement.

[0080] Technical effects: Improves the ability to capture high-speed moving targets and reduces axial motion blur.

[0081] (iii) Adaptive phase compensation unit 23; The adaptive phase compensation unit 23 is connected to the motion parameter sensing unit 21 and the wide beam coherent synthesis unit 22, respectively.

[0082] Its working principle is as follows: receiving axial traction velocity and radial torsion angle; combining the frame rate of the multispectral camera with ≥200fps, establishing a composite motion-phase deviation correlation model; calculating the beam phase deviation caused by radial torsion in real time; performing adaptive phase compensation on the wide-beam composite imaging beam through an electro-optic modulator; simultaneously performing time-domain motion blur correction in combination with axial traction velocity; and outputting the corrected image to the algorithm processing module 3.

[0083] Correction performance indicators: motion artifact elimination rate ≥80%; image blur reduction ≥85%.

[0084] Technical effect: It achieves two-dimensional correction of axial motion trailing and radial phase distortion, laying the foundation for accurate separation of texture and defects.

[0085] The objective configuration of the composite motion-phase deviation correlation model is as follows: A coupled mapping relationship between the axial traction speed and radial torsion angle of the steel strand and the phase deviation of the imaging beam under the constraint of multispectral sampling frame rate is established to achieve unified compensation for axial motion trailing and radial phase distortion.

[0086] The model logic configuration of the composite motion-phase deviation correlation model is as follows: acquire the time series of axial traction velocity and radial torsion angle; establish the phase drift integral expression within a unit exposure cycle by combining the acquisition frame rate; introduce the phase distortion caused by radial angular velocity into the Bessel function to describe the periodic modulation; use the Gamma function and Riemann Zeta function to characterize the high-order spectral disturbance under high speed conditions; implement Gaussian noise filtering through the error function; and construct a unified normalized fractional expression.

[0087] The integrated formula configuration of the composite motion-phase deviation correlation model is as follows: ; in, It is the phase compensation coefficient, which is the output value of the composite motion-phase deviation; It is the exposure cycle time, which is the exposure duration of a single frame; It is the axial attenuation coefficient, which is the velocity attenuation ratio constant; It is the axial traction speed, which is the real-time speed of the steel strand; It is an integral time variable, and is a time variable; It is a first-order Bessel function, a function that describes periodic modulation; It is the torsional modulation proportional coefficient, and the angular velocity modulation constant; It is the radial torsional angular velocity, which is the torsional angle per unit time; It is the error function, which is a Gaussian filter function; It is the noise suppression factor, which is the noise smoothing constant; It is the Gamma function, a higher-order spectral enhancement function; It is the velocity order adjustment parameter, which is the translation constant of the Gamma function; It is the Riemann Zeta function, which is a harmonic coupling function; is the spectral offset constant, which is the translation parameter of the Zeta function; ln is the natural logarithm function, which is a stable function; It is the velocity square adjustment constant, and the velocity normalization constant; It is the angular velocity square adjustment constant. It is a normalization function, also known as a standardized function. This is the system noise amplitude, which is the average noise value collected.

[0088] Value range description: ,when A value close to 1 indicates that the phase deviation is minimal and requires no compensation, while a value close to 0 indicates that the phase distortion is severe and requires enhanced compensation.

[0089] Dimensional explanation: The integral term is an integral over time, the exponential and Bessel terms are dimensionless functions, the denominator is a log-normalized expression, and the whole is a dimensionless coefficient.

[0090] This formula expresses the phase accumulation over the exposure cycle through time integration. The periodic torsional modulation is described by a first-order Bessel function. The function enhances the spectral perturbation weights in the high-speed range. The function characterizes the harmonic superposition effect caused by torsion, the error function suppresses random noise, and the denominator is logarithmically normalized with respect to velocity and angular velocity, thus outputting a unit dimensionless phase compensation coefficient. This solves the phase distortion problem caused by high-speed compound motion.

[0091] III. Structure and Working Principle of Algorithm Processing Module 3: The algorithm processing module 3 includes: a texture modeling unit 31, an adaptive masking unit 32, a micro-defect enhancement unit 33, and a defect localization unit 34. In this embodiment, the processing time for a single frame image is ≤0.8ms.

[0092] (a) Texture modeling unit 31; The texture modeling unit 31 is used to extract the strand pitch, wire diameter, helix angle and texture spectral reflectance characteristics of the steel strand, and to establish a standardized strand texture spectral model.

[0093] The construction steps and logical rule configuration for the standardized twisted texture spectral model are as follows: 1. Periodic feature extraction: Spatial sampling of the strand pitch, wire diameter and helix angle of the steel strand is performed to generate a periodic position sequence.

[0094] 2. Spectral mapping: Based on the reflectance data collected by the multispectral camera, calculate the normalized reflectance intensity of each pixel in each band.

[0095] 3. Texture spectrum modeling: High-order Fourier series are used to represent the multi-frequency features of the twisted texture. At the same time, elliptic integral function is used to characterize the nonlinear spiral structure, and a Bessel function is added for micro-angle modulation to ensure fine capture of periodic modulation.

[0096] 4. Normalization: The output spectrum values ​​are normalized to the [0, 1] interval to facilitate subsequent adaptive masking.

[0097] The formula configuration for the standardized twisted texture spectral model is as follows: ; in, It is the texture spectrum reference value, which is the standardized texture intensity output; For periodic points, and These are the Fourier series coefficients, calculated from historically collected samples; For the second type of elliptic integral function, it represents the nonlinear spatial mapping of the spiral structure; It is a zero-order Bessel function used for angle micro-amplitude modulation; The Fourier harmonic order; This is a periodic parameter, corresponding to the twist pitch.

[0098] The high-order spectral superposition of the texture can accurately reproduce the stranding period characteristics of the steel strand; Micro-angle modulation is applied to each harmonic to enhance the effect of the helix angle on the reflectivity. The non-linear spatial projection used to depict the twisting makes the texture model more closely resemble the actual twisting structure. It can be used as a texture masking reference to eliminate the interference of twisted textures on defect detection and ensure the visibility of micro-defect features.

[0099] Range description: Ψ(x)∈[0,1], 0 indicates no texture contribution, 1 indicates maximum texture contribution.

[0100] The technical effects of the standardized twisted texture spectral model: By periodically sampling the stranding pitch, wire diameter, and helix angle of the steel strand, and combining high-order Fourier series superposition and elliptic integral function modeling, the model can accurately restore the three-dimensional helical structure of the steel strand and its spectral reflectance characteristics, and realize the digital representation of the standard stranding texture. The standardized texture spectral values ​​output by the model can be used as a reference signal to perform layered differential operations with the actual acquired images, automatically shielding the spectral interference of the conventional stranded texture on the surface of the steel strand. By introducing Bessel function modulation of periodic harmonics, the model can maintain high-precision shielding even with small changes in texture, avoiding misjudging normal textures as defects; By masking out large areas of twisted texture, minute defect signals can be significantly highlighted, reducing false positive and false negative rates. Experiments show that this model can improve the contrast of minute defect signals by approximately 3-5 times, ensuring the reliability of pixel-level defect detection. Because the model is based on the superposition of periodic Fourier series and the elliptic integral function that can be calculated in real time, it can quickly process the multispectral images of high-speed conveying steel strands, ensuring that the texture masking processing time under the 8m / s high-speed production line is ≤0.8ms / frame, which meets the requirements of real-time detection.

[0101] (ii) Adaptive shielding unit 32; The adaptive shielding unit 32 connects the motion sensing and artifact correction module 2 and the texture modeling unit 31.

[0102] Its working process is as follows: receiving the corrected image; performing layered difference operations between the corrected image and the standardized stranded texture spectral model; and using an adaptive threshold algorithm to shield the stranded texture spectral signal on the steel strand surface. The adaptive threshold is dynamically adjusted according to the image grayscale distribution to ensure that the texture signal is effectively suppressed without accidentally damaging the defect signal.

[0103] Technical effect: Eliminates periodic twisted texture interference and improves the signal-to-noise ratio of defect signals.

[0104] An adaptive thresholding algorithm is used to distinguish normalized stranded texture signals from potential defect signals in high-speed acquired multispectral images, enabling preprocessing before micro-defect enhancement. This algorithm dynamically adjusts the threshold based on the local spectral distribution of each frame, rather than using a fixed threshold, thus adapting to variations in the surface texture of the steel strand and ambient lighting. The main functions of this algorithm include: 1. Local statistical feature calculation: Calculate the mean value within a local window (e.g., 5×5 pixels) for each pixel. and standard deviation .

[0105] 2. Dynamic Threshold Adjustment: Calculate pixel thresholds based on local statistical features. : ; in, The adjustment coefficient can be dynamically set according to the image noise level and light intensity fluctuation (α=1.2~1.5 in this embodiment).

[0106] 3. Texture signal masking: Masking the corrected image With threshold In comparison, the twisted texture signal is masked, and the processed image is output. : ; Among them, the pixel values ​​of the image When the value is less than the threshold, it is treated as texture background and masked (set to zero); when the value is greater than the threshold, the defect signal is retained and enhanced.

[0107] Image output process optimization after masking: 1. Multi-scale local analysis: To further adapt to variations in wire diameter and helix angle of steel strands, local mean and standard deviation can be calculated in multi-scale windows (such as 5×5, 7×7, and 9×9 pixels). By weighted averaging the threshold results at different scales, a final adaptive threshold is obtained, which is more robust to texture periodic fluctuations and local illumination changes.

[0108] 2. Dynamic spectral compensation: Before calculating the threshold, the multispectral band data of each pixel is normalized to ensure that the spectral values ​​of different bands can be directly compared, and to avoid threshold deviation caused by differences in light intensity of different bands.

[0109] 3. Output image optimization: For the masked image It can be directly fed into the micro-defect enhancement unit; at the same time, a binarized mask image is generated. : ; Mask image Used to label candidate micro-defect regions, improving the efficiency of subsequent Laplacian convolution kernel processing and reducing computational load.

[0110] Summary of the algorithm steps: Input: Corrected multispectral image ; Local statistics: Calculate the local mean for each pixel i and j. and standard deviation ; Calculate adaptive pixel threshold : ; Shielding twisted texture: With threshold Compare and generate masked images ; Generate a binary mask: Output candidate defect region mask ; Output: Image after masking and mask image This provides input for the micro-defect enhancement unit.

[0111] The technical effectiveness of this algorithm: By using a dynamic threshold to shield the spectral signal of the conventional stranding texture on the surface of the steel strand, the interference of texture with the detection of micro-defects is avoided, thereby improving the detection accuracy.

[0112] After shielding, the tiny defect pixels are preserved and enhanced, the signal contrast is improved, and it is convenient for further analysis by a dedicated spectral model for the tiny defects.

[0113] The adaptive algorithm can adjust the threshold in real time according to local spectral fluctuations, ensuring stable operation even under conditions of ambient light intensity fluctuations of 500-2000 lx and high-speed conveying at 8 m / s in industrial workshops.

[0114] Binary masks can effectively limit the candidate region for micro-defects, reduce the operation range of the Laplacian convolution kernel, and ensure that the processing time of a single frame is ≤0.8ms, meeting the requirements of online inspection in high-speed production lines.

[0115] (iii) Micro-defect enhancement unit 33; The micro-defect enhancement unit 33 is connected to the adaptive shielding unit 32.

[0116] Specifically, it is used to: employ an edge enhancement algorithm with a 5×5 Laplacian convolution kernel to perform feature magnification processing on the masked image in order to extract the spectral features of micro-defects and establish a micro-defect-specific spectral model.

[0117] Technical effect: Enhances the gradient changes at the edges of micro-cracks, indentations, and other minute defects, enabling them to form distinguishable features in multispectral space.

[0118] Steps and logical rules for constructing a micro-defect-specific spectral model: 1. Enhanced region selection: Extract candidate regions for micro-defects from the masked image.

[0119] 2. Local spectral enhancement: The micro-defect region is weighted and amplified using an exponential function and an error function to suppress surrounding background noise.

[0120] 3. Curvature Modulation: The local geometric curvature of microcracks or indentations is characterized using second-order Bessel functions.

[0121] 4. Cumulative calculation: Integrate the local spectral enhancement value along the defect region to obtain the micro-defect-specific spectral index.

[0122] Formula for the micro-defect-specific spectral model: ; ; in, For micro-defect specific spectral indices, For pixel index, For the pixel coordinates of the micro-defect, Normalized spectral reflectance; This is the exponential function decay coefficient, used to suppress edge noise; These are parameters for the error function, used to smooth the background; The second-order Bessel function is used to characterize the local curvature of the defect. This represents the distance from the (i, j)th pixel within the micro-defect region to the defect center. , The Euclidean distance of ). Used to attenuate the spectral contribution of defective edge pixels and suppress interfering background; Achieve Gaussian smoothing to improve the signal-to-noise ratio for micro-defects; Characterizing the local curvature modulation of defects makes the spectral features of microcracks or indentations more prominent; After summing up the pixels of all defective areas, we get As a spectral index specific to micro-defects, it provides a basis for defect quantification and classification.

[0123] Value range description: ≥0, >1 indicates a significant minor defect. ≤1 indicates no significant abnormality.

[0124] The technical effects of the micro-defect-specific spectral model: By using exponential and error functions to weight and enhance the defect region, the spectral features of the core defect pixels are amplified while background interference is suppressed. Furthermore, by characterizing the local curvature using a second-order Bessel function, the local structural features of microcracks or indentations can be further highlighted. The enhanced spectral signal can be used to accurately calibrate the pixel coordinates and area of ​​micro-defects, and in conjunction with the standardized texture model, it can achieve automatic classification of micro-defects (minor / moderate / serious), ensuring the accuracy of quantitative measurement and classification of micro-defects; The model's computational logic and formula design ensure that, even under high-speed multispectral image acquisition conditions, the single-frame image processing time can still be ≤0.8ms, with a significant improvement in signal-to-noise ratio. Experiments show that after enhancement of the micro-defect signal, its spectral response peak can be increased by approximately 2-4 times, enabling reliable detection of even minor defects with a depth ≤0.05mm. Output micro-defect specific spectral index It can be directly used for the quantification of defect depth and area, and can be linked with the production line PLC to fine-tune production parameters (such as stranding force and traction speed) to achieve active defect control and improve the quality and pass rate of steel strand production line.

[0125] (iv) Defect location unit 34; The defect location unit 34 is connected to the micro-defect enhancement unit 33.

[0126] Its working principle is as follows: pixel-level coordinate calibration is performed on the micro-defect region in the enhanced micro-defect image, and the output includes: defect plane coordinates, region range, preliminary defect type determination result, and corresponding multispectral features. The output data is transmitted to the multi-sensor fusion module 4.

[0127] Technical effect: Achieves pixel-level precise positioning of micro-defects, providing accurate trigger coordinates for subsequent physical parameter scanning.

[0128] IV. Overall Working Principle of Example 2: During the high-speed transport of steel strands: Phase 1: Multispectral image acquisition: The ring-shaped multispectral camera array 11, in conjunction with the polarization light source compensation unit 12, completes 360° circumferential multispectral acquisition, and the high-speed image acquisition card 13 transmits image data at a rate of ≥10Gbps.

[0129] Second stage, compound motion artifact correction: The motion parameter sensing unit 21 collects the axial traction speed and radial torsion angle, and the adaptive phase compensation unit 23 establishes a composite motion-phase deviation correlation model to dynamically compensate the wide-beam composite imaging beam and realize two-dimensional motion correction.

[0130] The third stage, texture stripping: The adaptive shielding unit 32 shields the twisted texture signal through layered differential operation and adaptive threshold algorithm.

[0131] The fourth stage, micro-defect enhancement and localization: The micro-defect enhancement unit 33 uses a 5×5 Laplacian convolution kernel to enhance the edges and establish a micro-defect-specific spectral model; the defect localization unit 34 completes pixel-level coordinate calibration and outputs defect features.

[0132] V. Overall Technical Effects: Through the synergistic effect of the above structures, this embodiment achieves: image artifact elimination rate ≥80% under high-speed motion conditions; image blur reduction ≥85%; single-frame image processing time ≤0.8ms; pixel-level precise stripping of multispectral textures and micro-defects; and provides high-confidence defect coordinates and spectral features for the subsequent multi-sensor fusion module 4.

[0133] Example 3, the third embodiment of the present invention, provides an online identification system for surface defects of steel strands based on multispectral imaging, applied to the online detection of micro-defects on a high-speed steel strand production line. Based on artifact correction and texture masking, the system achieves high-precision quantification of defect physical parameters through a multi-sensor fusion module 4, and combines this with the system's main control module 5 for real-time control and anomaly handling, achieving high-speed, accurate, and reliable online detection.

[0134] I. Working principle of multi-sensor fusion module 4: The multi-sensor fusion module 4 combines the pixel-level location information of micro-defects output by the algorithm processing module 3 with the physical quantity sensing data, realizing bidirectional mapping and verification between spectral features and defect physical parameters, ensuring accurate and reliable quantization results. Its main working principle is as follows: Laser force sensing unit 41 scanning: Based on the pixel-level position of the defect provided by the algorithm processing module 3, the non-contact laser phase reflection sensor scans the defect area and collects physical parameters such as defect depth, defect area and opening width. Equidistant micro-step scanning is adopted, with a step size ≤ 0.005 mm, to ensure complete capture of minute defects, while discarding data from non-defective areas to avoid noise interference.

[0135] Precise triggering linkage: The precise triggering linkage unit 42 receives pixel-level defect coordinates and moves the laser emitter head quickly and accurately to the defect area through the servo drive mechanism to achieve micro-step scanning; Linked control ensures that scanning is limited to the actual defect area, thereby significantly improving data acquisition efficiency (approximately 85% higher than full-area scanning) and reducing redundant data interference.

[0136] Data fusion modeling unit 43: Receives physical parameters and corresponding multispectral features obtained from laser scanning, and establishes a nonlinear correlation model to achieve bidirectional mapping through a machine learning algorithm that combines random forest and gradient boosting; the model fit is ≥0.98, and can be used for: Spectral prediction of physical parameters: Defect depth and area are estimated through multispectral features; Physical parameters are used to infer spectral authenticity: Physical parameters measured by laser are used to verify whether there are artifacts or texture interference in the spectral signal; By eliminating invalid data through bidirectional verification, accurate defect quantification data is output, including depth, area, opening width, and micro-defect classification.

[0137] Explanation of the construction of nonlinear correlation models: Nonlinear correlation models are used to establish a bidirectional mapping and verification relationship between multispectral features and laser force sensing physical parameters. Their core purpose is to achieve: Spectral prediction of physical parameters: Predicting defect depth, area, and opening width using multispectral features; Verifying Spectral Authenticity using Physical Parameters: The effectiveness of spectral features is verified by actually acquiring physical parameters, eliminating artifacts and texture interference. The construction logic is based on nonlinear mapping + high-dimensional integration + information entropy optimization + function regularization, and the specific steps are as follows: 1. Collect micro-defect data after artifact correction and texture masking to generate a training set; 2. Normalize each set of spectral features and corresponding physical parameters to the [0, 1] interval to reduce the influence of dimensional differences; 3. Use a fusion algorithm of random forest and gradient boosting to model the initial nonlinear relationship and output preliminary predictions; 4. Constructing a formulaic nonlinear function By combining advanced integral functions, Bessel functions, and error functions, continuous and nonlinear adjustment capabilities are introduced. 5. Spectral anomalies are suppressed through the error function (erf), and the sensitivity to local micro-defects is enhanced through the Bessel function, thus achieving accurate two-way verification.

[0138] Formula for nonlinear correlation model: ; ; in, For the predicted defect physical parameter vector , It is a nonlinear correlation model. Input spectral features (after normalization); This is a first-class Bessel function, used to enhance the response to local micro-defect features; The error function is in integral form, representing the cumulative probability characteristics of the spectral signal; This is an error function used to suppress anomalous spectral points. , The mean and standard deviation of the spectral characteristics; It is a logarithmic function used to dynamically compress high-intensity spectral values ​​and prevent high values ​​from dominating the prediction; This is the Riemann Zeta function, used for high-order tuning of the overall nonlinear coupling of spectral features; The normalization coefficient is... The number of spectral dimensions for fusion.

[0139] The range of the model formula: ∈[0, 5] mm, indicating the predicted range of defect depth; ∈[0, 10] mm, indicating the predicted range of defect area; ∈[0, 2]mm, representing the predicted range of defect opening width; the larger the value, the more severe the defect, corresponding to the production line warning level.

[0140] Algorithm details include: Local enhancement: Bessel function Amplify the response in micro-defect sensitive areas to ensure that small spectral fluctuations can significantly affect predictions; Anomaly Suppression: Error Function Extreme spectral values ​​are suppressed to [-1, 1] to avoid noise affecting the prediction of physical parameters; Nonlinear Coupling Regulation: Riemann Zeta Function It is responsible for handling higher-order interactions between spectra to ensure that multispectral features are used to predict physical parameters. Summation logic: Weighted summation is performed on all spectral dimensions to integrate global and local information and achieve a balance between prediction and verification.

[0141] Technical effects of the algorithm: High-precision prediction: By combining local micro-defect enhancement (Bessel function) and anomaly suppression (error function), the physical parameters of micro-defects can be accurately predicted; Two-way verification: The nonlinear coupling mechanism ensures that physical parameters and spectral features verify each other, eliminating artifacts or texture interference; High stability: Robust to spectral noise and illumination changes, ensuring the reliability of defect quantification under high-speed transmission; Balancing global and local considerations: The coupling of summation and Zeta function amplifies local feature responses while ensuring global consistency, thus optimizing defect identification and quantization efficiency.

[0142] II. Working principle of the system's main control module 5: The system's main control module 5 is the core control unit of the entire detection system, enabling coordination among modules, data scheduling, and anomaly handling. Its main functions are as follows: Anomaly handling: Real-time monitoring of the working status of motion sensing and artifact correction module 2, optical imaging module 1, algorithm processing module 3 and multi-sensor fusion module 4; When equipment malfunctions, image acquisition abnormalities occur, sensor data exceeds the range, or the motion parameters of the steel strand change suddenly, an audible and visual alarm (sound pressure ≥ 85dB, red flashing light) is immediately triggered and the detection process is suspended; all detection data before the abnormality occurs is saved to local storage to ensure the integrity of the detection data.

[0143] Adaptive control: The acquisition frame rate of the multispectral camera and the phase compensation frequency of the adaptive phase compensation unit 23 in the motion sensing and artifact correction module 2 are automatically adjusted according to the axial traction speed of the steel strand; to ensure that the system operating parameters are synchronized with the high-speed transport of the steel strand, and to eliminate artifacts or defects that are missed due to speed fluctuations.

[0144] High-precision clock synchronization: Provides a unified time reference for all modules, ensuring the timing consistency of image acquisition, laser scanning and algorithm processing, and adapting to high-speed online detection scenarios.

[0145] III. Testing Workflow: The steel strand enters the detection area under high-speed conveying. The motion sensing and artifact correction module 2 collects the axial velocity and radial torsion angle and eliminates motion artifacts. Optical imaging module 1 acquires multispectral images, and algorithm processing module 3 performs standardized twisted texture masking, adaptive thresholding, and micro-defect enhancement, outputting pixel-level defect coordinates and spectral features. The multi-sensor fusion module 4 triggers the laser force sensing unit 41 to perform micro-step scanning of the defect area and collect physical parameters. The data fusion modeling unit 43 performs bidirectional mapping and verification of spectral and physical parameters through a pre-trained nonlinear model and outputs defect quantification data. The system's main control module 5 generates alarms and reports based on quantitative data, or links with the production line PLC. When necessary, it adjusts the steel strand traction speed and detection parameters to achieve fully automatic online detection.

[0146] IV. Technical Effects: High-precision quantization: The laser force sensing unit 41 combines the algorithm to process the output pixel-level position information of micro-defects, realizing high-precision quantization of defect depth, area, and opening width (depth error ≤ 0.01mm). High-efficiency scanning: Precise triggering avoids full-area scanning, collecting data only from actual defect areas, improving scanning efficiency by approximately 85%; Spectral-physical bidirectional verification: The data fusion modeling unit 43 uses a nonlinear model to predict physical parameters from the spectrum and inversely infer the authenticity of the spectrum from the physical parameters, eliminating artifacts and texture interference data to ensure accurate and reliable defect detection. Adaptive system control: The main control module automatically adjusts the camera frame rate and phase compensation frequency according to the axial speed of the steel strand to ensure artifact elimination and micro-defect capture under high-speed transmission conditions; Anomaly protection and data security: The anomaly handling unit 51 can pause detection and save data when equipment fails or sensors malfunction, ensuring the continuity of the detection process and the integrity of the data.

[0147] Example 4 is the fourth embodiment of the present invention. Based on Examples 1 to 3, this embodiment further includes a quantitative evaluation module 6. Its core function is to determine the level of defects, generate reports, and control the production line in conjunction with the quantitative defect data output by the multi-sensor fusion module 4, so as to realize a quantifiable safety assessment of steel strand defects.

[0148] I. System Composition and Working Principle: (1) Function of defect level determination unit 61: Receive the defect physical parameters (d, A, w) output by multi-sensor fusion module 4, where d is the defect depth, A is the defect area, and w is the defect opening width.

[0149] Threshold definition (based on GB / T 5224 and workshop-specific safety specifications): Minor defects: d≤0.05mm, w≤0.2mm; Common defects: ; Critical defects: ; Determination Algorithm: Defect level L is determined according to the following logic: ; , , These are the level mapping functions corresponding to depth, area, and opening width, respectively. Exceeding the limit for any parameter will improve the overall defect level, thus achieving the principle of safety first.

[0150] Technical benefits: By comprehensively determining the defect level through multi-dimensional physical parameters, the objective quantification of defect classification is achieved, avoiding misjudgments caused by single-parameter judgment, and ensuring the safety and quality consistency of steel strand production.

[0151] (2) Report output unit 62; Function: Based on the defect level determination results and the physical parameters and multispectral characteristics output by the multi-sensor fusion module 4, an integrated inspection report is generated. The report content includes: Defect location (pixel coordinates and circumferential angle of steel strand), defect type (microcrack, pit, peeling, etc.), defect physical parameters (depth d, area A, opening width w), safety level (minor, moderate, severe), multispectral characteristics (UV, visible light, near-infrared spectral values), and steel strand motion parameters (axial traction speed, radial torsion angle).

[0152] Output methods: Supports local storage (SD card / hard drive) and remote network transmission (Ethernet, Modbus / TCP, Profinet).

[0153] Technical benefits: Enables complete traceability and historical record of testing data, providing a reliable data foundation for production line operation, quality traceability, and maintenance management.

[0154] (3) System linkage interface 63; Function: Transmits defect levels and inspection reports to the production line PLC or operation and maintenance management platform in real time via industrial communication protocols.

[0155] Application Scenarios: Production Process Adjustment: Adjust the traction speed, stranding force, or oiling dosage of steel strands based on defect levels. Equipment Maintenance Reminders: Trigger maintenance reminders for areas with recurring severe defects. Safety Rating: Generate quantifiable safety levels for steel strand batches, providing a reference for subsequent use in concrete structures.

[0156] Technical benefits: Enables closed-loop linkage between the detection module and the production line, providing real-time feedback on defect information, shortening response time, improving production efficiency, and reducing quality risks.

[0157] II. Workflow of Quantitative Assessment Module 6: After the multi-sensor fusion module 4 completes the defect quantification, it sends the physical parameters and spectral characteristics to the defect level determination unit 61. The defect level determination unit 61 automatically calculates the defect level L according to the threshold rule and outputs the level result; The report output unit 62 integrates the defect physical parameters, level, spectral characteristics and motion parameters to generate an integrated inspection report, and simultaneously stores and prepares it for transmission; The system linkage interface 63 transmits defect levels and inspection reports to the PLC control system or operation and maintenance platform in real time via industrial protocols, enabling production line process adjustment and safety management.

[0158] III. Summary of Technical Results: Automated level determination: Defects are assessed for safety level using multi-dimensional physical parameters, achieving full automation and rapid response; Information visualization and traceability: Generate structured inspection reports, including defect location, type, spectral characteristics, and motion parameters, facilitating quality control and operation and maintenance management; Production closed-loop optimization: By linking with PLC and operation and maintenance system, a closed loop of detection-feedback-control is realized to improve the safety and stability of steel strand production; Standardization and Customization: Thresholds and levels can be adjusted according to GB / T 5224 or workshop-customized safety specifications, taking into account both industry standards and actual on-site needs; High-efficiency response: Defect level determination and report generation are completed in real time (single defect processing time ≤1ms), ensuring real-time application on high-speed steel strand production lines.

[0159] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An online identification system for surface defects in steel strands based on multispectral imaging, characterized in that, include: An optical imaging module, connected to a motion sensing and artifact correction module, is used to synchronously acquire multispectral images of the steel strand in the circumferential direction and send the multispectral images to the motion sensing and artifact correction module. The motion sensing and artifact correction module is used to acquire the axial traction speed and radial torsion angle of the steel strand in real time under high-speed conveying conditions, and to receive the multispectral image acquired by the optical imaging module. Based on the axial traction speed and the radial torsion angle, the module performs two-dimensional correction of axial motion trailing and radial phase distortion on the multispectral image, and outputs the corrected image. The algorithm processing module, connected to the motion sensing and artifact correction module, is used to receive the corrected image, perform layered differential operations on the corrected image according to the pre-established standardized stranded texture spectral model, adaptively shield the stranded texture spectral signal on the surface of the steel strand, and perform micro-defect enhancement and localization on the shielded image, outputting the pixel-level location information of the defects and the corresponding multispectral features. The multi-sensor fusion module, connected to the algorithm processing module, is used to receive the pixel-level location information of the defect and the corresponding multispectral features, trigger a non-contact laser force sensing unit to scan the defect area based on the pixel-level location information, collect the depth, area and opening width of the defect as physical parameters, and establish a bidirectional mapping and mutual verification relationship between the multispectral features and the physical parameters based on the pre-trained fusion model, and output defect quantification data. The system main control module is connected to the motion sensing and artifact correction module, the optical imaging module, the algorithm processing module, and the multi-sensor fusion module, respectively, and is used to control the system's working sequence and data interaction.

2. The online identification system for steel strand surface defects based on multispectral imaging according to claim 1, characterized in that, The optical imaging module includes: A ring-shaped multispectral camera array, comprising a predetermined number of miniature multispectral cameras arranged in a ring around the center of the steel strand transport center; A polarization light source compensation unit includes a polarization LED light source corresponding to the miniature multispectral camera. Each polarization LED light source is used to dynamically adjust the emitted light intensity to a preset light intensity threshold range and adjust the polarization angle to a preset angle threshold range according to the ambient light intensity. A high-speed image acquisition card, with its input end connected to each of the miniature multispectral cameras and its output end connected to the motion sensing and artifact correction module, is used to convert the multispectral image into a digital signal and send it to the motion sensing and artifact correction module at a speed not lower than a preset transmission rate threshold.

3. The online identification system for steel strand surface defects based on multispectral imaging according to claim 1, characterized in that, The motion sensing and artifact correction module includes: The motion parameter sensing unit is used to collect the axial traction speed and radial torsion angle of the steel strand in real time under high-speed conveying conditions; The wide-beam coherent combining unit is used to coherently combine the imaging beams of each band in the multispectral image acquired by the optical imaging module to form a wide-beam composite imaging beam, so as to improve the ability to capture high-speed moving targets. An adaptive phase compensation unit is connected to the motion parameter sensing unit and the wide-beam coherent synthesis unit, respectively. It is used to receive the axial traction speed and the radial torsion angle, and, in combination with the camera acquisition frame rate of the optical imaging module, establish a composite motion-phase deviation correlation model. It calculates the beam phase deviation caused by radial torsion in real time, and performs adaptive phase compensation on the wide-beam composite imaging beam through an electro-optic modulator to eliminate axial motion trailing and radial phase distortion, and outputs the corrected image. The motion parameter sensing unit transmits the collected axial traction speed and radial torsion angle to the adaptive phase compensation unit and synchronizes them to the system main control module.

4. The online identification system for steel strand surface defects based on multispectral imaging according to claim 1, characterized in that, The algorithm processing module includes: The texture modeling unit is used to extract the strand pitch, wire diameter, helix angle and texture spectral reflectance characteristics of steel strands, and to establish a standardized strand texture spectral model. An adaptive shielding unit, connected to the motion sensing and artifact correction module and the texture modeling unit, is used to receive the corrected image, perform layered difference operations on the corrected image and the standardized stranded texture spectral model, and shield the stranded texture spectral signal on the surface of the steel strand based on a preset adaptive threshold algorithm, and output the shielded image. The micro-defect enhancement unit, connected to the adaptive shielding unit, is used to perform edge enhancement processing on the shielded image using a Laplacian convolution kernel of a preset size to amplify the spectral features of the micro-defect, establish a micro-defect-specific spectral model, and output the enhanced micro-defect image. The defect localization unit, connected to the micro-defect enhancement unit, is used to perform pixel-level coordinate calibration of the micro-defect region in the enhanced micro-defect image and output the pixel-level location information of the defect and the corresponding multispectral features.

5. The online identification system for steel strand surface defects based on multispectral imaging according to claim 4, characterized in that, The micro-defect enhancement unit is specifically used to: perform feature magnification processing on the masked image using a 5×5 Laplacian convolution kernel edge enhancement algorithm to extract the spectral features of the micro-defects and establish a micro-defect-specific spectral model.

6. The online identification system for surface defects of steel strand based on multispectral imaging according to claim 4, characterized in that, The multi-sensor fusion module includes: A laser force sensing unit is used to scan the defect area and collect the depth, area, and opening width of the defect as the physical parameters. The precise trigger linkage unit is connected to the algorithm processing module and the laser force sensing unit respectively, and includes a servo drive mechanism. The precise trigger linkage unit is used to receive the pixel-level position information of the defect, control the laser emitter to move to the defect area through the servo drive mechanism according to the pixel-level position information, and perform equidistant micro-step scanning on the defect area, and transmit the collected physical parameters to the data fusion modeling unit. The data fusion modeling unit, connected to the precise triggering linkage unit, is used to receive the physical parameters and corresponding multispectral features, establish a bidirectional mapping and mutual verification relationship between the multispectral features and the physical parameters based on the pre-trained fusion model, and output defect quantification data.

7. The online identification system for steel strand surface defects based on multispectral imaging according to claim 6, characterized in that, The data fusion modeling unit is specifically used to: based on a preset number of sets of experimental data on micro-defects in steel strands that have been corrected for artifacts, train and establish a nonlinear correlation model between the multispectral features of micro-defects in steel strands and the physical parameters of laser force sensing using a machine learning algorithm that combines random forest and gradient boosting, and the fitting degree of the nonlinear correlation model is not lower than a preset fitting degree threshold.

8. The online identification system for steel strand surface defects based on multispectral imaging according to claim 1, characterized in that, The system main control module includes: The anomaly handling unit is used to monitor the working status of each module and the motion parameters of the steel strand in real time. When at least one of the following abnormal events is detected: equipment failure, abnormal image acquisition, sensor data exceeding the range, or sudden change in the motion parameters of the steel strand, the unit immediately triggers the audible and visual alarm device to issue an alarm signal, suspends the detection process, and saves the detection data before the abnormal event occurred. An adaptive control unit is connected to the motion sensing and artifact correction module and the optical imaging module, respectively. It is used to automatically adjust the acquisition frame rate of the multispectral camera in the optical imaging module according to the axial traction speed of the steel strand, and to automatically adjust the phase compensation frequency of the adaptive phase compensation unit in the motion sensing and artifact correction module, so that the system operating parameters are adaptively matched with the steel strand conveying speed. A high-precision clock synchronization unit is used to provide a unified time reference for each module, ensuring the consistency of the working sequence of each module and adapting to the online detection requirements of steel strand in preset high-speed transmission scenarios.

9. The online identification system for surface defects of steel strand based on multispectral imaging according to claim 1, characterized in that, It also includes a quantitative evaluation module, which is connected to the multi-sensor fusion module and the system main control module, and is used to determine the level of the defect quantitative data and output the visualization.

10. The online identification system for steel strand surface defects based on multispectral imaging according to claim 9, characterized in that, The quantitative evaluation module includes: The defect level determination unit is used to receive the defect quantification data output by the multi-sensor fusion module, divide the defect into multiple preset safety levels according to the preset defect depth threshold, defect area threshold and opening width threshold, and output the defect level determination result. The report output unit, connected to the defect level determination unit, is used to generate an integrated inspection report based on the defect quantification data and the defect level determination result; the integrated inspection report includes at least the defect location, defect type, physical parameters, safety level, multispectral characteristics, and motion parameters; The system linkage interface connects the report output unit and the system main control module respectively, and is used to transmit the defect level judgment result and the integrated inspection report to the PLC control system or operation and maintenance management platform of the steel strand production line in real time through a preset industrial communication protocol.

Images