Leak detection method and system for threading passage of fabricated concrete precast

Abstract

The application discloses a method and system for detecting leakage of a threaded passage of a fabricated concrete prefabricated part, and relates to the field of building structure health monitoring.The method comprises: filling the threaded passage with pressurized gas; collecting a signal of a change in gas pressure inside the passage over time by means of a micro-pressure sensor to obtain a micro-pressure fluctuation time sequence; performing adaptive signal decomposition processing on the sequence to eliminate low-frequency trend components and extract high-frequency detail components; performing phase space reconstruction on the high-frequency detail components and extracting nonlinear chaotic characteristic parameters; and inputting the characteristic parameters into a transfer learning classification model to output a classification result of a leakage state of the threaded passage.The application solves the technical problems of the prior art, such as difficulty in effectively separating weak leakage characteristics from environmental interference signals and insufficient detection accuracy, and achieves the technical effect of effectively separating weak leakage characteristics from environmental interference signals and improving the detection accuracy of the threaded passage leakage.

Description

Technical Field

[0001] This application relates to the field of building structural health monitoring, and in particular to a method and system for detecting leakage in wiring channels of precast concrete components. Background Technology

[0002] The sealing performance of wiring channels in precast concrete components directly affects the operational stability of the electrical system and the overall quality of the building project, making it a crucial testing aspect in the acceptance of precast building construction. Currently, the industry commonly employs traditional methods for leak detection in these channels, such as manual visual inspection, hydrostatic leakage testing, and conventional pressure maintenance observation. This involves sealing both ends of the channel, applying constant pressure, and manually recording pressure changes to determine a leak. However, these conventional testing methods are susceptible to environmental interference and human error. Pressure signal fluctuations are difficult to accurately distinguish, and minute leaks are hard to identify effectively. The reliability and sensitivity of the test results are significantly insufficient.

[0003] Currently, the detection of leaks in the wiring channels of precast concrete components faces technical challenges, including the difficulty in effectively separating weak leakage characteristics from environmental interference signals and insufficient detection accuracy. Summary of the Invention

[0004] This application provides a method and system for detecting leaks in the wiring channels of precast concrete components. The method involves filling the wiring channels with pressurized gas, collecting the gas pressure changes over time using a micro-pressure sensor to obtain a micro-pressure fluctuation time series, performing adaptive signal decomposition on this time series to remove low-frequency trend components caused by ambient temperature drift, and extracting high-frequency detail components related to leakage flow characteristics. The high-frequency detail components are then reconstructed in phase space, and nonlinear chaotic feature parameters such as the maximum Lyapunov exponent, correlation dimension, and recursion graph quantitative analysis indicators are extracted. These extracted nonlinear chaotic feature parameters are input into a pre-trained transfer learning classification model, which outputs a classification result indicating whether the wiring channel is sealed or leaking. This approach solves the technical problems of existing methods for detecting leaks in wiring channels of precast concrete components, such as the difficulty in effectively separating weak leakage characteristics from environmental interference signals and insufficient detection accuracy. The method effectively separates weak leakage characteristics from environmental interference signals, improving the accuracy of wiring channel leak detection.

[0005] This application provides a method for detecting leakage in the wiring channel of precast concrete components, comprising: filling the wiring channel of the precast concrete component to be tested with pressurized gas; acquiring the time-varying signal of the gas pressure inside the channel through pre-deployed micro-pressure sensors to obtain a micro-pressure fluctuation time series; performing adaptive signal decomposition processing on the micro-pressure fluctuation time series to remove low-frequency trend components caused by ambient temperature drift and extracting high-frequency detail components related to leakage flow characteristics; reconstructing the phase space of the high-frequency detail components and extracting nonlinear chaotic feature parameters, including the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative analysis index; inputting the nonlinear chaotic feature parameters into a pre-trained transfer learning classification model, and outputting the leakage state classification result of the wiring channel through the transfer learning classification model.

[0006] In a possible implementation, the high-frequency detail components are reconstructed in phase space, and nonlinear chaotic feature parameters are extracted. The following processing is performed: a mutual information function is calculated for the high-frequency detail components, and the delay time corresponding to the first local minimum of the mutual information function is selected as the reconstruction delay time; based on the high-frequency detail components, the proportion of false neighbor points in the temporary phase space under different embedding dimensions is calculated using the false nearest neighbor method, and the dimension corresponding to the first time the proportion of false neighbor points is lower than a preset convergence threshold is selected as the embedding dimension; a phase space trajectory matrix is ​​constructed based on the reconstruction delay time and the embedding dimension; the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative index of the phase space trajectory matrix are calculated to form the nonlinear chaotic feature parameters.

[0007] In a possible implementation, the following processing is performed: the quantitative indicators of the recursion graph include recursion rate, determination rate, and laminar flow rate; wherein, the recursion rate is determined by the ratio of the number of recursive points in the phase space trajectory matrix to the total number of phase points; the determination rate is determined by the ratio of the number of recursive points constituting the diagonal structure to the total number of recursive points; and the laminar flow rate is determined by the ratio of the number of recursive points constituting the vertical structure to the total number of recursive points.

[0008] In a possible implementation, the mutual information function of the high-frequency detail components is calculated, and the delay time corresponding to the first local minimum of the mutual information function is selected as the reconstruction delay time. The following processing is performed: the high-frequency detail component sequence is divided into multiple equal-length intervals, and the mutual information function in each of the multiple equal-length intervals is calculated; the minimum position of the mutual information function in each interval is statistically analyzed, and the average value of the intervals after removing abnormal intervals that deviate from the mean by more than two standard deviations is taken as the reconstruction delay time.

[0009] In a possible implementation, based on the high-frequency detail components, the proportion of false neighbor points in the temporary phase space under different embedding dimensions is calculated using the false nearest neighbor method. The dimension corresponding to the first time the proportion of false neighbor points falls below a preset convergence threshold is taken as the embedding dimension. The following processing is performed: a preset upper limit for the embedding dimension search is set to determine each candidate dimension; based on the high-frequency detail components, the Euclidean distance change rate between adjacent phase points under each candidate dimension is calculated sequentially using the false nearest neighbor method; phase points whose Euclidean distance change rate between adjacent dimensions exceeds a preset multiple are determined as false neighbor points, and the proportion of false neighbor points under each candidate dimension is counted; the smallest candidate dimension corresponding to the first time the proportion of false neighbor points falls below the preset convergence threshold is selected as the embedding dimension.

[0010] In a possible implementation, the following processing is performed: The construction of the transfer learning classification model includes a meta-training stage and a scenario adaptation stage: In the meta-training stage, a micro-pressure fluctuation sample set collected under multiple working conditions, including different cable fill rates, different leakage apertures, and leakage working conditions, is used to train a feature extractor based on a one-dimensional convolutional neural network, enabling the feature extractor to learn leakage-related general feature representations; In the scenario adaptation stage, for the cable channel to be detected, on-site labeled samples are obtained, and the feature extractor is updated with gradients a limited number of times using the on-site labeled samples to form a classification model adapted to the current working condition, which serves as the transfer learning classification model.

[0011] In a possible implementation, the micro-pressure fluctuation time series is subjected to adaptive signal decomposition processing to remove low-frequency trend components caused by ambient temperature drift and extract high-frequency detail components related to leakage flow characteristics. The following processing is performed: an adaptive signal decomposition algorithm is executed on the micro-pressure fluctuation time series to obtain multiple intrinsic components with different frequency characteristics; multiple long-term signal dependence statistical indices of the multiple intrinsic components are calculated; the multiple long-term signal dependence statistical indices are identified according to a preset index threshold, low-frequency trend components dominated by ambient temperature drift are removed, and the remaining intrinsic components are linearly superimposed and reconstructed to form the high-frequency detail components.

[0012] In a possible implementation, the following processing is performed: the long-term dependence statistic of the signal is the Hearst exponent, which characterizes the strength of the signal’s temporal correlation.

[0013] In a possible implementation, before linearly superimposing and reconstructing the remaining intrinsic components, the following processing is performed: the proportion of the total energy of the remaining intrinsic components to the energy of the original micro-pressure fluctuation signal of the micro-pressure fluctuation time series is calculated. When the proportion is lower than a preset lower limit, it is determined that the leakage characteristic component in the current signal is insufficient, and the sampling time is extended for resampling.

[0014] This application also provides a leakage detection system for the wiring channel of precast concrete components, comprising: a pressure acquisition module for filling pressurized gas into the wiring channel of the precast concrete component to be tested, and acquiring the time-varying signal of the gas pressure inside the channel through pre-deployed micro-pressure sensors to obtain a micro-pressure fluctuation time series; a signal decomposition and processing module for performing adaptive signal decomposition processing on the micro-pressure fluctuation time series, removing low-frequency trend components caused by ambient temperature drift, and extracting high-frequency detail components related to leakage flow characteristics; a nonlinear chaotic feature parameter extraction module for reconstructing the phase space of the high-frequency detail components and extracting nonlinear chaotic feature parameters, including the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative analysis index; and a leakage state classification module for inputting the nonlinear chaotic feature parameters into a pre-trained transfer learning classification model, and outputting the leakage state classification result of the wiring channel through the transfer learning classification model.

[0015] The proposed method and system for detecting leaks in the wiring channels of precast concrete components involves first filling the wiring channels of the precast concrete component under test with pressurized gas. Using pre-deployed micro-pressure sensors, the system collects the time-varying signals of the gas pressure inside the channel, obtaining a micro-pressure fluctuation time series. Next, the micro-pressure fluctuation time series undergoes adaptive signal decomposition processing to remove low-frequency trend components caused by ambient temperature drift and extract high-frequency detail components related to leakage flow characteristics. Then, the high-frequency detail components are reconstructed in phase space, and nonlinear chaotic feature parameters are extracted. These nonlinear chaotic feature parameters include the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative analysis indicators. Finally, the nonlinear chaotic feature parameters are input into a pre-trained transfer learning classification model, which outputs the leakage state classification result of the wiring channel. The proposed method and system effectively separate weak leakage characteristics from environmental interference signals, improving the technical accuracy of wiring channel leak detection. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the method for detecting leakage in the wiring channels of precast concrete components provided in this application.

[0018] Figure 2 This is a schematic diagram of the structure of the wire passage leakage detection system for precast concrete components provided in this application embodiment.

[0019] Figure labeling: 10 pressure acquisition module, 20 signal decomposition and processing module, 30 nonlinear chaotic feature parameter extraction module, 40 leakage state classification module. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments; however, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a method for detecting leakage in the wiring channels of precast concrete components, such as... Figure 1 As shown, the method includes:

[0024] Step S100: Pressurized gas is injected into the wiring channel of the precast concrete component to be tested. The gas pressure inside the channel is collected over time by a pre-installed micro-pressure sensor to obtain a micro-pressure fluctuation time series.

[0025] Specifically, the wiring channels of the precast concrete components to be tested are sealed using sealing plugs to completely block both ends of the channels, ensuring complete isolation between the channel and the external environment and preventing external gases from entering or internal pressurized gas from leaking and affecting testing accuracy. After sealing, pressurized gas is injected into the wiring channels using a high-pressure air pump. Dry air is used to avoid moisture interference with the testing. Pressure changes are monitored in real time during inflation until the pressure inside the channel reaches a preset value, at which point inflation stops. Before inflation, micro-pressure sensors are pre-installed inside the wiring channels. These are miniature piezoelectric sensors, with their probes fixed in the middle of the channel to accurately collect the real-time changes in gas pressure. The sensors are connected to a data acquisition instrument via a data transmission line. After inflation, the data acquisition instrument is activated, controlling the micro-pressure sensors to collect the time-varying signals of the gas pressure inside the channel at a preset sampling frequency. The data acquisition instrument converts the collected pressure signals into digital signals and organizes them chronologically to form a micro-pressure fluctuation time series. This time series contains the gas pressure values ​​inside the channel at each sampling moment.

[0026] Step S200: Perform adaptive signal decomposition processing on the micro-pressure fluctuation time series, remove low-frequency trend components caused by ambient temperature drift, and extract high-frequency detail components related to leakage flow characteristics.

[0027] Specifically, the adaptive signal decomposition algorithm uses ensemble empirical mode decomposition (EMD). First, the micro-pressure fluctuation time series is initialized by removing outliers. Outliers are defined as values ​​deviating from the series mean by more than three standard deviations. These outliers are replaced with the average of two adjacent normal values. Then, the number of decomposition iterations is set, for example, 12. During each iteration, intrinsic mode components (IMCs) are extracted from the series through a filtering process until the remaining components are monotonic. This yields multiple IMCs with different frequency characteristics, each corresponding to a signal component within a different frequency range. Low-frequency components are mainly caused by ambient temperature drift, while high-frequency components are mainly related to gas flow characteristics caused by leakage in the wiring channel. After decomposition, further filtering and reconstruction yield pure high-frequency detail components.

[0028] In one possible implementation, the micro-pressure fluctuation time series is subjected to adaptive signal decomposition processing to remove low-frequency trend components caused by ambient temperature drift and extract high-frequency detail components related to leakage flow characteristics. Step S200 further includes step S210, which executes an adaptive signal decomposition algorithm on the micro-pressure fluctuation time series to obtain multiple intrinsic components with different frequency characteristics. Specifically, the first step is to preprocess the micro-pressure fluctuation time series to remove outliers. Outliers are identified by calculating the mean and standard deviation of the micro-pressure fluctuation time series, and identifying points with values ​​greater than the mean plus three standard deviations or less than the mean minus three standard deviations as outliers. These outliers are then replaced with the arithmetic mean of the values ​​of the two preceding and following sampling points. The second step is to initialize the intrinsic component set to an empty set, using the preprocessed micro-pressure fluctuation time series as the initial raw sequence. The third step is to filter the initial raw sequence, identifying all local maxima and local minima. Cubic spline interpolation is used to fit these local maxima and local minima to obtain the upper and lower envelopes. The average of the upper and lower envelopes is then calculated to obtain the envelope mean line. The fourth step is to subtract the envelope mean line from the raw sequence. The process involves several steps: First, a candidate intrinsic component is obtained. Then, it is determined whether this candidate intrinsic component meets the criteria for intrinsic modal components. The criteria are: in the entire sequence, the number of local maxima and local minima is equal or differs by no more than one; and at any given time, the upper and lower envelopes are symmetrically distributed about the time axis. Second, if the candidate intrinsic component meets the criteria, it is added to the intrinsic component set, and the original sequence is updated to the remaining sequence after subtracting the candidate intrinsic component. If the criteria are not met, the candidate intrinsic component is used as the new original sequence, and the screening process of steps three and four is repeated until an intrinsic component that meets the criteria is obtained. Third, steps three and five are repeated until the remaining sequence is a monotonic trend component and cannot be further decomposed into intrinsic components that meet the criteria. Finally, multiple intrinsic components with different frequency characteristics are obtained.

[0029] Step S220: Calculate multiple long-term signal dependence statistical indices for the multiple intrinsic components. These long-term signal dependence statistical indices are the Hearst exponents, which characterize the strength of the signal's temporal correlation. Specifically, the Hearst exponent is used to characterize the strength of the signal's temporal correlation, and its value ranges from 0 to 1. Generally, when the Hearst exponent is greater than 0.5, it indicates that the signal has a long-term positive correlation, meaning that the future trend of the signal is consistent with the past; when the Hearst exponent is equal to 0.5, it indicates that the signal is a random sequence with no long-term correlation; when the Hearst exponent is less than 0.5, it indicates that the signal has a long-term negative correlation, meaning that the future trend of the signal is opposite to the past. The Hearst exponent is calculated using rescaled range analysis, specifically as follows: First, for each intrinsic component, it is divided into multiple subsequences of equal length. If the total number of sampling points for an intrinsic component is not divisible by the subsequence length, the last sampling point less than the length of a subsequence is discarded. Second, the mean of each subsequence is calculated, and then the deviation of each sampling point within each subsequence from the mean of that subsequence is calculated, resulting in a deviation sequence. Third, the cumulative deviation of each deviation sequence is calculated, and then the range of each subsequence is calculated, which is the difference between the maximum and minimum values ​​of the cumulative deviation. Fourth... The first step is to calculate the standard deviation of each subsequence, and then divide the range of each subsequence by its standard deviation to obtain the rescaled range. The fifth step is to calculate the average of the rescaled ranges of all subsequences as the mean of the rescaled range of the intrinsic component. The sixth step is to change the length of the subsequence and repeat the process from the first step to the fifth step to obtain the mean of the rescaled range corresponding to different subsequence lengths. The seventh step is to take the logarithm of each subsequence length and the corresponding mean of the rescaled range, and then use the least squares method to perform linear fitting on the logarithmized values. The slope of the fitted line is the Hearst exponent of the intrinsic component.

[0030] Step S230: Identify the long-term dependence statistical indicators of the multiple signals based on a preset index threshold, remove low-frequency trend components dominated by ambient temperature drift, and reconstruct the remaining intrinsic components by linear superposition to form the high-frequency detail components. Specifically, a preset index threshold for the Hurst exponent is set. This threshold is determined based on experimental data, i.e., statistically analyzing the range of Hurst exponents of low-frequency components caused by ambient temperature drift under leak-free conditions, and using the maximum value of this range as the preset index threshold. For example, experiments have shown that the Hurst exponents of low-frequency components dominated by ambient temperature drift are all greater than 0.55, so the preset index threshold is set to 0.55. Then, the relationship between the Hurst exponent of each intrinsic component and the preset index threshold is determined one by one. Intrinsic components with a Hurst exponent greater than the preset index threshold are identified as low-frequency trend components dominated by ambient temperature drift and are removed. Intrinsic components with a Hurst exponent less than or equal to the preset index threshold are identified as components related to leakage flow characteristics and are retained. Finally, the remaining intrinsic components are linearly superimposed and reconstructed, that is, each retained intrinsic component is directly added together according to its original amplitude to obtain the reconstructed signal, which is the high-frequency detail component related to the leakage flow characteristics.

[0031] In one possible implementation, before linearly superimposing and reconstructing the remaining intrinsic components, step S200 further includes step S240, calculating the proportion of the total energy of the remaining intrinsic components to the energy of the original micro-pressure fluctuation signal in the micro-pressure fluctuation time series. When the proportion is lower than a preset lower limit, it is determined that the leakage characteristic component in the current signal is insufficient, and the sampling time is extended for resampling. Specifically, the energy of the original micro-pressure fluctuation signal is calculated by squaring the value of each sampling point in the original micro-pressure fluctuation time series, and then summing all the squared values ​​to obtain the original signal energy. Then, the total energy of the remaining intrinsic components is calculated by calculating the energy of each remaining intrinsic component separately, using the same method as for the original signal energy, i.e., summing the squares of the sampling point values ​​of each intrinsic component, and then summing the energies of all remaining intrinsic components to obtain the total energy of the remaining intrinsic components. Finally, the ratio of the total energy of the remaining intrinsic components to the energy of the original signal is calculated to obtain the energy proportion. The preset lower limit is determined based on experimental data. When the calculated energy ratio is lower than the preset lower limit, it is determined that the leakage characteristic component in the current signal is insufficient. At this time, it is necessary to extend the sampling time for resampling. The resampling time is set to 1.5 to 2 times the original sampling time. The resampling process is consistent with the sampling process in step S100. The same micro-pressure sensor and sampling frequency are used to re-acquire the micro-pressure fluctuation time series. Then, the processing process from step S200 to step S230 is repeated until the total energy ratio of the remaining intrinsic components is not lower than the preset lower limit.

[0032] Step S300: The phase space of the high-frequency detail components is reconstructed, and nonlinear chaotic feature parameters are extracted. The nonlinear chaotic feature parameters include the maximum Lyapunov exponent, correlation dimension, and recursion graph quantitative analysis index.

[0033] Specifically, phase space reconstruction maps a one-dimensional high-frequency detail component time series into a high-dimensional space by setting a delay time and embedding dimension, constructing a phase space trajectory that reflects the chaotic characteristics of the time series, thus more clearly presenting the variation patterns of the high-frequency detail components. Nonlinear chaotic characteristic parameters are quantitative indicators used to characterize the chaotic characteristics of the high-frequency detail components. These parameters can effectively distinguish the signal differences between leaky and non-leaking states. Specifically, they include the maximum Lyapunov exponent, correlation dimension, and recursion graph quantitative analysis indicators. The maximum Lyapunov exponent characterizes the divergence or convergence speed of the phase space trajectory, reflecting the strength of the signal's chaotic characteristics; the maximum Lyapunov exponent of leaked signals typically has a specific range. The correlation dimension characterizes the complexity of the phase space trajectory, reflecting the spatial distribution characteristics of leak-related signals. The recursion graph quantitative analysis indicators are quantitative parameters extracted after constructing a recursion graph based on the phase space trajectory, used to further characterize the recursive characteristics of the signal and assist in determining the leak state. This step first determines the key parameters for phase space reconstruction by calculating the delay time and embedding dimension, then constructs the phase space trajectory matrix based on these parameters, and finally calculates various nonlinear chaotic characteristic parameters based on this matrix.

[0034] In one possible implementation, the high-frequency detail components are reconstructed in phase space, and nonlinear chaotic feature parameters are extracted. Step S300 further includes step S310, calculating the mutual information function for the high-frequency detail components, and selecting the delay time corresponding to the first local minimum of the mutual information function as the reconstruction delay time. Specifically, the mutual information function is used to measure the correlation between two signals, and here it is used to measure the correlation between the high-frequency detail component sequence and its own delayed sequence. The delay time is the number of sampling points of the sequence delay. The mutual information function is calculated as follows: First, determine the search range of the delay time; Second, for each candidate delay time, delay the high-frequency detail component sequence by that number of sampling points to obtain the delayed sequence. For example, if the candidate delay time is 5 sampling points, delay the value of each sampling point in the high-frequency detail component sequence by 5 sampling points to obtain the delayed sequence. The first 5 sampling points of the delayed sequence are empty and filled with 0; Third, divide the high-frequency detail component sequence and the corresponding delayed sequence into multiple equal-length intervals; Fourth, calculate the joint probability distribution of the high-frequency detail component sequence and the delayed sequence and their respective marginal probability distributions within each interval; Fifth, based on the joint... The mutual information value for the interval is calculated using the probability distribution and the marginal probability distribution. The calculation of the mutual information value follows the mutual information formula in information theory, that is, the mutual information value is equal to the log-weighted sum of the joint probability distribution minus the log-weighted sum of the two marginal probability distributions. The sixth step is to calculate the average of the mutual information values ​​of all intervals as the mutual information function value corresponding to the candidate delay time. The seventh step is to repeat the process from the second to the sixth step to calculate the mutual information function value corresponding to each candidate delay time, and obtain the curve of the mutual information function changing with the delay time. The eighth step is to find the position where the mutual information function first reaches a local minimum from the curve, and the delay time corresponding to this position is the reconstruction delay time.

[0035] Step S320: Based on the high-frequency detail components, the proportion of false neighbor points in the temporary phase space under different embedding dimensions is calculated using the false nearest neighbor method. The dimension corresponding to the first time the proportion of false neighbor points falls below a preset convergence threshold is taken as the embedding dimension. Specifically, the embedding dimension is used to determine the dimension of the phase space reconstruction, ensuring that the phase space can accurately reflect the chaotic characteristics of the high-frequency detail components. The specific execution is as follows: First, determine the search range of the embedding dimension; Second, for each candidate embedding dimension, construct a temporary phase space based on the high-frequency detail component sequence and the reconstruction delay time. The method is as follows: For a high-frequency detail component sequence of length N, when the embedding dimension is m and the reconstruction delay time is τ, the temporary phase space consists of N-(m-1)τ phase points, and the coordinates of each phase point are x(i), x(i+τ), x(i+2τ), ..., x[i+(m-1)τ], where i ranges from 1 to N-(m-1)τ, and x(i) is the value of the i-th sampling point of the high-frequency detail component sequence; Third, use the false nearest neighbor method to calculate the proportion of false neighbor points in the temporary phase space. Specifically, for each phase point, find its nearest neighbor in the temporary phase space. The process involves six steps: 1) Calculate the Euclidean distance between two phase points, then increase the embedding dimension by one to construct a new temporary phase space. Calculate the Euclidean distance between the phase point and its original nearest neighbor in the new phase space, and calculate the ratio of the two Euclidean distances. If this ratio is greater than a preset threshold, the nearest neighbor is determined to be a false neighbor. 2) Count the number of false neighbors in the temporary phase space and calculate the ratio of false neighbors to the total number of phase points to obtain the false neighbor ratio. 3) Repeat steps two through four to calculate the false neighbor ratio corresponding to each candidate embedding dimension. 4) Set a preset convergence threshold, determined based on experimental data, and find the minimum candidate embedding dimension corresponding to the first time the false neighbor ratio falls below the preset convergence threshold. This dimension is the embedding dimension.

[0036] Step S330: Construct a phase space trajectory matrix based on the reconstruction delay time and the embedding dimension. Specifically, obtain a high-frequency detail component sequence, which contains multiple sampling point values ​​arranged in chronological order. Calculate the number of rows and columns of the phase space trajectory matrix. The number of rows = total number of sampling points - (embedding dimension - 1) × reconstruction delay time. This number of rows represents the number of phase points. The number of columns equals the embedding dimension. Each row of the matrix corresponds to a phase point, and each column corresponds to a coordinate of the phase point. The element in the nth row and jth column of the matrix is ​​the value of the [n+(j-1)×reconstruction delay time]th sampling point in the high-frequency detail component sequence. The value of n ranges from 1 to [total number of sampling points - (embedding dimension - 1)×reconstruction delay time], and the value of j ranges from 1 to the embedding dimension.

[0037] Step S340: Calculate the maximum Lyapunov exponent, correlation dimension, and recursion graph quantitative indicators of the phase space trajectory matrix to form the nonlinear chaotic characteristic parameters. The recursion graph quantitative indicators include recursion rate, determination rate, and laminar flow rate. The recursion rate is determined by the ratio of the number of recursive points in the phase space trajectory matrix to the total number of phase points. The determination rate is determined by the ratio of the number of recursive points forming a diagonal structure to the total number of recursive points. The laminar flow rate is determined by the ratio of the number of recursive points forming a vertical line structure to the total number of recursive points.

[0038] Specifically, the calculation of the maximum Lyapunov exponent employs a small-data-volume method: Multiple initial phase points are selected from the phase space trajectory matrix, for example, 15 initial phase points are set and evenly distributed throughout the phase space trajectory. For each initial phase point, its nearest neighbor phase point in phase space is found, and the initial distance between the two phase points is calculated. The evolution trajectory of the two phase points over time is tracked, and the distance between the two phase points at different times is calculated. The natural logarithm of the distance at each time point is taken, and the relationship between the logarithmic distance and time is linearly fitted using the least squares method. The slope of the fitted line is the Lyapunov exponent corresponding to that initial phase point. The average value of the Lyapunov exponents corresponding to all initial phase points is calculated as the maximum Lyapunov exponent of the phase space trajectory matrix.

[0039] The correlation dimension is calculated using the correlation integral method, specifically: A search range for the correlation radius is set, ranging from 0.1 to 10 times the standard deviation of the high-frequency detail component sequence. For each correlation radius r, the Euclidean distance between all phase pairs in the phase space trajectory matrix is ​​calculated, and the number of phase pairs with an Euclidean distance less than or equal to r is counted, denoted as C(r). The correlation integral is calculated as the ratio of C(r) to the total number of phase pairs, which is M×(M-1) / 2, where M is the number of rows in the phase space trajectory matrix. For different correlation radii r and their corresponding correlation integrals, the logarithm is taken, and the logarithmized values ​​are linearly fitted using the least squares method. The slope of the fitted line is the correlation dimension.

[0040] The calculation of quantitative indicators of the recursion graph first involves constructing a recursion graph. The method for constructing the recursion graph is as follows: the number of rows M of the phase space trajectory matrix is ​​used as the horizontal and vertical axes, and both the horizontal and vertical axes correspond to the index of the phase points. When the Euclidean distance between two phase points is less than or equal to the preset recursion threshold, the corresponding coordinate point is marked in black; otherwise, it is marked in white, thus forming a recursion graph. The preset recursion threshold is set to 0.5 to 1 times the standard deviation of the high-frequency detail component sequence. Then, based on the recursion graph, the recursion rate, the determination rate, and the laminar flow rate are calculated: The recursion rate is determined by the ratio of the number of recursive points in the phase space trajectory matrix to the total number of phase points. The recursive points are the coordinate points marked in black in the recursion graph. The calculation method is to divide the number of recursive points by the total number of phase points. The determination rate is determined by the ratio of the number of recursive points forming a diagonal structure to the total number of recursive points. A diagonal structure refers to a diagonal line formed by three or more consecutive black points in the recursion graph. The determination rate is obtained by counting the number of all recursive points forming a diagonal structure and dividing it by the total number of recursive points. The laminar flow rate is determined by the ratio of the number of recursive points forming a vertical line structure to the total number of recursive points. A vertical line structure refers to a vertical line formed by three or more consecutive black points in the recursion graph. The laminar flow rate is obtained by counting the number of all recursive points forming a vertical line structure and dividing it by the total number of recursive points.

[0041] In one possible implementation, the mutual information function is calculated for the high-frequency detail components, and the delay time corresponding to the first local minimum of the mutual information function is selected as the reconstruction delay time. Step S310 further includes step S311, which involves dividing the high-frequency detail component sequence into multiple equal-length intervals and calculating the mutual information function within each interval; statistically analyzing the minimum position of the mutual information function in each interval, removing outlier intervals that deviate from the mean by more than two standard deviations, and taking the average as the reconstruction delay time. Specifically, for each interval, the mutual information function corresponding to that interval is calculated according to the mutual information function calculation method in step S310, and the delay time corresponding to the first local minimum of the mutual information function in each interval is found, i.e., each interval corresponds to a candidate delay time. Calculate the mean and standard deviation of all candidate delay times. Compare each candidate delay time with the mean. If a candidate delay time deviates from the mean by more than twice the standard deviation, the interval is considered an outlier and is removed. Take the arithmetic mean of the remaining candidate delay times after removing outliers as the final reconstructed delay time. This improves the accuracy of delay time determination and avoids the influence of outlier data from a single interval on the results.

[0042] In one possible implementation, based on the high-frequency detail components, the proportion of false neighbors in the temporary phase space under different embedding dimensions is calculated using the false nearest neighbor method. The dimension corresponding to the first time the proportion of false neighbors falls below a preset convergence threshold is taken as the embedding dimension. Step S320 further includes step S321, which presets an upper limit for the embedding dimension search and determines each candidate dimension. Specifically, based on the chaotic characteristics of the high-frequency detail components, an upper limit for the embedding dimension search is preset to ensure that the search range can cover the true embedding dimension and avoid the inability to find a suitable embedding dimension due to insufficient search range. Then, each candidate dimension is determined, starting from one dimension and increasing sequentially to the preset upper limit. Each candidate dimension is used as a candidate embedding dimension for calculating the proportion of false neighbors.

[0043] Step S322: Based on the high-frequency detail components, the pseudo-nearest neighbor method is used to calculate the rate of change of Euclidean distance between adjacent phase points under each candidate dimension. Specifically, the first step is to construct a temporary phase space for each candidate dimension m based on the reconstruction delay time τ. The phase point composition of the temporary phase space is the same as in step S320. The second step is to use the k-nearest neighbor algorithm to find its nearest neighbor phase point q for each phase point p in the temporary phase space. In the k-nearest neighbor algorithm, k is set to 1, that is, only the closest phase point is found. The third step is to calculate the Euclidean distance between phase point p and its nearest neighbor phase point q, denoted as d1. The fourth step is to increase the embedding dimension to m+1 and construct an m+1 dimensional temporary phase space. Similarly, the nearest neighbor phase point q' of phase point p in this new phase space is found. The fifth step is to calculate the Euclidean distance between phase point p and its nearest neighbor phase point q' in the m+1 dimensional temporary phase space, denoted as d2. The sixth step is to calculate the rate of change of the Euclidean distance, which is the ratio of d2 to d1, i.e., the rate of change = d2 / d1. The seventh step is to repeat the process from the second to the sixth step to calculate the rate of change of the Euclidean distance for all phase points under the candidate dimension.

[0044] Step S323: Phase points whose Euclidean distance change rate exceeds a preset multiple in adjacent dimensions are identified as false neighbors, and the proportion of false neighbors under each candidate dimension is calculated. Specifically, a preset multiple is set based on experimental data to distinguish between false neighbors and true neighbors. For example, if it is set to 1.1, it means that when the Euclidean distance change rate is greater than 1.1, it indicates that the distance change of the neighbor point is large after the embedding dimension increases, and it is not a true neighbor point, thus being identified as a false neighbor point. For each candidate dimension m, the Euclidean distance change rate of all phase points under that dimension is checked one by one, and the number of phase points whose Euclidean distance change rate exceeds the preset multiple is counted, i.e., the number of false neighbors. Finally, the proportion of false neighbors is calculated by dividing the number of false neighbors by the total number of phase points in the temporary phase space under that dimension, thus obtaining the proportion of false neighbors corresponding to that candidate dimension.

[0045] Step S324: Select the smallest candidate dimension corresponding to the first time the proportion of false neighbors falls below the preset convergence threshold as the embedding dimension. Specifically, the preset convergence threshold is determined based on experimental data to determine whether the embedding dimension is sufficient to reflect the chaotic characteristics of high-frequency detail components. Following the candidate dimensions in ascending order, compare the proportion of false neighbors corresponding to each candidate embedding dimension with the preset convergence threshold. When a candidate embedding dimension is found whose proportion of false neighbors falls below the preset convergence threshold for the first time, and this candidate dimension is the smallest among all dimensions that meet the conditions, this candidate dimension is determined as the final embedding dimension. If the proportion of false neighbors corresponding to all candidate dimensions is not lower than the preset convergence threshold, appropriately expand the upper limit of the embedding dimension search, and repeat steps S321 to S324 until an embedding dimension that meets the conditions is found.

[0046] Step S400: Input the nonlinear chaotic feature parameters into a pre-trained transfer learning classification model, and output the leakage state classification result of the threading channel through the transfer learning classification model.

[0047] Specifically, the input to the transfer learning classification model consists of five feature parameters: maximum Lyapunov exponent, correlation dimension, recursion rate, determination rate, and laminar flow rate. The output is a classification label corresponding to the leakage state. The classification label is represented by a numerical code. For example, the sealed state corresponds to the number 0, and the leakage state corresponds to the number 1. The sealed state means that there is no gas leakage in the wire threading channel, and the pressure of the pressurized gas inside the channel can be kept stable. The leakage state means that there is gas leakage in the wire threading channel, and regardless of the size of the leakage hole, it is uniformly classified as a leakage state.

[0048] In one possible implementation, step S400 further includes step S410, wherein the construction of the transfer learning classification model includes a meta-training stage and a scenario adaptation stage: In the meta-training stage, a micro-pressure fluctuation sample set collected under multiple working conditions, including different cable fill rates, different leakage apertures, and leakage working conditions, is used to train a feature extractor based on a one-dimensional convolutional neural network, enabling the feature extractor to learn leakage-related general feature representations; In the scenario adaptation stage, for the cable channel to be detected, on-site labeled samples are obtained, and the feature extractor is updated with gradients a limited number of times using the on-site labeled samples to form a classification model adapted to the current working condition, which serves as the transfer learning classification model.

[0049] Specifically, in the meta-training phase, a micro-pressure fluctuation sample set under multiple operating conditions is first constructed. This sample set is used to train the feature extractor, enabling it to learn general feature representations related to leakage. The sample set contains two categories of sample labels: intact sealing and leaking states. The sampling conditions for the sample set include different cable fill rates, different leakage orifice diameters, and different leakage conditions. The intact sealing samples are collected under no-leakage conditions with the same cable fill rate. At least 100 sets of samples are collected under each condition. Each set of samples includes: maximum Lyapunov exponent, correlation dimension, recursion rate, determination rate, and laminar flow rate, as well as the corresponding state label: intact sealing or leaking state. A classification model is constructed with the following structure: the input layer receives five nonlinear chaotic feature parameters, with a five-dimensional input dimension; the first fully connected layer has 128 neurons, using the ReLU activation function; the second fully connected layer has 64 neurons, also using the ReLU activation function; and the output layer is the classification label, with a two-dimensional dimension, corresponding to the two categories: intact sealing and leaking states. During training, a stochastic gradient descent optimizer was used, with a learning rate of 0.001, a batch size of 32, and 50 training epochs. The cross-entropy loss function was selected. The sample set was divided into training and validation sets at a ratio of 80% and 20%, respectively. The classification model was trained using the training set, and its performance was verified using the validation set. Training was stopped when the accuracy of the validation set reached over 90%, resulting in a trained basic classification model. This model can learn the correlation between five feature parameters and two states, effectively distinguishing between a sealed and leaking state.

[0050] In the scene adaptation phase, for the cable channel to be inspected, on-site labeled samples are acquired. The acquisition environment of the on-site labeled samples is consistent with the actual working conditions of the cable channel to be inspected, including the same cable fill rate, the same prefabricated material and size. The number of samples collected is 20 to 50 sets. Each set of on-site labeled samples contains five nonlinear chaotic feature parameters and the corresponding actual state label. Then, the basic classification model obtained in the meta-training phase is updated with gradients a limited number of times using the on-site labeled samples to adapt to the current detection conditions. The gradient update process is as follows: the five nonlinear chaotic feature parameters from the on-site labeled samples are directly input into the classification model. The model calculates the five input parameters, outputs the classification result, and calculates the classification loss between the classification result and the actual label. A stochastic gradient descent optimizer is used, with a learning rate set to 0.0001, a batch size set to 10, and 5 to 10 gradient update rounds. After each round of updates, the validation accuracy is calculated. When the validation accuracy reaches 95% or higher, the gradient update is stopped. At this point, a classification model adapted to the current working conditions is obtained. This classification model is the transfer learning classification model. This scenario adaptation phase addresses the issue of limited on-site sample quantity and improves the model's accuracy in distinguishing between intact seals and leaks in specific detection scenarios.

[0051] This application employs a technique of filling the wiring channel of precast concrete components with pressurized gas, collecting the signal of gas pressure change over time using a micro-pressure sensor to obtain a micro-pressure fluctuation time series, performing adaptive signal decomposition processing on this time series to remove low-frequency trend components caused by ambient temperature drift, extracting high-frequency detail components related to leakage flow characteristics, reconstructing the phase space of these high-frequency detail components, and extracting nonlinear chaotic feature parameters such as the maximum Lyapunov exponent, correlation dimension, and recursion graph quantitative analysis index. These extracted nonlinear chaotic feature parameters are then input into a pre-trained transfer learning classification model, which outputs a classification result indicating whether the wiring channel is sealed or leaking. This technique solves the technical problem of insufficient detection accuracy in existing precast concrete component wiring channel leak detection methods, which struggle to effectively separate weak leakage features from environmental interference signals. It achieves the technical effect of effectively separating weak leakage features from environmental interference signals and improving the accuracy of wiring channel leak detection.

[0052] In the above text, refer to Figure 1 A method for detecting leakage in the wiring channels of precast concrete components according to an embodiment of the present invention is described in detail. Next, reference will be made to... Figure 2 A system for detecting leakage in the wiring channels of precast concrete components according to an embodiment of the present invention is described.

[0053] The wire-threading channel leakage detection system for precast concrete components according to embodiments of the present invention addresses the technical problems of insufficient detection accuracy in existing wire-threading channel leak detection methods, which struggle to effectively separate weak leakage characteristics from environmental interference signals. The system achieves the technical effect of effectively separating weak leakage characteristics from environmental interference signals, thereby improving the accuracy of wire-threading channel leak detection. The wire-threading channel leakage detection system for precast concrete components includes: a pressure acquisition module 10, a signal decomposition and processing module 20, a nonlinear chaotic feature parameter extraction module 30, and a leakage state classification module 40.

[0054] The pressure acquisition module 10 is used to fill the threading channel of the precast concrete component to be tested with pressurized gas, and acquire the time-varying signal of the gas pressure inside the channel through pre-deployed micro-pressure sensors to obtain a micro-pressure fluctuation time series. The signal decomposition and processing module 20 is used to perform adaptive signal decomposition processing on the micro-pressure fluctuation time series, remove low-frequency trend components caused by ambient temperature drift, and extract high-frequency detail components related to leakage flow characteristics. The nonlinear chaotic feature parameter extraction module 30 is used to reconstruct the phase space of the high-frequency detail components and extract nonlinear chaotic feature parameters, including the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative analysis index. The leakage state classification module 40 is used to input the nonlinear chaotic feature parameters into a pre-trained transfer learning classification model, and output the leakage state classification result of the threading channel through the transfer learning classification model.

[0055] The detailed description of the specific configuration of the nonlinear chaotic feature parameter extraction module 30 is explained as follows: As described above, the phase space is reconstructed for the high-frequency detail components, and nonlinear chaotic feature parameters are extracted. The nonlinear chaotic feature parameter extraction module 30 may further include: a mutual information function calculation unit for calculating the mutual information function for the high-frequency detail components, and selecting the delay time corresponding to the first local minimum of the mutual information function as the reconstruction delay time; a false neighbor ratio calculation unit for calculating the false neighbor ratio in the temporary phase space under different embedding dimensions based on the high-frequency detail components using the false neighbor method, and taking the dimension corresponding to the first time the false neighbor ratio is lower than the preset convergence threshold as the embedding dimension; a phase space trajectory matrix construction unit for constructing a phase space trajectory matrix based on the reconstruction delay time and the embedding dimension; and a nonlinear chaotic feature parameter calculation unit for calculating the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative index of the phase space trajectory matrix to form the nonlinear chaotic feature parameters.

[0056] The nonlinear chaotic characteristic parameter calculation unit may further include: the quantitative indicators of the recursion graph include recursion rate, determination rate, and laminar flow rate; wherein, the recursion rate is determined by the ratio of the number of recursion points in the phase space trajectory matrix to the total number of phase points; the determination rate is determined by the ratio of the number of recursion points constituting the diagonal structure to the total number of recursion points; and the laminar flow rate is determined by the ratio of the number of recursion points constituting the vertical line structure to the total number of recursion points.

[0057] Specifically, the mutual information function is calculated for the high-frequency detail components, and the delay time corresponding to the first local minimum of the mutual information function is selected as the reconstruction delay time. The mutual information function calculation unit may further include: a high-frequency detail component sequence partitioning subunit for dividing the high-frequency detail component sequence into multiple equal-length intervals and calculating the mutual information function in each of the multiple equal-length intervals; and a statistical analysis subunit for statistically analyzing the minimum position of the mutual information function in each interval, removing abnormal intervals that deviate from the mean by more than two standard deviations, and taking the mean as the reconstruction delay time.

[0058] Specifically, based on the high-frequency detail components, the proportion of false neighbor points in the temporary phase space under different embedding dimensions is calculated using the false nearest neighbor method. The dimension corresponding to the first time the proportion of false neighbor points falls below a preset convergence threshold is taken as the embedding dimension. The false neighbor point proportion calculation unit may further include: a candidate dimension determination subunit for setting a preset upper limit for the embedding dimension search and determining each candidate dimension; an Euclidean distance change rate calculation subunit for calculating the Euclidean distance change rate between adjacent phase points under each candidate dimension based on the high-frequency detail components and using the false nearest neighbor method; a false neighbor point determination subunit for determining phase points whose Euclidean distance change rate exceeds a preset multiple as false neighbor points and calculating the proportion of false neighbor points under each candidate dimension; and an embedding dimension determination subunit for selecting the smallest candidate dimension corresponding to the first time the proportion of false neighbor points falls below the preset convergence threshold as the embedding dimension.

[0059] The detailed description of the specific configuration of the leakage state classification module 40 is explained as follows: As mentioned above, the leakage state classification module 40 may further include: a transfer learning classification model construction unit for constructing the transfer learning classification model, including a meta-training stage and a scenario adaptation stage: In the meta-training stage, a micro-pressure fluctuation sample set containing different cable fill rates, different leakage apertures, and leakage conditions is collected under multiple working conditions, and a feature extractor is trained based on a one-dimensional convolutional neural network, so that the feature extractor learns the general feature representation related to leakage; In the scenario adaptation stage, for the cable channel to be detected, on-site labeled samples are obtained, and the feature extractor is updated with gradients a limited number of times using the on-site labeled samples to form a classification model adapted to the current working condition, which serves as the transfer learning classification model.

[0060] The specific configuration of the signal decomposition processing module 20 is described in detail below: As mentioned above, the micro-pressure fluctuation time series is subjected to adaptive signal decomposition processing to remove low-frequency trend components caused by ambient temperature drift and extract high-frequency detail components related to leakage flow characteristics. The signal decomposition processing module 20 may further include: an adaptive signal decomposition unit for performing an adaptive signal decomposition algorithm on the micro-pressure fluctuation time series to obtain multiple intrinsic components with different frequency characteristics; a signal long-term dependence statistical index calculation unit for calculating multiple signal long-term dependence statistical indices of the multiple intrinsic components; and a linear superposition reconstruction unit for identifying the multiple signal long-term dependence statistical indices according to a preset index threshold, removing low-frequency trend components dominated by ambient temperature drift, and linearly superimposing and reconstructing the remaining intrinsic components to form the high-frequency detail components.

[0061] The signal long-term dependence statistical index calculation unit may further include: the signal long-term dependence statistical index is the Hearst exponent, which characterizes the temporal correlation strength of the signal.

[0062] Before linearly superimposing and reconstructing the remaining intrinsic components, the signal decomposition processing module 20 may further include: a resampling unit for calculating the ratio of the total energy of the remaining intrinsic components to the energy of the original micro-pressure fluctuation signal of the micro-pressure fluctuation time series; when the ratio is lower than a preset lower limit, it is determined that the leakage characteristic components in the current signal are insufficient, and the sampling time is extended for resampling.

[0063] The wire passage leakage detection system for precast concrete components provided in this embodiment of the invention can execute the wire passage leakage detection method for precast concrete components provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0064] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0065] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for detecting leakage in the wiring channels of precast concrete components, characterized in that, include: Pressurized gas is filled into the threading channel of the precast concrete component to be tested. The time-varying signal of the gas pressure inside the channel is collected by the pre-deployed micro-pressure sensors to obtain the micro-pressure fluctuation time series. The micro-pressure fluctuation time series is subjected to adaptive signal decomposition processing to remove low-frequency trend components caused by ambient temperature drift and extract high-frequency detail components related to leakage flow characteristics. The high-frequency detail components are reconstructed in phase space, and nonlinear chaotic feature parameters are extracted. These nonlinear chaotic feature parameters include the maximum Lyapunov exponent, correlation dimension, and quantitative analysis index of the recurrence graph. The nonlinear chaotic feature parameters are input into a pre-trained transfer learning classification model, and the leakage state classification result of the threading channel is output through the transfer learning classification model.

2. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 1, characterized in that, Phase space reconstruction is performed on the high-frequency detail components, and nonlinear chaotic feature parameters are extracted, including: The mutual information function is calculated for the high-frequency detail components, and the delay time corresponding to the first local minimum value of the mutual information function is selected as the reconstruction delay time. Based on the high-frequency detail components, the false nearest neighbor method is used to calculate the proportion of false neighbor points in the temporary phase space under different embedding dimensions. The dimension corresponding to the first time the proportion of false neighbor points is lower than the preset convergence threshold is taken as the embedding dimension. Construct a phase space trajectory matrix based on the reconstruction delay time and the embedding dimension; The maximum Lyapunov exponent, correlation dimension, and recursion graph quantitative index of the phase space trajectory matrix are calculated to form the nonlinear chaotic characteristic parameters.

3. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 2, characterized in that, The quantitative indicators of the recursion graph include recursion rate, determination rate, and laminar flow rate; The recursion rate is determined by the ratio of the number of recursive points in the phase space trajectory matrix to the total number of phase points; the determination rate is determined by the ratio of the number of recursive points constituting the diagonal structure to the total number of recursive points; and the laminar flow rate is determined by the ratio of the number of recursive points constituting the vertical structure to the total number of recursive points.

4. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 2, characterized in that, The mutual information function is calculated for the high-frequency detail components, and the delay time corresponding to the first local minimum of the mutual information function is selected as the reconstruction delay time, including: The high-frequency detail component sequence is divided into multiple equal-length intervals, and the mutual information function within each of the multiple equal-length intervals is calculated. Statistical analysis was performed on the minimum value of the mutual information function in each interval. After removing abnormal intervals that deviated from the mean by more than two standard deviations, the mean value was taken as the reconstruction delay time.

5. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 2, characterized in that, Based on the high-frequency detail components, the proportion of false neighbor points in the temporary phase space under different embedding dimensions is calculated using the false nearest neighbor method. The dimension corresponding to the first time the proportion of false neighbor points falls below a preset convergence threshold is taken as the embedding dimension, including: Preset the upper limit of the embedding dimension search to determine each candidate dimension; Based on the high-frequency detail components, the pseudo nearest neighbor method is used to calculate the rate of change of Euclidean distance between adjacent phase points under each candidate dimension. Points whose Euclidean distance change rate in adjacent dimensions exceeds a preset multiple are identified as false neighbors, and the proportion of false neighbors in each candidate dimension is statistically analyzed. The smallest candidate dimension corresponding to the first time the proportion of false neighbor points falls below the preset convergence threshold is selected as the embedding dimension.

6. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 1, characterized in that, The construction of the transfer learning classification model includes a meta-training phase and a scene adaptation phase: During the meta-training phase, a micro-pressure fluctuation sample set containing different cable fill rates, different leakage apertures, and leakage conditions is collected under multiple operating conditions. A feature extractor is trained based on a one-dimensional convolutional neural network, enabling the feature extractor to learn leakage-related general feature representations. In the scene adaptation phase, for the threading channel to be detected, on-site labeled samples are obtained, and the feature extractor is updated with gradients a limited number of times using the on-site labeled samples to form a classification model adapted to the current working conditions, which serves as the transfer learning classification model.

7. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 1, characterized in that, The micro-pressure fluctuation time series is subjected to adaptive signal decomposition processing to remove low-frequency trend components caused by ambient temperature drift and extract high-frequency detail components related to leakage flow characteristics, including: An adaptive signal decomposition algorithm is performed on the micro-pressure fluctuation time series to obtain multiple intrinsic components with different frequency characteristics; Calculate multiple signal long-term dependence statistics for the multiple intrinsic components; Based on preset threshold values, the long-term dependence statistical indicators of the multiple signals are identified, low-frequency trend components dominated by ambient temperature drift are removed, and the remaining intrinsic components are linearly superimposed and reconstructed to form the high-frequency detail components.

8. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 7, characterized in that, The long-term dependence statistic for the signal is the Hearst exponent, which characterizes the strength of the signal's temporal correlation.

9. The method for detecting leakage in the wiring channel of precast concrete components as described in claim 1, characterized in that, Before reconstructing the remaining eigencomponents by linear superposition, the following steps are included: The proportion of the total energy of the remaining intrinsic components to the energy of the original micro-pressure fluctuation signal of the micro-pressure fluctuation time series is calculated. When the proportion is lower than a preset lower limit, it is determined that the leakage characteristic component in the current signal is insufficient, and the sampling time is extended for resampling.

10. A leakage detection system for wiring channels in precast concrete components, characterized in that, The system is used to implement the method for detecting leakage in the wiring channels of precast concrete components according to any one of claims 1-9, and the system comprises: The air pressure acquisition module is used to fill the wiring channel of the precast concrete component to be tested with pressurized gas, and to acquire the signal of the change of gas pressure inside the channel over time through the pre-deployed micro-pressure sensors to obtain the micro-pressure fluctuation time series. The signal decomposition and processing module is used to perform adaptive signal decomposition processing on the micro-pressure fluctuation time series, remove low-frequency trend components caused by ambient temperature drift, and extract high-frequency detail components related to leakage flow characteristics. The nonlinear chaotic feature parameter extraction module is used to reconstruct the phase space of the high-frequency detail components and extract nonlinear chaotic feature parameters, including the maximum Lyapunov exponent, correlation dimension, and recursive graph quantitative analysis index. The leakage state classification module is used to input the nonlinear chaotic feature parameters into a pre-trained transfer learning classification model, and output the leakage state classification result of the wire-threading channel through the transfer learning classification model.

Images