Method and system for evaluating corrosion resistance of composite board based on spectral analysis

Abstract

The invention relates to the field of multispectral imaging analysis and industrial automation control, in particular to a composite board corrosion resistance evaluation method and system based on spectral analysis. The method comprises the following steps: configuring a multispectral imaging unit to synchronously acquire first and second reference wave bands and characteristic wave band data on the surface of a plate; calculating a nonlinear attenuation factor representing the flattening pressing texture by using the logarithmic difference ratio of the reference wave band, and constructing a theoretical reference reflectivity of the characteristic wave band under the pure physical texture based on a power law model; and physical texture interference is eliminated by calculating a logarithmic residual error between a measured value and a theoretical reference reflectivity, so that the corrosion resistance of the composite board is evaluated. By utilizing the optical transmission and absorption characteristics of the metal surface passivation layer, the distinguishing between physical texture and chemical corrosion is realized, the false alarm problem of traditional optical detection is effectively solved, and high-precision online corrosion resistance evaluation is realized.

Description

Technical Field

[0001] This invention relates to the fields of multispectral imaging analysis and industrial automation control, specifically to a method and system for evaluating the corrosion resistance of composite plates based on spectral analysis. Background Technology

[0002] Bimetallic composite plates made of titanium steel, nickel steel, etc. are widely used in extreme working environments such as petrochemical containers, marine engineering platforms and nuclear power heat exchangers due to their excellent corrosion resistance and structural strength. These plates are usually manufactured by explosive welding or hot rolling processes. After the composite is completed, in order to eliminate the huge residual stress and ensure the flatness of the plate surface, it must be subjected to high-intensity physical leveling treatment by a large multi-roll leveling machine. However, this necessary machining process poses a significant challenge to the evaluation of the corrosion resistance of composite panels. High-intensity rolling will produce micron-level anisotropic pressure textures or fine roller marks on the surface of the metal coating. From an optical perspective, these physical textures will cause strong diffuse reflection and scattering of incident light, resulting in a significant attenuation of local reflected light intensity. In the existing industrial inspection system, it mainly relies on manual visual inspection or traditional machine vision technology based on RGB visible light. Traditional colorimetric methods or grayscale threshold algorithms determine defects solely based on changes in light intensity, failing to differentiate between physical light intensity attenuation caused by surface roughness and chemical light intensity attenuation caused by oxidation and corrosion. This leads to a common industry pain point: when harmless physical indentations or processing textures exist on the board surface, traditional methods often misjudge them as severe corrosion points or surface contamination, resulting in extremely high false alarm rates and unnecessary rework, severely impacting production efficiency. Furthermore, while traditional electrochemical probe detection methods are accurate, they are contact-based offline detection methods, which cannot meet the needs of high-speed continuous production lines for comprehensive online evaluation. Summary of the Invention

[0003] To address the problem that existing technologies cannot efficiently analyze the physical texture interference and chemical corrosion characteristics of composite board surfaces in real time, this invention provides a method and system for evaluating the corrosion resistance of composite boards based on spectral analysis.

[0004] In a first aspect, the present invention provides a method for evaluating the corrosion resistance of composite plates based on spectral analysis, employing the following technical solution: A multispectral imaging unit is configured to simultaneously acquire spectral response data of each detection point on the surface of the test plate in the first reference band, second reference band, and characteristic band, and perform preprocessing to obtain reflectance data of the detection point in the first reference band, second reference band, and characteristic band. The nonlinear attenuation factor of the detection point is calculated using the logarithmic difference ratio between the reflectance data of the detection point in the first reference band and the second reference band. Based on a preset power-law model, the theoretical reference reflectance of the detection point in the characteristic band, under conditions of only physical texture interference, is constructed using the nonlinear attenuation factor and the reflectance data of the detection point in the first reference band. The logarithmic residual between the reflectance data of the detection point in the characteristic band and the theoretical reference reflectance is calculated. Based on the logarithmic residual, the corrosion resistance of the test plate is evaluated using a preset threshold.

[0005] This invention constructs a theoretical reference reflectance of characteristic bands under pure physical texture interference by collecting multi-band data and calculating the nonlinear attenuation factor using the logarithmic difference ratio. Then, it calculates the logarithmic residual, enabling the system to accurately distinguish physical texture interference and retain only the spectral absorption characteristics caused by changes in chemical composition. This effectively solves the problem that traditional optical detection cannot distinguish between physical defects and chemical defects and has a high false alarm rate.

[0006] Furthermore, the first reference band and the second reference band are selected from the spectral range in which the passivation layer on the metal surface exhibits optical transmission characteristics; the characteristic band is selected from the spectral range in which the passivation layer on the metal surface exhibits optical absorption characteristics.

[0007] This invention establishes the basis for signal separation from a physical perspective by defining the reference band as the optical transmission range of the passivation layer and the characteristic band as the optical absorption range. This ensures that subsequent calculations have clear physical meaning rather than being mere data fitting, thereby improving the reliability of the evaluation results.

[0008] Furthermore, the preprocessing includes: acquiring dark background noise data and standard white board response data of the multispectral imaging unit, performing radiometric correction on the spectral response data using the dark background noise data and the standard white board response data, and converting the spectral response data into the reflectance data.

[0009] This invention introduces a radiometric correction preprocessing based on a dark background and a standard white board, which converts the original spectral response data into reflectance data. This process eliminates the dark background noise of the multispectral imaging unit, the illumination inhomogeneity between the center and edge of the field of view, and the differences in the spectral response of the sensor in different bands, ensuring the purity and authenticity of the results and preventing hardware system errors from being mistakenly amplified into defective signals by the algorithm.

[0010] Furthermore, the nonlinear attenuation factor satisfies the following relationship:

[0011] in, This refers to the nonlinear attenuation factor; The reflectance data for the first reference band; The reflectance data is for the second reference band; The center wavelength value of the first reference band; This is the center wavelength value of the second reference band; It is a very small positive number.

[0012] Furthermore, the theoretical reference reflectivity satisfies the following relationship:

[0013] in, The theoretical reference reflectivity; The reflectance data is for the first reference band; The center wavelength value of the characteristic band; The center wavelength value of the first reference band; is the nonlinear attenuation factor.

[0014] This invention uses a power-law model to construct a theoretical benchmark reflectance. Since the micro-rough texture of the metal surface scatters light in a manner similar to Mie scattering or Rayleigh scattering, it usually exhibits a non-linear power-law decay law with wavelength. This model can more accurately describe the trend of light intensity variation in different wavelength bands and improve the ability to detect weak defects.

[0015] Furthermore, the logarithmic residuals satisfy the following relationship:

[0016] in, The logarithmic residual; The reflectance data for the characteristic band; The theoretical reference reflectivity; It is a very small positive number.

[0017] Furthermore, the preset threshold is obtained through a physicochemical signal decoupling calibration experiment.

[0018] Furthermore, the evaluation of the corrosion resistance of the test plate using a preset threshold specifically includes: traversing each test point, comparing the logarithmic residual with the preset threshold; when the absolute value of the logarithmic residual is less than or equal to the preset threshold, marking the test point as a normal point; when the absolute value of the logarithmic residual is greater than the preset threshold, marking the test point as a suspected defect point; performing connected component analysis on all suspected defect points, counting the number of suspected defect points contained in each independent connected component and converting it into a physical defect area; when the physical defect area is greater than a preset defect tolerance value, determining that the corrosion resistance of the test plate is unqualified.

[0019] This invention achieves a leap from pixel-level numerical calculation to industrial-grade quality assessment through connected component analysis and physical defect area conversion mechanism. It not only effectively filters out isolated noise points, but also combines defect tolerance values ​​to achieve quantitative and standardized evaluation of corrosion resistance performance, avoiding excessive rejection due to extremely small defects.

[0020] Furthermore, the defect tolerance value is set based on the resolution of the acquisition device and international composite board standards.

[0021] Secondly, the present invention provides a composite plate corrosion resistance evaluation system based on spectral analysis, which adopts the following technical solution: A composite plate corrosion resistance evaluation system based on spectral analysis includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned composite plate corrosion resistance evaluation method based on spectral analysis is implemented. By adopting the above technical solution, a computer program for evaluating the corrosion resistance of composite plates based on spectral analysis is generated and stored in a container for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0022] The present invention has the following technical effects: This invention eliminates the need to collect massive defect samples to train deep learning models. Instead, it directly models based on spectral physical properties, making it adaptable to composite panels of different batches and textures. Furthermore, it successfully solves the problem of strong straightening of texture interference without damaging the surface of the panels, reducing the false alarm rate on the production line and providing reliable technical support for the automated quality control of high-end metal composite panels. This invention combines materials science mechanisms and utilizes the difference between the high transmittance of the passivation layer in the characteristic band and the strong light absorption of iron-based contaminants to directly pinpoint the logarithmic residual calculated by the algorithm as the two root causes of corrosion resistance failure: passivation layer damage or foreign metal contamination. This makes the evaluation results no longer vague numerical anomalies, but quality judgments with clear electrochemical significance. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for evaluating the corrosion resistance of composite plates based on spectral analysis, provided in an embodiment of the present invention. Figure 2 This is a comparison diagram of the effects of the prior art provided in the embodiments of the present invention and the present invention. Detailed Implementation

[0024] Before describing the specific embodiments of the present invention in detail, it should be noted that the chemical composition characteristics or chemical composition changes mentioned in this embodiment do not refer to any changes in any chemical element, but specifically to the surface passivation layer state and surface contamination characteristics that directly determine the corrosion resistance of the metal composite plate. Specifically, the corrosion resistance of titanium steel or stainless steel composite plates mainly comes from a dense, continuous oxide passivation layer, such as titanium dioxide or chromium oxide, covering its surface. Physical optical experiments show that, under the condition of an intact passivation layer, the film exhibits low light absorption or specific transmission characteristics in the selected characteristic wavelength bands, and its reflectivity is mainly affected by the physical texture characteristics of the substrate. However, when corrosion resistance defects appear on the surface of the sheet material, they are usually accompanied by the following two specific microscopic chemical changes: Physical damage and depletion of the passivation layer leads to direct exposure of the active metal in the substrate, disrupting the electrochemical balance of the surface and altering the negative refractive index and spectral absorption rate of the surface. Foreign metal oxide contamination: During the high-pressure leveling process, if a small amount of iron filings on the surface of the roll are pressed into the surface of the active metal, they will be rapidly oxidized to form iron oxides, such as Fe2O3 and Fe3O4. These oxides have obvious characteristic absorption peaks in specific wavelengths and have strong light absorption properties. Iron contamination is the main cause of pitting corrosion and galvanic corrosion. Therefore, under the specific controlled process environment of plate leveling and shipment, this invention directly maps the changes in chemical composition in the spectral dimension to the corrosion resistance performance under industrial standards.

[0025] This invention provides a method for evaluating the corrosion resistance of composite plates based on spectral analysis, referring to... Figure 1 This includes steps S1-S4: S1: Data Acquisition and Preprocessing.

[0026] Specifically, by configuring a multispectral imaging unit, the near-infrared dual reference band, which is sensitive to physical texture, and the visible light characteristic band, which is sensitive to chemical composition, are selected to simultaneously acquire spectral response data at each detection point on the surface of the test board. Subsequently, the spectral response data are radiometrically corrected and normalized using pre-acquired dark background noise data and standard white board response data to eliminate interference from sensor noise and uneven illumination, and finally output reflectance data that can characterize the true physicochemical properties of the test board surface.

[0027] First, hardware configuration and band selection are carried out. In this embodiment, a multispectral imaging unit is installed at the entrance and exit of the composite board leveling production line, perpendicular to the movement direction of the board to be tested. The multispectral imaging unit adopts a three-channel narrowband filter industrial camera, combined with a highly uniform diffused dome LED light source to eliminate the high light overflow caused by specular reflection on the surface of the board to be tested. Next, configure three detection channels as follows: The first reference band has a center wavelength of 980nm, which is located in the near-infrared region. The oxide passivation layer on the surface of the test material exhibits high optical transmission characteristics in this band, allowing light to penetrate the film layer and reach the substrate directly. Therefore, it mainly reflects the physical texture characteristics of the test material, which include all surface features that cause non-chemical light scattering, such as roller marks, roughness changes, and micro-morphological undulations. The second reference band, with a center wavelength of 850nm, is also located in the transmission range of the passivation layer. It is used in conjunction with the first reference band to analyze the nonlinear scattering characteristics of the physical texture through the difference between the two bands. The characteristic band, with a center wavelength of 450nm, is located in the short-wave region of visible light. Iron oxides and organic pollutants have strong optical absorption characteristics in this band, while the absorption rate of a well-preserved passivation layer is low in this band.

[0028] It should be noted that the selection of the above three specific detection bands is based on the energy band structure of the passivation film on the metal surface and the physical mechanism of light-matter interaction, as detailed below: Regarding the selection of the first and second reference bands, since the dense oxide passivation layer covering the surface of the titanium steel or stainless steel composite plate is a wide bandgap semiconductor material with a bandgap greater than 3 eV, according to the semiconductor optoelectronic theory, when the photon energy of the incident light is less than the bandgap of the material, the photon cannot excite electron transitions, and the material exhibits optical transparency. 980nm and 850nm are located in the near-infrared spectral region, and their photon energy is about 1.2 eV, which is much lower than the bandgap of the passivation layer. Therefore, in these two bands, the passivation layer on the surface of the plate under test is almost transparent to the incident light. The reflection signal collected by the multispectral imaging unit mainly comes from the reflection that occurs on the substrate after the light penetrates the passivation layer. This means that the reflectivity changes in these two bands are determined only by the physical texture of the surface of the plate under test and are not affected by the surface chemical composition. In this embodiment, 980nm and 850nm are preferred as the center wavelengths, which are the standard response bands for industrial-grade photoelectric sensors. The corresponding narrowband filters and high-power LED light sources have the highest quantum efficiency and luminous stability. At the same time, the transmittance curve of the passivation layer is relatively flat in the 800nm-1000nm range. The above two bands can fully represent the physical optical characteristics of this range. Compared with the adjacent non-standard response bands, using 980nm and 850nm can obtain data with a higher signal-to-noise ratio. In addition, two reference bands are selected instead of one in order to calculate the attenuation slope of light intensity as a function of wavelength. Since the scattering effect of the micro-texture of the metal surface on light is similar to Mie scattering, it exhibits a non-linear change with wavelength. Single-point measurement cannot fit this scattering curve, and it must be calculated by logarithmic difference of two bands. Regarding the selection of characteristic wavelength bands, due to the failure of corrosion resistance on the surface of titanium steel or stainless steel, such as when foreign metal contamination or matrix corrosion occurs, the generated iron oxides exhibit a sharp decrease in reflectivity at 400nm-500nm. In contrast, in the long wavelength region, such as 650nm or infrared light, the absorption rate of iron oxides is significantly reduced, and the reflectivity is increased, making it difficult to form an effective contrast with the matrix background. Furthermore, within the 400nm-500nm range, 450nm is currently the most mature standard wavelength band for industrial-grade blue LED technology, with the highest radiant power density and the smallest emission wavelength temperature drift. At the same time, mainstream silicon-based CMOS or CCD image sensors have a quantum efficiency exceeding 60% at 450nm. Compared with other wavelength bands, selecting 450nm can obtain the best signal-to-noise ratio data while ensuring detection sensitivity, meeting the requirements of long-term online detection in industrial settings.

[0029] To ensure that the data collected on the high-speed production line are spatially identical across the three different frequency bands, the following acquisition strategy is adopted: A photoelectric encoder is installed on the conveyor roller of the composite board leveling production line. When the photoelectric encoder detects that the board to be tested has moved to the center of the field of view of the multispectral imaging unit, it sends a synchronous trigger signal to the multispectral imaging unit. When the synchronous trigger signal is received, the internal photosensitive chip is controlled to synchronously collect the spectral response data of each detection point on the surface of the board to be tested, that is, each pixel point, in the above three bands. The spectral response data is stored in the form of a two-dimensional digital matrix. Each element value in the matrix represents the photoelectric conversion intensity value of the detection point in a specific band. The spectral response data collected above has not yet undergone physical dimension conversion. Its magnitude depends not only on the reflectivity of the surface of the test plate, but also on the ambient light intensity, the difference in light transmission between the center and edge of the lens, and the difference in quantum efficiency of the sensor at different wavelengths. In order to obtain the true physical reflectivity, this embodiment uses the black-and-white plate calibration method to perform radiometric correction and normalization on the spectral response data, as follows: It should be noted that, to ensure the mathematical rigor of subsequent calculations, this embodiment defines the spectral response data acquired by the multispectral imaging unit as a two-dimensional numerical matrix. Each element in the matrix is ​​defined as an independent data pixel, and each data pixel physically uniquely corresponds to a tiny area on the surface of the material to be tested, i.e., a detection point. To cover a sufficiently wide detection field of view, the spatial resolution of the multispectral imaging unit is set to 0.1 mm / pixel. The number of rows and columns of this matrix is ​​related to the resolution of the multispectral imaging unit. Since this embodiment uses a high-resolution industrial camera, the matrix contains 2048×2048 data elements. Since the data acquisition of the first reference band, the second reference band, and the characteristic band is carried out simultaneously, each detection point has three sets of independent data. In the preprocessing stage, the 2048×2048 matrices corresponding to these three bands will be processed separately.

[0030] To acquire dark background noise data, in a completely dark environment with the light source turned off and the camera covered, the multispectral imaging unit collects data. The spectral response data obtained at this time comes from the thermal noise and circuit noise inside the sensor, and is denoted as dark background noise data, which is a 2048×2048 matrix. To acquire standard whiteboard response data, with the measurement light source turned on and the lighting conditions consistent with the actual test, a metrologically certified standard diffuse reflection whiteboard is placed on the plane where the test board is located. In this embodiment, polytetrafluoroethylene (PTFE) is selected as the data acquisition object, and a multispectral imaging unit is used to acquire data from it to obtain standard whiteboard response data, which characterizes the maximum light intensity response distribution under the current light field. Since the energy distribution of the light source is different in different wavelength bands, it is necessary to acquire standard whiteboard response data corresponding to the first reference band, the second reference band, and the characteristic band, all of which are 2048×2048 matrices.

[0031] According to the empirical linear method and flat-field correction described in "Reflectance and radiance-based methods for the in-flight absolute radiometric calibration of multispectral sensors", since the data output by the photoelectric sensor includes the target's true reflected radiation, sensor dark current noise, and optical gain, in order to convert the electrical signal into physical reflectivity, it is necessary to first calculate the ratio of the target's electrical signal to the reference electrical signal, and then multiply it by the known physical reflectivity of the reference target. Based on this, the processor in this embodiment performs the same radiometric correction operation three times on the matrices of the first reference band, the second reference band, and the characteristic band. Taking any band as an example, for the coordinates in the matrix... For each detection point, its reflectance data is calculated, and the specific relationship is as follows:

[0032] in, This is reflectivity data; For spectral response data; Data with a dark background and high noise levels; Standard whiteboard response data; This is a mapping factor used to map the relative ratio to the actual physical reflectance value. Since this embodiment selects polytetrafluoroethylene (PTFE) as the data acquisition object when obtaining standard whiteboard response data, based on the optical properties of this material, it is set... ; This is the row index for the direction of movement of the material to be tested. This is the column index in the width direction of the material to be tested.

[0033] Finally, the reflectance data of each detection point on the surface of the test plate were obtained under the first reference band, the second reference band, and the characteristic band, denoted as . , , Since subsequent calculations are performed separately for each detection point and are not affected by different detection points, the reflectance data is abbreviated as... , , .

[0034] It should be noted that the above-mentioned specific implementation details are only a preferred embodiment of the present invention and do not constitute a limitation on the scope of protection of the present invention. In practical applications, the multispectral imaging unit is not limited to a three-channel narrowband filter industrial camera. Hyperspectral cameras, liquid crystal tunable filter imaging systems, acousto-optic tunable filter imaging systems, and combinations of multiple single-band cameras can also be used. As long as the spectral image data of the surface of the test board under a specific band can be acquired simultaneously, they all fall within the scope of protection of the present invention. In addition, the acquisition object selected for acquiring standard whiteboard response data is not limited to polytetrafluoroethylene. Optical standard materials such as barium sulfate and diffuse reflection ceramics can also be selected.

[0035] S2: Calculation of nonlinear attenuation factor.

[0036] Specifically, a general mathematical model describing the scattering law of texture is established based on the principle of physical optics. The reflectivity data of the first reference band and the second reference band are substituted into the model to establish a system of simultaneous equations. Irrelevant variables are eliminated through mathematical transformation, and the nonlinear attenuation factor characterizing the physical texture features of each detection point is calculated.

[0037] Based on the band selection criteria in S1, since both the first and second reference bands are located in the optical transmission region of the passivation layer, light can directly reach the substrate. Therefore, the reflectivity data changes of these two bands are only affected by the physical texture characteristics of the surface of the test material and are unrelated to the surface chemical composition. Physical optics studies show that the light scattering intensity of such micro-rough surfaces changes with wavelength according to a power-law decay, similar to the Mie scattering mechanism, and is suitable for situations where the surface roughness is much smaller than or close to the wavelength. Based on this, this embodiment constructs the following light intensity wavelength mapping model:

[0038] in, Representative wavelength Reflectance data at the location; This is a geometrical proportionality constant, related to the optical path distance of the detection point, the basic reflectivity, and the local micro-facet angle. For the same fixed detection point, it varies across different wavelengths. It can be considered a constant; It is a non-linear decay factor.

[0039] According to the paper "On the Atmospheric Transmission of Sun Radiation and on Dust in the Air," the Oombström exponent can be used to solve for the scattering characteristics of microscopic rough surfaces. This involves measuring reflectivity at two different wavelengths, taking the logarithmic difference, and then dividing by the logarithmic difference between the two wavelengths. Based on this, to calculate the nonlinear attenuation factor in the above model, the processor substitutes the reflectivity data of each detection point in the first and second reference bands into the power-law model, takes the logarithm of both sides of the model, and transforms the nonlinear power series relationship into a linear additive relationship, as follows:

[0040] in, The center wavelength value of the first reference band is 980. The center wavelength of the second reference band is 850; subtracting the following formula from the above formula yields:

[0041] This leads to the nonlinear attenuation factor. The relation is:

[0042] in, To ensure the logarithmic term is a very small positive number and to prevent it from being zero, this embodiment uses 10. -6 .

[0043] S3: Construct the theoretical reference reflectance and calculate the logarithmic residual.

[0044] Specifically, using the nonlinear attenuation factor obtained from S2 and combined with a preset power-law model, the physical texture features of the near-infrared band are transformed into visible light feature bands to construct a theoretical reference reflectance containing only physical scattering information. Then, the logarithmic residual is obtained by calculating the difference between the measured value and the theoretical value, thus achieving complete decoupling of physical and chemical signals.

[0045] According to the physical scaling mapping rule described in "Light Scattering by Small Particles", when the optical scattering response at a certain wavelength, such as reflectivity, is known, to solve for the response at another wavelength, the known optical scattering response can be multiplied by the k-th power of the ratio of these two wavelengths. Since the reflection signal in the characteristic band is actually the result of the combined effect of physical texture and chemical corrosion, in order to separate these two features, the processor first assumes that the detection point has not been subjected to any chemical corrosion or contamination, and only physical texture features exist. Based on the power-law model defined in S2, the following simultaneous equations can be obtained:

[0046] in, The theoretical reference reflectivity; The center wavelength of the characteristic band is 450; dividing the two equations and rearranging the terms, we get:

[0047] It should be noted that although this embodiment calculates the theoretical reference reflectance based on the first reference band, mathematically, using data from the second reference band is completely equivalent. This is because the nonlinear attenuation factor is obtained by fitting data from both bands. If the second reference band is used, the relationship becomes:

[0048] The two formulas above regarding the theoretical reference reflectance are completely consistent in mathematical principle. In this embodiment, the first reference band is preferably used as the calculation reference, and the results are obtained. .

[0049] The processor reads reflectance data from the characteristic bands. Based on the theoretical reference reflectance and the spectral absorption principle of Beer-Lambert law, changes in chemical composition mainly manifest as exponential attenuation in the optical path. Therefore, this embodiment uses logarithmic difference to extract the residual signal, with the specific relationship as follows:

[0050] in, The logarithmic residuals; To ensure the logarithmic term is a very small positive number and to prevent it from being zero, this embodiment uses 10. -6 .

[0051] The processor iterates through each detection point, calculates the logarithmic residual for each point, and obtains the logarithmic residual matrix, denoted as... Each data point in the matrix represents the corrosion resistance of the surface of the plate being tested.

[0052] S4: Corrosion resistance assessment.

[0053] Specifically, based on the logarithmic residual matrix obtained in S3, the continuous numerical signal is transformed into discrete defect judgment results through statistical methods, and then the final corrosion resistance performance evaluation is completed according to industrial standards.

[0054] Each data point in the logarithmic residual matrix is ​​compared with a preset threshold. If the absolute value of the data is less than or equal to the preset threshold, the detection point corresponding to the data is marked as a normal point and recorded as 0. If the absolute value of the data is greater than the preset threshold, the detection point corresponding to the data is marked as a suspected defect point and recorded as 1. Finally, a binary matrix marking all suspected defect locations is obtained.

[0055] It should be noted that the preset thresholds mentioned above were obtained through a physicochemical signal decoupling calibration experiment, the specific steps of which are as follows: Twenty qualified plates, each 200mm x 200mm in size, made of the same material as the plate to be tested, with an intact surface passivation layer and no corrosion or iron contamination, were selected as experimental samples and divided into two groups. They were subjected to different degrees of high-intensity roller pressing and leveling treatment. Group A, physical texture interference samples, 10 experimental samples were selected, of which 5 retained normal leveling roller marks, and the other 5 were manually sanded with sandpaper in a directional manner. After metallographic microscopy, it was confirmed that the surface of the Group A samples had no oxidation corrosion or foreign metal contamination. Group B, chemical corrosion defect samples: 10 experimental samples were selected and the drop corrosion method was used. A 5% FeCl3 solution was dripped onto the surface of the experimental samples at fixed points. The reaction time was controlled at 10 seconds, 20 seconds, 30 seconds, 40 seconds, 50 seconds, 60 seconds, 70 seconds, 80 seconds, 90 seconds, and 100 seconds, forming corrosion spots of different diameters and color intensities on the surface of the experimental samples. The 20 experimental samples from groups A and B are placed sequentially into the acquisition field of view of the multispectral imaging unit, and the operation is carried out according to the process of steps S1-S3 to obtain the logarithmic residual matrix. For group A, the absolute values ​​of the logarithmic residuals for all detection points were calculated. Experimental data showed that, due to the effective suppression of physical texture by the algorithm model, the absolute values ​​of the logarithmic residuals for the experimental samples in group A were generally concentrated in the low range of 0 to 0.15, exhibiting a Gaussian white noise distribution. According to statistical principles, the maximum logarithmic residual value at the 99.7% confidence interval in group A was recorded and denoted as [value missing]. The measurements in this embodiment are as follows ; For group B, detection points in the corrosion defect area were extracted, and the absolute values ​​of their logarithmic residuals were calculated. Experimental data showed that the values ​​were generally distributed in the high range of 0.5 to 2.5. The minimum logarithmic residual value of the weakest corrosion point was recorded and denoted as . The measurements in this embodiment are as follows ; The safety margin method is used to determine the preset threshold, and the safety factor is set to 0.5 to balance the risks of missed detections and false alarms. The specific relationship is as follows:

[0056] in, The preset threshold can be calculated to obtain ; For safety factors.

[0057] Connectivity analysis is performed on the binarized matrix. In this embodiment, the eight-neighbor connectivity algorithm is used to scan the binarized matrix, obtaining several independent connected components. The number of suspected defect points contained in each independent connected component is counted, denoted as . ,in, An index for an independent connected component; Next, defect tolerance values ​​are set, and the corrosion resistance of the test plate is evaluated: In machine vision inspection, to prevent random thermal noise on the photosensitive chip or dust particles in the air from being misidentified as defects, a valid defect must contain at least a 4×4 pixel connected region. Combined with the resolution of the multispectral imaging unit, the minimum reliable detection area is 0.04 mm². 2 This means that the defect tolerance value must be greater than 0.04, otherwise the system will not be able to distinguish between real minute corrosion and random photoelectric noise; Referring to GB / T 8547-2019 and NB / T 47002.3-2019 standards, for Class I or Class II composite panels, isolated point defects with a diameter less than 0.5 mm are allowed to exist without affecting corrosion resistance. Converting the specified negligible upper limit of 0.5 mm into area yields 0.196 mm². 2 After rounding, it is 0.2mm. 2 ; Combining the above two points, this embodiment sets the defect tolerance value to 0.2, which is much larger than the minimum reliable detection area and is benchmarked against the industry standard baseline, thus balancing detection sensitivity and production yield.

[0058] The number of suspected defects Multiplying the area by the square of the spatial resolution of the multispectral acquisition unit yields the area of ​​the physical defect, denoted as . ; Will Compare with the defect tolerance value; if there is a defect... If a connected region exceeds the defect tolerance value, the corrosion resistance of the tested material is deemed unqualified; if no such region exists... If the connected regions are greater than the defect tolerance value, the corrosion resistance of the tested material is deemed to be qualified.

[0059] Figure 2The figure shows a comparison between the prior art and the present invention provided in the embodiments of the present invention. As shown in the figure, the prior art is affected by the strong leveling process of the composite board, which shows a violent mid-term fluctuation, causing false alarms in the system. However, the present invention can eliminate false alarms caused by physical texture interference by introducing dual reference bands to construct a theoretical benchmark and performing logarithmic difference operation, thus achieving high signal-to-noise ratio detection of the corrosion resistance performance of the composite board.

Claims

1. A method for evaluating the corrosion resistance of composite plates based on spectral analysis, characterized in that, include: A multispectral imaging unit is configured to synchronously acquire spectral response data of each detection point on the surface of the plate under test in the first reference band, the second reference band, and the characteristic band, and perform preprocessing to obtain the reflectance data of the detection point in the first reference band, the second reference band, and the characteristic band. The nonlinear attenuation factor of the detection point is calculated by using the ratio of the logarithmic difference between the reflectance data of the detection point in the first reference band and the second reference band. Based on a preset power-law model, the nonlinear attenuation factor and the reflectance data of the detection point in the first reference band are used to construct the theoretical reference reflectance of the detection point in the characteristic band under the condition of only physical texture interference. The logarithmic residual between the reflectance data of the detection point in the characteristic band and the theoretical reference reflectance is calculated. Based on the logarithmic residual, the corrosion resistance of the test plate is evaluated using a preset threshold.

2. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The first reference band and the second reference band are selected from the spectral range in which the passivation layer on the metal surface exhibits optical transmission characteristics; the characteristic band is selected from the spectral range in which the passivation layer on the metal surface exhibits optical absorption characteristics.

3. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The preprocessing includes: acquiring dark background noise data and standard white board response data of the multispectral imaging unit; performing radiometric correction on the spectral response data using the dark background noise data and the standard white board response data; and converting the spectral response data into reflectance data.

4. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The nonlinear attenuation factor satisfies the following relationship: in, This refers to the nonlinear attenuation factor; The reflectance data is for the first reference band; The reflectance data is for the second reference band; The center wavelength value of the first reference band; This is the center wavelength value of the second reference band; It is a very small positive number.

5. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The theoretical reference reflectivity satisfies the following relationship: in, The theoretical reference reflectivity; The reflectance data is for the first reference band; The center wavelength value of the characteristic band; The center wavelength value of the first reference band; is the nonlinear attenuation factor.

6. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The logarithmic residuals satisfy the following relationship: in, The logarithmic residual; The reflectance data for the characteristic band; The theoretical reference reflectivity; It is a very small positive number.

7. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The preset threshold was obtained through a physicochemical signal decoupling calibration experiment.

8. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 1, characterized in that, The evaluation of the corrosion resistance of the test material using a preset threshold specifically includes: traversing each test point, comparing the logarithmic residual with the preset threshold; marking the test point as a normal point when the absolute value of the logarithmic residual is less than or equal to the preset threshold; marking the test point as a suspected defect point when the absolute value of the logarithmic residual is greater than the preset threshold; performing connected component analysis on all suspected defect points, counting the number of suspected defect points contained in each independent connected component and converting it into physical defect area; and determining that the corrosion resistance of the test material is unqualified when the physical defect area is greater than a preset defect tolerance value.

9. The method for evaluating the corrosion resistance of composite plates based on spectral analysis according to claim 8, characterized in that, The defect tolerance value is set based on the resolution of the acquisition equipment and international composite panel standards.

10. A system for evaluating the corrosion resistance of composite plates based on spectral analysis, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a method for evaluating the corrosion resistance of composite plates based on spectral analysis according to any one of claims 1-9.

Images