Fireproof coating spraying quality monitoring method and system

Abstract

The application discloses a fireproof coating spraying quality monitoring method and system, and belongs to the field of coating quality detection.The method comprises the following steps: emitting coherent laser to a fireproof coating wet film on a target base material, collecting a dynamic laser speckle image sequence; determining a speckle particle reference size, and adaptively dividing each frame of image in the dynamic laser speckle image sequence into a plurality of non-overlapping pixel sub-regions according to the speckle particle reference size; extracting a speckle decorrelation time, generating global decorrelation time field data covering the fireproof coating wet film; calculating the spatial gradient of the global decorrelation time field data to obtain a gradient amplitude, and when the gradient amplitude is greater than a preset fault judgment threshold, it is determined that an abnormality exists at a corresponding spatial position and an abnormality mark is triggered.The application realizes zero delay, interpretable defect classification early warning, and can cover various typical microscopic defect types in fireproof coating construction.

Description

Technical Field

[0001] This invention relates to the field of coating quality inspection, and in particular to a method and system for monitoring the quality of fire-retardant coating spraying. Background Technology

[0002] Fire-retardant coatings are a type of functional coating applied to the surface of building components or steel structure substrates. Under fire conditions, they can expand and foam to form a heat-insulating carbonized layer to delay the temperature rise of the structure. The quality of fire-retardant coating application directly affects whether the coating can effectively perform its heat insulation and protection functions under fire conditions. Therefore, quality control of the construction process is of great engineering significance.

[0003] Currently, the quality inspection of fire-retardant coatings mainly relies on offline testing methods after the coating has fully cured. Common methods include dry film thickness gauges, pull-out tests for bond strength, and visual inspection. Dry film thickness gauges measure the coating thickness point-by-point using magnetic or eddy current probes after complete curing. This method only obtains thickness information at discrete sampling points and cannot reflect the uniformity of the coating's internal microstructure and rheological state. The pull-out test for bond strength requires attaching test spools to the cured coating and applying tension until the coating peels off; it is a destructive test and cannot be applied on a large scale and in real-time at the construction site. Visual inspection can only detect obvious defects that have already developed at the macroscopic level, such as large-area sagging, cracking, and blistering, but it is completely ineffective for latent defects still in their microscopic infancy.

[0004] In recent years, dynamic laser speckle imaging technology has gained some application in the biomedical field (such as blood perfusion monitoring). This technology utilizes the physical principle that the speckle pattern generated after coherent laser irradiation of a sample surface dynamically evolves with the movement of internal scattering particles, and infers the microscopic motion information inside the sample by analyzing the temporal correlation of the speckle pattern. However, applying this technology to the novel industrial application scenario of monitoring the quality of fire-retardant coatings faces completely different technical challenges compared to the biomedical field, including but not limited to: the mechanical vibration interference in industrial environments is much stronger than in medical laboratory environments; the scattering characteristics of fire-retardant coating wet films are fundamentally different from those of biological tissues; and the identification of coating defects requires the establishment of a completely new judgment logic based on field theory analysis. Currently, there is no systematic technical solution for applying dynamic laser speckle technology to the real-time monitoring of the micro-rheological state of fire-retardant coating wet films and the early classification and warning of defects. Summary of the Invention

[0005] One of the objectives of this invention is to provide a method and system for monitoring the quality of fire-retardant coating spraying, in order to solve the problems in the prior art that it is impossible to perceive the internal micro-rheological state of the coating in real time when it is in a wet film state, the detection timing is seriously delayed after the defect formation process, there is a lack of automatic classification and early warning capability for different types of coating defects, and it is difficult to adapt to the inspection needs of complex curved workpieces.

[0006] This invention is achieved through the following technical solution: a method for monitoring the quality of fire-retardant coating spraying, comprising the following steps: emitting coherent laser light onto a wet film of fire-retardant coating on a target substrate using a laser emitting device, and continuously acquiring a dynamic laser speckle image sequence formed by the interference of scattered light from the wet film of the fire-retardant coating using an image acquisition device; selecting the first frame speckle image in the dynamic laser speckle image sequence based on the light intensity of the speckle images in the sequence, and calculating a normalized two-dimensional spatial autocorrelation function based on the light intensity value of each pixel in the first frame speckle image; setting the side length of the pixel sub-region to a preset multiple of the speckle pattern reference size, and then adjusting the speckle pattern according to the side length of the pixel sub-region. Each frame in the dynamic laser speckle image sequence is adaptively divided into multiple non-overlapping pixel sub-regions. The minimum radial offset distance, where the autocorrelation function value is no greater than half of the peak value, is determined as the reference size of the speckle pattern. For each pixel sub-region, the speckle decorrelation time is extracted, and the speckle decorrelation times of all pixel sub-regions are stitched together according to their corresponding spatial positions to generate global decorrelation time field data covering the wet film of the fireproof coating. The spatial gradient of the global decorrelation time field data is calculated to obtain the gradient magnitude. When the gradient magnitude is greater than a preset tomography threshold, an anomaly is determined to exist at the corresponding spatial position, and an anomaly marker is triggered.

[0007] Further, determining the spatial scale corresponding to the decay of the main lobe of the normalized two-dimensional spatial autocorrelation function to half of its peak value includes: taking the null offset position of the normalized two-dimensional spatial autocorrelation function as the center, determining the minimum radial offset distance that ensures the autocorrelation function values ​​in multiple preset orientation directions are not greater than half of the peak value.

[0008] Further, determining the speckle particle reference diameter based on the main lobe width of the spatial autocorrelation function specifically includes: performing an ergodic analysis of the spatial autocorrelation function along each azimuth direction; determining the minimum radial distance corresponding to when the value of the spatial autocorrelation function decays from the peak value to less than half of the peak value in all azimuth directions; and using the minimum radial distance as the speckle particle reference diameter.

[0009] Furthermore, the preset multiplier is determined by a multiplier coefficient, the value of which ranges from 5 to 10.

[0010] Furthermore, before extracting the speckle decorrelation time for each pixel sub-region, a step of performing rigid body displacement compensation preprocessing on the dynamic laser speckle image sequence is included to eliminate signal contamination caused by environmental vibration.

[0011] Further, the rigid body displacement compensation preprocessing step includes: setting a stationary reference point in the edge region of the substrate without fire-retardant coating; acquiring a speckle image that simultaneously includes the wet film surface of the fire-retardant coating and the stationary reference point; calculating the two-dimensional cross-correlation function of the stationary reference point region image between adjacent image frames in the dynamic laser speckle image sequence; extracting the offset of the peak coordinates of the two-dimensional cross-correlation function relative to the center as a rigid body displacement vector; and performing reverse subpixel translation compensation on each frame image of the wet film surface of the fire-retardant coating according to the rigid body displacement vector.

[0012] Furthermore, the two-dimensional cross-correlation function is obtained by performing two-dimensional Fourier transforms on adjacent frame images in the frequency domain, taking the complex conjugate of the Fourier transform result of the current frame image, multiplying it element-by-element with the Fourier transform result of the next frame image, and then performing an inverse Fourier transform.

[0013] Furthermore, the reverse subpixel translation compensation is achieved using bicubic interpolation; wherein, the offset is obtained by performing two-dimensional surface fitting on discrete sampling points near the peak of the two-dimensional cross-correlation function to achieve subpixel accuracy.

[0014] Further, the speckle decorrelation time is extracted, including: extracting the intensity time series of each pixel within the pixel sub-region in multiple consecutive frames of images; calculating the time mean of the intensity time series and calculating the deviation of the light intensity value at each moment from the time mean; calculating a normalized second-order time autocorrelation function based on the intensity time series, wherein the normalized second-order time autocorrelation function describes the statistical similarity between light intensity signals at the same spatial location under different time delays; and extracting the speckle decorrelation time based on the normalized second-order time autocorrelation function.

[0015] Further, the speckle decorrelation time is extracted based on the normalized second-order time autocorrelation function, specifically including: constructing a preset exponential decay model, wherein the exponential decay model includes a coherence factor and a characteristic decay time as fitting parameters; performing nonlinear least squares fitting between the normalized second-order time autocorrelation function and the exponential decay model to minimize the sum of squared residuals between the output of the exponential decay model and the measured data of the normalized second-order time autocorrelation function; and extracting the fitted characteristic decay time as the speckle decorrelation time.

[0016] Furthermore, after obtaining the global decorrelation time field data and before determining the gradient magnitude based on the global decorrelation time field data, a surface curvature correction step is included to eliminate gradient calculation deviations caused by workpiece surface curvature projection deformation. The surface curvature correction step includes: acquiring pre-stored three-dimensional point cloud data and constructing a normal vector field of the workpiece surface based on the three-dimensional point cloud data; calculating the projection angle between the normal vector at each position in the normal vector field and the optical axis of the image acquisition device; constructing a Jacobian coordinate transformation matrix based on the projection angle, the Jacobian coordinate transformation matrix being used to describe the linear transformation relationship between a small displacement on the physical surface of the workpiece and the corresponding projected displacement on the image projection plane; multiplying the gradient value calculated on the image projection plane with the inverse of the Jacobian coordinate transformation matrix to restore the true tangential gradient of the physical surface of the workpiece.

[0017] Further, constructing the Jacobian coordinate transformation matrix based on the projection angle specifically includes: in a locally orthogonal coordinate system, constructing the Jacobian coordinate transformation matrix as a diagonal matrix, wherein the diagonal elements in the locally orthogonal coordinate system are the cosine values ​​of the projection angle, and the diagonal elements along the direction perpendicular to the tilt are 1; wherein multiplying the gradient value calculated on the image projection plane with the inverse matrix of the Jacobian coordinate transformation matrix includes: dividing the image plane gradient component along the surface tilt direction by the cosine value of the projection angle to restore the true physical gradient in that direction.

[0018] Furthermore, after triggering the anomaly marker, the method further includes the following steps: classifying the defect type of the region corresponding to the anomaly marker: performing spatial gradient calculation on the global decorrelation time field data to obtain the gradient direction vector at each location; normalizing the gradient direction vector to obtain a normalized gradient direction vector field; calculating the divergence of the normalized gradient direction vector field; and determining the defect type based on the divergence and the relationship between the gradient direction vector and the gravity direction.

[0019] Further, the determination of the defect type based on the divergence and the relationship between the gradient direction vector and the gravity direction includes: when the absolute value of the divergence in the region corresponding to the anomaly mark is greater than a preset divergence determination threshold, and the gradient direction vector exhibits a radial divergence or convergence distribution with the center of the anomaly mark as the origin, a first warning signal is output; wherein, the first warning signal characterizes the tendency of microcrack formation.

[0020] Furthermore, determining the defect type based on the divergence and the relationship between the gradient direction vector and the gravity direction also includes: when the absolute value of the divergence in the region corresponding to the anomaly marker is less than or equal to a preset divergence determination threshold, and the angle between the gradient direction vector and the gravity direction is less than a preset direction tolerance threshold, a second warning signal is output; wherein, the second warning signal represents a local sag tendency.

[0021] Furthermore, the quality monitoring method also includes a step of tracking and analyzing the evolution trend of the speckle decorrelation time over a long time scale to diagnose chemical evolution defects.

[0022] Furthermore, the evolution trend of the speckle decorrelation time over a long time scale is tracked and analyzed, including: constructing an evolution curve of the speckle decorrelation time as a function of curing time for the same spatial region; calculating the derivative of the evolution curve with respect to curing time; when the derivative is continuously negative within a preset time window during the first third of the curing cycle, it is determined that secondary liquefaction caused by surface drying has occurred in the spatial region, and a surface drying defect warning is output.

[0023] Furthermore, the method also includes a step of calculating the speckle spatial contrast of each pixel sub-region, wherein the speckle spatial contrast is obtained by calculating the ratio of the standard deviation to the mean of the light intensity values ​​of each pixel in the pixel sub-region; wherein, tracking and analyzing the evolution trend of the speckle decorrelation time over a long time scale further includes: when the speckle spatial contrast of the same spatial region shows a decreasing trend along the time direction, and the speckle decorrelation time shows a non-linear accelerating increasing trend along the time direction, it is determined that particle agglomeration and precipitation has occurred in the spatial region, and a particle agglomeration early warning signal is output.

[0024] Another aspect of the present invention provides a fire-retardant coating spraying quality monitoring system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the fire-retardant coating spraying quality monitoring methods described above.

[0025] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0026] 1. This invention emits coherent laser light onto the surface of a freshly sprayed fire-retardant coating wet film and continuously acquires dynamic laser speckle image sequences. Utilizing the temporal evolution characteristics of the speckle pattern caused by the movement of microscopic scattering particles within the wet film, spatiotemporal correlation processing is performed on the speckle image sequences to extract global decorrelation time field data. Thus, microscopic rheological state information of each local area within the entire field of view of the coating can be obtained non-contactly immediately after the coating is applied, opening a time window for early warning and timely process intervention during the microscopic defect initiation stage.

[0027] 2. This invention calculates the spatial gradient from global decorrelation time field data to obtain the gradient magnitude and gradient direction vector. Furthermore, it utilizes the spatial structure characteristics of the gradient field, especially the magnitude of the gradient magnitude, the divergence of the normalized gradient direction vector field, and the consistency between the gradient direction and the gravity direction, to achieve automatic classification and judgment of different types of coating defects. This judgment logic is entirely based on vector calculus and field theory tools, has a rigorous physical and mechanical basis, and does not require accumulating a large number of defect samples for machine learning training, thus achieving zero-latency and interpretable defect classification and early warning.

[0028] 3. This invention detects abnormal particle aggregation by jointly encoding the spatial contrast of speckles and the decorrelation time of speckles to establish a two-dimensional feature matrix. It also captures secondary liquefaction defects caused by surface drying and internal incomplete drying by performing long-term time-series evolution analysis of speckle decorrelation time and monitoring anomalous changes in the first-order time derivative. This forms a complete coating quality early warning system covering both spatial and temporal domains, which can cover various typical micro-defect types in fireproof coating construction. Attached Figure Description

[0029] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0030] Figure 1 This is a flowchart of the method provided in Embodiment 1 of the present invention.

[0031] Figure 2 The original image of simulated dynamic laser speckle provided in Embodiment 1 of the present invention.

[0032] Figure 3 This is a schematic diagram of two-dimensional distribution and reference diameter extraction provided in Embodiment 1 of the present invention.

[0033] Figure 4 This is a schematic diagram of the rigid body displacement compensation effect provided in Embodiment 1 of the present invention.

[0034] Figure 5 This is a decorrelated time field comparison heatmap provided in Embodiment 1 of the present invention.

[0035] Figure 6 This is a comparison curve of speckle decorrelation time estimates provided in Embodiment 1 of the present invention.

[0036] Figure 7 This is a two-dimensional heat map of the decorrelated time field provided in Embodiment 1 of the present invention.

[0037] Figure 8 This is a schematic diagram comparing the measured curve of the autocorrelation function with the fitted curve of exponential decay provided in Embodiment 1 of the present invention.

[0038] Figure 9 This is a schematic diagram of the radiation divergence characteristics of the microcrack region provided in Embodiment 1 of the present invention.

[0039] Figure 10 This is a schematic diagram of the unidirectional parallel features of the sag region provided in Embodiment 1 of the present invention.

[0040] Figure 11 This is a schematic diagram of the defect classification and early warning overlay output provided in Embodiment 1 of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0042] Example 1

[0043] This embodiment discloses a method for monitoring the quality of fire-retardant coating spraying. Figure 1 The flowchart of the method in this embodiment is shown. As can be seen from the flowchart, this embodiment includes the following steps:

[0044] Step 1: Emit coherent laser light onto the wet film surface of the freshly sprayed fireproof coating using a laser emitter, and continuously acquire dynamic laser speckle image sequences formed by the interference of scattered light from the wet film surface using an image acquisition device. Simultaneously, adaptively determine the optimal observation sub-region size based on the spatial statistical characteristics of the speckle images.

[0045] Among them, a fire-retardant coating wet film refers to the state of the fire-retardant coating after spraying but before it has fully cured. This wet film contains a large number of suspended scattering particles (such as titanium dioxide and filler particles), which undergo microscopic Brownian motion and directional rheological motion within the liquid matrix. Coherent laser refers to a beam of light with high spatial and temporal coherence generated by a laser emitter; for example, a coherent laser can be a single-mode continuous beam emitted by a helium-neon laser with a wavelength of 632.8 nm, or a coherent beam generated by a semiconductor laser or a solid-state laser. A dynamic laser speckle image sequence refers to a set of dynamically evolving speckle interference patterns formed on the imaging surface of an image acquisition device when a coherent laser irradiates the surface of a fire-retardant coating wet film. This is caused by the microscopic motion of scattering particles within the wet film, resulting in a continuous change in the phase difference between scattered light over time. For example, a dynamic laser speckle image sequence can be grayscale image frames continuously captured by a high-frame-rate industrial camera at a acquisition rate of hundreds to thousands of frames per second.

[0046] It is understandable that the light intensity value of each pixel in a speckle image depends on the phase superposition relationship between multiple scattering light paths at that point. When the scattering particles inside the coating are displaced, the phase relationship of each scattering light path changes accordingly, which macroscopically manifests as the brightness and darkness of the speckle pattern changing continuously over time. Therefore, the dynamic change characteristics of the speckle image directly encode information about the micro-rheological motion inside the coating.

[0047] Image acquisition equipment refers to photoelectric sensors used to continuously acquire speckle images at a predetermined frame rate. For example, image acquisition equipment can be a CMOS industrial area scan camera or a CCD industrial area scan camera. Figure 2 The original image of the simulated dynamic laser speckle in this embodiment is shown. Figure 2 The image shows a typical speckle interference pattern (randomly distributed bright and dark granules) formed on the imaging surface after coherent laser irradiation of a wet film.

[0048] In this embodiment, before performing subsequent analysis on the acquired dynamic laser speckle image sequence, a step of adaptively determining the size of pixel sub-regions is included to address the core issue of how to determine the appropriate size of the local observation window.

[0049] Specifically, when partitioning speckle images, if the window size of the local observation sub-region is set too small, the number of speckle particles contained in each sub-region will be insufficient, resulting in inadequate statistical sampling. Consequently, the calculated decorrelation time and other statistical quantities will fluctuate drastically, severely affecting the accuracy of the analysis. Conversely, if the window size is set too large, the spatial resolution will be significantly reduced, making it impossible to capture the fine microscopic changes in local areas within the coating. Traditional methods usually rely on operators manually specifying a fixed window size based on experience. However, under different spraying distances and different coating roughness conditions, the actual physical scale of speckle particles will change significantly, and a fixed window size is prone to failure when actual working conditions change.

[0050] To address the aforementioned issues, this embodiment utilizes the inherent spatial autocorrelation characteristics of speckle images to automatically extract the intrinsic physical scale of speckle particles, thereby achieving adaptive determination of the sub-region window size.

[0051] The spatial autocorrelation function (SAC) is a two-dimensional function obtained by performing normalized autocorrelation on an image in its spatial dimension. This function describes the statistical similarity of light intensity fluctuations between any point in the image and its spatial neighbors. The SAC reaches its maximum value at zero offset. As the spatial offset increases, the function value gradually decreases. The spatial offset distance corresponding to the function value decaying to half its peak value is the full width at half maximum (FWHM), which defines the reference diameter of the average speckle grains.

[0052] In this embodiment, the step of adaptively determining the size of the pixel sub-region may specifically include:

[0053] Calculate the global spatial autocorrelation function of the first frame speckle image; extract the full width at half maximum (FWHM) of the main lobe of the spatial autocorrelation function as the reference diameter of the average speckle particles; dynamically set the side length of the pixel sub-region to a multiple of the reference diameter.

[0054] For example, in this embodiment, the normalized spatial autocorrelation function of the initial frame global image is assumed to be: :

[0055]

[0056] in, Image in spatial coordinates The light intensity value at that location; This represents the average light intensity of the entire image. and This refers to the spatial offset along the horizontal and vertical directions; This is the normalized spatial autocorrelation function, which takes the value 1 at the zero offset position and decreases monotonically as the offset increases.

[0057] After calculating the spatial autocorrelation function, the reference diameter of the average speckle grains is defined by solving for the full width at half maximum (FWHM) of the main lobe of the function. :

[0058]

[0059] in, The average speckle grain reference diameter is defined by the full width at half maximum (FWHM). For radial distance variables; As an angle variable, iterate through all azimuth angles to ensure that the full width at half maximum (WHM) condition is met in all directions.

[0060] It is understandable that the above formula determines the equivalent average diameter of speckle particles by finding the minimum radial distance when the autocorrelation function value decays to less than half of the peak value in all azimuth directions. Its physical significance is that the average scale of speckle particles determines the typical spatial spacing between adjacent bright and dark units in the speckle pattern. Therefore, the main lobe width of the spatial autocorrelation function directly reflects this inherent physical scale.

[0061] Therefore, by defining the side length of the local observation sub-region as ,in This is the multiplier, and its value range is... .in, This is a dimensionless scaling factor used to balance statistical sufficiency and spatial resolution. When When the lower limit is set to 5, each sub-region contains approximately 25 speckle particles (estimated by area). The statistical sample size meets the basic requirements for statistical sufficiency while maintaining high spatial resolution, which is beneficial for capturing fine local variations within the coating. When the upper limit of 10 is used, each sub-region contains approximately 100 speckle particles, further improving statistical accuracy, but reducing spatial resolution. In practical applications, an appropriate value within the above range can be selected based on specific detection accuracy requirements and computational resources. value. Figure 3 This diagram illustrates the two-dimensional distribution of the spatial autocorrelation function and the extraction of the reference diameter in this embodiment. Figure 3 The spatial autocorrelation function is displayed in the form of a two-dimensional heat map, showing the decay pattern from the center to the outside. The full width at half maximum (FWHM) isopleth isopleth isopleth isopleth, and the extracted speckle particle reference diameter and the sub-region boundary grid determined by it are marked.

[0062] Understandably, in the actual numerical calculation of the above formula, the spatial autocorrelation function is solved in the two-dimensional Fourier transform domain through the Wiener-Khinchin theorem. That is, a two-dimensional Fourier transform is first performed on the image to obtain the power spectral density, and then an inverse Fourier transform is performed on the power spectral density to obtain the spatial autocorrelation function. This significantly improves the computational efficiency compared to performing sliding cross-correlation calculation pixel by pixel in the spatial domain.

[0063] Using the method described above, the entire speckle image is adaptively divided into... There are non-overlapping sub-matrix blocks, among which and These represent the number of sub-regions along the horizontal and vertical directions, respectively. The physical dimensions of each sub-region strictly match the inherent physical scale of the current speckle, fundamentally solving the problem of the failure of the traditional fixed window method due to changes in speckle particle size under different spraying conditions.

[0064] Step 2: Perform rigid body displacement compensation preprocessing on the dynamic laser speckle image sequence to eliminate signal pollution caused by industrial environmental vibration.

[0065] After determining the adaptive spatial resolution and completing the image sub-region segmentation, it is still necessary to address the signal contamination problem caused by vibrations in the industrial environment. In actual painting workshops, low-frequency vibrations generated by factors such as the operation of mechanical equipment, the movement of painting robots, and structural vibrations of the factory building cause rigid body translation of the entire coating and imaging system. This rigid body translation superimposes on the actual micro-rheological motion within the coating in the time domain. If the rigid body translation component is not separated, the decorrelation time calculated subsequently based on the temporal fluctuation characteristics of speckle light intensity will be mixed with the contribution of external vibrations, thus seriously interfering with the analysis and judgment of the actual micro-rheological state within the coating.

[0066] Among them, rigid body displacement refers to the overall relative translational motion between the coating surface and the image acquisition device caused by environmental vibration. This translational motion does not change the relative positional relationship between scattering particles inside the coating, but only causes the entire speckle image to undergo overall spatial translation on the imaging surface.

[0067] Understandably, traditional methods might attempt to eliminate the effects of vibration using low-pass filters in the frequency domain. However, this approach has an inherent flaw: the movement of microscopic scattering particles in the actual rheological process within the coating also generates low-frequency fluctuations in the speckle light intensity signal. While the low-pass filter blocks the vibration signal, it inevitably obliterates valuable microscopic rheological information. Therefore, this embodiment employs a rigid body displacement compensation method based on cross-correlation. This method sets a known stationary reference region in the image, uses cross-correlation calculations to accurately extract the frame-by-frame translation caused by environmental vibration, and then performs precise reverse translation correction on each frame of the wet film region. This completely removes the external vibration displacement component while fully preserving the relative microscopic rheological information within the coating.

[0068] In this embodiment, the specific steps of rigid body displacement compensation preprocessing include:

[0069] A reflective stationary reference point is set at the edge area of ​​the substrate without fire-retardant coating, so that the image acquisition device can simultaneously acquire speckle images containing the wet film surface and the stationary reference point; the two-dimensional cross-correlation function of the images of the stationary reference point area between adjacent frames is calculated, and the offset of the peak coordinate of the two-dimensional cross-correlation function relative to the center is extracted as the rigid body displacement vector caused by the environment; each frame image of the wet film surface is compensated by reverse subpixel translation according to the rigid body displacement vector.

[0070] In this context, a stationary reference point refers to a highly reflective marker that is fixedly affixed or etched onto the exposed edge of a substrate that has not been coated with fire-retardant paint. This marker remains absolutely stationary throughout the monitoring process (i.e., it does not generate any inherent micro-motion), therefore its frame-by-frame displacement in the speckle image is entirely and solely caused by rigid body translation due to environmental vibrations. For example, a stationary reference point could be a small retroreflective patch affixed to the surface of an uncoated area of ​​a steel structure.

[0071] The two-dimensional cross-correlation function is a two-dimensional function obtained by performing a cross-correlation operation on the light intensity distribution of the same reference point region in two frames of images. The peak position of this function corresponds to the best matching offset of the reference point region between the two frames, which reflects the overall translation vector caused by environmental vibration between the two acquisition times.

[0072] Specifically, for the image of the stationary reference point region Calculate the current time With the next moment Two-dimensional cross-correlation function between :

[0073]

[0074] in, It is a two-dimensional Fourier transform operator; This is the complex conjugate operation of the Fourier transform result; This is an element-wise product operation (i.e., the Hadamard product). For a moment Light intensity image of the region at the stationary reference point; The cross-correlation function measures the spatial offset between two frames of images. The similarity of the matches.

[0075] Understandably, calculating the cross-correlation function in the frequency domain using Fourier transform and its conjugate has significantly higher computational efficiency than calculating it pixel by pixel directly in the spatial domain. This is especially crucial for industrial applications that require real-time processing of high frame rate speckle image sequences.

[0076] After obtaining the cross-correlation function, the translation vector caused by the environment is determined by extracting the sub-pixel precision coordinates of the cross-correlation function peak:

[0077]

[0078] in, The extracted rigid body translation vector is the frame-by-frame vector caused by environmental vibrations. Its components can be non-integer values ​​(i.e., have sub-pixel precision).

[0079] Understandably, subpixel precision offset extraction can be achieved by performing two-dimensional parabolic or Gaussian fitting on discrete sampling points near the peak of the cross-correlation function, thereby obtaining displacement estimation results that are superior to those with single-pixel precision.

[0080] Subsequently, reverse subpixel translation interpolation correction is performed on each frame of the original image in the wet film region:

[0081]

[0082] in, This is the original speckle image of the wet film region; The clean image after rigid body displacement compensation; It is a bicubic interpolation function used to achieve spatial translation correction with sub-pixel accuracy.

[0083] Understandably, bicubic interpolation, by using the light intensity values ​​of 16 known pixels in a 4×4 neighborhood around the target pixel to perform a cubic polynomial weighted summation, can achieve image quality far superior to nearest neighbor interpolation and bilinear interpolation when performing spatial translations with non-integer pixel offsets. This ensures that even tiny vibration offsets smaller than one pixel can be accurately compensated and eliminated without introducing additional interpolation artifacts.

[0084] The rigid body displacement compensation preprocessing described above ensures that all subsequent calculations and processing are purely relative micro-rheological signals within the coating, enabling the entire monitoring algorithm to maintain extremely high robustness and analytical accuracy even in industrial spraying workshops with severe environmental vibrations.

[0085] Figure 4 A schematic diagram of the rigid body displacement compensation effect in this embodiment is shown. Figure 4 The subpixel displacement trajectory (with irregular low-frequency drift) of the stationary reference point tracked by the cross-correlation function over hundreds of consecutive frames is displayed in the form of a two-dimensional scatter plot. It can be seen that this method can accurately extract frame-by-frame vibration displacement with subpixel precision, providing a reliable basis for reverse compensation. Figure 5 The diagram shows a heatmap comparing the decorrelation time field before and after rigid body displacement compensation in this embodiment. Figure 5 The two side-by-side heatmaps in the image show the global decorrelation time field before and after rigid body displacement compensation, respectively. It can be seen that the image before compensation has obvious false fault fringes, while the image after compensation is smooth and truly reflects the coating state. This shows that the rigid body displacement compensation in this embodiment can effectively eliminate signal pollution caused by environmental vibration, avoid false fault false alarms, and verify the effectiveness of the anti-vibration interference method. Figure 6 This embodiment shows a comparison curve of speckle decorrelation time estimates before and after rigid body displacement compensation. Figure 6 This demonstrates that rigid body displacement caused by environmental vibration leads to a systematic underestimation of the decorrelation time. After rigid body displacement compensation based on cross-correlation, the measured value is restored to a level close to the true value, eliminating the interference of environmental vibration.

[0086] Step 3: Perform spatiotemporal correlation processing on the preprocessed dynamic laser speckle image sequence to construct global decorrelation time field data characterizing the micro-rheological state.

[0087] After establishing an adaptive spatial resolution and eliminating environmental vibration interference, the next core task is to convert the purified optical speckle signal into a physical quantity that can quantitatively reflect the micro-rheological state inside the coating, namely, the decorrelation time. This step is the key to converting the optical signal into a mechanical and rheological signal.

[0088] Among them, decorrelation time refers to the characteristic time constant required for the normalized second-order time autocorrelation function of speckle intensity to decay to a specific level within a certain spatial sub-region. This time constant quantitatively reflects the degree of microscopic motion activity of scattering particles within the sub-region.

[0089] It is understandable that decorrelation time is inversely proportional to the velocity of the microscopic scattering particles within the coating. The faster the flow within the coating (e.g., when sagging is about to occur), the more intense the Brownian and shear motions of the suspended scattering particles, the faster the speckle pattern reconstructs, and the shorter the decorrelation time. The better the coating's curing degree, the slower the movement of the scattering particles, the slower the speckle pattern changes, and the longer the decorrelation time. When the coating is fully cured, the internal particles almost stop moving, and the decorrelation time tends towards infinity. Therefore, decorrelation time can be used as a core physical quantity characterizing the microscopic curing and rheological rate of the coating.

[0090] Global decorrelation time field data refers to a two-dimensional scalar field formed by stitching together the decorrelation time values ​​calculated independently for all pixel sub-regions according to their original spatial positions. This scalar field, with spatial coordinates as the independent variable and decorrelation time as the dependent variable, comprehensively depicts the distribution of the micro-rheological state at various locations on the wet film surface of the coating. Scattering particles such as titanium dioxide suspended in the fire-retardant coating wet film undergo micro-Brownian and directional rheological motion. When irradiated by coherent laser, the micro-displacement of these particles induces changes in the phase difference of the scattered light, leading to drastic fluctuations in the intensity of the macroscopic speckle image. The rate of fluctuation of the speckle intensity over time directly reflects the activity level of particle motion within the coating. Therefore, by calculating the time autocorrelation function of the speckle intensity signal in each sub-region and extracting the characteristic decay time, the optical signal can be quantized and converted into physical parameters reflecting the micro-mechanical state.

[0091] In this embodiment, the step of constructing global decorrelational temporal field data may specifically include:

[0092] For each pixel sub-region, the intensity time series of each pixel in the continuous multi-frame image is extracted; the normalized second-order intensity autocorrelation function is calculated; the autocorrelation function curve is fitted with the preset exponential decay mathematical model using nonlinear least squares method, and the fitting parameters are extracted as the speckle decorrelation time of the sub-region; the speckle decorrelation times of all sub-regions are stitched together according to their original spatial positions to generate global decorrelation time field data.

[0093] Specifically, for a sub-region after translation compensation (Center coordinates are) First, extract the pixels in the continuous range of this sub-region. Intensity time series in frame images And calculate the time mean of pixel intensity. With intensity pulsation .

[0094] Among them, intensity time series This refers to the continuous acquisition of the same pixel. A one-dimensional discrete signal composed of light intensity values ​​recorded in a frame speckle image arranged in chronological order; For this pixel in all The arithmetic mean of light intensity on the frame; Intensity pulsation represents the degree of deviation of the actual light intensity value at each moment from the time mean.

[0095] Then, the normalized second-order intensity autocorrelation function is calculated. :

[0096] ,

[0097] in, This is the normalized second-order time autocorrelation function, which describes the time delay at the same spatial location. The statistical similarity between light intensity signals at two different times; For time delay variables; This is the length of the time window, i.e., the total duration of the light intensity sequence used to calculate the autocorrelation function; For the sub-region center at time The speckle light intensity value.

[0098] Understandably, the normalized second-order time autocorrelation function actually reflects the time interval... When the time interval is short, the speckle patterns at the two time points are highly similar because the scattering particles have not yet undergone significant displacement, resulting in high autocorrelation function values; as the time interval increases... As the particle size increases, the continuous motion of the scattering particles causes the speckle pattern to constantly reshape, and the similarity between the speckle patterns at two different times gradually decreases, resulting in a monotonically decaying autocorrelation function value. The rate of this decay process directly encodes the velocity information of the scattering particles.

[0099] The autocorrelation function is calculated. Subsequently, a nonlinear least squares optimization objective functional was constructed based on the Siegert relation. In order to solve the relevant time .

[0100] The Siegert relation is a fundamental theorem in speckle optics that establishes a deterministic mathematical relationship between the second-order autocorrelation function of speckle intensity and the first-order autocorrelation function of the electric field. Theoretical predictions from this theorem indicate that for a speckle field generated by a large number of independently moving scatterers, its second-order intensity autocorrelation function should obey... It exhibits an exponential decay form.

[0101] For example, in this embodiment, the target functional The specific expression can be:

[0102]

[0103] Decorrelation time is extracted by minimizing the above objective functional:

[0104]

[0105] in, The coherence factor (also known as the instrument constant) of the optical system depends on the light-gathering geometry of the detector and is usually between 0 and 1. The speckle decorrelation time for this sub-region is the core physical quantity characterizing the micro-rheological state in this algorithm; This represents the number of time delay sampling points used for fitting.

[0106] Understandably, the essence of the aforementioned nonlinear least squares fitting process is to optimally match the autocorrelation function curve obtained from actual measurements with the theoretical exponential decay model predicted by the Siegert relationship, by adjusting the parameters. and This minimizes the sum of squared residuals between the theoretical model curve and the measured data, thereby accurately extracting the characteristic time constant that characterizes the activity level of scattering particles. This method is based on an explicit physical model of exponential decay, relies entirely on deterministic mathematical fitting, and has clear physical interpretability.

[0107] After extracting the decorrelation time of all pixel sub-regions, the decorrelation time values ​​of each sub-region are... According to its spatial coordinates in the original image By stitching the data together, a global decorrelation time field covering the entire wet film surface is generated. This two-dimensional scalar field comprehensively depicts the microscopic curing and rheological rate state of the coating at various locations through the spatial distribution of decorrelation time: regions with larger decorrelation times indicate that the coating is well cured and the scattering particles move slowly; regions with smaller decorrelation times suggest that the coating is still in a strong rheological state or that there are abnormal microscopic movements.

[0108] Figure 7 The two-dimensional heatmap of the decorrelation time field in this embodiment is shown (including normal regions and defect precursor regions). Figure 7 The decorrelation temporal and spatial distribution of the entire wet film surface is shown using a pseudo-color thermal map, with most areas exhibiting a uniform intermediate distribution. The value (normal curing) includes a radially abnormally low value area (a precursor to microcracks) and a strip-shaped low value area along the direction of gravity (a precursor to sagging). Figure 8 This diagram illustrates a comparison between the measured curves of the autocorrelation function and the fitted curves of the exponential decay model for different curing stages in this embodiment. Each group includes measured data points (scatter points) and the fitted exponential decay model curve (solid line). Figure 8 This demonstrates that the exponential decay model in this embodiment can accurately fit the autocorrelation curves of different curing stages, and the decorrelation time systematically increases with the increase of coating curing degree.

[0109] Step 4: Perform Jacobi correction mapping of surface curvature on the global decorrelation time field data to eliminate gradient calculation bias caused by workpiece surface curvature projection deformation.

[0110] After obtaining the global decorrelation time field data, if we further perform spatial gradient calculations to analyze the stress distribution characteristics inside the coating, we need to consider the projection distortion problem caused by imaging geometry. In actual industrial spraying scenarios, the objects sprayed with fire-retardant coatings are often not ideal planes, but irregularly shaped components with complex three-dimensional curvatures, such as large steel pipes, irregularly shaped steel structure nodes, and curved shells. When the image acquisition device images such curved workpieces, due to the perspective projection effect of the camera, the three-dimensional physical distance of the workpiece surface will be geometrically compressed when projected onto the two-dimensional image plane. The degree of compression depends on the angle between the workpiece surface normal vector and the camera optical axis.

[0111] Projection distortion refers to the geometrical scale change that occurs when the actual physical distance on the surface of an object in three-dimensional space is mapped to a two-dimensional image plane through perspective projection. In areas where the angle between the surface normal vector and the camera optical axis is large, the spatial distance on the image plane is significantly compressed, resulting in a larger gradient value calculated directly on the image plane.

[0112] Understandably, if the spatial gradient of the decorrelational time field is calculated directly on the two-dimensional image plane without projection correction, a systematically high gradient amplitude will be generated in the inclined region of the curved surface. These high gradient values ​​are not caused by the actual non-uniform stress within the coating, but are purely geometric projection artifacts, which will trigger a large number of false tomography alarms, severely reducing the alarm accuracy of the monitoring system.

[0113] In this embodiment, for the spraying scenario of large irregular curved workpieces, a surface curvature correction step is included before calculating the spatial gradient, specifically including:

[0114] The normal vector field of the workpiece surface is constructed by acquiring pre-stored 3D laser scanning point cloud data; the Jacobian coordinate transformation matrix is ​​constructed based on the angle between the 3D normal vector of the sprayed surface and the optical axis of the image acquisition device; the gradient value calculated on the 2D image projection surface is multiplied by the inverse of the Jacobian coordinate transformation matrix to restore the true tangential gradient of the physical surface of the curved workpiece.

[0115] Among them, 3D laser scanning point cloud data refers to the set of dense spatial point coordinates obtained by pre-measuring the workpiece surface using a 3D laser scanner. This point cloud data provides precise 3D geometric information for various locations on the workpiece surface. The normal vector field refers to the vector field calculated from the 3D point cloud data, describing the local normal directions at various spatial locations on the workpiece surface. The Jacobian coordinate transformation matrix is ​​the matrix that establishes the differential mapping relationship between the 3D physical surface coordinate system and the 2D image projection plane coordinate system. This matrix describes the linear transformation relationship from infinitesimal displacements on the physical surface to the corresponding projected displacements on the image plane.

[0116] Specifically, in this embodiment, the surface curvature correction step may include:

[0117] The normal vector field of the workpiece surface is constructed by acquiring pre-stored 3D laser scanning point cloud data.

[0118] The 3D laser scanning point cloud data refers to the set of 3D spatial coordinate points obtained after pre-scanning the workpiece surface using a 3D laser scanner. Each point contains its coordinate information in 3D space and its corresponding surface normal vector information. This point cloud data is pre-collected and stored before the spraying operation, serving as prior knowledge for subsequent geometric correction. (Normal vector field) This refers to the component of the unit outward normal vector at each point on the workpiece surface in three-dimensional space, describing the orientation of the surface at that location. It is determined by the angle between the three-dimensional normal vector of the sprayed surface and the optical axis of the image acquisition device. Construct the Jacobian coordinate transformation matrix .

[0119] Among them, the optical axis direction vector of the image acquisition device is In the camera coordinate system, it can usually be simplified to represent as Projection angle Defined as the workpiece surface normal vector relative to the camera optical axis direction vector The included angle between them:

[0120]

[0121] The Jacobian coordinate transformation matrix is ​​a local linear transformation matrix that describes the mapping relationship between a three-dimensional physical surface coordinate system and a two-dimensional image projection coordinate system. Under a local orthogonal basis, this transformation matrix can be simplified to:

[0122]

[0123] It is understandable that when the surface normal vector is parallel to the camera's optical axis ( ), The transformation matrix is ​​the identity matrix, producing no correction because the physical surface is parallel to the image plane, and there is no projection compression; when As the surface becomes larger and more tilted, The smaller the value, the more severely the distance on the image plane along that direction is compressed relative to the distance on the real physical surface, requiring a larger correction factor to restore it.

[0124] The calculated gradient magnitude on the two-dimensional image projection plane is then compared with the Jacobian coordinate transformation matrix. Multiplying the inverse matrix, we can restore the true tangential gradient of the physical surface of the curved workpiece.

[0125] For example, in this embodiment, the true tangential gradient after surface correction can be calculated using the following formula:

[0126]

[0127] in, and These are the decorrelation time field edges directly calculated on the image plane. and The gradient component in the direction; It is the inverse matrix of the Jacobian projection transformation matrix; This is a corrected time gradient that reflects the decorrelation on the real physical surface.

[0128] Understandably, by using differential geometric transformation to restore the gradient values ​​on the image plane that are inflated due to projection compression to the gradient values ​​on the real physical surface, the illusion of inflated gradients caused by projection compression due to the tilt of the target surface is eliminated. This makes the algorithm not only applicable to flat workpieces, but also compatible with the monitoring of complex industrial components such as pipes and irregularly shaped steel structures, effectively eliminating false tomography alarms.

[0129] Step 5: Perform spatial gradient calculation and vector calculus analysis on the global decorrelation time field data to achieve the identification of abnormal faults and the classification and early warning of microcracks and sag defects.

[0130] After obtaining the true decorrelation time field data after surface correction, it is necessary to transform the spatial variation characteristics of this scalar field into vector field characteristics with physical discriminative significance, thereby achieving accurate classification of precursors to different types of defects in the coating. The core idea is that the magnitude of the gradient reflects the intensity of the stress abrupt change within the coating, while the directional distribution pattern of the gradient reveals the spatial organization of the stress abrupt change; different spatial organization patterns correspond to different physical defect occurrence mechanisms.

[0131] The spatial gradient field refers to the vector field formed by calculating the spatial partial derivatives of the decorrelation time scalar field point by point in two-dimensional space. The gradient vector at each location in this vector field points in the direction of the fastest increase in decorrelation time, and its magnitude characterizes the degree of spatial variation of decorrelation time at that location. An anomalous fault refers to a local spatial region appearing in the decorrelation time field, where the decorrelation time value differs significantly from its surrounding neighborhood, manifested as an abnormally high gradient amplitude. Physically, this corresponds to a rapid, discontinuous change in the spatial micro-rheological state within the coating.

[0132] In this embodiment, the operations of calculating the spatial gradient and performing anomaly detection may specifically include:

[0133] Obtaining spatial location The speckle decorrelation time corresponding to the pixel sub-region The forward difference algorithm is used to calculate respectively. Spatial partial derivatives of the direction and Calculate the spatial partial derivative of the direction; calculate the gradient magnitude and gradient direction vector at that location; when the gradient magnitude is greater than the preset fault determination threshold, trigger the primary anomaly marker.

[0134] For example, in this embodiment, the amplitude of the real physical gradient field after surface correction is calculated. With normalized directional field :

[0135]

[0136]

[0137] in, The magnitude of the decorrelation temporal gradient field, i.e., the Euclidean norm 2 of the gradient vector, quantifies the intensity of the decorrelation temporal spatial abrupt change at that spatial location. It is a normalized gradient direction unit vector that retains only the gradient direction information while removing the magnitude information.

[0138] in, Spatial partial derivatives of direction and Spatial partial derivatives of direction The results were obtained by forward difference algorithm respectively:

[0139]

[0140]

[0141] Understandably, the forward difference algorithm uses the difference in decorrelation time between adjacent sub-regions as a numerical approximation of the spatial partial derivative. Although its accuracy is first-order, since the spatial spacing of the sub-regions has been matched to the inherent physical scale of the speckle particles through the aforementioned adaptive method, this discretization accuracy is sufficient to meet engineering requirements for the macroscopic determination of coating defects.

[0142] The fault determination threshold refers to a pre-set reference value for gradient amplitude, used to distinguish between normal decorrelational temporal and spatial fluctuations and anomalous stress abrupt changes. When the gradient amplitude at a certain location... When the value exceeds this threshold, it is considered that there is an abnormal spatial fault in the time-decorrelated region at that location, triggering a primary anomaly marker.

[0143] Understandably, the threshold for determining tomography needs to be calibrated based on the specific coating system, spraying process parameters, and historical monitoring data. For example, statistical analysis of the decorrelation time gradient field of a large number of normal spraying samples can be performed, and the upper limit of the normal fluctuation range (such as the mean plus three times the standard deviation) can be taken as the threshold for determining tomography.

[0144] After completing the initial anomaly labeling, in order to further distinguish whether the region triggering the anomaly labeling is a precursor to microcracks or a precursor to sagging, this embodiment calculates the divergence of the gradient direction vector field and combines it with the relationship between the gradient direction and the gravity direction to achieve accurate classification of defect types. Here, divergence is a fundamental operator in vector calculus, describing the degree of spatial divergence or convergence of a vector field at a certain spatial location.

[0145] For example, in this embodiment, for the gradient direction vector field The divergence can be defined as follows:

[0146]

[0147] in, The normalized gradient direction vector field at position The divergence value at; and The normalized gradient direction vectors are respectively in direction and Components in direction.

[0148] It is understandable that the physical meaning of the divergence value is: when the absolute value of the divergence is large, it indicates that the gradient vectors around the point exhibit a significant radial divergence or convergence distribution, that is, the gradient direction radiates outward from the center point or converges towards the center point; when the absolute value of the divergence is small, it indicates that the gradient vectors around the point are highly parallel and consistent, and there are no significant divergence or convergence characteristics.

[0149] In this embodiment, the specific judgment rule for distinguishing between microcrack tendency and sagging tendency using deterministic logic is as follows: Let the warning threshold be... (i.e., fault determination threshold), divergence determination threshold is The unit vector of the direction of gravity is When the primary anomaly marker region meets the following conditions, a first warning signal (microcrack formation tendency) is output: gradient amplitude. Greater than the fault determination threshold And the absolute value of divergence Greater than the preset divergence threshold Furthermore, the gradient direction vector exhibits radial divergence or convergence characteristics with the anomaly marker center as the origin.

[0150] The first warning signal characterizes the tendency for microcrack formation. Its physical basis is that in the precursor of microcracks, there is a highly uneven deep stress concentration inside the coating. The decorrelation time shows a violent gradient radiation around the stress concentration point. The large divergence indicates that the local area is being violently torn by the micro-mechanical field due to the extremely uneven deep stress, which is a typical micro-stress field characteristic of the precursor of cracking.

[0151] When the primary anomaly marker region meets the following conditions, a second warning signal (local sag tendency) is output: gradient magnitude Greater than the fault determination threshold And the absolute value of divergence Less than or equal to the preset divergence threshold And the gradient direction vector Unit vector relative to the direction of gravity inner product Greater than That is, the angle between the gradient direction and the gravity direction is less than 15 degrees, exhibiting a uniform unidirectional vector characteristic.

[0152] The second warning signal characterizes a localized sagging tendency. Its physical basis lies in the fact that, in the precursory sagging phase, the coating fluid undergoes a unified microscopic shear slip under its own gravity. The gradient vectors of the decorrelation time are highly parallel and consistent within the sagging region, and their direction closely matches the direction of gravity. Small divergence indicates that the vector field is parallel and consistent, and there are no radial stress-tearing characteristics. The 15-degree directional tolerance threshold considers the directional estimation error in actual measurements and the slight deviation between the microscopic fluid flow direction and the macroscopic gravity direction.

[0153] Understandably, the above-mentioned judgment logic quantifies the intensity of stress mutation by the magnitude of the gradient, quantifies the spatial distribution pattern of stress mutation by the divergence of the gradient direction vector field, and quantifies the consistency between the flow direction and gravity by the inner product of the gradient direction and the gravity direction. With pure field theory and vector calculus logic, it achieves real-time, zero-latency, and fully interpretable classification and early warning of different types of defect precursors without relying on neural network training or a large number of historical defect image samples. Figure 9 This diagram illustrates the radiative divergence characteristics of the microcrack region in the gradient direction vector field of this embodiment. Figure 9 The gradient direction vector field is magnified and displayed in the microcrack precursor region. The arrows radiate outwards from the center, and the absolute value of the divergence is large. Figure 10 This illustration shows the unidirectional parallel feature of the sag region of the gradient direction vector field in this embodiment. Figure 10 The gradient direction vector field is magnified in the precursor region of the sag, with the arrows being parallel and aligned with the direction of gravity, and the divergence being close to zero. Figure 9 and Figure 10 The background is a two-dimensional heatmap of the decorrelated time field. Figure 11 This diagram illustrates the defect classification and early warning overlay output in this embodiment. Figure 11 The heatmap shows the spatial distribution of the divergence of the gradient direction vector field. The absolute value of the divergence is high in the microcrack region (red / blue hot spots), while the divergence in the sag region is close to zero (grayish-white). Two types of warning marker boxes are superimposed on the heatmap.

[0154] Step 6: By tracking the evolution trend of decorrelation time and speckle contrast over a long time scale, a time-series anomaly diagnosis mechanism is established to capture chemical evolution defects unique to fire-retardant coatings, such as surface drying, secondary liquefaction, and particle agglomeration and precipitation.

[0155] After establishing an instant defect classification and early warning system based on spatial gradient and divergence, it is necessary to further focus on the physicochemical evolution behavior of the coating over a longer curing timescale. Fire-retardant coatings have unique chemical compositions and curing kinetics; some defects do not exhibit abrupt spatial gradient changes but rather show abnormal evolutionary trends over time, requiring long-term time-series tracking for diagnosis.

[0156] Among them, long-range temporal evolution refers to the continuous tracking and trend analysis of physical quantities such as decorrelation time and speckle contrast in the same spatial region throughout the entire curing cycle from the start of coating application. Its time scale ranges from several minutes to several hours, which is much longer than the acquisition window of a single speckle decorrelation time.

[0157] In this embodiment, long-range time series evolution analysis specifically includes the following two types of anomaly diagnosis:

[0158] Category 1: Diagnosis of secondary liquefaction caused by surface drying. During the normal coating curing process, the solvent and water in the coating gradually and uniformly evaporate from the inside out, the viscosity of the coating continuously increases, and the microscopic movement speed of the scattering particles continuously slows down. Therefore, the relevant time... It should be based on the macro solidification time The progress increases monotonically. However, when the coating surface forms a skin and dries too quickly under high temperature or low humidity conditions, the solidified film formed on the surface will prevent the normal evaporation channels of the inner solvent, causing the inner coating to undergo abnormal secondary liquefaction due to the inability of the solvent to be discharged—that is, the internal viscosity decreases, the movement of scattering particles accelerates again, and the decorrelation time decreases abnormally.

[0159] Secondary liquefaction refers to an abnormal chemical-physical phenomenon during the curing process of a coating where the rapid formation of a skin on the surface seals off the solvent evaporation channels of the inner layer, causing the sealed inner layer of coating to return to a highly fluid state. This phenomenon is extremely difficult to detect with the naked eye (because the surface has already cured and formed a skin), but it leads to uneven internal structure of the coating and is a typical cause of the surface-drying-in-the-inner-layer quality defect in fire-retardant coating projects.

[0160] Specifically, it involves extracting decorrelation time data for the same local spatial region at different macroscopic time points. Construction with curing time Evolutionary curve of change And calculate the first time derivative of the evolution curve:

[0161]

[0162] in, It is a macroscopic time scale, representing the total curing time calculated from the start of coating application; The derivative of the relevant time with respect to macroscopic time represents the trend of the curing rate.

[0163] Understandably, under normal curing conditions, It should always remain positive (i.e., the decorrelation time continuously increases, reflecting the coating's continued curing and hardening). If detected... (i.e., the decorrelation time decreases anomalously), which indicates that the coating in that region is undergoing an anomalous process of particle re-acceleration.

[0164] If, during the first third of the curing cycle, a certain region exhibits an abnormal decrease in the derivative, which remains negative for an consecutive preset time window, then the following criteria are met:

[0165]

[0166] in, This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. and The start and end times of the monitoring period of interest are within the first third of the curing cycle; This is a time window threshold for persistent anomalies, used to exclude transient, accidental disturbances.

[0167] The system determines that the area has experienced secondary liquefaction due to the deep solvent being encased by the surface sealing layer, and immediately issues a quality defect warning indicating that the surface sealing caused the interior to not be completely dried.

[0168] Understandably, limiting the monitoring period to the first third of the curing cycle is because secondary liquefaction caused by surface drying usually occurs in the early stages of curing—at this time, the coating surface still has a high solvent content, making it prone to rapid skin formation under external high temperature and low humidity conditions. In the middle and later stages of curing, a brief decrease in decorrelation time is more likely a normal phenomenon caused by the exothermic effect of the coating's chemical reaction or the release of volume shrinkage stress, and should not be misjudged as secondary liquefaction.

[0169] Category 2: Diagnosis of local density abnormalities caused by particle agglomeration and precipitation.

[0170] While constructing the global decorrelation time field, this embodiment also simultaneously calculates the speckle spatial contrast of each pixel sub-region and combines the decorrelation time to establish a bivariate state space for collaborative analysis.

[0171] Speckle spatial contrast refers to a normalized measure of the statistical dispersion of speckle intensity distribution within a specific pixel sub-region. The calculation formula is:

[0172]

[0173] in, is the standard deviation of pixel intensity within the sub-region; (Right now () represents the mean pixel intensity within the sub-region; This represents the total number of pixels within the sub-region. This represents the light intensity value of each pixel.

[0174] It is understandable that the spatial contrast of speckle patterns reflects the uniformity of the density distribution of scattering particles within a sub-region: when the scattering particles are uniformly distributed, the speckle pattern has high contrast (clear distinction between light and dark); when particles agglomerate and precipitate, resulting in uneven density distribution (some areas are too dense and others are too sparse), the speckle pattern tends to become blurred and the contrast decreases.

[0175] Combined with the relevant time for the removal of speckles Establish a two-dimensional feature matrix When the following anomalous cooperative change patterns are identified, it is determined that density anomalies caused by particle aggregation and precipitation have occurred:

[0176] If a local area is in time Simultaneously satisfy speckle contrast It shows a downward trend along the time direction (i.e.) ), and go to the relevant time It shows a steep increasing trend along the time direction (i.e.) If the increase in the relevant time exhibits a non-linear accelerating characteristic (rather than the linear and slow increase during normal curing), then it is determined that a local density anomaly caused by particle agglomeration and precipitation has occurred in that area, and a particle agglomeration early warning signal is output. It is a unit vector in the time direction; This represents the rate of change of speckle contrast over time. It represents the rate of change of the relevant time along the time direction.

[0177] Understandably, the physical basis of the aforementioned bivariate judgment logic lies in the following: when functional fillers in the coating (such as expanded graphite, silicate particles, etc.) agglomerate and precipitate during the coating curing process, the scattered particles in local areas aggregate from their originally uniformly dispersed state into larger clusters. On the one hand, the decrease in particle density distribution uniformity leads to a decrease in speckle contrast; on the other hand, because the inertia of large particle clusters is much greater than that of dispersed small particles, their Brownian motion velocity is significantly lower than that of dispersed particles, resulting in a sharp increase in decorrelation time. The anomalous synergy between the decrease in speckle contrast and the sharp increase in decorrelation time constitutes a characteristic fingerprint signal of particle agglomeration and precipitation, perfectly complementing the aforementioned spatial gradient anomaly warning in the temporal dimension.

[0178] In this embodiment, to comprehensively cover all possible quality defects, the aforementioned long-range temporal evolution analysis and the aforementioned spatial gradient anomaly detection are performed in parallel: spatial gradient and divergence analysis is responsible for capturing precursors of spatially abrupt defects (microcracks, sagging) that can be identified at a single moment, while long-range temporal evolution analysis is responsible for capturing chemically evolved defects (secondary liquefaction, particle agglomeration) that only manifest on a longer timescale. The two work together to form a complete quality monitoring system covering both the spatial and temporal domains.

[0179] Example 2

[0180] This embodiment discloses a fire-retardant coating spraying quality monitoring system. Specifically, this system can be integrated into an electronic device, such as an industrial edge computing terminal, a Field-Programmable Gate Array (FPGA) processing platform, an industrial control computer, or a server. The industrial edge computing terminal can be an embedded image processing module, a smart industrial tablet, or a personal computer (PC); the server can be a single server or a server cluster composed of multiple servers. In some embodiments, the system can also be integrated into multiple electronic devices. For example, the image acquisition part of the fire-retardant coating spraying quality monitoring system can be deployed in an edge computing terminal at the spraying site, while the core spatiotemporal correlation processing and anomaly detection module can be integrated into a remote server, with multiple electronic devices collaboratively implementing the fire-retardant coating spraying quality monitoring method of Embodiment 1 of this application. In some embodiments, the server can also be implemented in the form of a terminal.

[0181] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of monitoring the quality of fireproof paint spraying, characterized in that, The quality monitoring method includes: A coherent laser is emitted onto the wet film of fire-retardant coating on the target substrate using a laser emitting device, and a dynamic laser speckle image sequence formed by the interference of scattered light from the wet film of fire-retardant coating is continuously acquired using an image acquisition device. Based on the light intensity of the speckle image in the dynamic laser speckle image sequence, the first frame speckle image in the dynamic laser speckle image sequence is selected, and the normalized two-dimensional spatial autocorrelation function is calculated based on the light intensity value of each pixel in the first frame speckle image. The side length of the pixel sub-region is set as a preset multiple of the speckle pattern reference size, and each frame of the dynamic laser speckle image sequence is adaptively divided into multiple non-overlapping pixel sub-regions according to the side length of the pixel sub-region. The minimum radial offset distance whose autocorrelation function value is not greater than half of the peak value is determined as the speckle pattern reference size. For each pixel sub-region, the speckle decorrelation time is extracted, and the speckle decorrelation times of all pixel sub-regions are stitched together according to their corresponding spatial positions to generate global decorrelation time field data covering the wet film of the fireproof coating. The spatial gradient of the global decorrelation time field data is calculated to obtain the gradient magnitude. When the gradient magnitude is greater than the preset fault determination threshold, an anomaly is determined to exist at the corresponding spatial location and an anomaly marker is triggered.

2. The method of claim 1, wherein Determining the spatial scale corresponding to the main lobe of the normalized two-dimensional spatial autocorrelation function decaying to half of its peak value includes: Using the null offset position of the normalized two-dimensional spatial autocorrelation function as the center, determine the minimum radial offset distance that ensures the autocorrelation function values ​​in multiple preset orientation directions are not greater than half of the peak value.

3. The method of claim 1, wherein, Before extracting the speckle decorrelation time for each pixel sub-region, the method further includes: A rigid body displacement compensation preprocessing step is performed on the dynamic laser speckle image sequence to eliminate signal contamination caused by environmental vibration.

4. The method of claim 3, wherein The steps of the rigid body displacement compensation preprocessing include: Set a stationary reference point at the edge of the substrate that has not been coated with fire retardant. Acquire speckle images that simultaneously include the wet film surface of the fire-retardant coating and the stationary reference point; Calculate the two-dimensional cross-correlation function of the static reference point region image between adjacent image frames in the dynamic laser speckle image sequence; The offset of the peak coordinates of the two-dimensional cross-correlation function relative to the center is extracted and used as the rigid body displacement vector; Each frame of the image on the wet film surface of the fireproof coating is compensated for by reverse subpixel translation according to the rigid body displacement vector.

5. The method of claim 1, wherein, The extraction speckle decorrelation time includes: Extract the intensity time series of each pixel within the pixel sub-region in multiple consecutive frames of images; Calculate the time mean of the intensity time series, and calculate the deviation of the light intensity value at each moment from the time mean; Based on the intensity time series, a normalized second-order time autocorrelation function is calculated. The normalized second-order time autocorrelation function describes the statistical similarity between light intensity signals at the same spatial location under different time delays. The speckle decorrelation time is extracted based on the normalized second-order time autocorrelation function.

6. The method of claim 5, wherein the method further comprises: Extracting the speckle decorrelation time based on the normalized second-order time autocorrelation function specifically includes: Construct a pre-defined exponential decay model, which includes a coherence factor and a characteristic decay time as parameters to be fitted. The normalized second-order time autocorrelation function is fitted with the exponential decay model using nonlinear least squares fitting to minimize the sum of squared residuals between the output of the exponential decay model and the measured data of the normalized second-order time autocorrelation function. The feature decay time obtained from the fitting is extracted as the speckle decorrelation time.

7. The method of claim 1, wherein After obtaining the global decorrelation time field data, and before determining the gradient magnitude based on the global decorrelation time field data, a surface curvature correction step is also included: To eliminate gradient calculation deviations caused by surface curvature projection deformation of the workpiece; The surface curvature correction step includes: Acquire pre-stored 3D point cloud data and construct the normal vector field of the workpiece surface based on the 3D point cloud data; Calculate the projection angle between the normal vector at each position in the normal vector field and the optical axis of the image acquisition device; The Jacobian coordinate transformation matrix is ​​constructed based on the projection angle. The Jacobian coordinate transformation matrix is ​​used to describe the linear transformation relationship between the minute displacement on the physical surface of the workpiece and the corresponding projected displacement on the image projection plane. The gradient value calculated on the image projection plane is multiplied by the inverse of the Jacobian coordinate transformation matrix to restore the true tangential gradient of the physical surface of the workpiece.

8. The method of claim 7, wherein the method further comprises: The construction of the Jacobian coordinate transformation matrix based on the projection angle specifically includes: In a locally orthogonal coordinate system, the Jacobian coordinate transformation matrix is ​​constructed as a diagonal matrix, where the diagonal elements in the locally orthogonal coordinate system are the cosine values ​​of the projection angles, and the diagonal elements perpendicular to the tilt direction are 1. The step of multiplying the gradient value calculated on the image projection plane with the inverse of the Jacobian coordinate transformation matrix includes: Divide the image plane gradient component along the surface tilt direction by the cosine of the projection angle to restore the true physical gradient in that direction.

9. The method of claim 1, wherein, After triggering the anomaly marker, the method further includes: classifying the defect type of the region corresponding to the anomaly marker. Based on the spatial gradient of the global decorrelation time field data, the gradient direction vector at each location is obtained; The gradient direction vector is normalized to obtain a normalized gradient direction vector field. Calculate the divergence of the normalized gradient direction vector field; The defect type is determined based on the divergence and the relationship between the gradient direction vector and the gravity direction.

10. A fireproof coating spray quality monitoring system characterized by, The quality monitoring system includes: processor; A memory storing a computer program, which, when executed by a processor, implements the fire-retardant coating spraying quality monitoring method as described in any one of claims 1 to 9.

Images