Intelligent detection system and method for performance of waterproof coating

Abstract

The application discloses a waterproof coating performance intelligent detection system and method, and relates to the waterproof coating detection technical field.The system comprises the following modules: a multi-source sensing data acquisition module that acquires multi-source sensing fusion data in real time; a water content data noise compensation module that adopts Kalman filtering and non-local mean denoising to generate a denoised water content signal; a penetration path atlas construction module that generates a penetration path atlas through interpolation and gradient field calculation based on the spatial positions of the sensor array; an edge interference identification and correction module that identifies and eliminates edge path data and outputs corrected penetration behavior parameters; a coating surface defect detection module that generates a defect detection signal through image enhancement and threshold segmentation; and an evaluation module that receives the corrected penetration behavior parameter signal and the defect detection signal for comprehensive evaluation and generates a waterproof coating comprehensive performance evaluation signal.The application solves the problem of evaluation distortion caused by multi-source sensing noise and edge interference in waterproof coating detection, and realizes precise and intelligent detection.

Description

Technical Field

[0001] This invention relates to the field of intelligent testing technology for waterproof coating performance, specifically to an intelligent testing system and method for waterproof coating performance. Background Technology

[0002] Waterproof coatings are widely used in building and underground engineering, and their waterproof performance directly affects structural durability. Traditional impermeability testing involves applying pressure in stages using an impermeability tester and manually observing water seepage, providing only a qualitative assessment. Permeability testing requires dissecting the substrate to measure the penetration depth, which is cumbersome and highly subjective. Defects such as bubbles, pinholes, and cracks in the coating rely mainly on visual inspection, lacking objective quantification. With the development of sensor technology, some devices embed humidity sensors in the substrate to monitor changes in moisture content in real time. However, the acquired signals are susceptible to temperature fluctuations, mechanical vibrations, and electromagnetic interference, containing a large amount of non-stationary noise that existing simple filtering methods cannot effectively suppress. Waterproof coating specimens must be sealed around their perimeter. If the seal is flawed, water will seep in from the edge interface, causing the measured permeability data to contain false components and failing to accurately reflect the coating's water-blocking ability. Currently, there is a lack of quantitative identification and correction methods for edge sealing interference, often relying on manual experience or discarding the entire data set for retesting, resulting in a waste of testing resources. Existing systems typically collect only single types of data, failing to integrate and analyze multi-dimensional information such as water pressure, moisture content distribution, substrate strain, and surface images, making it difficult to comprehensively and objectively evaluate the overall performance of waterproof coatings. Therefore, an intelligent detection solution is needed that can simultaneously acquire multi-source sensor data, effectively suppress environmental noise, automatically identify and correct edge sealing interference, and simultaneously achieve automatic detection of surface defects. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of this invention is to provide an intelligent detection system and method for waterproof coating performance, which solves the problem of evaluation distortion caused by multi-source sensor noise and edge interference in waterproof coating detection.

[0004] This invention simultaneously acquires multi-source signals, performs two-step noise reduction on the moisture content signal using Kalman filtering and nonlocal mean denoising, constructs a moisture content field based on spatial location interpolation of the humidity sensor, and tracks the permeation path along the gradient. It uses statistical analysis of the permeation initiation distribution and strain time-series correlation to identify and eliminate edge sealing interference. At the same time, it employs limited contrast enhancement and adaptive threshold segmentation to detect coating surface defects, and finally outputs corrected permeation behavior parameters and defect information to achieve an objective evaluation of the comprehensive performance of waterproof coatings.

[0005] This invention provides an intelligent testing system for the performance of waterproof coatings, comprising: The multi-source sensor data acquisition module acquires multi-source sensor fusion data sequence signals in real time; The moisture content data noise compensation module receives the moisture content signal from the multi-source sensor fusion data sequence signal, and generates a denoised moisture content signal through Kalman filtering and nonlocal mean denoising processing. The permeation path map construction module receives the denoised moisture content signal, performs interpolation and gradient field calculation based on the spatial location of the humidity sensor array, and generates the permeation path map signal. The edge interference identification and correction module receives strain signals from the permeation path map signal and the multi-source sensor fusion data sequence signal, identifies edge paths whose starting point is located in the edge region, removes the data corresponding to the edge paths, and generates a corrected permeation behavior parameter signal. The coating surface defect detection module receives image signals from multi-source sensor fusion data sequence signals and generates defect detection signals through contrast-limited adaptive histogram equalization and adaptive threshold segmentation. The evaluation module receives and comprehensively evaluates the modified penetration behavior parameter signals and defect detection signals to generate a comprehensive performance evaluation signal for the waterproof coating.

[0006] In one embodiment of the present invention, the permeation path mapping module acquires the spatial position of each sensor in the humidity sensor array, including the horizontal distance of the sensor from the central axis of the sample and the depth of the sensor embedded in the substrate; using the horizontal distance as the first coordinate and the embedding depth as the second coordinate, each sensor is mapped as a two-dimensional grid node to construct a spatial grid; for any non-node position within the spatial grid, the four nodes closest to that non-node position are selected, and using the denoised moisture content values ​​of the four nodes, a first intermediate value is obtained by linear interpolation along the first coordinate direction, and then a first intermediate value is obtained by linear interpolation along the second coordinate direction to obtain the moisture content interpolation value of that non-node position, thus forming an interpolation. The water content field after interpolation is calculated. For each grid node, using the water content values ​​of the preceding and following nodes along the first coordinate direction, the partial derivative of the water content at that node along the first coordinate direction is calculated. Similarly, using the water content values ​​of the preceding and following nodes along the second coordinate direction, the partial derivative of the water content at that node along the second coordinate direction is calculated. The partial derivatives are calculated using the central difference method. The value of the partial derivative is equal to the difference in water content between two adjacent nodes in the corresponding direction divided by twice the distance between adjacent nodes. The vector formed by the two partial derivatives is used as the water content gradient vector for that node, and the direction of the gradient vector is determined as the seepage propagation direction. The formula for calculating the water content gradient vector is as follows: ;

[0007] in, Let H(x,y) be the partial derivative of the water content at grid node (x,y) along the i-th coordinate direction, and let H(x,y) be the denoised water content value at grid node (x,y). The distance between adjacent nodes in the i-th coordinate direction is... These are the second-order difference correction coefficients. The distance between adjacent nodes is perpendicular to the i-th coordinate direction.

[0008] In one embodiment of the present invention, the permeation path mapping module scans the moisture content values ​​at different times at various locations along the interface between the coating and the substrate from the center to the edge in the interpolated moisture content field. The location where the first increase in moisture content relative to the initial dry state exceeds a preset abrupt change threshold and continues to show an upward trend for multiple subsequent detection times is determined as the permeation starting point. Using the determined permeation starting point as the current point, the module moves along the gradient vector direction of the current point at a preset step size. After reaching the new position, the gradient vector of that point is obtained by interpolating the denoised moisture content of the adjacent nodes at the new position, and the moving direction is updated accordingly. This process is repeated until a preset threshold is met. The termination conditions include moving to the sample boundary or the magnitude of the gradient vector being lower than a preset lower limit, thereby tracing a continuous permeation path curve. The above tracing process is performed on each determined permeation starting point to obtain multiple permeation path curves. The set of coordinate sequences of each permeation path curve is used as the permeation path spectrum signal. When the moisture content data noise compensation module outputs the denoised moisture content signal corresponding to the next level of water pressure loading, the permeation starting point determination and permeation path tracing are re-executed based on the updated denoised moisture content signal to generate a new permeation path spectrum signal, which is output in sequence form to form a dynamic evolution spectrum signal of the permeation path in time series.

[0009] In one embodiment of the present invention, the edge interference identification and correction module receives a permeation path map signal from the permeation path map construction module. The permeation path map signal contains a sequence of starting coordinates of each permeation path curve. The starting coordinates are the position coordinates recorded by the permeation path map construction module when determining the permeation starting point. Based on the preset geometric dimensions of the effective detection area of ​​the sample and the preset edge width, an edge region and a central region are defined on a two-dimensional coordinate plane. The edge region is an annular region less than the preset edge width from the sample boundary, and the central region is the remaining region inside the boundary. The preset edge width is set according to the width of the sample edge sealing strip. The starting coordinates of each permeation path are compared with the positions of the edge region and the central region. Permeation paths whose starting coordinates fall within the edge region are classified as edge paths, and permeation paths whose starting coordinates fall within the central region are classified as central paths. The total number of all permeation paths and the number of edge paths among them are counted, and the ratio of the number of edge paths to the total number of permeation paths is calculated to obtain the edge path proportion. The edge path proportion is compared with a preset interference judgment threshold. When the edge path proportion exceeds the preset interference judgment threshold, an edge sealing interference presence flag signal is output, and subsequent correction operations are triggered.

[0010] In one embodiment of the present invention, after the edge interference identification and correction module outputs an edge sealing interference presence flag signal, it determines the corresponding sample surface measurement point for each permeation path classified as an edge path in the permeation path map signal based on its starting point coordinates; it extracts the strain time series data of the surface measurement point as a function of detection time from the strain signal in the multi-source sensor fusion data sequence signal. The strain signal is collected by a strain gauge installed on the sample substrate surface near the edge position to monitor the minute deformation of the substrate caused by moisture penetration; it extracts the moisture content time series data corresponding to the shallowest humidity sensor at the same starting point position from the denoised moisture content signal output by the moisture content data noise compensation module. The shallowest humidity sensor is the sensor embedded in the substrate closest to the coating. Align the time axes of strain time-series data and moisture content time-series data to the same baseline, and calculate the linear correlation coefficient between them. The calculation of the linear correlation coefficient includes: calculating the covariance and standard deviation of the strain time-series data and the moisture content time-series data, and dividing the covariance by the product of the two standard deviations to obtain the correlation coefficient value; when the absolute value of the correlation coefficient is greater than a preset correlation threshold, it is determined that the seepage behavior of the edge path is causally related to the strain change caused by insufficient edge sealing, and the edge path is identified as an interference path caused by edge sealing defects; when the absolute value of the correlation coefficient is not greater than the preset correlation threshold, it is determined that the seepage behavior of the edge path is not related to insufficient edge sealing, and the edge path is retained. The formula for calculating the covariance of strain time-series data and moisture content time-series data is as follows: ;

[0011] in, For strain time series data With moisture content time series data The linear correlation coefficient, Let be the strain value at time t. The mean of the strain time series data, Let be the denoised moisture content value at time t. denoted as the mean of the moisture content time series data, and T represents the total number of sampling points in the time series data.

[0012] In one embodiment of the present invention, after receiving confirmed interference path information, the edge interference identification and correction module deletes all edge path curves confirmed to be caused by edge sealing defects and all sampling point data contained therein from the permeation path map signal. The sampling point data includes the coordinates, moisture content, and permeation depth of each point on the path. Simultaneously, it removes the moisture content and water pressure records corresponding to the timestamps of the edge path sampling points from the moisture content signal and the corresponding water pressure signal in the multi-source sensor fusion data sequence signal. Using the retained permeation depth data of all center paths, a common time axis is established at preset fixed time intervals to rank the permeation depth of each center path. Linear interpolation is performed on the columns to obtain the penetration depth values ​​of each center path on the common time axis, and the arithmetic mean of the penetration depths of all center paths at each time point is calculated as the effective penetration depth. The effective penetration rate is obtained by numerically differentiating the effective penetration depth over time. The numerical differentiation uses the central difference method to calculate the difference in effective penetration depth between two adjacent time points divided by the time interval. When the number of center paths involved in the calculation is zero at a certain time point, the effective penetration depth and effective penetration rate at that time point are marked as invalid values. The effective penetration depth, effective penetration rate, and the number of retained center paths are combined to form a corrected penetration behavior parameter signal, which is output to the external performance evaluation unit.

[0013] In one embodiment of the present invention, the moisture content data noise compensation module models the moisture content signal as a linear dynamic system. The state vector contains the true moisture content and its rate of change, and the observed value is the output of the moisture content sensor. A state transition equation and an observation equation are constructed. The moisture content signal is then denoised in the first stage using Kalman filtering recursive calculation. The Kalman filtering recursive calculation includes predicting the current state using the state estimate from the previous time step, and then correcting and updating the predicted value based on the current observation value. The process noise covariance and observation noise covariance are pre-set based on sensor static calibration data and environmental disturbance levels. The Kalman-filtered signal is then processed... The moisture content signal sequence undergoes a second-stage nonlocal mean denoising process. In this stage, for each sampling point in the sequence, a signal segment of a first preset length is taken as the local neighborhood centered on that point. Within a search window containing that sampling point, the weighted Euclidean distance between the local neighborhoods of other points and the local neighborhood of the center point is calculated. Based on the weighted Euclidean distance, the similarity weight of each point is calculated. The moisture content values ​​of all points within the search window are then weighted and averaged according to their similarity weights to obtain the final denoised moisture content value of that sampling point, thus generating the denoised moisture content signal. The length of the search window is a second preset length. The formula for calculating the final denoised moisture content value is as follows: ;

[0014] in, Let j be the similarity weight of point j relative to the center point i within the search window. The value of the local moisture content signal segment at center point i is taken. Let j be the value of the local moisture content signal segment. It is the L2 norm. The parameters are Gaussian kernel smoothing parameters. The set of all sampled points within the search window. It is an exponential function.

[0015] In one embodiment of the present invention, the coating surface defect detection module converts the image signal from the red-green-blue color space into a grayscale image; divides the grayscale image into several non-overlapping sub-blocks, performs histogram equalization on each sub-block, and truncates the portion of the histogram of each sub-block that exceeds a preset contrast limiting threshold. The truncated portion is then evenly distributed across all grayscale levels and equalized again to enhance local image contrast and suppress noise, resulting in an enhanced grayscale image. The adaptive binarization threshold of the enhanced grayscale image is calculated using the maximum inter-class variance method, and the image is divided into foreground defect regions and normal background regions based on this threshold. Region; calculate geometric feature parameters for each connected component in the binarized image, including the area, perimeter, roundness, and aspect ratio of the connected component; according to preset classification rules, connected components with roundness greater than a preset upper limit threshold and area less than a preset upper limit threshold for pinhole area are identified as pinhole defects; connected components with roundness less than a preset lower limit threshold for roundness and aspect ratio greater than a preset aspect ratio threshold are identified as crack defects; connected components with area greater than a preset lower limit threshold for bubble area and roundness between a preset lower limit threshold and an upper limit threshold for roundness are identified as bubble defects, and a defect detection signal containing defect type, location, and geometric parameters is generated.

[0016] In one embodiment of the present invention, the evaluation module includes a spatial registration unit, a correlation analysis unit, and a grade determination unit. The spatial registration unit receives a modified penetration behavior parameter signal and a defect detection signal, and maps the two-dimensional coordinates of each sampling point in the modified penetration behavior parameter signal and the position coordinates of each defect region in the defect detection signal to the same sample coordinate system, establishing a spatial correspondence between the penetration path sampling points and the defect regions on the coating surface. The correlation analysis unit calculates the distance between each penetration path sampling point and the boundary of the nearest defect region based on the spatial correspondence. When the distance is less than a preset correlation threshold, the penetration path sampling point is marked as a defect-related penetration point. The proportion of defect-related penetration points among all penetration path sampling points is statistically analyzed. When the proportion of defects exceeds a preset threshold, it is determined that surface defects have a significant impact on penetration behavior, and a defect penetration correlation indicator signal is output. At the same time, the defect severity index is calculated based on the geometric parameters of the defects in the defect detection signal. The defect severity index is a weighted sum of the defect area and the defect aspect ratio. The grade determination unit receives the effective penetration depth and effective penetration rate from the corrected penetration behavior parameter signal, as well as the defect penetration correlation indicator signal and defect severity index output by the correlation analysis unit. Based on the preset comprehensive performance determination rules, the comprehensive performance grade of the waterproof coating is determined, and a comprehensive performance evaluation signal of the waterproof coating containing the comprehensive performance grade, effective penetration depth, effective penetration rate, defect type, and defect severity index is generated.

[0017] This invention also includes an intelligent detection method for the performance of waterproof coatings, comprising: S1: Real-time acquisition of water pressure signal, water content signal, image signal and strain signal, time synchronization processing of each signal to form a multi-source sensor fusion data sequence signal; S2: Extract the moisture content signal from the multi-source sensor fusion data sequence signal, perform recursive denoising through Kalman filtering, and then perform non-local mean denoising to generate a denoised moisture content signal. S3: Based on the spatial position of each sensor in the humidity sensor array within the substrate, the noise-reduced moisture content signal is interpolated to form a moisture content field, the gradient of the moisture content field is calculated, the permeation path is traced along the gradient direction, and a permeation path map signal is generated. S4: Receive the strain signal from the permeation path map signal and the multi-source sensor fusion data sequence signal, identify the edge path whose starting point is located in the preset edge area, combine the strain signal to remove the edge path data caused by edge sealing interference, and use the permeation depth of the retained path to generate a corrected permeation behavior parameter signal; S5: Extract image signals from multi-source sensor fusion data sequence signals, enhance them by limiting contrast adaptive histogram equalization, and obtain defect detection signals by adaptive threshold segmentation; S6: The modified penetration behavior parameter signal and the defect detection signal are comprehensively evaluated to generate a comprehensive performance evaluation signal for the waterproof coating.

[0018] This invention provides an intelligent detection system and method for waterproof coating performance. By simultaneously acquiring multi-source signals, the moisture content signal is denoised in two steps using Kalman filtering and nonlocal mean denoising. A moisture content field is constructed based on the spatial location interpolation of the humidity sensor, and the permeation path is tracked along the gradient. Permeation initiation distribution statistics and strain time-series correlation are used to confirm and eliminate edge sealing interference. At the same time, contrast enhancement and adaptive threshold segmentation are used to detect coating surface defects. Finally, the corrected permeation behavior parameters and defect information are output, realizing an objective evaluation of the comprehensive performance of the waterproof coating. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A system architecture diagram of an intelligent performance testing system for waterproof coatings; Figure 2 A diagram illustrating the overall system data flow architecture of an intelligent performance testing system for waterproof coatings. Figure 3 Flowchart for penetration path construction and edge interference correction; Figure 4 A flowchart for parallel processing of moisture content noise reduction and defect detection; Figure 5 This is a flowchart of a method for intelligent testing of the performance of waterproof coatings. Detailed Implementation

[0021] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0023] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0024] Please see Figure 1-5 The image illustrates an intelligent detection system and method for waterproof coating performance according to the present invention. The intelligent detection system for waterproof coating performance includes a multi-source sensor data acquisition module, which acquires multi-source sensor fusion data sequence signals in real time; a moisture content data noise compensation module, which receives the moisture content signal from the multi-source sensor fusion data sequence signal, performs Kalman filtering and non-local mean denoising to generate a denoised moisture content signal, and transmits it to a permeation path map construction module; a permeation path map construction module, which receives the denoised moisture content signal, performs interpolation and gradient field calculation based on the spatial position of the humidity sensor array to generate a permeation path map signal, and transmits it to an edge interference identification and correction module; an edge interference identification and correction module, which receives the permeation path map signal and strain signal from the multi-source sensor fusion data sequence signal, identifies edge paths whose starting point is located in the edge region, removes the data corresponding to the edge paths, and generates a corrected permeation behavior parameter signal; and a coating surface defect detection module, which receives the image signal from the multi-source sensor fusion data sequence signal, and generates a defect detection signal through contrast-limited adaptive histogram equalization and adaptive threshold segmentation.

[0025] like Figure 1 As shown, an intelligent testing system for waterproof coating performance includes a multi-source sensor data acquisition module, a moisture content data noise compensation module, a permeation path map construction module, an edge interference identification and correction module, a coating surface defect detection module, and an evaluation module.

[0026] The multi-source sensor data acquisition module is the system's front-end sensing unit. This module synchronously acquires multi-dimensional raw physical signals reflecting the performance of the waterproof coating through various sensors deployed in the testing area. This module includes four types of sensors: a water pressure sensor, a humidity sensor array, an image acquisition device, and a strain sensor. The water pressure sensor is installed in the water pressure loading pipeline in the testing area. During the testing process, as the water pressure loading device applies water pressure to the surface of the waterproof coating in stages, the sensor collects the water pressure signal acting on the coating surface in real time. The water pressure signal reflects the applied load conditions during the testing process and is a fundamental parameter for judging the waterproof performance of the coating under different water pressure levels. The humidity sensor array consists of multiple humidity sensors embedded at different depths within the sample substrate at preset intervals. During the testing process, when moisture penetrates the waterproof coating and enters the substrate, sensors at different depths capture the distribution and dynamic changes of the moisture content within the substrate along the depth direction, thereby generating a moisture content signal. This signal is the core basis for evaluating the coating's water-blocking ability and penetration process. The image acquisition device is positioned facing the surface of the sample coated with the waterproof coating and captures images of the sample surface in real time during the testing process, generating image signals to provide raw image data for subsequent coating surface defect detection. Strain sensors are attached to the surface near the edge of the sample substrate to monitor minute deformations caused by moisture penetration or edge effects, generating strain signals. These strain signals serve as auxiliary information for the identification and verification of edge interference. Within the multi-source sensor data acquisition module, a time synchronization unit is also included. This unit uses a unified system clock as a reference to annotate each data frame of the water pressure, moisture content, image, and strain signals with a corresponding acquisition timestamp. After timestamp annotation, data frames belonging to the same acquisition time or with time deviations within a preset tolerance range are combined into a single multi-source sensor data frame. These frames are then arranged chronologically to form a multi-source sensor fusion data sequence signal. Through this time synchronization and data fusion processing, the multi-source sensor data acquisition module outputs a composite signal that is aligned in the time dimension and fused in the data dimension. This composite signal unifies the four heterogeneous data types—water pressure, moisture content, image, and strain—under the same time reference, providing consistent data input for subsequent processing modules.

[0027] like Figure 2As shown, the overall data flow of this system starts with multi-source sensor data acquisition, constructing a multi-branch, cyclical intelligent detection data processing link. Each module works collaboratively to achieve comprehensive and accurate detection of waterproof coating performance. The multi-source sensor data acquisition module, as the system's data input, integrates four types of sensors: water pressure, humidity array, image, and strain. It collects four types of raw data in real time: surface water pressure of the waterproof coating, internal moisture content of the substrate, coating surface image, and substrate edge strain. This provides a comprehensive and fundamental data source for subsequent analysis, and its acquisition frequency and synchronization accuracy directly determine the reliability of subsequent data processing. The time synchronization processing module receives the raw data output from the acquisition module. Using a unified system clock as a reference, it marks each data frame with a precise timestamp, integrating asynchronous data from different sensors into a time-aligned, format-uniform multi-source sensor fusion data sequence. This completely solves the problem of time sequence misalignment in multi-source data and ensures the effectiveness of subsequent cross-module data correlation analysis. As the core of the system's data scheduling, the data branch distribution module receives the fused data sequence and performs parallel splitting according to the data type. It sends the moisture content signal to the noise compensation module, the image signal to the image preprocessing module, and the strain signal to the designated storage unit, thereby realizing the professional classification and processing of data, avoiding mutual interference between different types of data, and improving the system's data processing efficiency.

[0028] The moisture content data noise compensation module receives the split moisture content signal and sequentially processes it through Kalman filtering and nonlocal mean denoising algorithms to remove random noise from the sensor and environmental disturbance noise, outputting a high-fidelity denoised moisture content signal to provide clean data support for permeation path analysis. The coating image preprocessing module receives the image signal and performs basic processing such as grayscale conversion and contrast enhancement to weaken image noise and enhance defect features, simplifying calculations and improving detection accuracy for subsequent defect detection. The strain data caching module temporarily stores the split strain signal and, after the permeation path map is generated, synchronously sends it to the edge interference identification and correction module for correlation verification of edge interference behavior, enabling on-demand data retrieval. The permeation path map construction module receives the denoised moisture content signal and, based on the spatial location of the humidity sensor array, generates a permeation path map signal that intuitively reflects the moisture permeation trend through interpolation, gradient field calculation, and path tracing algorithms, clearly presenting the internal moisture permeation trajectory and dynamic evolution process of the substrate. The edge interference identification and correction module simultaneously receives the penetration path map signal and the buffered strain signal. Through path region classification, proportion statistics, and correlation analysis, it accurately identifies and eliminates interference paths caused by edge sealing defects, outputting corrected penetration behavior parameters after interference removal to ensure that penetration performance evaluation is not affected by edge factors. The coating surface defect detection module receives the pre-processed image signal, performs threshold segmentation, feature extraction, and classification, identifies defects such as pinholes, cracks, and bubbles on the coating surface, and outputs defect detection signals to complete the coating surface quality inspection. The defect detection signal output module summarizes the defect type, location, and geometric parameters to form standardized defect detection results; the corrected penetration behavior parameter output module outputs core indicators such as effective penetration depth and penetration rate, providing quantitative basis for performance evaluation. The system also sets up a feedback loop from the edge interference identification and correction module to the multi-source sensor data acquisition module. When edge interference is detected, the subsequent data acquisition strategy can be dynamically adjusted to form a closed-loop optimization mechanism, continuously improving the stability and adaptability of the detection system and ensuring the intelligent and accurate operation of the entire waterproof coating performance testing process.

[0029] The moisture content data noise compensation module receives the moisture content signal from the multi-source sensor fusion data sequence and performs two-stage noise suppression processing on this signal to eliminate the influence of environmental disturbances and sensor noise on the infiltration behavior analysis. The first stage is Kalman filtering denoising, modeling the moisture content signal as a linear dynamic system. The state vector of this system contains at least the true moisture content value and the rate of change of moisture content, while the observed value is the actual moisture content reading output by the humidity sensor. In this model, the state transition equation describes the physical law of the evolution of the true moisture content value over time, and the observation equation describes the mapping relationship between the sensor reading and the true moisture content value. The recursive process of Kalman filtering includes two steps: prediction and update. In the prediction step, the prior state estimate for the current time step is calculated using the posterior state estimate and state transition matrix from the previous time step. Simultaneously, the prior error covariance for the current time step is calculated using the posterior error covariance and process noise covariance from the previous time step. In the update step, the Kalman gain is calculated using the prior error covariance, observation matrix, and observation noise covariance. Then, the state estimate is corrected using the residual between the current sensor observation and the prior state estimate, as well as the Kalman gain, to obtain the posterior state estimate for the current time step. The posterior error covariance is also updated for use in the next time step. The process noise covariance and observation noise covariance are pre-set based on the sensor's static calibration data and the actual disturbance level of the detection environment. After Kalman filtering, the linear Gaussian noise component in the moisture content signal is effectively suppressed, and the signal waveform tends to be smoother, but some non-stationary disturbance components may still remain. The second stage is nonlocal mean denoising. Using the water content signal sequence output by the Kalman filter as input, further refined denoising is performed on each sampling point in the sequence. For the current sampling point in the sequence, the signal segment consisting of a preset number of sampling points before and after that point is taken as the local neighborhood of that point. Simultaneously, within a search area centered on the current sampling point and within a preset search window length, the corresponding local neighborhood of each candidate point is extracted. The Gaussian weighted Euclidean distance between the local neighborhood of the candidate point and the local neighborhood of the current point is calculated, and the similarity weight of the candidate point is calculated based on this distance. The smaller the distance, the larger the weight, indicating a higher waveform similarity between the two local neighborhoods. The water content values ​​of all candidate points are weighted and averaged according to their respective similarity weights. The result is the final denoised water content value of the current sampling point. After traversing the above nonlocal mean denoising process for all sampling points, the output sequence is the denoised water content signal. The water content data noise compensation module transmits this denoised water content signal to the permeation path map construction module as input data for permeation path construction. The final formula for calculating the denoised moisture content value is as follows:

[0030] Accurate denoising of moisture content signals is fundamental to the reliability of data throughout the entire process of waterproof coating performance testing. It directly affects the accuracy of subsequent gradient calculation, path tracing, and correlation analysis. The nonlocal mean denoising algorithm, as the core of secondary denoising after Kalman filtering, relies on similarity weight calculation to achieve effective denoising and preserve the true characteristics of moisture content changes. This weight calculation formula is designed based on the principle of signal local feature similarity, adapting to the characteristics of low signal-to-noise ratio, gradual local changes, and the need to retain abrupt changes in moisture content time-series signals. In the formula, w(i,j) is the similarity weight of the j-th sampling point relative to the central sampling point i within the search window, with a value range of [0,1]. A larger weight indicates that the local moisture content characteristics of point j are more similar to those of point i, contributing more to weighted average denoising. , These are local moisture content signal segments extracted with the central sampling point i and the sampling point j within the search window as the core. The segment length is a first preset length, covering the local variation features of the moisture content signal and avoiding the influence of single-point noise on feature matching. The L2 norm is used to quantify the Euclidean distance between two local signal segment vectors. The smaller the distance, the closer the local water content change trends are. This is a Gaussian kernel smoothing parameter, whose value is positively correlated with the noise level of the water content signal. The higher the noise, the larger the value. It is used to control the weight decay rate and balance the denoising intensity and feature retention. The set of all sampling points within a search window containing the central sampling point i is defined, with a window length of a second preset length, balancing the comprehensiveness of local feature matching with computational efficiency. During calculation, the signal segment distances are first converted into similarity weights using a Gaussian kernel function; smaller distances result in larger weights. Then, all weights within the window are normalized to ensure a total weight sum of 1. This design fully utilizes the non-local similarity of the water content time-series signal, differing from the limitations of traditional local denoising algorithms that rely solely on single-point neighborhoods. It effectively suppresses sensor thermal noise, impulse noise from random environmental disturbances, and Gaussian noise, while fully preserving key features such as sudden changes and gradual shifts in water content during the infiltration process, avoiding the loss of true infiltration information due to excessive denoising. The water content signal, after weight calculation and weighted averaging using this formula, provides high signal-to-noise ratio and high fidelity foundational data for water content gradient calculation and strain-water content correlation analysis, ensuring the stability and accuracy of the waterproof coating infiltration performance detection algorithm chain from the source.

[0031] like Figure 3As shown, the penetration path construction and edge interference correction process is the core step in the penetration performance testing of waterproof coatings. Through step-by-step algorithm calculation, branching logic judgment, and iterative path tracing, the accurate restoration of the water penetration trajectory and the effective elimination of edge interference are achieved. Each processing module undertakes key calculation or judgment functions, ensuring the authenticity and reliability of the penetration analysis results. The noise-reduced moisture content signal receiving module serves as the starting point of the process, receiving pure moisture content data output from the moisture content data noise compensation module. This data has eliminated various noise interferences and truly reflects the internal moisture content distribution of the substrate, laying a high-quality data foundation for subsequent spatial modeling and gradient calculation. The two-dimensional spatial mesh construction module extracts the horizontal distance of each sensor from the central axis of the sample and the embedding depth in the substrate as spatial coordinates based on the actual installation layout of the humidity sensor array. The physical positions of the sensors are mapped to two-dimensional mesh nodes, constructing a regular mesh model that matches the internal space of the substrate, realizing the transformation of discrete sensor data into continuous spatial data, and providing spatial framework support for interpolation calculations. The interpolation module for generating the moisture content field selects the four nearest grid nodes for sensorless blank locations within the grid. It then uses a bidirectional linear interpolation algorithm to calculate the moisture content value at these blank locations, expanding the discrete node moisture content data into a continuous, smooth two-dimensional moisture content field. This fully restores the spatial distribution characteristics of the moisture content within the substrate, eliminating data gaps caused by sparse sensor distribution and ensuring that the penetration path analysis covers the entire detection area of ​​the substrate. The gradient calculation and penetration initiation point determination module, based on the continuous moisture content field, calculates the partial derivatives of the moisture content of each grid node along the horizontal distance and embedding depth using the central difference method. These are combined to form a moisture content gradient vector, clarifying the potential direction of moisture penetration. Simultaneously, it scans the time-series moisture content data along the coating-substrate interface, identifying locations where the moisture content abruptly changes and continuously increases as the penetration initiation point. This accurately locates the initial position of moisture penetration, providing starting coordinates for path tracking.

[0032] The gradient direction tracking permeation path module starts from the determined permeation initiation point and moves along the current gradient vector direction at a preset step size. Upon reaching a new position, it recalculates the local gradient vector and updates the movement direction, continuously iterating to track the permeation extension trajectory. This simulates the natural permeation process of water within the substrate, achieving dynamic generation of a single permeation path. The path termination condition module, acting as the logical control node for iterative tracking, determines in real time whether the path movement reaches the physical boundary of the sample or whether the gradient vector magnitude is below a preset lower limit. When either condition is met, the current path tracking terminates, avoiding invalid calculations and ensuring that each generated permeation path is a valid permeation trajectory, improving path generation efficiency and effectiveness. The permeation path map generation module summarizes the coordinate sequences of all valid permeation paths, integrating them into a complete permeation path map. This visually presents the distribution, extension direction, and spatial relationships of multiple water permeation trajectories within the substrate, providing visualized and data-driven map support for edge interference identification. The edge / center path classification module delineates the sample edge and center regions based on a preset edge width. Paths with permeation initiation points in the edge region are classified as edge paths, and paths in the center region are classified as center paths, achieving regional attribute classification of the paths and providing a classification basis for subsequent interference identification. The edge path proportion exceeding threshold module calculates the ratio of edge paths to total paths and compares this ratio with a preset interference judgment threshold to determine the presence of edge sealing interference. This initial screening for interference risks allows for rapid identification of detection scenarios with edge interference risks, avoiding resource waste caused by indiscriminate fine-grained verification. The strain-moisture content correlation verification module, after determining the presence of edge interference risks, extracts the substrate edge strain time-series data and shallow moisture content time-series data corresponding to the edge path starting point. After aligning the time axis, it calculates the linear correlation coefficient between the two, quantifying the temporal correlation between edge strain and moisture penetration. This distinguishes between interference penetration caused by edge sealing defects and actual coating penetration behavior, providing data basis for interference path elimination. The interference path elimination and parameter correction module removes interference path data caused by edge sealing defects based on the correlation verification results, retains the central path data that truly reflects the coating's impermeability performance, constructs a unified time axis through interpolation, calculates core parameters such as effective penetration depth and penetration rate, completes the correction of penetration behavior parameters, and finally outputs accurate and reliable quantitative indicators of penetration performance. This completely eliminates the impact of edge interference on the waterproof coating's penetration performance evaluation results, ensuring that the test results can truly reflect the coating's own impermeability.

[0033] The permeation path mapping module receives the denoised moisture content signal output by the moisture content data noise compensation module and generates a permeation path mapping signal based on the spatial positions of each sensor in the humidity sensor array within the substrate through spatial interpolation and gradient field calculation. First, the spatial position information of each sensor in the humidity sensor array is obtained. The spatial position of each sensor is determined by two coordinates: its horizontal distance from the sample's central axis and its embedding depth within the substrate. Using the horizontal distance as the first coordinate axis and the embedding depth as the second coordinate axis, each sensor is mapped as a node on a two-dimensional grid, constructing a spatial grid coordinate system for subsequent calculations. On the spatial grid, for any non-node location, its moisture content value is obtained by bilinear interpolation of the denoised moisture content values ​​of the four sensors closest to that location. The specific operation of bilinear interpolation is as follows: first, linear interpolation is performed along the first coordinate axis for two nodes at the same depth to obtain two intermediate moisture content values; then, linear interpolation is performed along the second coordinate axis for these two intermediate moisture content values ​​to finally obtain the interpolated moisture content value for that non-node location. After performing the above interpolation operation on all non-node locations within the entire spatial grid, a continuous water content field distribution is formed. On the interpolated water content field, for each grid node, the partial derivative of the water content at that node along the first coordinate axis is calculated using the water content values ​​of its adjacent nodes along the first coordinate axis. This calculation is done using central difference, i.e., the difference between the water content of the forward adjacent node and the backward adjacent node divided by twice the node spacing. Similarly, the partial derivative of the water content at that node along the second coordinate axis is calculated using the water content values ​​of its adjacent nodes along the second coordinate axis. The vector formed by these two partial derivatives is used as the water content gradient vector at that node. The direction of this gradient vector indicates the direction in which the water content increases most rapidly at that node, representing the direction of infiltration during water permeation. After obtaining the water content gradient field, infiltration path tracing is performed. The noise-reduced moisture content values ​​at different detection times are scanned sequentially from the center to the edge along the interface between the coating and the substrate. When the moisture content value at a certain location at a certain time shows an increase exceeding a preset abrupt change threshold relative to its initial dry state, and maintains an upward trend for multiple consecutive detection times thereafter, that location is identified as a penetration starting point. Using the identified penetration starting point as the starting point for tracking, the system moves along the gradient vector direction of that point at a preset tracking step size. Upon reaching a new location, the gradient vector at that point is recalculated using the moisture content interpolation of adjacent nodes, and the movement direction is updated accordingly. This process is repeated until a preset tracking termination condition is met. The tracking termination condition includes the current point reaching the sample boundary, or the gradient vector magnitude at the current point falling below a preset lower limit, indicating that the moisture content change at that point is no longer significant and penetration has essentially stopped. The sampling points obtained through the above process are connected to form a continuous penetration path curve.The above tracking operation is performed on each determined infiltration starting point to obtain multiple permeability path curves. The set of coordinate sequences of all these permeability path curves constitutes the permeability path map signal. The permeability path map construction module transmits this permeability path map signal to the edge interference identification and correction module. The formula for calculating the water content gradient vector is as follows:

[0034] In the formula, H(x,y) represents the partial derivative of the moisture content at the node with coordinates (x,y) in the two-dimensional space grid along the i-th coordinate direction (horizontal distance or embedding depth direction). This value directly reflects the rate of change of moisture content in the corresponding direction and is the core feature quantity for judging the trend of moisture penetration. H(x,y) is the true value of the moisture content inside the substrate corresponding to the grid node (x,y) after Kalman filtering and nonlocal mean denoising processing. It eliminates abnormal fluctuations caused by sensor random noise and environmental temperature and humidity disturbances, ensuring the reliability of the basic calculation data. α is the spacing between adjacent humidity sensor nodes in the i-th coordinate direction. Its value is determined according to the preset installation spacing of the humidity sensor array. The uniformity of the spacing directly affects the consistency of the partial derivative calculation. α is the second-order difference correction coefficient, which is usually set between 0.05 and 0.2. It is used to compensate for the error caused by the single central difference only considering the first-order change and ignoring the local moisture content curvature change, and to adapt to the actual characteristics of the nonlinear change of moisture content inside the waterproof coating substrate. The distance between adjacent nodes perpendicular to the i-th coordinate direction, and... A spatial scale benchmark for a two-dimensional grid is jointly constructed. During the calculation, the first-order partial derivative is obtained through central difference, and then a second-order difference term is introduced to correct the local moisture content gradient. This avoids computational redundancy caused by higher-order differences and effectively suppresses computational biases at grid edge nodes and regions with abrupt changes in moisture content. The combination of partial derivatives generated by this formula forms the moisture content gradient vector, whose direction directly serves as the direction of water penetration, providing precise directional guidance for subsequent penetration initiation determination and path tracking. From an algorithmic perspective, this ensures that the penetration path map accurately reflects the actual physical process of water penetration within the waterproof coating, avoiding path misjudgment and omissions caused by gradient calculation distortion. This provides reliable basic map data for subsequent edge interference identification and quantitative evaluation of penetration performance.

[0035] The edge interference identification and correction module receives the permeation path map signal output by the permeation path map construction module and simultaneously obtains strain signals from the multi-source sensor fusion data sequence signal. By analyzing the spatial distribution characteristics of the permeation paths and the correlation between strain and water content, it identifies and corrects edge sealing interference. This module first extracts the starting coordinates of each permeation path curve from the permeation path map signal. These starting coordinates are recorded by the permeation path map construction module during the permeation starting point determination process. Based on the pre-set geometric dimensions of the effective detection area of ​​the sample and the preset edge width set according to the width of the sample edge sealing strip, edge regions and central regions are delineated on a two-dimensional coordinate plane. The edge region is an annular area less than the preset edge width from the sample boundary, and the central region is the remaining area inside the sample boundary excluding the edge region. The starting coordinates of each permeation path are compared with the edge and central regions respectively. Permeation paths with starting coordinates located within the edge region are classified as edge paths, and those with starting coordinates located within the central region are classified as central paths. The total number of all permeation paths and the number of edge paths are counted, and the ratio of the number of edge paths to the total number of permeation paths is calculated to obtain the edge path proportion. The proportion of edge paths is compared with a preset interference threshold. When the proportion of edge paths exceeds the preset interference threshold, an edge sealing interference presence indicator signal is output. After outputting the edge sealing interference presence indicator signal, for each penetration path classified as an edge path, the strain measurement point corresponding to the sample surface is determined based on its starting coordinates. The strain time-series data of that point changing with the detection time is extracted from the strain signal in the multi-source sensor fusion data sequence signal. Simultaneously, the moisture content time-series data corresponding to the shallowest moisture sensor at the same starting position is extracted from the denoised moisture content signal output by the moisture content data noise compensation module. This sensor is the one embedded in the substrate closest to the coating position. The time axes of the strain time-series data and the moisture content time-series data are aligned to the same reference, and the linear correlation coefficient between the two sets of time-series data is calculated. The calculation process of the correlation coefficient includes: calculating the covariance and standard deviation of the strain time-series data and the moisture content time-series data, and dividing the covariance by the product of the two standard deviations to obtain the value of the correlation coefficient. When the absolute value of the correlation coefficient is greater than the preset correlation threshold, it indicates that the strain change and moisture content change at the edge path are significantly consistent in time, and there is a causal relationship between the two. Therefore, the seepage behavior at the edge path is determined to be caused by edge sealing defects, and the edge path is identified as an interference path. When the absolute value of the correlation coefficient is not greater than the preset correlation threshold, it indicates that there is no significant correlation between the seepage behavior at the edge path and insufficient edge sealing, and the edge path is retained.After confirming the interference paths, the edge interference identification and correction module deletes all edge path curves identified as interference paths and all their included sampling point data from the permeation path map signal. The deleted sampling point data includes the coordinates, water content, and permeation depth of each point on the path. Simultaneously, it removes the water content and water pressure records corresponding to the timestamps of the aforementioned edge path sampling points from the water content and corresponding water pressure signals in the multi-source sensor fusion data sequence signal. Using only the retained permeation depth data of all central paths, a common time axis is established at preset fixed time intervals. Linear interpolation is performed on the permeation depth sequences of each central path to obtain the permeation depth values ​​of each central path at each moment on the common time axis. The arithmetic mean of the permeation depths of all central paths at each moment is calculated as the effective permeation depth. The effective permeation rate is obtained by numerically differentiating the effective permeation depth sequence over time. The numerical differentiation uses the central difference method, i.e., the difference in effective permeation depth between two adjacent moments divided by the time interval. When the number of central paths involved in the calculation at a certain moment is zero, both the effective permeation depth and effective permeation rate at that moment are marked as invalid values. The effective permeation depth, effective permeation rate, and the number of retained center paths are combined to form a modified permeation behavior parameter signal, which is output to the external performance evaluation unit. The formula for calculating the covariance between strain time series data and water content time series data is as follows:

[0036] In the interference path verification stage of the edge interference identification and correction module, the calculation of the linear correlation coefficient between strain time-series data and moisture content time-series data is the core algorithmic support for distinguishing between edge sealing defect interference penetration and the actual penetration behavior of the coating, directly determining the accuracy and rationality of edge path elimination. This correlation coefficient calculation formula is designed based on the statistical Pearson correlation coefficient principle, adapting to the temporal correlation characteristics of substrate deformation and moisture penetration in the edge area of ​​the waterproof coating. In the formula, Let ε be the linear correlation coefficient between the strain time series and the moisture content time series H, with a value range of [−1, 1]. The closer the absolute value is to 1, the stronger the consistency of the temporal change trend of the two; the closer it is to 0, the weaker the correlation. The strain monitoring value of the substrate edge surface at the t-th sampling time is collected in real time by a strain sensor attached to the edge of the sample, reflecting the minute deformation of the substrate caused by moisture penetration or sealing defects. The arithmetic mean of the strain time series data is used to eliminate the DC component interference of the strain signal and focus on the dynamic change characteristics of the deformation. For the denoised moisture content value of the shallow layer of the substrate at the starting point of the edge path at the t-th sampling time, select the humidity sensor data closest to the coating to accurately correspond to the moisture penetration state of the coating-substrate interface. The arithmetic mean of the moisture content time-series data is used to eliminate the static baseline offset of the moisture content signal and highlight the dynamic changes in moisture penetration. T is the total number of synchronous sampling points for strain and moisture content time-series data, uniformly calibrated by the system time synchronization unit to ensure that the time axes of the two sets of data are completely aligned and to avoid correlation distortion caused by time misalignment. During calculation, the covariance of the two sets of data is first solved to quantify the degree of coordinated change between the two, and then normalized by dividing by the product of their respective standard deviations to eliminate the influence of differences in data dimensions. In actual testing, insufficient edge sealing can cause moisture to seep in along the edge gaps, simultaneously causing deformation of the substrate edge. At this time, the absolute value of the correlation coefficient will be significantly greater than the preset correlation threshold (usually set to 0.7), which can accurately identify the interference path. However, the moisture content change caused by actual coating penetration has no strong time correlation with edge strain, and the absolute value of the correlation coefficient is lower than the threshold, so the corresponding path is retained. This formula scientifically determines interference paths from the perspective of data correlation, avoiding the one-sidedness of judging solely by spatial location, significantly reducing the interference of edge sealing defects on the test results of waterproof coating permeability, and ensuring that the test data truly reflects the coating's own impermeability.

[0037] like Figure 4As shown, the parallel workflow for moisture content denoising and defect detection adopts a multi-task synchronous processing architecture, advancing the two core tasks of moisture content signal purification and coating surface defect detection in parallel. Through independent algorithm links, precise logical judgments, and iterative optimization mechanisms, it simultaneously achieves the purification of basic data for penetration analysis and the detection of coating surface quality. Each module has a clear division of labor and works together efficiently, significantly improving the overall detection efficiency and data quality of the system. The multi-source fusion data receiving module serves as the unified input to the parallel workflow, receiving the multi-source sensor fusion data sequence output by the time synchronization processing module. This data has undergone time alignment and format standardization, containing multiple key information such as moisture content and images, providing a unified and reliable data source for the parallel tasks and ensuring the consistency of the data foundation for the two parallel tasks. The parallel task triggering module, as the workflow scheduling hub, synchronously triggers the two independent task links of moisture content processing and image processing upon receiving the fused data, realizing real-time data splitting and parallel computation, avoiding the inefficiency caused by single-task serial processing, fully leveraging the parallel processing capabilities of the system's multiple modules, and shortening the overall detection cycle. The moisture content signal extraction module precisely filters moisture content-related data frames from the fused data sequence, removing irrelevant data such as water pressure, images, and strain, and outputs a pure raw moisture content signal. This provides target data for subsequent denoising processing, reduces interference from irrelevant information on the denoising algorithm, and improves the purification effect of the moisture content signal. Simultaneously, the image signal extraction module filters image data frames from the fused data sequence, eliminating other types of data interference and outputting the raw image signal of the coating surface. This provides a clear image data source for defect detection, ensuring the accuracy of defect feature extraction. The first-order Kalman filter denoising module, as the primary denoising unit for the moisture content signal, models the moisture content signal as a linear dynamic system, constructs state transition equations and observation equations, and predicts and corrects the estimated moisture content state value through recursive calculation. This effectively suppresses sensor observation noise and low-frequency environmental disturbance noise, completing the first round of noise filtering for the moisture content signal and significantly reducing the overall signal noise level, laying a good foundation for secondary denoising. The nonlocal mean weighted denoising module, as a deep denoising unit for the water content signal, extracts local signal segments centered on each sampling point for the signal after Kalman filtering. Within a preset search window, it calculates the weighted Euclidean distance between the local segments at each point and the segment at the center point. Based on the distance, it calculates similarity weights and performs weighted averaging. This fully utilizes the nonlocal similarity of the water content signal to remove impulse noise and high-frequency random noise, while completely preserving the real change characteristics such as sudden changes and gradual changes in water content. This avoids the loss of effective information due to excessive denoising and completes the deep purification of the water content signal.

[0038] The noise level compliance module serves as a quality control node for moisture content denoising. It evaluates the signal-to-noise ratio (SNR) of the denoised moisture content signal in real time to determine if the noise level meets the accuracy requirements of subsequent permeability analysis. If it fails to meet the standard, it feeds back to the first-order Kalman filter denoising module, initiating an iterative denoising process that dynamically optimizes denoising parameters until the signal quality meets the standard, ensuring that the output moisture content signal fully complies with the high-precision requirements of permeability path analysis. Once the noise level meets the standard, the output denoised moisture content signal module outputs a high SNR, high-fidelity denoised moisture content signal, providing core data support for permeability path mapping and ensuring the accuracy of permeability trajectory analysis. The grayscale conversion and contrast enhancement module, serving as a preprocessing unit for image processing, converts color image signals into grayscale images, simplifying data dimensions. Then, an adaptive histogram equalization algorithm with limited contrast is used to divide the grayscale image into multiple sub-blocks. The histograms of these sub-blocks are equalized and contrast limited, enhancing the grayscale difference between defective and normal areas on the coating surface, weakening background noise, and making defects such as pinholes, cracks, and bubbles more clearly prominent, creating favorable conditions for subsequent threshold segmentation and feature extraction. The adaptive threshold segmentation module uses the maximum inter-class variance method to calculate the optimal binarization threshold for the grayscale image. Based on the threshold, the image is divided into foreground defect areas and background normal areas, achieving accurate segmentation of defective regions and separating defects from complex backgrounds, facilitating subsequent quantitative analysis of defect features. The connected component feature matching module extracts geometric feature parameters such as area, perimeter, roundness, and aspect ratio of each connected component in the segmented binary image, establishing a defect feature database. The extracted feature parameters are matched with preset defect classification rules, providing a quantitative basis for defect type determination and avoiding subjective errors from manual judgment. The pinhole / crack / bubble defect classification module, based on feature matching results and preset threshold standards, classifies connected regions with high roundness and small area as pinhole defects, connected regions with low roundness and large aspect ratio as crack defects, and connected regions with moderate area and medium roundness as bubble defects. This achieves accurate classification of coating surface defects, clearly defining the defect type, location, and geometric parameters. The defect detection signal output module summarizes the defect classification results and related parameters, generating standardized defect detection signals that visually present the distribution of defects on the coating surface. This provides quantitative and visualized detection results for the assessment of the waterproof coating's surface quality. Ultimately, it achieves the parallel completion of basic data purification for the waterproof coating's permeability performance and surface defect detection, balancing detection accuracy and efficiency, and comprehensively supporting the intelligent and integrated detection of waterproof coating performance.

[0039] The coating surface defect detection module receives image signals from multi-source sensor fusion data sequences and generates defect detection signals through image enhancement and segmentation processing. First, the module converts the image signal from the red-green-blue color space to a grayscale image. A weighted average method is used to synthesize the brightness values ​​of the red, green, and blue color channels into a single grayscale value according to preset weights. The converted grayscale image is divided into several non-overlapping sub-blocks, and histogram equalization is performed on each sub-block to enhance local contrast. During histogram equalization, when the frequency of a certain grayscale level in the sub-block histogram exceeds a preset contrast threshold, the frequencies exceeding the threshold are truncated, and the truncated frequencies are evenly distributed across all grayscale levels. Then, equalization transformation is performed on the redistributed histogram. This process effectively suppresses excessive noise amplification while enhancing contrast. After processing each sub-block, bilinear interpolation is performed on the pixel values ​​at the boundaries of each sub-block to eliminate stitching artifacts between sub-blocks, resulting in an enhanced grayscale image. An adaptive binarization threshold is calculated using the maximum inter-class variance (MOV) method for the enhanced grayscale image. This method iterates through all possible grayscale values ​​in the image and automatically determines the optimal segmentation threshold based on maximizing the inter-class variance, thus maximizing the separation between the foreground defect region and the background normal region after segmentation. Based on this adaptive threshold, pixels with grayscale values ​​greater than the threshold are classified as background normal regions, and pixels with grayscale values ​​less than or equal to the threshold are classified as foreground defect regions, generating a binarized defect image. Geometric feature parameters are calculated for each connected region in the binarized defect image. These parameters include the area of ​​the connected region (the total number of pixels within it); the perimeter (the number of pixels at the boundary of the connected region); roundness (equal to four times pi multiplied by the area and then divided by the square of the perimeter); and the aspect ratio (the ratio of the longer side to the shorter side of the smallest bounding rectangle of the connected region). According to preset classification rules, connected components with roundness greater than a preset upper limit threshold and area less than a preset upper limit threshold for pinhole area are classified as pinhole defects; connected components with roundness less than a preset lower limit threshold and aspect ratio greater than a preset aspect ratio threshold are classified as crack defects; and connected components with area greater than a preset lower limit threshold for bubble area and roundness between the preset lower and upper limit thresholds for roundness are classified as bubble defects. The defect type, location coordinates in the image, and corresponding geometric feature parameters of all identified defects are combined to form a defect detection signal, which is then output.

[0040] In practical implementation, the evaluation module receives corrected permeation behavior parameter signals from the edge interference identification and correction module and defect detection signals from the coating surface defect detection module. Through internal spatial registration, correlation analysis, and grade determination units, it achieves a quantitative evaluation of the comprehensive performance of the waterproof coating, thus organically integrating the previously relatively independent permeation performance testing and surface defect detection into a unified technical solution. The primary task of the spatial registration unit is to align the coordinate systems of the two signals. The permeation path sampling points in the corrected permeation behavior parameter signal all have two-dimensional coordinates with the geometric center of the sample as a reference, typically including radial distance and depth. The defect area positions in the defect detection signal originate from image coordinates, which, after camera calibration and physical size conversion, can be mapped to the same sample coordinate system. The spatial registration unit uses a pre-calibrated transformation matrix to unify the coordinates of the permeation path sampling points and the contour coordinates of the coating surface defect areas into a rectangular coordinate system based on the sample substrate plane, and establishes a spatial correspondence between each permeation path sampling point and each defect area.

[0041] After coordinate alignment, the correlation analysis unit calculates the Euclidean distance between each penetration path sampling point and its nearest spatial defect region boundary. The defect region boundary can be represented by a set of defect contour points. The nearest distance reflects whether the penetration path is within the direct influence range of the defect. The system presets a correlation threshold, which is set comprehensively based on the coating type, substrate porosity, and test conditions. It is generally taken as a multiple of the defect feature size to ensure effective identification of abnormal penetration caused by defects such as coating cracking, pinholes, or bubbles. When the nearest distance corresponding to a penetration path sampling point is less than the preset correlation threshold, the correlation analysis unit marks that sampling point as a defect-correlated penetration point. After traversing all penetration path sampling points, the proportion of defect-correlated penetration points to the total number of penetration path sampling points is counted. If this proportion exceeds a preset threshold, such as 30%, a defect penetration correlation flag signal is output to indicate that coating surface defects have a significant impact on the overall penetration behavior.

[0042] Meanwhile, to quantify the severity of the defects themselves, the correlation analysis unit also calculates a defect severity index based on the geometric parameters in the defect detection signal. This index integrates the area and shape characteristics of the defect and can be characterized by a weighted sum of the defect area and the defect aspect ratio. The larger the area of ​​the defect and the more its aspect ratio deviates from the normal range, the greater the damage to the coating's protective ability. Therefore, the defect severity index can more comprehensively reflect the degree of harm caused by the defect. The rating unit, based on the defect penetration correlation indicator signal and defect severity index provided by the correlation analysis unit, combined with the effective penetration depth and effective penetration rate in the corrected penetration behavior parameter signal, determines the final performance rating of the waterproof coating according to preset comprehensive performance rating rules. The judgment rule has a clear logical hierarchy: First, the defect penetration association indicator signal is checked. If the signal indicates that the surface defects of the coating have a significant impact on the penetration behavior, the overall performance level of the waterproof coating is directly judged as unqualified, regardless of the current penetration depth and rate values. This is because the presence of defects has destroyed the continuity and barrier properties of the coating, which is very likely to lead to waterproofing failure under actual working conditions. If the defect penetration association indicator signal indicates that there is no significant impact, it means that the surface defects are not associated with the penetration path. At this time, the parameters of the penetration behavior itself are further examined. The effective penetration depth is compared with the preset depth threshold, and the effective penetration rate is compared with the preset rate threshold. Only when both are less than their respective thresholds is the overall performance level judged as qualified; otherwise, it is judged as unqualified.

[0043] This dual-judgment mechanism ensures that even if surface defects do not cause significant associated penetration, the coating's water-blocking ability still fails to meet usage requirements if the penetration depth or rate exceeds the standard's allowable range. The comprehensive performance evaluation signal of the waterproof coating ultimately generated by the rating judgment unit not only includes a pass / fail comprehensive performance rating but also integrates detailed data such as effective penetration depth, effective penetration rate, defect type, and defect severity index. This information can be received by external display devices or data management systems, allowing testing personnel to intuitively understand the coating's performance in different dimensions. Through the above-mentioned evaluation module settings, this invention deeply correlates the dynamic monitoring results of the penetration path with the static detection results of coating surface morphology defects at both spatial and logical levels, overcoming the drawbacks of traditional testing methods that treat the two separately. For example, in some cases, samples with small penetration depths may quickly lose their waterproof function in actual use due to surface cracking defects. Relying solely on penetration parameters for evaluation may overestimate the coating's performance, while this system can promptly identify such risks through correlation analysis. Furthermore, this evaluation module outputs a comprehensive judgment result rather than an isolated list of parameters, reducing the subjectivity and reliance on experience in human comprehensive analysis, and improving the objectivity, consistency, and automation of the detection. It is suitable for rapid quality inspection and performance rating in fields such as building waterproof coatings and industrial protective coatings. The entire evaluation process requires no manual intervention and can be seamlessly embedded into the multi-source sensor data acquisition and processing workflow, significantly improving the integrity and practicality of the intelligent detection system for waterproof coating performance. It allows the multi-source data obtained by the penetration performance detection subsystem and the surface defect detection subsystem to be integrated under a unified evaluation framework, ultimately generating a single comprehensive evaluation conclusion for the coating, thus overcoming the technical obstacle of singularity.

[0044] The six modules described above form a hierarchical progression in the signal flow. The multi-source sensor data acquisition module, located at the front end of the system, is responsible for synchronously acquiring various raw physical signals and performing time alignment and data fusion to form a unified multi-source sensor fusion data sequence signal as the system's common data source. The moisture content data noise compensation module receives the moisture content signal from the fused data sequence, performs two-stage denoising, and outputs a high-quality denoised moisture content signal. This denoised moisture content signal serves as the core input to the permeation path mapping construction module, generating a permeation path mapping signal through spatial interpolation and gradient field tracking. The permeation path mapping signal and the strain signal from the fused data sequence are jointly input to the edge interference identification and correction module. Edge interference is identified and removed through distribution statistics and correlation verification, outputting a corrected permeation behavior parameter signal. Simultaneously, the image signal from the fused data sequence is independently processed by the coating surface defect detection module, outputting a defect detection signal after enhancement and segmentation. The corrected permeation behavior parameter signal and the defect detection signal are ultimately transmitted to the system's performance evaluation stage, jointly supporting an objective evaluation of the waterproof coating's comprehensive performance from two dimensions: permeability and water-blocking capability, and surface integrity.

[0045] like Figure 5 As shown, this invention also provides an intelligent detection method for the performance of waterproof coatings, comprising: S1: real-time acquisition of water pressure signals, moisture content signals, image signals, and strain signals, and time synchronization processing of each signal to form a multi-source sensor fusion data sequence signal; S2: extraction of moisture content signals from the multi-source sensor fusion data sequence signal, recursive denoising through Kalman filtering, and then non-local mean denoising processing to generate a denoised moisture content signal; S3: interpolation of the denoised moisture content signal to form a moisture content field based on the spatial position of each sensor in the humidity sensor array within the substrate, calculation of the moisture content field gradient, tracing the permeation path along the gradient direction, and generating a permeation path map. S4: Receive the strain signal from the permeation path map signal and the multi-source sensor fusion data sequence signal, identify the edge path whose starting point is located in the preset edge area, combine the strain signal to remove the edge path data caused by edge sealing interference, and generate a corrected permeation behavior parameter signal using the permeation depth of the retained path; S5: Extract the image signal from the multi-source sensor fusion data sequence signal, enhance it by limiting contrast adaptive histogram equalization, and obtain the defect detection signal by adaptive threshold segmentation; S6: Comprehensively evaluate the corrected permeation behavior parameter signal and the defect detection signal to generate a comprehensive performance evaluation signal for the waterproof coating.

[0046] This invention provides an intelligent detection system and method for waterproof coating performance. By simultaneously acquiring multi-source signals, the moisture content signal is denoised in two steps using Kalman filtering and nonlocal mean denoising. A moisture content field is constructed based on the spatial location interpolation of the humidity sensor, and the permeation path is tracked along the gradient. Permeation initiation distribution statistics and strain time-series correlation are used to confirm and eliminate edge sealing interference. At the same time, contrast enhancement and adaptive threshold segmentation are used to detect coating surface defects. Finally, the corrected permeation behavior parameters and defect information are output to achieve an objective evaluation of the comprehensive performance of the waterproof coating.

[0047] Therefore, the intelligent detection system and method for waterproof coating performance of the present invention solves the problem of evaluation distortion caused by multi-source sensor noise and edge interference in the detection of waterproof coatings.

[0048] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An intelligent testing system for the performance of waterproof coatings, characterized in that, include: The multi-source sensor data acquisition module acquires multi-source sensor fusion data sequence signals in real time; The moisture content data noise compensation module receives the moisture content signal from the multi-source sensor fusion data sequence signal and generates a denoised moisture content signal through Kalman filtering and nonlocal mean denoising processing. The infiltration path map construction module receives the denoised moisture content signal, performs interpolation and gradient field calculation based on the spatial position of the humidity sensor array, and generates an infiltration path map signal. The edge interference identification and correction module receives the strain signal from the permeation path map signal and the multi-source sensor fusion data sequence signal, identifies the edge path whose starting point is located in the edge region, removes the data corresponding to the edge path, and generates a corrected permeation behavior parameter signal. The coating surface defect detection module receives the image signal from the multi-source sensor fusion data sequence signal and generates a defect detection signal by limiting contrast adaptive histogram equalization and adaptive threshold segmentation. The evaluation module receives the modified penetration behavior parameter signal and the defect detection signal, performs a comprehensive evaluation, and generates a comprehensive performance evaluation signal for the waterproof coating.

2. The intelligent detection system for waterproof coating performance according to claim 1, characterized in that, The permeation path mapping module obtains the spatial position of each sensor in the humidity sensor array. The spatial position includes the horizontal distance of the sensor from the central axis of the sample and the depth of the sensor embedded in the substrate. Using the horizontal distance as the first coordinate and the embedding depth as the second coordinate, each sensor is mapped as a two-dimensional grid node to construct a spatial grid. For any non-node location within the spatial grid, select the four nodes closest to that location. Using the denoised moisture content values ​​of these four nodes, first perform linear interpolation along the first coordinate direction to obtain a first intermediate value, then perform linear interpolation along the second coordinate direction on the first intermediate value to obtain the interpolated moisture content value for that non-node location, forming an interpolated moisture content field. In this interpolated moisture content field, for each grid node, use the moisture content values ​​of the node preceding and following it along the first coordinate direction to calculate the partial derivative of the moisture content at that node along the first coordinate direction. Similarly, use the moisture content values ​​of the node preceding and following it along the second coordinate direction to calculate the partial derivative of the moisture content at that node along the second coordinate direction. The partial derivative is calculated using the central difference method, and its value is equal to the difference in moisture content between two adjacent nodes in the corresponding direction divided by twice the distance between adjacent nodes. The vector formed by the two partial derivatives is used as the moisture content gradient vector for that node, and the direction of the gradient vector is determined as the infiltration propagation direction. The formula for calculating the moisture content gradient vector is as follows: ； in, Let H(x,y) be the partial derivative of the water content at grid node (x,y) along the i-th coordinate direction, and let H(x,y) be the denoised water content value at grid node (x,y). The distance between adjacent nodes in the i-th coordinate direction is... These are the second-order difference correction coefficients. The distance between adjacent nodes is perpendicular to the i-th coordinate direction.

3. The intelligent detection system for waterproof coating performance according to claim 2, characterized in that, The permeation path mapping module scans the moisture content values ​​at different times along the interface between the coating and the substrate from the center to the edge in the interpolated moisture content field. The position where the first increase in moisture content relative to the initial dry state exceeds a preset abrupt change threshold and continues to rise for multiple subsequent detection times is determined as the permeation start point. Using the determined permeation start point as the current point, the module moves along the gradient vector direction of the current point at a preset step size. Upon reaching a new position, the gradient vector of that point is obtained by interpolating the denoised moisture content of adjacent nodes, and the moving direction is updated accordingly. This process is repeated until a preset termination condition is met. Termination conditions include moving to the sample boundary or the magnitude of the gradient vector being lower than a preset lower limit, thereby tracing a continuous permeation path curve; the above tracing process is performed on each determined permeation starting point to obtain multiple permeation path curves, and the set of coordinate sequences of each permeation path curve is used as a permeation path spectrum signal; when the moisture content data noise compensation module outputs the denoised moisture content signal corresponding to the next level of water pressure loading, the permeation starting point determination and permeation path tracing are re-executed according to the updated denoised moisture content signal to generate a new permeation path spectrum signal, which is output in sequence form to form a dynamic evolution spectrum signal of permeation path in time series.

4. The intelligent detection system for waterproof coating performance according to claim 1, characterized in that, The edge interference identification and correction module receives a permeation path map signal from the permeation path map construction module. This permeation path map signal contains a sequence of starting coordinates for each permeation path curve. The starting coordinates are the position coordinates recorded by the permeation path map construction module when determining the permeation starting point. Based on the preset geometric dimensions of the effective detection area of ​​the sample and the preset edge width, an edge region and a central region are defined on a two-dimensional coordinate plane. The edge region is an annular region less than the preset edge width from the sample boundary, and the central region is the remaining region inside the boundary. The preset edge width is set according to the width of the sample edge sealing strip. The starting coordinates of each permeation path are compared with the edge region and the central region. Permeation paths whose starting coordinates fall within the edge region are classified as edge paths, and those whose starting coordinates fall within the central region are classified as central paths. The total number of all permeation paths and the number of edge paths among them are counted, and the ratio of the number of edge paths to the total number of permeation paths is calculated to obtain the edge path proportion. The edge path proportion is compared with a preset interference judgment threshold. When the edge path proportion exceeds the preset interference judgment threshold, an edge sealing interference presence flag signal is output, and subsequent correction operations are triggered.

5. The intelligent detection system for waterproof coating performance according to claim 4, characterized in that, After outputting the edge sealing interference indicator signal, the edge interference identification and correction module determines the corresponding sample surface measurement point based on the starting coordinates of each permeation path classified as an edge path in the permeation path map signal. From the strain signal in the multi-source sensor fusion data sequence signal, the strain time series data of the surface measurement point changing with the detection time is extracted. The strain signal is collected by a strain gauge installed on the surface of the sample substrate near the edge, and is used to monitor the small deformation of the substrate caused by moisture penetration. From the denoised moisture content signal output by the moisture content data noise compensation module, extract the moisture content time series data corresponding to the shallowest humidity sensor at the same starting point position. The shallowest humidity sensor is the sensor embedded in the substrate closest to the coating. Align the time axes of the strain time series data and the moisture content time series data to the same reference, and calculate the linear correlation coefficient between them. The calculation of the linear correlation coefficient includes: calculating the covariance and standard deviation of the strain time series data and the moisture content time series data, and dividing the covariance by the product of the two standard deviations to obtain the correlation coefficient value; when the absolute value of the correlation coefficient is greater than a preset correlation threshold, it is determined that the seepage behavior of the edge path is causally related to the strain change caused by insufficient edge sealing, and the edge path is identified as an interference path caused by edge sealing defects; when the absolute value of the correlation coefficient is not greater than the preset correlation threshold, it is determined that the seepage behavior of the edge path is not related to insufficient edge sealing, and the edge path is retained. The formula for calculating the covariance of the strain time series data and the moisture content time series data is as follows: ； in, For strain time series data With moisture content time series data The linear correlation coefficient, Let be the strain value at time t. The mean of the strain time series data, Let be the denoised moisture content value at time t. denoted as the mean of the moisture content time series data, and T represents the total number of sampling points in the time series data.

6. The intelligent detection system for waterproof coating performance according to claim 5, characterized in that, After receiving confirmed interference path information, the edge interference identification and correction module deletes all edge path curves confirmed to be caused by edge sealing defects and all their sampling point data from the permeation path map signal. The sampling point data includes the coordinates, moisture content, and permeation depth of each point on the path. Simultaneously, it removes the moisture content and water pressure records corresponding to the timestamps of the edge path sampling points from the moisture content and corresponding water pressure signals in the multi-source sensor fusion data sequence signal. Using the retained permeation depth data of all center paths, a common time axis is established at preset fixed time intervals to analyze the permeation depth sequence of each center path. Linear interpolation is used to obtain the penetration depth values ​​of each center path on the common time axis, and the arithmetic mean of the penetration depths of all center paths at each time point is calculated as the effective penetration depth. The effective penetration rate is obtained by numerically differentiating the sequence of the effective penetration depth over time. The numerical differentiation uses the central difference method to calculate the difference in effective penetration depth between two adjacent time points divided by the time interval. When the number of center paths involved in the calculation is zero at a certain time point, the effective penetration depth and effective penetration rate at that time point are marked as invalid values. The effective penetration depth, effective penetration rate, and the number of retained center paths are combined to form a corrected penetration behavior parameter signal, which is output to the evaluation unit.

7. The intelligent detection system for waterproof coating performance according to claim 1, characterized in that, The moisture content data noise compensation module models the moisture content signal as a linear dynamic system. The state vector contains the true moisture content and its rate of change, while the observed value is the output of the moisture content sensor. It constructs state transition equations and observation equations, and performs a first-stage denoising of the moisture content signal through Kalman filtering recursive calculation. This Kalman filtering recursive calculation includes predicting the current state using the state estimate from the previous time step, and then correcting and updating the predicted value based on the current observed value. The process noise covariance and observation noise covariance are pre-set based on sensor static calibration data and environmental disturbance levels. The module then processes the Kalman-filtered moisture content signal sequence. The second stage of nonlocal mean denoising is performed. In this stage, for each sampling point in the sequence, a signal segment of a first preset length is taken as the local neighborhood centered on that point. Within a search window containing that sampling point, the weighted Euclidean distance between the local neighborhoods of other points and the local neighborhood of the center point is calculated. Based on the weighted Euclidean distance, the similarity weight of each point is calculated. The moisture content values ​​of all points within the search window are weighted and averaged according to the similarity weight to obtain the final denoised moisture content value of that sampling point, thereby generating the denoised moisture content signal. The length of the search window is a second preset length, and the formula for calculating the final denoised moisture content value is as follows: ； in, Let j be the similarity weight of point j relative to the center point i within the search window. The value of the local moisture content signal segment at center point i is taken. Let j be the value of the local moisture content signal segment. It is the L2 norm. The parameters are Gaussian kernel smoothing parameters. The set of all sampled points within the search window. It is an exponential function.

8. The intelligent detection system for waterproof coating performance according to claim 1, characterized in that, The coating surface defect detection module converts the image signal from the red-green-blue color space into a grayscale image; divides the grayscale image into several non-overlapping sub-blocks, performs histogram equalization on each sub-block, and truncates the portion of the histogram of each sub-block that exceeds the preset contrast limit threshold. The truncated portion is then evenly distributed to all grayscale levels and equalized again to enhance the local contrast of the image and suppress noise, resulting in an enhanced grayscale image. The adaptive binarization threshold of the enhanced grayscale image is calculated using the maximum inter-class variance method, and the image is divided into foreground defect region and background normal region based on the threshold. Geometric feature parameters are calculated for each connected component in the binarized image. These parameters include the area, perimeter, roundness, and aspect ratio of the connected component. According to preset classification rules, connected components with roundness greater than a preset upper limit threshold and area less than a preset upper limit threshold for pinhole area are identified as pinhole defects. Connected components with roundness less than a preset lower limit threshold and aspect ratio greater than a preset aspect ratio threshold are identified as crack defects. Connected components with area greater than a preset lower limit threshold for bubble area and roundness between the preset lower and upper limit thresholds for roundness are identified as bubble defects. A defect detection signal containing defect type, location, and geometric parameters is generated.

9. The intelligent detection system for waterproof coating performance according to claim 1, characterized in that, The evaluation module includes a spatial registration unit, a correlation analysis unit, and a grade determination unit. The spatial registration unit receives the modified penetration behavior parameter signal and the defect detection signal, and maps the two-dimensional coordinates of each sampling point in the modified penetration behavior parameter signal and the position coordinates of each defect area in the defect detection signal to the same sample coordinate system, thereby establishing a spatial correspondence between the penetration path sampling points and the defect areas on the coating surface. The correlation analysis unit calculates the distance between each penetration path sampling point and the boundary of the nearest defect region based on the spatial correspondence. When the distance is less than a preset correlation threshold, the penetration path sampling point is marked as a defect-related penetration point. The proportion of defect-associated penetration points among all penetration path sampling points is statistically analyzed. When the proportion exceeds a preset threshold, it is determined that the coating surface defect has a significant impact on the penetration behavior, and a defect penetration association signal is output. Simultaneously, a defect severity index is calculated based on the geometric parameters of the defect in the defect detection signal. The defect severity index is a weighted sum of the defect area and the defect aspect ratio. The level determination unit receives the effective penetration depth and effective penetration rate from the corrected penetration behavior parameter signal, as well as the defect penetration association signal and defect severity index output by the association analysis unit. Based on preset comprehensive performance determination rules, the comprehensive performance level of the waterproof coating is determined, and a comprehensive performance evaluation signal of the waterproof coating containing the comprehensive performance level, the effective penetration depth, the effective penetration rate, the defect type, and the defect severity index is generated.

10. A method for intelligent testing system of waterproof coating performance according to any one of claims 1-9, characterized in that, include: S1: Real-time acquisition of water pressure signal, water content signal, image signal and strain signal, time synchronization processing of each signal to form a multi-source sensor fusion data sequence signal; S2: Extract the moisture content signal from the multi-source sensor fusion data sequence signal, perform recursive denoising through Kalman filtering, and then perform non-local mean denoising to generate a denoised moisture content signal. S3: Based on the spatial position of each sensor in the humidity sensor array within the substrate, the denoised moisture content signal is interpolated to form a moisture content field, the gradient of the moisture content field is calculated, the permeation path is traced along the gradient direction, and a permeation path map signal is generated. S4: Receive the strain signal from the permeation path map signal and the multi-source sensor fusion data sequence signal, identify the edge path whose starting point is located in the preset edge area, combine the strain signal to remove the edge path data caused by edge sealing interference, and use the permeation depth of the retained path to generate a corrected permeation behavior parameter signal. S5: Extract image signals from multi-source sensor fusion data sequence signals, enhance them by limiting contrast adaptive histogram equalization, and obtain defect detection signals by adaptive threshold segmentation; S6: The modified penetration behavior parameter signal and the defect detection signal are comprehensively evaluated to generate a comprehensive performance evaluation signal for the waterproof coating.

Images