A method for detecting, identifying, and locating building leaks in construction engineering.

Abstract

This application relates to the field of non-destructive testing technology for buildings, and discloses a method for detecting, identifying, and locating leak points in buildings for construction engineering. The method includes: first, adaptively identifying the set of resonant frequencies of the building structure to be tested; then, based on this set of resonant frequencies, applying normal and tangential excitations sequentially through a composite orthogonal dual-modal excitation unit, and simultaneously acquiring surface micro-motion video data of the structure by a high frame rate optical imaging unit; next, decomposing the video data to generate P-wave velocity fields and S-wave velocity fields, and constructing a wave velocity ratio field sensitive to leakage characteristics; finally, fusing multi-dimensional features to construct a high-dimensional feature vector, and using an unsupervised learning algorithm to calculate anomaly scores, thereby achieving automated identification and precise location of leak points. This invention utilizes a resonance enhancement mechanism and a wave velocity ratio diagnostic basis sensitive to water media, significantly improving the sensitivity and accuracy of detection, and achieving automation of the entire detection process and objectification of result interpretation through the introduction of unsupervised learning.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for buildings, specifically a method for detecting, identifying, and locating leak points in buildings for construction engineering. Background Technology

[0002] Building leaks are a common and extremely harmful problem in the construction industry. They not only damage the building's appearance and living environment but also corrode structural steel bars, affecting the building's structural safety and lifespan, and causing significant economic losses. Therefore, timely and accurate detection and location of building leaks are crucial for building maintenance and repair.

[0003] Currently, there are various technical methods used for detecting building leaks, mainly including infrared thermal imaging, acoustic emission detection, resistivity methods, and microwave methods. Infrared thermal imaging identifies leaks by detecting abnormal surface temperatures in the leaking area caused by water evaporation or differences in specific heat capacity; acoustic emission detection locates leaks by capturing weak acoustic signals generated by the flow or dripping of leaking water; resistivity methods and microwave methods detect leaks by utilizing changes in conductivity and dielectric constant in water-bearing areas, respectively.

[0004] However, these conventional detection methods generally have inherent limitations in practical applications. On the one hand, these methods have limited detection sensitivity, especially for early, minute leaks or internal leaks with complex seepage paths. The physical signals (such as temperature difference and acoustic signals) generated are extremely weak and easily drowned out by complex environmental noise and structural background signals, resulting in a low signal-to-noise ratio and a high risk of missed detection.

[0005] On the other hand, the diagnostic criteria of many existing technologies are vague and indirect, thus affecting the accuracy of identification. For example, temperature anomalies detected by infrared thermography can have multiple causes, such as non-leakage factors like uneven structural materials, internal cavities, or thermal bridging effects; the source of acoustic emission signals can also be confused with other friction and vibration sources. This non-uniqueness of diagnostic criteria leads to a high false positive rate in the test results, making it difficult to provide conclusive evidence of leakage.

[0006] Furthermore, the implementation process and result interpretation of existing detection technologies largely rely on the professional knowledge and subjective experience of the testing personnel. Operators need to manually set complex instrument parameters according to the site conditions and interpret the detection patterns or signals through visual observation and experience. This not only leads to low detection efficiency and poor result repeatability, but also lacks an objective and standardized evaluation system, making it difficult to guarantee the reliability of the test results. Therefore, developing a highly sensitive, highly accurate, and automated building leakage detection method has become an urgent technical challenge to be solved in this field. Summary of the Invention

[0007] The technical problem to be solved by this invention is that existing building leakage detection technologies are not sensitive to early minor leaks, are easily affected by environmental noise, rely on human experience for detection results, and are difficult to accurately locate and visualize the leakage area.

[0008] To address the aforementioned technical problems, the first aspect of this invention provides a method for detecting, identifying, and locating leakage points in buildings for construction engineering. The core innovation of this method lies in amplifying the micro-motion response of the structure through orthogonal dual-mode resonance excitation, utilizing the difference in propagation speed of different mechanical waves (P-waves and S-waves) in a water-containing medium as a highly sensitive diagnostic criterion, and finally combining this with an unsupervised machine learning algorithm to achieve automated and high-precision identification of leakage areas.

[0009] The method includes the following steps:

[0010] S1. System setup and parameter initialization: Deploy a high frame rate optical imaging unit so that its field of view covers the region of interest of the building structure under test, and couple a composite orthogonal dual-modal excitation unit to the building structure under test.

[0011] S2. Adaptive Resonant Frequency Identification: Identify the set of resonant frequencies of the building structure under test. The significance of this step is that by applying excitation at the structural resonant frequency, the maximum structural response can be obtained with the minimum energy input, thereby greatly improving the signal-to-noise ratio of the detection.

[0012] S3. Orthogonal Dual-Mode Resonance Data Acquisition: Based on the set of resonant frequencies, normal and tangential excitations are applied sequentially to the building structure under test, and the high frame rate optical imaging unit is used to simultaneously acquire surface micro-motion video data of the building structure under test under the normal and tangential excitations. The normal excitation mainly excites P-waves, while the tangential excitation mainly excites S-waves, providing mutually orthogonal raw data for subsequent multi-dimensional feature calculations.

[0013] S4. Calculate the multidimensional micro-motion characteristic field: Solve the surface micro-motion video data to generate P-wave velocity fields and S-wave velocity fields corresponding to the normal excitation and tangential excitation, respectively, and calculate the wave velocity ratio field based on the P-wave velocity fields and S-wave velocity fields. Due to the incompressibility of water, the bulk modulus of the medium in the leakage area increases significantly while the shear modulus remains unchanged, resulting in a significant increase in P-wave velocity and a slight decrease in S-wave velocity. Therefore, the ratio of the two, i.e., the wave velocity ratio field, can greatly amplify the abnormal characteristics of the leakage area.

[0014] S5. Feature Fusion and Leakage Area Identification: Based on the aforementioned wave velocity ratio field, a high-dimensional feature vector representing the physical state of each pixel is constructed. An anomaly score for each pixel is calculated using an unsupervised learning-based anomaly detection algorithm to identify and locate leakage points. This method eliminates the need for pre-labeled samples and can automatically discover abnormal patterns from high-dimensional data, achieving intelligent and objective leakage identification.

[0015] In one specific embodiment, S2 specifically includes: applying a broadband sweep frequency excitation to the building structure under test and acquiring its dynamic response; constructing a structural transfer function by performing Fourier transform and spatial accumulation on the dynamic response; and applying a peak search algorithm to the structural transfer function to automatically and accurately identify the set of resonant frequencies.

[0016] Preferably, in step S4, the generation of the P-wave velocity field and the S-wave velocity field specifically includes:

[0017] The P-wave phase field was calculated from the surface micro-motion video data. and S-wave phase field The P-wave number field is obtained by calculating the spatial gradient of the phase field. and S-wave number field The P-wave velocity field is calculated according to the following formula. and S-wave velocity field :

[0018] ;

[0019] ;

[0020] in, The resonant frequency used for excitation.

[0021] Furthermore, the P-wave phase field and S-wave phase field are obtained by orthogonal demodulation of the time-domain displacement signal extracted from the surface micro-motion video data.

[0022] Preferably, in step S5, the construction of the high-dimensional feature vector specifically includes:

[0023] For each pixel within the test area, the wave velocity ratio field characteristic value, P-wave amplitude field characteristic value, and S-wave amplitude field characteristic value calculated at each frequency of the resonant frequency set are concatenated to construct the high-dimensional feature vector. This method integrates information from multiple frequencies and multiple physical dimensions, enhancing the robustness and accuracy of the identification.

[0024] In one specific implementation, the unsupervised learning anomaly detection algorithm is the Isolation Forest algorithm.

[0025] Preferably, the identification and location of the leakage point in S5 specifically includes: generating an abnormal score map based on the abnormal scores of each pixel, and performing adaptive threshold segmentation on the abnormal score map to generate a binary mask image identifying the leakage point.

[0026] Furthermore, after generating the binary mask image, morphological processing operations are performed on it to eliminate noise and optimize the integrity of the leakage area.

[0027] Preferably, after identifying and locating the leak point, the method further includes a step of visually presenting the identified leak point area by overlaying it with pseudo-color onto the original optical image, providing users with intuitive diagnostic results.

[0028] A second aspect of the present invention provides a building leakage point detection, identification, and location system for building engineering, the system comprising:

[0029] A composite orthogonal bimodal excitation unit;

[0030] A high frame rate optical imaging unit;

[0031] And an information processing and control unit.

[0032] This invention provides a method for detecting, identifying, and locating leak points in buildings for construction engineering. It has the following beneficial effects:

[0033] 1. This invention, through adaptive identification and application of excitation at the structural resonant frequency, can achieve maximized micro-motion response of the structural surface with extremely low excitation energy, thereby significantly improving the signal-to-noise ratio. This resonant enhancement mechanism makes the ability to capture early, weak leakage signals far superior to traditional non-resonant detection methods, and also has a strong ability to suppress vibration noise in the field environment.

[0034] 2. This invention does not rely on a single amplitude or phase anomaly, but innovatively utilizes the differential response of P-wave and S-wave velocities to changes in the physical parameters of the water-bearing medium. By constructing a "wave velocity ratio field" that is extremely sensitive to leakage characteristics as the core diagnostic indicator, it can effectively eliminate interference from non-leakage factors such as material inhomogeneity and internal microcavities, thus fundamentally ensuring the reliability and accuracy of the diagnostic results.

[0035] 3. This invention constructs a high-dimensional feature vector by fusing multi-dimensional feature information and introduces an unsupervised learning algorithm to detect anomalies in the data, thus eliminating the heavy reliance on the professional experience of operators in traditional detection methods. The entire identification process requires no manual intervention or preset thresholds, and can automatically and objectively locate leakage areas, significantly improving detection efficiency and standardization. The results are ultimately presented in a visual manner, making them intuitive and easy to understand. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of a building leakage point detection, identification and location system for building engineering according to an embodiment of the present invention;

[0037] Figure 2 This is a flowchart illustrating a method for detecting, identifying, and locating building leakage points in a construction project, according to an embodiment of the present invention.

[0038] Figure 3 This is a schematic diagram of the adaptive resonant frequency identification process according to an embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the orthogonal dual-mode resonance data acquisition process according to an embodiment of the present invention;

[0040] Figure 5 This is a schematic diagram illustrating the process of solving a multidimensional micro-motion feature field according to an embodiment of the present invention.

[0041] Figure 6 This is a schematic diagram illustrating the feature fusion and leakage area identification process according to an embodiment of the present invention.

[0042] Among them, 100 is a composite orthogonal dual-modal excitation unit; 110 is a normal excitation component; 120 is a tangential excitation component; 130 is a coupling interface; 200 is a high frame rate optical imaging unit; 210 is an image sensor; 220 is an optical lens; 230 is a synchronous trigger interface; 300 is an information processing and control unit; 310 is a signal generation and control module; 320 is a data acquisition and synchronization module; 330 is a feature calculation module; and 340 is a fusion recognition and visualization module. Detailed Implementation

[0043] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] See attached document Figure 1 The present invention provides a building leakage point detection, identification and location system for building engineering. The system includes: a composite orthogonal dual-modal excitation unit 100, a high frame rate optical imaging unit 200 and an information processing and control unit 300.

[0045] The information processing and control unit 300 is electrically or signal-connected to both the composite orthogonal bimodal excitation unit 100 and the high frame rate optical imaging unit 200. During operation, the information processing and control unit 300 sends excitation control signals to the composite orthogonal bimodal excitation unit 100 and synchronous triggering and acquisition commands to the high frame rate optical imaging unit 200. The image sequence data acquired by the high frame rate optical imaging unit 200 is then transmitted to the information processing and control unit 300 for further processing.

[0046] Specifically, the composite orthogonal dual-modal excitation unit 100 functions to apply controllable, orthogonal normal and tangential micro-vibrations to the surface of the measured building structure. The composite orthogonal dual-modal excitation unit 100 may include a normal excitation component 110, a tangential excitation component 120, and a coupling interface 130.

[0047] The normal excitation assembly 110 is used to generate vibrations perpendicular to the surface of the structure under test, and it may be composed of a stack of piezoelectric ceramics.

[0048] The tangential excitation assembly 120 is used to generate vibrations parallel to the surface of the structure under test, and it can be composed of a shear-mode piezoelectric sheet or an electromagnetic exciter.

[0049] The coupling interface 130 is used to ensure good mechanical energy transfer between the excitation unit and the surface of the structure under test.

[0050] The composite orthogonal dual-mode excitation unit 100 responds to instructions from the information processing and control unit 300 and can selectively output a wideband excitation signal or a single-frequency harmonic excitation signal of a specific frequency.

[0051] The high frame rate optical imaging unit 200 is designed to non-contactly acquire video sequences of the micro-motion response of the surface of a building structure under excitation. The high frame rate optical imaging unit 200 includes an image sensor 210, an optical lens 220, and a synchronous trigger interface 230.

[0052] The image sensor 210 is an industrial-grade high-speed CMOS or CCD sensor with a frame rate and spatial resolution sufficient to capture the high-frequency vibration response of the measured structure under excitation.

[0053] The optical lens 220 is a high-resolution lens used to clearly image the area to be measured onto the image sensor 210;

[0054] The synchronization trigger interface 230 is connected to the information processing and control unit 300 to receive synchronization commands, ensuring precise alignment between the start time of image sequence acquisition and the application of vibration excitation. The digital video sequence acquired by this unit can be represented as... ,in These are the pixel coordinates of the image. It is a time variable.

[0055] The information processing and control unit 300, as the central core of the system, is responsible for executing all control, data processing, algorithm calculations, and result presentation. This unit is a hardware computing platform, and its internal functions are implemented through software modules. These modules include: a signal generation and control module 310, a data acquisition and synchronization module 320, a feature calculation module 330, and a fusion recognition and visualization module 340.

[0056] The signal generation and control module 310 is responsible for generating the digital waveform signals required to drive the composite orthogonal dual-mode excitation unit 100, including a wideband signal for pre-scanning and a single-frequency sine signal for main detection, and controlling the switching of excitation modes (normal or tangential).

[0057] The data acquisition and synchronization module 320 is responsible for sending acquisition commands to the high frame rate optical imaging unit 200 through the synchronization trigger interface 230, and for receiving and buffering the video sequences transmitted by the unit. .

[0058] Feature calculation module 330 is responsible for executing the core algorithm and processing the cached video sequences. The process is performed to calculate multi-dimensional physical characteristic fields, including but not limited to amplitude field, phase field, P-wave velocity field, S-wave velocity field, and wave velocity ratio field.

[0059] The fusion recognition and visualization module 340 is responsible for fusing multiple feature field data output by the feature calculation module 330 to construct a high-dimensional feature vector, and using an unsupervised learning algorithm to analyze the feature vector to identify abnormal areas. Finally, the identified leakage area location information is superimposed with the original optical image to generate a visualization result.

[0060] See attached document Figure 2 This invention provides a method for detecting, identifying, and locating building leaks in construction projects. The method may include the following steps:

[0061] S201. System setup and parameter initialization: The setup includes arranging a composite orthogonal dual-modal excitation unit 100 and a high frame rate optical imaging unit 200 on the surface area of ​​the building structure to be measured, and establishing a mapping relationship between the pixel coordinate system and the physical coordinate system.

[0062] S202 Adaptive resonant frequency identification: By applying a wideband excitation signal and acquiring the preliminary dynamic response of the structural surface, the structural transfer function of the measured area is calculated and constructed, thereby identifying one or more resonant frequencies;

[0063] S203, Orthogonal dual-mode resonance data acquisition: that is, excitation is performed at each identified resonant frequency using normal vibration mode and tangential vibration mode respectively, and the corresponding micro-motion response video sequence is synchronously acquired by the high frame rate optical imaging unit 200.

[0064] S204. Solve the multidimensional micro-motion feature field: Process the acquired video sequences to calculate and construct multiple feature field maps, including amplitude field, phase field, P-wave velocity field, S-wave velocity field and velocity ratio field.

[0065] S205. Feature Fusion and Leakage Area Identification: The data of the multiple feature field maps are fused into a high-dimensional feature vector. An unsupervised anomaly detection algorithm is used to analyze the feature vector to locate the feature anomaly area, and finally a visualization result map with leakage location information superimposed is generated.

[0066] The following will provide a detailed description of each of the above-described steps in the method of the embodiments of the present invention.

[0067] S201. System Setup and Parameter Initialization:

[0068] When implementing the detection and positioning method of the present invention, the system setup and parameter initialization are performed first. This step provides the basis for subsequent accurate measurements. This step involves the on-site arrangement of the composite orthogonal dual-modal excitation unit 100 and the high frame rate optical imaging unit 200, and the necessary calibration of the system.

[0069] Specifically, a region of interest (ROI) is selected on the surface of the building structure to be tested. This region is the area suspected of having leakage or requiring screening. A composite orthogonal bimodal excitation unit 100 is positioned at the edge or center of this ROI. To ensure efficient and stable transmission of excitation energy to the building structure, an acoustic coupling agent is applied between the coupling interface 130 of the excitation unit and the structural surface. This coupling agent can be an ultrasonic coupling agent, glycerin, or high-viscosity grease to fill minute gaps and achieve good acoustic impedance matching. Simultaneously, it is ensured that the normal excitation direction of the excitation unit is perpendicular to the structural surface, and the tangential excitation direction is parallel to the structural surface.

[0070] The high frame rate optical imaging unit 200 is positioned directly facing the region of interest, and the focal length of its optical lens 220 is adjusted to ensure that the natural texture or artificial markers on the structural surface are clearly imaged on the image sensor 210. The distance between the high frame rate optical imaging unit 200 and the structural surface, as well as the lens angle, are adjusted to ensure that its field of view completely covers the entire region of interest. In cases of unstable or insufficient ambient light, an auxiliary light source can be activated to provide uniform and constant illumination to the region of interest, thereby eliminating interference from ambient light fluctuations on image acquisition.

[0071] To ensure the accuracy of subsequent physical quantity calculations, it is necessary to establish a mapping relationship between the image pixel coordinate system and the physical world coordinate system of the measured structure surface. This process is achieved through camera calibration. A specific implementation method is as follows: a planar calibration plate with a known-sized array pattern (e.g., a checkerboard or dot array) is placed tightly against the surface of the region of interest to be measured, and a static image of the calibration plate is acquired by a high-frame-rate optical imaging unit 200.

[0072] The information processing and control unit 300 then invokes the camera calibration algorithm. By identifying the positions of the corner points or centers of the calibration pattern in the image within the pixel coordinate system, and using the known precise coordinates of these points in the physical world coordinate system, it calculates the transformation matrix between the two coordinate systems. This transformation relationship can be represented by a 3x3 homography matrix. To describe it, it establishes physical world coordinates. Its corresponding image pixel coordinates The mapping between them. This mapping relationship can be expressed in homogeneous coordinates as:

[0073] ;

[0074] It is the physical coordinate of a point on the surface of the structure being measured, in meters or millimeters.

[0075] It is the pixel coordinate of that point on the image sensor.

[0076] It is a non-zero scale factor.

[0077] It is the homography transformation matrix obtained through the calibration process.

[0078] After calibration, remove the calibration plate. The calculated transformation matrix... The information is stored in the information processing and control unit 300 and used in the subsequent S204 to convert the spatial gradient calculated in pixels into a spatial gradient expressed in physical units, thereby ensuring that the calculation results of physical characteristic quantities such as wave velocity field have accurate physical meaning.

[0079] After completing system setup and parameter initialization, the system proceeds to adaptive resonant frequency identification. This step is a key technical feature of this invention, aiming to intelligently determine the optimal excitation frequency by conducting a preliminary investigation of the dynamic characteristics of the structure under test, thereby replacing traditional fixed-frequency or blind frequency sweep detection methods.

[0080] The physical basis of this step lies in the fact that any building structure possesses inherent vibrational characteristics, manifested as a series of natural frequencies or resonant frequencies. When the frequency of external excitation coincides with one of the structure's resonant frequencies, resonance occurs. In this case, the structure requires only a small amount of excitation energy to produce a vibrational response far greater than that at the non-harmonic frequency. The existence of a leakage area, due to its alteration of local mass, stiffness, and damping, will change the overall or local resonant characteristics of the structure, or generate new local resonance modes related to leakage defects. Therefore, by identifying and utilizing these resonant frequencies most sensitive to changes in the internal medium of the structure for excitation, the response difference between the leakage area and the intact area can be greatly amplified, thereby significantly improving the signal-to-noise ratio of the detection signal and the sensitivity to weak leakage defects.

[0081] See attached document Figure 3 To achieve automatic identification of the resonant frequency, this embodiment first applies a wideband excitation signal to the structure and acquires its preliminary response. The signal generation and control module 310 of the information processing and control unit 300 generates a preset wideband excitation signal and drives the composite orthogonal dual-mode excitation unit 100 to excite the structure. The excitation mode here can be a normal excitation mode with relatively uniform energy distribution.

[0082] The broadband excitation signal is a linear frequency modulated signal, also known as a chirp signal. The instantaneous frequency of this signal changes linearly with time. In time interval The interior can be represented as:

[0083] ;

[0084] in:

[0085] The amplitude of the excitation signal is controlled within a tiny range that will not cause any damage to the building structure.

[0086] It is the starting frequency of the scan;

[0087] It is the scan termination frequency;

[0088] The frequency range The selection is based on the general response characteristics of the material (such as concrete, brick wall) and thickness of the structure being tested, and can be set from 10 Hz to 1000 Hz.

[0089] It is the duration of the linear frequency modulated signal.

[0090] While the composite orthogonal dual-modal excitation unit 100 applies the broadband excitation signal, the data acquisition and synchronization module 320 of the information processing and control unit 300 simultaneously triggers the high frame rate optical imaging unit 200, ensuring that it operates throughout the entire excitation duration. The system continuously acquires dynamic response video sequences of the tested area and processes these sequences. Stored in the cache for use in subsequent structure transfer function construction steps.

[0091] After acquiring the initial dynamic response video sequence Subsequently, the feature calculation module 330 of the information processing and control unit 300 processes it to construct the structure transfer function and identify the resonant frequency.

[0092] First, a micro-motion analysis algorithm is invoked to analyze the video sequence. The temporal displacement signal of each pixel is extracted. In one specific implementation, the micro-motion analysis algorithm is a phase-based spatial optical flow algorithm. This algorithm estimates the sub-pixel-level displacement by analyzing the temporal changes in the local phase information of each pixel's neighborhood in the video sequence, exhibiting high robustness to illumination changes. The output of this step is the temporal displacement signal of each pixel covering the entire region of interest. To analyze the frequency response characteristics of the measured area, the feature calculation module 330 calculates the time-domain displacement signal of each pixel. Applying the Fast Fourier Transform (FFT), the frequency response of each pixel is transformed from the time domain to the frequency domain to obtain the complex frequency response of each pixel. This transformation can be expressed as:

[0093] ;

[0094] in: It is a frequency variable; It is a time variable; It is the imaginary unit; It is the duration of the broadband excitation signal.

[0095] To obtain the structure transfer function that characterizes the overall dynamic properties of the measured region This requires spatially accumulating the frequency response amplitudes of all pixels. The structure transfer function... By analyzing the frequency response magnitude of all pixels within the entire region of interest (ROI). Construct by integration or summation:

[0096] ;

[0097] The structure transfer function This is a curve describing the magnitude of the structure's response to excitations at different frequencies. The local peaks on this curve correspond to the resonant frequencies of the structure.

[0098] This algorithm ultimately yields a set containing one or more resonant frequencies. This set is passed to the signal generation and control module 310 to guide the subsequent orthogonal dual-mode resonant excitation main detection step.

[0099] See attached document Figure 4 The set of resonant frequencies was determined through adaptive resonant frequency identification. Next, the method proceeds to orthogonal dual-mode resonance data acquisition. In this step, the signal generation and control module 310 of the information processing and control unit 300, based on the identified set of resonant frequencies... It automatically generates and executes a predetermined stimulus control sequence for master data acquisition.

[0100] The generation and execution process of the excitation control sequence is as follows: The information processing and control unit 300 first sets the excitation mode to the normal excitation mode. Then, it iterates through the set of resonant frequencies. Each frequency in the set. For the first resonant frequency in the set. The signal generation and control module 310 generates a frequency of The steady-state single-frequency sinusoidal harmonic signal is generated and output to the composite orthogonal dual-mode excitation unit 100 via a digital-to-analog converter. Simultaneously, a control command is issued to activate only its normal excitation component 110. This single-frequency harmonic signal... It can be represented as:

[0101] ;

[0102] in:

[0103] It is the amplitude of the resonant excitation, which is set to ensure that the surface of the structure produces a micro-motion response that can be clearly captured by the high frame rate optical imaging unit 200, but is far below any level that may cause structural fatigue or damage.

[0104] From a set The current resonant frequency selected in the middle;

[0105] It is a time variable.

[0106] In terms of frequency After performing normal excitation and completing data acquisition, the signal generation and control module 310 continues to select the next resonant frequency from the set. Generate the corresponding sinusoidal harmonic signals and excite them until the set All frequencies are used for normal excitation.

[0107] After completing the normal excitation at all frequencies, the signal generation and control module 310 switches the excitation mode to tangential excitation mode. This switch is achieved by sending a specific control signal to the composite orthogonal bimodal excitation unit 100, which disables the normal excitation component 110 and activates the tangential excitation component 120. Subsequently, the set of resonant frequencies is traversed again. Each resonant frequency is used sequentially in the same order as the normal excitation. A single-frequency harmonic signal is generated and a tangential excitation is applied.

[0108] In this way, the excitation control sequence ensures that dynamic data caused by normal excitation (generating a response dominated by compression waves) and tangential excitation (generating a response dominated by shear waves) are collected at each resonant frequency point that is sensitive to structural defects, providing a complete data foundation for the subsequent comprehensive solution of multi-dimensional characteristic fields.

[0109] For each single-frequency resonant excitation event in the aforementioned excitation control sequence, the data acquisition and synchronization module 320 of the information processing and control unit 300 performs a data acquisition operation that is precisely synchronized with it to ensure that the acquired video data and the excitation signal have a definite time and phase relationship.

[0110] Specifically, at the start of an excitation event, for example when the composite orthogonal dual-mode excitation unit 100 starts at its resonant frequency... After normal excitation, the system will wait for a preset settling time. The stabilization time The setting is to ensure that the forced vibration of the structural surface reaches a steady state and to eliminate the influence of transient response.

[0111] During the stable period After completion, the data acquisition and synchronization module 320 generates a digital trigger pulse signal through a hardware trigger circuit. The trigger pulse signal is a TTL level signal. This trigger pulse signal is simultaneously sent to two targets:

[0112] First, the pulse is sent to the high frame rate optical imaging unit 200 through the synchronous trigger interface 230. The rising or falling edge of the pulse serves as the start command for image acquisition, causing it to immediately begin capturing video sequences according to the preset frame rate and total number of frames acquired.

[0113] Secondly, this pulse signal also serves as a reference signal for the zero-time point. The signal is recorded by the signal generation and control module 310 and used to mark the excitation sine waveform. The initial phase.

[0114] By using a hardware triggering mechanism instead of a software command triggering mechanism, the time delay jitter between excitation and acquisition can be controlled at the nanosecond or microsecond level, thus providing a reliable time reference for subsequent high-precision phase field calculation.

[0115] After receiving the trigger signal, the high frame rate optical imaging unit 200 continuously acquires data for a preset duration. The video data. To ensure the accuracy of subsequent Fourier analysis, the acquisition duration... Set as excitation signal period This is an integer multiple of the standard value. This measure can effectively suppress spectral leakage and improve the signal-to-noise ratio.

[0116] The synchronous data acquisition process is repeated for each excitation event (i.e., each resonant frequency under each excitation mode) in the excitation control sequence.

[0117] Finally, the information processing and control unit 300 acquires and stores a series of video sequence data corresponding one-to-one with the excitation events, denoted as the normal excitation video set. and tangentially stimulated video set ,in Iterate through all identified resonant frequencies. These video sequences serve as direct input for subsequent multidimensional feature field solving steps.

[0118] See attached document Figure 5 After data acquisition, the multidimensional micro-motion feature field is solved. The goal of this step is to accurately extract multidimensional features that characterize the local physical properties of the structure from the acquired raw video data. This step is executed by the feature calculation module 330 of the information processing and control unit 300. First, the normal excitation video set is processed... and tangentially stimulated video set For each video sequence, a micro-motion extraction operation is performed. This operation is consistent with the phase-based spatial optical flow field algorithm used in S202 above, and it extracts the video sequence... Converted into pixel-level time-domain displacement signal .

[0119] Since the excitation is a single-frequency steady-state harmonic signal, at any point on the structure surface during the steady-state response step... displacement response It can also be modeled as a pair with the same frequency The harmonic function, in its general form is:

[0120] ;

[0121] in:

[0122] Is this point? The vibration amplitude at a point reflects the distribution of vibration energy at that point.

[0123] Is this point? The phase delay of the vibration response relative to the excitation source, and its spatial distribution, reflects the delay characteristics of wave propagation in the medium.

[0124] To obtain the time domain signal The amplitude was accurately solved in the middle. and phase The feature calculation module 330 employs an orthogonal demodulation algorithm. This algorithm effectively suppresses noise at non-excitation frequencies and directly solves for amplitude and phase information by processing the measured signal with a pair of reference signals with the same frequency but orthogonal phases.

[0125] Specifically, for the displacement signal of each pixel The algorithm first calculates its in-phase components. and orthogonal components This calculation will be performed by... Respectively with reference signal and Multiply, and throughout the collection time To achieve this, we need to perform integration within the inner quadrant:

[0126] ;

[0127] ;

[0128] According to the product-to-sum formula for trigonometric functions, the above integration operation is performed in... When the frequency is an integer multiple of the excitation period, it can filter out frequencies of The high-frequency components are extracted, retaining only the DC components that are related to amplitude and phase. The resulting in-phase components... and orthogonal components The relationship with amplitude and phase is as follows and Therefore, the final amplitude and phase It can be calculated in the following way:

[0129] ;

[0130] ;

[0131] in, The function is a four-quadrant arctangent function, and its calculation result can determine the phase angle. Clearly determined in Within the complete interval, the quadrant ambiguity problem caused by a single arctangent function is avoided.

[0132] Feature calculation module 330 calculates each excitation frequency The above orthogonal demodulation calculation is performed on all pixels in both the normal and tangential excitation videos. The final output of this process is, for each... Four two-dimensional characteristic field maps were generated: P-wave amplitude field P-wave phase field S-wave amplitude field and S-wave phase field These feature fields form the basis for constructing subsequent higher-order diagnostic features.

[0133] After generating the amplitude and phase fields of the P-wave and S-wave, the feature calculation module 330 further processes the phase field to estimate and construct the dynamic wave velocity field. The physical principle of this estimation is based on wave theory, specifically the propagation speed of plane harmonics. It is its angular frequency With wavenumber The ratio of angular frequency to angular frequency. The excitation parameters are known, while the wavenumber is... As a measure of the phase change of a wave within a unit physical distance, its magnitude is equal to the phase field. The magnitude of the spatial gradient.

[0134] Since the phase field exists as two-dimensional data in the form of a discrete pixel grid, its spatial gradient needs to be approximated using numerical methods. One specific implementation method is to use the central difference method to calculate the gradient for each pixel. The partial derivative at that point. For the P-wave phase field. , its in and The partial derivative in the direction can be approximated as:

[0135] ;

[0136] ;

[0137] in and These represent the horizontal and vertical spacing of pixels in the physical world coordinate system, respectively. This spacing value was obtained during the camera calibration process with S201. To enhance the robustness of the computation to noise, another implementation can use an image gradient operator incorporating smoothing filtering, such as the Sobel operator, to calculate these two partial derivative components. After calculating the partial derivative components of the phase field, the value of the point can be obtained. Local P-wave wavenumber Its value is the magnitude of the phase gradient vector:

[0138] ;

[0139] Similarly, for the S-wave phase field Performing the same gradient calculation yields the local wavenumber of the S-wave. Finally, based on the relationship between angular frequency and wavenumber, the value of each pixel can be calculated. P-wave velocity at the location and S-wave speed :

[0140] ;

[0141] ;

[0142] For each resonant frequency The above calculations were performed on both the P-wave phase field and the S-wave phase field, ultimately generating the P-wave velocity field. and S-wave velocity field These two wave velocity fields are key intermediate physical quantities for constructing the final diagnostic features.

[0143] After obtaining the P-wave velocity field and S-wave velocity field Then, the feature calculation module 330 performs combined calculations on these two physical quantities to generate a key diagnostic feature field that has higher sensitivity and stronger indicativeness for the leakage area.

[0144] A core technical solution of this invention lies in constructing and utilizing a wave velocity ratio field. As a core indicator for identifying leaks, this wave velocity ratio is calculated by dividing the P-wave velocity by the S-wave velocity point by point for each pixel.

[0145] ;

[0146] The wave velocity is greater than the field It has a clear and sensitive physical meaning; an abnormal increase in its value directly indicates an increase in the water content in the pores of building materials. This physical meaning is rooted in the basic principle of elastic wave propagation in porous media (such as concrete and masonry structures).

[0147] Specifically, the propagation speed of P-waves (compression waves) is primarily determined by the bulk modulus and shear modulus of the medium, reflecting its ability to resist volume changes. Water, as a nearly incompressible fluid, has a high bulk modulus. When water infiltrates and fills the internal pores of materials such as concrete, it significantly enhances the overall bulk modulus of the water-solid composite medium. Therefore, in aquifers, the propagation speed of P-waves... It usually does not decrease significantly, and may even remain at a high level.

[0148] In contrast, the propagation speed of S-waves (shear waves) is determined solely by the shear modulus of the medium, reflecting its ability to resist shape changes. Water, as a fluid, lacks the ability to resist shear deformation, and its shear modulus is zero. When water fills the pores of a material, it not only fails to contribute to the shear stiffness of the composite medium but also acts as a lubricant between solid particles, significantly reducing the overall shear modulus in that region. This reduction in shear modulus directly affects the propagation speed of S-waves. A sharp decline occurred in the water-bearing area.

[0149] In summary, in water-bearing areas caused by leakage, due to Remaining relatively stable or slightly declining, while A significant decrease occurred; the combined effect of these two factors led to their ratio. A localized, significant increase occurs. Therefore, the wave velocity is higher than the field. The physicochemical change of leakage is transformed into a two-dimensional physical field feature with high contrast and high signal-to-noise ratio that can be accurately measured.

[0150] In addition to the wave velocity ratio field, in this embodiment, the previously calculated P-wave amplitude field and S-wave amplitude field It is also retained as an auxiliary diagnostic feature. Aqueous regions increase the internal damping of the material, leading to accelerated energy attenuation of waves during propagation, which manifests in localized rapid changes in the amplitude field.

[0151] Regarding rapid attenuation, the feature calculation module 330 calculates each resonant frequency... All of these generate the aforementioned set of feature fields, providing a rich and complementary set of multidimensional data for the final feature fusion and recognition steps.

[0152] See attached document Figure 6 After completing the calculation of all feature fields, the method of this invention proceeds to S205. This step is executed by the fusion recognition and visualization module 340 of the information processing and control unit 300. Its primary task is to integrate the multiple discrete feature field map data generated in the previous step into a unified, high-dimensional feature space, providing input for subsequent intelligent recognition algorithms. This integration process is performed at the pixel level. For any pixel in the region of interest (ROI) being measured... The system will apply this to all different excitation conditions (i.e., different resonant frequencies). The feature vector is constructed by combining multiple feature values ​​generated by different physical mechanisms (i.e., P-wave and S-wave) and concatenating them in a predetermined order to construct a multidimensional feature vector specific to that pixel.

[0153] Specifically, for each pixel Its high-dimensional feature vector The construction method is as follows:

[0154] This vector is composed of all One resonant frequency The calculated feature values ​​are concatenated. In one embodiment, the feature selected for each frequency includes the P-wave amplitude. S-wave amplitude and wave speed ratio Then the feature vector of that pixel. It can be represented as: ;

[0155] in: It is a pixel. The high-dimensional feature vector. It is in the The P-wave amplitude at that point at a resonant frequency. It is in the The S-wave amplitude at that point at a resonant frequency. It is in the The wave velocity ratio at that point at each resonant frequency. This indicates that all elements are concatenated into a single column vector.

[0156] If adopted One resonant frequency and Types of features (in this example) =3), then each pixel will be mapped to a The feature vectors are 3D. Before constructing the feature vectors, to eliminate the influence of differences in dimensions and numerical ranges between different physical features, the fusion recognition and visualization module 340 normalizes each feature field map. A specific implementation is min-maximum normalization, which linearly scales all pixel values ​​of each feature field to 1D. Within the interval. For any eigenvalue... Its normalized value The calculation is as follows:

[0157] ;

[0158] in and These are the maximum and minimum values ​​in the feature field, respectively.

[0159] By constructing high-dimensional feature vectors, the method of this invention effectively integrates previously isolated, single-dimensional information. The feature vector of each pixel provides a comprehensive, multi-faceted description of its physical state. This rich information expression greatly enhances the distinguishability between normal areas and abnormal areas caused by leakage, laying a data foundation for subsequent high-precision, automated identification using unsupervised learning algorithms.

[0160] After constructing a high-dimensional feature vector for each pixel within the tested area, the fusion recognition and visualization module 340 employs an unsupervised learning-based anomaly detection algorithm to analyze these feature vectors and identify abnormal areas related to leakage. The fundamental reason for using unsupervised learning is that in actual detection scenarios, the pattern, size, and location of leakage areas are unknown, and they typically only occupy a small portion of the entire detection area. This perfectly matches the typical characteristics of anomaly detection problems: "scarce and unknown abnormal samples."

[0161] In this embodiment, the unsupervised anomaly detection algorithm is an isolation forest algorithm. The core idea of ​​this algorithm is that outliers are more easily isolated in the feature space due to their "sparse and distinct" characteristics. Isolation forest realizes this idea by constructing an ensemble model composed of multiple "isolation trees" (iTrees).

[0162] The specific execution process of the algorithm is as follows: First, a subsample is randomly selected from the high-dimensional feature vector set of all pixels to construct an isolation tree. During the construction of this tree, at each node, the algorithm randomly selects a feature dimension from the data points contained in the current node, and then randomly selects a split value between the maximum and minimum values ​​of that dimension, dividing the data point into two parts, which are then assigned to the left and right child nodes respectively. This process is executed recursively until each data point is isolated to a leaf node, or the depth of the tree reaches a preset upper limit.

[0163] Due to the pixels in the leakage area, their feature vectors (especially the wave velocity ratio) The anomalous feature vectors (which are significantly higher in size) are far removed from the dense data clusters composed of pixels from normal, intact structural regions. Therefore, in the above random segmentation process, these anomalous feature vectors are likely to be separated from the main data group with fewer segmentations. This is reflected in the structure of the isolation tree, where the leaf nodes corresponding to the anomalous points are usually located very close to the root node, and their path length is short.

[0164] The algorithm constructs multiple such isolated trees, forming a "forest". For the feature vector of each pixel to be evaluated... The algorithm calculates the average path length across all trees in the forest. .

[0165] Finally, based on this average path length, the anomaly score for each pixel is calculated. The score is normalized to between 0 and 1, and its calculation formula is as follows:

[0166] ;

[0167] in:

[0168] It is an eigenvector The average path length across all isolated trees in the forest.

[0169] This is the number of sample points used to construct the tree.

[0170] It is a normalization factor, representing the normalization factor for a given number of samples. Below, the average path length of the binary search tree is used to standardize the results. Its calculation formula is:

[0171] ;

[0172] in It is a harmonic series.

[0173] This abnormal score The value has a clear indicative meaning:

[0174] When the average path length of a point much smaller At that time, its score The value will approach 1, indicating that the point is very likely an outlier;

[0175] When the average path length is close to At that time, the score was... Approaching 0.5;

[0176] When the average path length is much greater than At that time, the score was... A value close to 0 indicates that the point is highly likely to be a normal point.

[0177] By performing the above calculations on all pixels within the tested area, the fusion recognition and visualization module 340 ultimately generates a two-dimensional anomaly score map. In this map, the grayscale value or pseudocolor of each pixel directly corresponds to its anomaly score; areas with higher scores are suspected leakage areas.

[0178] This transformation process is achieved by thresholding the anomaly score map. The system sets an anomaly score threshold. Convert the score map into a binary mask image. For each pixel in the graph Its value in the binary mask image is determined according to the following rules:

[0179] ;

[0180] in, This is an abnormal score for that pixel. Pixels with a value of 1 are initially identified as belonging to the leakage area, while those with a value of 0 are identified as normal areas.

[0181] The threshold There are several ways to determine the threshold. One specific implementation is that the operator can preset a fixed value based on experience, such as 0.75, that is, pixels with abnormal scores higher than 0.75 are considered to be leakage points.

[0182] To automate this process, another preferred implementation is to employ an adaptive threshold segmentation algorithm, such as the Otsu method. This algorithm automatically finds an optimal threshold on the histogram of the anomaly score map that best distinguishes between "abnormal" and "normal" pixels by maximizing the inter-class variance, thus avoiding the subjectivity of manual setting.

[0183] The binary mask image obtained from the initial segmentation To eliminate isolated misjudged pixels caused by noise or fill in possible tiny holes within a region, a series of morphological image processing operations can be performed. These morphological operations may include an opening operation to remove small noise points, followed by a closing operation to fill in the holes within the region, thereby making the final identified leakage area boundary smoother and the area more complete.

[0184] After the above processing, the system obtains a final mask image that accurately represents the contour of the leakage area. The final step is to fuse this recognition result with an original optical reference image (e.g., the first still image acquired by the high frame rate optical imaging unit 200) to achieve a visual representation of the result. One specific presentation method is to overlay the area marked as leaking in the final mask image onto the original optical image with a striking, semi-transparent pseudo-color (e.g., red). Alternatively, the contour line of the leakage area can be extracted and drawn on the original image with a highlight color.

[0185] The final visualization results are presented on the user interface of the information processing and control unit 300, providing users with intuitive and clear indications of the location, shape, and extent of the leak.

[0186] To further illustrate the practical application effect of the present invention, the following is a detailed description of a leakage detection project on the exterior wall of a high-rise residential building in a certain city.

[0187] 1. Project Background and Inspection Object: The inspection object is an area of ​​approximately 2m x 2m located on the 5th floor of the east facade of this residential building. The wall structure is a reinforced concrete shear wall, and the exterior finish is glazed tiles. Dampness is observed in the corresponding location inside this area, requiring precise location of the external wall leakage point.

[0188] 2. System Deployment:

[0189] Excitation unit arrangement: The composite orthogonal dual-mode excitation unit 100 is installed on the edge of the windowsill below the area to be measured, and fixed to the wall base surface by high viscosity vacuum grease as a coupling agent to ensure effective transmission of normal and tangential vibration energy.

[0190] Imaging unit setup: The high frame rate optical imaging unit 200 is mounted on a ground-mounted lifting platform approximately 8 meters above the wall. A 50mm focal length industrial lens is selected, providing complete field of view coverage of a 2m × 2m region of interest.

[0191] Environmental Assistance: Since the detection took place in the afternoon with ample natural light, no auxiliary light source was used. A checkerboard calibration board was used to calibrate the camera to the physical plane, obtaining the homography matrix.

[0192] 3. Testing process:

[0193] Frequency identification: The system first transmits a linear frequency modulated signal from 100Hz to 800Hz. Fast Fourier Transform analysis identifies a significant local resonant frequency in the finish layer of this area. .

[0194] Data Acquisition: The system automatically locks to 420Hz, first controlling the normal excitation component to operate for 0.5 seconds, then switching to the tangential excitation component to operate for 0.5 seconds. The camera synchronously acquires two sets of video sequences at a rate of 2000 frames per second.

[0195] Feature Calculation: The information processing and control unit 300 uses an orthogonal demodulation algorithm to process video data. Calculations revealed that near the upper left corner of the region (x=0.5m, y=1.8m), the S-wave velocity field... It shows a clear low-value area (reduced by about 40%), while the P-wave velocity field Not much has changed.

[0196] Wave velocity ratio calculation: Wave velocity ratio field generated by the system The graph shows that the wave velocity ratio in the upper left region jumps from 1.6 in the background to 2.1, strongly indicating the presence of water-filled pores at that location.

[0197] 4. Identification Results and Verification:

[0198] Intelligent Decision Making: The anomaly score map output by the Isolation Forest algorithm is highlighted in the upper left corner (score > 0.85). The automatically generated red semi-transparent mask accurately covers this irregular area.

[0199] Engineering Verification: Based on the visualization results output by the system, the engineers removed the tiles in the marked area for inspection and found that the bonding mortar layer behind the tiles was completely wet and saturated, and there were obvious hollow water accumulation channels, which was highly consistent with the detection conclusion of this system.

[0200] Differentiation between micro-crack leakage and dry crack detection in basement concrete shear walls:

[0201] This embodiment aims to demonstrate the invention's ability to distinguish between "leakage (water)" and "dry cracks (air)".

[0202] 1. Project Background and Testing Object: The object to be tested is a concrete shear wall in the underground parking garage of a commercial center. The surface has visible micro-cracks, and it is necessary to determine whether groundwater has seeped into the cracks.

[0203] 2. System Deployment: Due to the dim lighting in the basement, a high-frequency flicker-free LED surface light source was installed next to the imaging unit for supplemental lighting. The excitation unit was directly attached to the concrete wall.

[0204] 3. Testing process:

[0205] Frequency identification: The sweep frequency range is set to 500Hz to 2000Hz (concrete structures have high rigidity, and their resonant frequencies are usually high). The system automatically identifies the main resonant frequency as 1150Hz.

[0206] Dual-modal imaging: Micro-motion videos under normal and tangential excitations are acquired at a frequency of 1150Hz.

[0207] Physical field analysis:

[0208] Region A (upper section of the crack): Characteristic calculations show that the P-wave amplitude... and S-wave amplitude All showed significant attenuation (due to the cracks blocking wave propagation), and the wave velocity ratio was lower than that of the cracks. It remains within the range of normal concrete (around 1.5). This indicates that the area is a dry medium (an air-filled crack).

[0209] Region B (lower section of the crack): Feature analysis shows that the S-wave velocity... The intensity drops sharply, but P waves can still propagate relatively well due to the incompressibility of water. (Less attenuation), resulting in a lower wave velocity ratio at that location. The level abnormally increased to above 1.9.

[0210] 4. Identification Results and Verification: Based on the wave velocity ratio characteristics, the system's integrated identification module marks only "Area B" as an abnormal leakage area, while excluding "Area A" (judging it as ordinary structural damage).

[0211] Verification: On-site drilling and sampling verified that the powder drilled from area A was dry, while a trace amount of water seeped out after drilling from area B. This proves that the present invention, utilizing an orthogonal dual-mode wave velocity ratio field, can effectively eliminate interference from dry cracks and achieve accurate identification of true and false leaks.

Claims

1. A method for detecting, identifying, and locating building leakage points in construction engineering, characterized in that, Includes the following steps: S1. System setup and parameter initialization: Deploy a high frame rate optical imaging unit to cover the building structure under test with its field of view, and couple a composite orthogonal dual-modal excitation unit to the building structure under test. S2, Adaptive Resonance Frequency Identification: Identifying the set of resonance frequencies of the building structure under test; S3. Orthogonal dual-mode resonance data acquisition: Based on the set of resonant frequencies, normal excitation and tangential excitation are applied to the building structure under test in sequence, and the high frame rate optical imaging unit is used to synchronously acquire surface micro-motion video data of the building structure under test under the normal excitation and tangential excitation. S4. Solve the multidimensional micro-motion feature field: Solve the surface micro-motion video data to generate P-wave velocity field and S-wave velocity field corresponding to the normal excitation and tangential excitation respectively, and calculate the wave velocity ratio field based on the P-wave velocity field and S-wave velocity field. S5. Feature Fusion and Leakage Area Identification: Based on the wave velocity ratio field, a high-dimensional feature vector representing the physical state of each pixel is constructed, and the anomaly score of each pixel is calculated through an unsupervised learning anomaly detection algorithm to identify and locate the leakage point.

2. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 1, characterized in that, S2 specifically includes: A broadband sweep frequency excitation is applied to the building structure under test and its dynamic response is collected. The structure transfer function is constructed by performing Fourier transform and spatial accumulation on the dynamic response, and a peak search algorithm is applied to the structure transfer function to identify the set of resonant frequencies.

3. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 1, characterized in that, The generation of the P-wave velocity field and S-wave velocity field in S4 specifically includes: The P-wave phase field and S-wave phase field are calculated from the surface micro-motion video data. The P-wave wavenumber field and S-wave wavenumber field are obtained by calculating the spatial gradient of the P-wave phase field and S-wave phase field. The P-wave velocity field and S-wave velocity field are calculated according to the relationship between wave velocity and wave number.

4. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 3, characterized in that, The P-wave phase field and S-wave phase field are obtained by orthogonal demodulation of the time-domain displacement signal extracted from the surface micro-motion video data.

5. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 1, characterized in that, The construction of the high-dimensional feature vector in S5 specifically includes: For each pixel within the test area, the wave velocity ratio field characteristic value, P-wave amplitude field characteristic value, and S-wave amplitude field characteristic value calculated at each frequency of the set of resonant frequencies are concatenated to construct the high-dimensional feature vector.

6. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 1, characterized in that, The unsupervised learning anomaly detection algorithm is the Isolation Forest algorithm.

7. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 1, characterized in that, The identification and location of leak points in S5 specifically includes: An anomaly score map is generated based on the anomaly scores of each pixel, and adaptive threshold segmentation is performed on the anomaly score map to generate a binary mask image that identifies the leakage point.

8. The method for detecting, identifying, and locating building leakage points in construction engineering according to claim 7, characterized in that, After generating the binary mask image, morphological processing operations are also performed on it to eliminate noise and optimize the integrity of the leakage area.

9. A method for detecting, identifying, and locating building leakage points in construction engineering according to claim 1, characterized in that, After identifying and locating the leak points, the process also includes a step of visualizing the identified leak point areas by overlaying them in pseudo-color onto the original optical image.

10. A building leakage point detection, identification, and location system for building engineering, applied to the building leakage point detection, identification, and location method for building engineering as described in any one of claims 1-9, characterized in that, include: A composite orthogonal dual-mode excitation unit is used to apply controllable, orthogonal normal and tangential micro-vibrations to the surface of the building structure under test. A high frame rate optical imaging unit is used to non-contactly acquire video sequences of the micro-motion response of the surface of the tested building structure under excitation. The information processing and control unit is electrically connected or signal-connected to the composite orthogonal bimodal excitation unit and the high frame rate optical imaging unit, respectively. The information processing and control unit is used to send excitation control signals to the composite orthogonal bimodal excitation unit, send synchronization trigger and acquisition commands to the high frame rate optical imaging unit, and receive and process the micro-motion response video sequence acquired by the high frame rate optical imaging unit.

Images