An engineering structure vibration real-time intelligent monitoring method and system

Abstract

The application discloses an engineering structure vibration real-time intelligent monitoring method and system, and belongs to the technical field of engineering structure vibration monitoring. The technical problem to be solved is the contradiction between long-term monitoring and calculation resource consumption of existing visual monitoring, resulting in high calculation load and energy consumption. The technical solution points are as follows: S1. Multiple cameras are arranged to form a monitoring network for long-term guarding; S2. When the displacement pixel amplitude of a camera continuously exceeds a preset first displacement pixel threshold, it is determined that an effective vibration event occurs, and a high-precision analysis mode is switched to; S3. A main visual angle wakes up one or more cameras as a slave visual angle to synchronously track a vibration target; three-dimensional real displacement of the vibration target at different moments and vibration components in each direction are solved, and three-dimensional vibration displacement time history is obtained; and S4. Multi-dimensional vibration parameters are solved based on measured modal parameters, and engineering structure health state evaluation and dynamic early warning are carried out.

Description

Technical Field

[0001] This invention relates to the field of engineering structure vibration monitoring technology, specifically to a real-time intelligent monitoring method and system for engineering structure vibration, and particularly to a non-contact engineering structure vibration monitoring method and system based on a fixed camera and multi-view collaborative measurement. Background Technology

[0002] Engineering structures (such as high-rise buildings, long-span bridges, and historical pagodas) will vibrate under wind loads, earthquakes, traffic loads, etc. Long-term monitoring of structural vibration parameters (frequency, amplitude, damping ratio, mode shape) is of great significance for assessing the health status of structures and providing early warning of potential damage.

[0003] Currently, the mainstream structural vibration monitoring technologies include: (1) Traditional contact sensor method: accelerometers, strain gauges, displacement gauges and other sensors are installed on the structure, or vibration data is collected through wired or wireless networks.

[0004] (2) Existing visual monitoring methods: The structure vibration video was captured by a camera, and the displacement time history was extracted by algorithms such as digital image correlation (DIC) and optical flow method. The modal parameters were then further analyzed.

[0005] Traditional contact sensors require fixed contact installation on the structure being measured, which is costly and complex to install. Furthermore, point-based measurements make it difficult to obtain full-field vibration information of the structure. The sensors are also prone to aging and difficult to maintain in the long term, making them unsuitable for special scenarios such as historical buildings and high-temperature and high-pressure environments.

[0006] Existing visual monitoring continuously runs high-precision algorithms, which have a large computational load and high power consumption, making them unsuitable for long-term online monitoring. Single-view measurement can only acquire two-dimensional displacement within the plane and cannot detect out-of-plane vibration components. Multi-camera solutions mostly involve continuous synchronous acquisition, which consumes a lot of resources and lacks intelligent collaborative mechanisms. Summary of the Invention

[0007] The purpose of this invention is to provide a real-time intelligent monitoring method for vibration of engineering structures, and related technologies, to solve the technical problems of high computational load and energy consumption caused by the contradiction between long-term monitoring and computational resource consumption in existing visual monitoring, or a combination thereof.

[0008] In a first aspect, the present invention provides a method for real-time intelligent monitoring of vibration in engineering structures, comprising the following steps: S1. Multiple cameras are deployed around the engineering structure to form a monitoring network for long-term monitoring in a low-power monitoring mode; each camera continuously collects time-series video streams from the surface of the engineering structure. S2. In the time-series video stream of each camera, the displacement pixel amplitude of feature points in the monitoring area is calculated in real time. When the displacement pixel amplitude of a camera continuously exceeds the preset first displacement pixel threshold, it is determined that a valid vibration event has occurred, and the monitoring network switches from low-power guard mode to high-precision analysis mode. S3. In high-precision analysis mode, the camera that first detects vibration is used as the main perspective to track the vibration target; and one or more cameras are activated as subordinate perspectives to synchronously track the vibration target; two-dimensional displacement data of the vibration target under different perspectives are obtained; the two-dimensional displacement data of each perspective are fused to calculate the three-dimensional real displacement of the vibration target in three-dimensional space at different times and its vibration components in each direction, so as to obtain the three-dimensional vibration displacement time history. S4. Based on the three-dimensional vibration displacement time history, identify the measured modal parameters of the engineering structure, use the measured modal parameters and their preset benchmark values ​​to obtain multi-dimensional vibration parameters, and conduct health status assessment and dynamic early warning of the engineering structure based on the comparison of the multi-dimensional vibration parameters with their preset thresholds.

[0009] The first preferred solution is as follows: In step S2, the method for determining the first displacement pixel threshold is: based on the long-term monitoring data of the engineering structure under normal environmental vibration, the baseline value and standard deviation of the displacement pixel amplitude are statistically obtained, and the baseline value is added to 2-3 times the standard deviation as the first displacement pixel threshold.

[0010] The second preferred option: In step S2, the low-power guard mode is used: a lightweight optical flow method or frame difference method is adopted to calculate only the displacement amplitude and not to perform tracking.

[0011] The third preferred embodiment: In step S3, under the high-precision analysis mode, both the main-view camera and the subordinate-view camera track the vibration target based on the high-precision feature tracking algorithm; the high-precision feature tracking algorithm extracts the displacement of the vibration target based on the sub-pixel-level feature tracking algorithm.

[0012] The fourth preferred option: In step S3, the sub-pixel level feature tracking algorithm, the algorithm execution logic includes the following steps: For each frame of image, a local window ROI containing the vibration target of the monitored object is selected, centered on the coordinates of the target feature point in the current frame. Within this window, the spatial gradient of image grayscale in the horizontal and vertical directions, as well as the grayscale temporal change rate between adjacent frames, are calculated, and optical flow constraint equations are established. The optical flow constraint equation is solved by the least squares method to obtain the optical flow vector within the local window ROI. The displacement increment of the feature point in the next frame is obtained, the coordinates of the target feature point are updated, and the local window is extracted again. The above calculation is repeated until the displacement increment is less than its corresponding preset threshold. The final output is the sub-pixel coordinates of the target feature points in each frame, which are 0.1 pixels in unit, thus obtaining the sub-pixel coordinate sequence of the vibrating target in continuous time sequence.

[0013] The fifth preferred option: In step S3, the two-dimensional displacement data of the vibrating target under different perspectives is obtained. Specifically, based on the sub-pixel level feature tracking algorithm, different sub-pixel coordinate sequences of multiple cameras are obtained. The sub-pixel coordinates of the target feature points of each camera at the same time in the different sub-pixel coordinate sequences are extracted, and the sub-pixel coordinates of multiple target feature points are used as the two-dimensional displacement data of the vibrating target.

[0014] The sixth preferred solution: In step S3, the two-dimensional displacement data from various perspectives are fused: the two-dimensional displacement data from various perspectives are fused through multi-view geometric calculation; wherein, multi-view geometric calculation includes: Each camera is pre-calibrated to establish a viewpoint mapping relationship between image coordinates from different viewpoints, and a viewpoint mapping matrix is ​​obtained. The main viewpoint maps the image coordinates to preset bit commands of the subordinate viewpoint in real time through the viewpoint mapping matrix.

[0015] The seventh preferred option: Calculate the true three-dimensional displacement of the vibrating target, specifically: Based on the pinhole imaging model, the projection matrix is ​​used to represent the projection relationship between the three-dimensional point of the vibration target acquired by each camera in the world coordinate system and the sub-pixel coordinates of the imaging point in the image, and an overdetermined set of equations for the vibration target captured by multiple cameras is constructed. The intrinsic and extrinsic parameter matrices of each camera, which were pre-calibrated using Zhang's method, and the sub-pixel coordinates of the target feature points of the vibrating target at the same moment extracted from different perspectives, were used as inputs to the overdetermined equation system. The weighted least squares method was used to solve the overdetermined equation system to obtain the three-dimensional coordinates of the vibrating target in the world coordinate system at a certain moment. Substituting the subpixel coordinates at all times into the overdetermined equations, the three-dimensional coordinates of the vibrating target in the world coordinate system at different times are repeatedly calculated to obtain the three-dimensional coordinate sequence of the vibrating target in the world coordinate system. Using the initial 3D coordinates in the 3D coordinate sequence as reference coordinates, the displacement of the 3D coordinates relative to the reference coordinates at each time step is calculated as the true 3D displacement; ultimately, a true 3D displacement sequence is formed.

[0016] Eighth preferred option: Calculate the vibration components of the vibrating target: The three vibration components of the three-dimensional true displacement at each moment are obtained, namely the vibration components in the rightward direction with the wind, the rightward direction with the crosswind, and the vertical upward direction, to obtain the three-dimensional vibration displacement time history.

[0017] The ninth preferred option: In step S4, the measured modal parameters of the engineering structure are identified based on the vibration components extracted from the three-dimensional vibration displacement time history, including the identification of natural frequency, damping ratio and mode shape.

[0018] Eleventh preferred option: Identification of inherent frequencies, specifically: The three-dimensional vibration displacement time history data is segmented, with 50% overlap between adjacent segments. Calculate the square of the fast Fourier transform amplitude of each data segment to obtain the power spectrum of that segment; The final power spectral density curve is obtained by arithmetically averaging the power spectra of all segments. In the power spectral density curve, significant peak frequencies are automatically identified through a peak search algorithm, with each peak corresponding to a natural frequency.

[0019] The twelfth preferred option: Damping ratio identification: The damping ratio at each natural frequency is calculated using the half-power bandwidth method.

[0020] Thirteenth preferred option: Mode shape identification: At the natural frequency, obtain the Fourier transform amplitude and phase of the displacement signal of the engineering structure at each measuring point, and after normalization, obtain the r-th order mode shape vector. .

[0021] The fourteenth preferred option: multidimensional vibration parameters, including the rate of change of frequency and the mode shape correlation coefficient.

[0022] The fifteenth preferred option: Based on the comparison of the multidimensional vibration parameters with their preset thresholds, an assessment of the structural health status is performed. When the rate of change of frequency exceeds its preset threshold, it is determined that the stiffness of the engineering structure may be degraded. Modal correlation coefficient This indicates a significant change in the vibration mode of the engineering structure, suggesting possible localized damage.

[0023] Secondly, the present invention provides a real-time intelligent monitoring system for vibration of engineering structures, used to perform the aforementioned method, including: The data sensing module includes a monitoring network consisting of multiple fixed cameras, used to perform data acquisition and preprocessing of the engineering structure; Collaborative control central module: used to perform displacement pixel amplitude and first displacement pixel threshold calculation and comparison on the data collected by the monitoring network; and to control the linkage of multi-view cameras, switching from low-power monitoring mode to high-precision analysis mode. Data processing and analysis module: In high-precision analysis mode, it integrates two-dimensional displacement data from various perspectives, uses the intrinsic and extrinsic parameter matrices of each camera, and the two-dimensional displacement data to calculate the true displacement of the vibrating target at different times and its vibration components in each direction; identifies the measured modal parameters of the engineering structure, and uses the measured modal parameters and their preset benchmark values ​​to obtain multi-dimensional vibration parameters for assessing the health status of the engineering structure. Early warning module: Based on the health status assessment information of the engineering structure, it generates and publishes early warning information, records video summaries and timestamps of abnormal vibration moments, and circles the abnormal vibration area on key frames.

[0024] The first preferred option is a camera, including a high-definition camcorder, a telephoto lens, or an industrial camera.

[0025] Compared with the prior art, the present invention has the following beneficial effects: (1) High-efficiency and energy-saving long-term monitoring capability: By introducing a displacement pixel threshold triggering mechanism, the system operates in a low-power standby mode most of the time, and only starts high-precision analysis when a significant vibration event is detected, which greatly reduces the computational load and energy consumption, enabling efficient and energy-saving long-term monitoring capabilities. It is particularly suitable for long-term application scenarios that require continuous monitoring 24 / 7.

[0026] The vibration monitoring of engineering structures is divided into two working modes: "monitoring" and "analysis". The displacement pixel amplitude calculated by the video stream is used as the trigger condition. The "displacement pixel threshold triggering" mechanism is used to start high-precision analysis as needed, realizing the switching of the monitoring network from low-power monitoring mode to high-precision analysis mode. This solves the contradiction between long-term monitoring and computing resource consumption.

[0027] This invention achieves non-contact, three-dimensional, and intelligent vibration monitoring of engineering structures by combining a displacement pixel threshold triggering mechanism with multi-view linkage and coordination, thus balancing low power consumption for long-term monitoring with high precision for event triggering.

[0028] (2) Improve the accuracy of vibration target perception in engineering structures: By using multi-view linkage measurement, the limitation of a single camera being able to measure only two-dimensional displacement in a plane is overcome. It can accurately acquire the real motion trajectory and vibration components in each direction in three-dimensional space, realize three-dimensional multi-directional vibration perception, provide more complete input data for structural dynamics analysis, and improve the accuracy of vibration target perception in engineering structures.

[0029] (3) Multiple cameras forming a monitoring network result in a fast response speed: Through the visual sensor network collaborative mode of "primary view triggering + subordinate view automatic wake-up", the monitoring network achieves self-organized linkage without manual intervention, with fast response speed and high reliability.

[0030] "Master-slave multi-view automatic linkage": The triggering view is the master view, and the subordinate view is automatically awakened for collaborative tracking, realizing the self-organized collaborative perception of the visual sensor network.

[0031] (4) Non-contact, full-field measurement: No sensors need to be installed on the structure; the upgrade can be achieved using existing surveillance cameras, making it especially suitable for scenarios where it is not advisable to access historical buildings, tall structures, dilapidated buildings, or other unsafe locations. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method of the present invention.

[0033] Figure 2 This is a timing diagram of the method of the present invention.

[0034] Figure 3 This is a flowchart of the sub-pixel-level feature tracking algorithm of the present invention.

[0035] Figure 4 This is a schematic diagram of the system structure of the present invention.

[0036] Figure 5 This is a schematic diagram of the camera deployment according to Embodiment 1 of the present invention.

[0037] Figure 6 This is a schematic diagram of camera deployment according to Embodiment 2 of the present invention. Detailed Implementation

[0038] The following non-limiting embodiments are intended to enable those skilled in the art to gain a more comprehensive understanding of the present invention, but do not limit the invention in any way. The following content is merely an exemplary description of the scope of protection claimed by the present invention, and those skilled in the art can make various changes and modifications to the present invention based on the disclosed content, and such changes should also fall within the scope of protection claimed by the present invention.

[0039] Combination Figures 1-3 As shown, this invention provides a real-time intelligent monitoring method for vibration of engineering structures, comprising the following steps: S1: Fixed camera network deployment and long-term monitoring: Multiple fixed cameras with complementary viewing angles are deployed at key locations around the engineering structure to form a monitoring network. Each camera continuously collects a time-series video stream of the structure's surface.

[0040] S2: Intelligent triggering of vibration events based on displacement pixel thresholds: In the time-series video streams of each camera, the displacement pixel amplitude of feature points within the monitoring area is calculated in real time. When the displacement pixel amplitude of a camera continuously exceeds the preset first displacement pixel threshold, it is determined that a valid vibration event has occurred, and the monitoring network switches from low-power monitoring mode to high-precision analysis mode.

[0041] Low-power monitoring mode: Employs lightweight optical flow or frame difference methods, only calculating the displacement pixel amplitude, without performing high-precision tracking.

[0042] High-precision analysis mode: Activate sub-pixel-level feature tracking algorithms (such as improved LK optical flow method or deep learning feature matching) to obtain high-precision displacement time history.

[0043] The average optical flow amplitude of feature points within the monitoring area is calculated in real time using the optical flow method as the displacement pixel amplitude. The method for determining the first displacement pixel threshold is as follows: based on long-term monitoring data of the structure under normal environmental vibration, the baseline value and standard deviation of the displacement amplitude are statistically obtained, and the baseline value is added to 2-3 times the standard deviation as the first displacement pixel threshold.

[0044] S3: Multi-view collaborative tracking and three-dimensional vibration displacement time history extraction: First, the camera that detects the vibration is designated as the primary viewpoint. A high-precision feature tracking algorithm is then initiated, and this camera is locked as the primary viewpoint throughout the vibration event. Based on a preset viewpoint mapping relationship, one or more subordinate viewpoint cameras are automatically activated. Through feature point matching between multiple viewpoints, the corresponding image points of the same vibrating target in the images of each camera are determined, thereby guiding the subordinate viewpoints to synchronously track the same vibrating target using the high-precision feature tracking algorithm; and acquiring two-dimensional displacement data of the vibrating target from different viewpoints.

[0045] By performing multi-view geometric calculations and integrating two-dimensional displacement data from various perspectives, the true displacement and vibration components in each direction of the vibrating target in three-dimensional space at different times are calculated, thus obtaining the three-dimensional vibration displacement time history.

[0046] Among them, multi-view geometric calculation: Each camera is pre-calibrated to establish a viewpoint mapping relationship between image coordinates from different viewpoints, and a viewpoint mapping matrix is ​​obtained. The main viewpoint maps image coordinates to preset position commands of subordinate viewpoints in real time using a viewpoint mapping matrix. S3.1 high-precision feature tracking of the main viewpoint acquires two-dimensional displacement data of the vibrating target from the main viewpoint: When the displacement pixel amplitude continuously exceeds the preset first displacement pixel threshold, the main view automatically switches from low power consumption guard mode to high precision analysis mode. In high precision analysis mode, a sub-pixel-level feature tracking algorithm is launched to extract the displacement of the vibration target in the main view image with high precision and obtain the two-dimensional displacement data of the vibration target under the main view.

[0047] Subpixel-level feature tracking algorithm execution logic: For each frame of image, the system selects a local window ROI containing the monitored object, centered on the coordinates of the target feature points in the current frame.

[0048] Within this window, the image grayscale is calculated in the horizontal direction ( ), vertical direction ( The spatial gradient and the rate of change of grayscale over time between adjacent frames ( ) Based on the assumption of grayscale invariance, an optical flow constraint equation is established. The optical flow vector within the window is solved using the least squares method to obtain the displacement increment of the feature point in the next frame.

[0049] To achieve sub-pixel accuracy, an iterative solution method is adopted: after each calculation, the feature point coordinates are updated, the local window is re-extracted, and the above calculation is repeated until the displacement increment is less than a preset threshold (such as 0.05 pixels).

[0050] The final output is the sub-pixel coordinates (accurate to 0.1 pixels) of the target feature points in each frame. The continuous temporal sequence of sub-pixel coordinates is denoted as... .

[0051] S3.2 Automatic wake-up from the subordinate perspective to obtain two-dimensional displacement data of the vibrating target from the subordinate perspective: Once the main viewpoint switches to high-precision analysis mode, a "Start High-Precision Tracking" command is sent to all subordinate viewpoint cameras via the control command bus. This command does not contain any position adjustment information.

[0052] The subordinate cameras are all fixedly installed, and their respective imaging ranges and overlapping fields of view are pre-calibrated. Upon receiving a command, each subordinate camera synchronously begins to execute the same high-precision feature tracking algorithm (same as S3.1) on the spatial region in its own frame corresponding to the target in the main view (i.e., through hardware or pre-calibrated ROI parameters) to acquire the two-dimensional displacement data of the vibrating target from the subordinate viewpoint. This process is completed automatically without manual intervention.

[0053] S3.3 Multi-view synchronous acquisition and timestamp alignment: To ensure temporal consistency of displacement data across different viewpoints, all cameras receive the same external synchronization signal (such as a GPS second pulse or PTP clock) in hardware and expose simultaneously at a fixed frame rate (e.g., 30fps). Each frame is recorded with a uniform timestamp (accurate to the millisecond level). The video streams captured by each camera are aligned with the timestamps.

[0054] S3.4 Integrate the two-dimensional displacement data from various perspectives to perform a three-dimensional vibration displacement time history solution: At the same time, the three cameras each extract the sub-pixel coordinates of the target feature points of the vibrating target (i.e., the two-dimensional displacement data of each viewpoint). As input, each camera has been pre-calibrated using Zhang's method to obtain its intrinsic parameter matrix (including focal length, principal point coordinates, and distortion coefficients) and extrinsic parameter matrix (rotation matrix and translation vector relative to the world coordinate system).

[0055] By combining the sub-pixel coordinates pi=(u,v) of the target feature points recorded by different cameras at various times with the corresponding intrinsic and extrinsic parameter matrices K and t, P in three-dimensional world coordinates can be obtained by solving the system of equations.

[0056] Detailed solution steps: Each camera satisfies the pinhole imaging model, which is a three-dimensional point of the vibrating target in the world coordinate system. Rather than the imaging point in the image The projection relationship can be represented by a projection matrix. For three cameras, each camera gives two independent equations, for a total of six equations, while the unknowns are... The three variables form an overdetermined system of equations.

[0057] The overdetermined system of equations is solved using the weighted least squares method. The intrinsic parameter matrices for each camera are known. Distortion coefficient extrinsic parameter matrix (rotation matrix) Translation vector The above parameters can be obtained before monitoring begins using Zhang's calibration method. The intrinsic and extrinsic parameter matrices of each camera, which have been pre-calibrated using Zhang's method, as well as the sub-pixel coordinates of the target feature points of the vibrating target extracted from different viewpoints at the same time, are used as inputs to the overdetermined equation system. The overdetermined equation system is solved using the weighted least squares method to obtain the three-dimensional coordinates of the vibrating target in the world coordinate system at a certain time.

[0058] 3D coordinates in the world coordinate system satisfy:

[0059] Repeat the above calculation for all time points to obtain the three-dimensional coordinate sequence. Using the three-dimensional coordinates at the initial moment (i.e., when the structure is at rest or in its reference state). For reference, the displacement of the three-dimensional coordinates relative to the reference coordinates at each moment is calculated as the three-dimensional true displacement, ultimately forming a three-dimensional true displacement sequence:

[0060] : Three-dimensional true displacement sequence; The three components are denoted as follows: (To the right with the wind) (Crosswind to the right) and (Vertical upward direction), to obtain the final three-dimensional vibration displacement time history.

[0061] S4: Vibration Parameter Identification and Intelligent Diagnosis Based on three-dimensional vibration time-series data, multidimensional frequencies, mode shapes, and damping ratios of the structure are extracted. The extracted modal parameters are then input into a pre-trained damage diagnosis model to assess the structural health status.

[0062] S4: Vibration Parameter Identification and Intelligent Diagnosis Based on the three-dimensional vibration displacement time history, the measured modal parameters of the engineering structure are identified. Multidimensional vibration parameters are obtained by using the measured modal parameters and their preset benchmark values. The health status assessment and dynamic early warning of the engineering structure are carried out by comparing the change ratio with its preset threshold.

[0063] S4.1 Identification of measured modal parameters: Based on the extracted three-dimensional vibration displacement time history Identify the measured modal parameters of engineering structures, including the natural frequencies, damping ratios, and mode shapes.

[0064] S4.1.1 Power spectral density: The three-dimensional vibration displacement time history data is segmented, with a 50% overlap between adjacent segments. The square of the Fast Fourier Transform (FFT) amplitude of each segment is calculated to obtain the power spectrum of that segment. The power spectra of all segments are then arithmetically averaged to obtain the final power spectral density curve.

[0065] S4.1.2 Natural frequency: In the power spectral density curve, significant peak frequencies are automatically identified using a peak search algorithm. For typical structures such as high-rise buildings, each peak corresponds to a natural frequency, and first-order frequencies, second-order frequencies, etc., can usually be obtained.

[0066] S4.1.3 Damping ratio: The damping ratio at each natural frequency is calculated using the half-power bandwidth method. (Peak frequency is used as the reference.) Using the center as the reference point, find the two frequency points on the power spectrum curve corresponding to the point before and after where the power is half of the peak frequency. and The formula for calculating the damping ratio is: ; S4.1.4 Mode shape (if multiple measurements are taken): If multiple sets of feature points are arranged at different heights (or different measuring points) in the engineering structure, the above-mentioned three-dimensional true displacement extraction is repeated for each measuring point to obtain the three-dimensional vibration displacement time history of each measuring point. At the natural frequency, the Fourier transform amplitude and phase of the displacement signal of each measuring point are taken, and after normalization, the r-th order mode vector is obtained. .

[0067] S4.2 Structural Health Status Diagnosis The system compares the measured modal parameters with preset benchmark values ​​(from the finite element model or the first measurement under the healthy state of the structure).

[0068] Rate of change of frequency: Calculation .when If the value exceeds a preset threshold (e.g., 5%), it is determined that the stiffness of the engineering structure may be degraded.

[0069] Modal correlation coefficient (MAC): A modal confidence criterion for calculating the relationship between the measured mode shape and the reference mode shape.

[0070] when When this occurs, it indicates a significant change in the vibration mode, which may indicate localized damage.

[0071] S5: Dynamic Early Warning When multidimensional vibration parameters ("rate of change of frequency and / or mode shape correlation coefficient") exceed the preset threshold or show abnormal trends, a multi-level early warning is triggered.

[0072] like Figure 4 As shown, the present invention also provides a real-time intelligent monitoring system for vibration of engineering structures, comprising: The data sensing module includes a monitoring network consisting of multiple fixed cameras, used to perform data acquisition and preprocessing of the engineering structure; Collaborative control central module: used to perform displacement pixel amplitude and first displacement pixel threshold calculation and comparison on the data collected by the monitoring network; and to control the linkage of multi-view cameras, switching from low-power monitoring mode to high-precision analysis mode. Data processing and analysis module: In high-precision analysis mode, it integrates two-dimensional displacement data from various perspectives, uses the intrinsic and extrinsic parameters of each camera and the two-dimensional displacement data to calculate the three-dimensional real displacement of the vibrating target at different times and its vibration components in each direction; identifies the measured modal parameters of the engineering structure, obtains multi-dimensional vibration parameters, and conducts health status assessment of the engineering structure. Early warning module: Based on the health status assessment information of the engineering structure, it generates and publishes early warning information, records video summaries and timestamps of abnormal vibration moments, and circles the abnormal vibration area on key frames.

[0073] The data sensing module consists of a monitoring network composed of cameras (including high-definition cameras, telephoto lenses, industrial cameras, etc.) and includes a data acquisition and preprocessing unit.

[0074] The collaborative control central module includes a threshold judgment unit, a view scheduling unit, and a data synchronization and fusion unit.

[0075] Threshold judgment unit: used to perform displacement pixel amplitude calculation and comparison with the first displacement pixel threshold; the average optical flow amplitude of feature points in the monitoring area is calculated in real time using the optical flow method as the displacement pixel amplitude; the method for determining the first displacement pixel threshold is: based on long-term monitoring data of the structure under normal environmental vibration, the baseline value and standard deviation of the displacement amplitude are statistically obtained, and the baseline value is added to 2-3 times the standard deviation as the first displacement pixel threshold.

[0076] Viewpoint Scheduling Unit: Used to execute multi-viewpoint linkage control logic; pre-calibrates each camera to establish a mapping relationship between image coordinates from different viewpoints, forming a viewpoint mapping matrix; converts the image coordinates of the main viewpoint into preset bit commands of the subordinate viewpoints in real time through the viewpoint mapping matrix. Data synchronization and fusion unit: responsible for timestamp alignment and 2D displacement data fusion of multiple video streams. Specifically, using the intrinsic and extrinsic parameter matrices of each camera, it establishes projection equations between 3D spatial points and 2D image coordinates from each viewpoint. It then solves the 3D vibration displacement time history using the least squares method or Kalman filtering, outputting the vibration components of the structure in the X (horizontal transverse), Y (horizontal longitudinal), and Z (vertical) directions.

[0077] Example 1: High-rise building wind-induced vibration monitoring Application scenario: Long-term vibration monitoring of a 300-meter super high-rise building under wind load. Deployment Plan: Install three fixed long-focus cameras on the rooftops of adjacent buildings, all aimed at key vibration-prone areas of the buildings. For high-rise buildings, aim at the rooftop structure. The three cameras should be positioned at different angles: Camera 1 (main view) should be positioned across the wind, Camera 2 (main view) should be positioned with the wind at a 90-degree angle to Camera 1, and the remaining cameras (secondary view) should be positioned at a certain angle to Cameras 1 and 2. Figure 5 As shown.

[0078] Workflow: During the monitoring phase, the three cameras continuously acquire video streams at a frame rate of 15fps. Each camera independently runs a lightweight optical flow algorithm to calculate the average optical flow amplitude of the monitored area. The first displacement pixel threshold is preset to 0.5 pixels (based on the historical quiet period statistical baseline of 0.2 pixels + 3 × 0.1 pixel standard deviation).

[0079] During the triggering phase, under typhoon weather, camera 1 detected that the average optical flow amplitude rose to 0.7 pixels, exceeding the 0.5 pixel threshold. The threshold was exceeded for 5 consecutive frames, and the system determined it to be a valid vibration event. Camera 1 automatically switched from low-power monitoring mode to high-precision analysis mode.

[0080] During the linkage phase, camera 1 sends the image coordinates to the collaborative control center module. The collaborative control center module calculates the preset position parameters of camera 2 and camera 3 through the pre-calibrated viewpoint mapping matrix, and sends linkage commands to camera 2 and camera 3. The two cameras synchronously switch to high-precision analysis mode, and the three cameras acquire data synchronously at 30fps, with timestamp alignment accuracy reaching the millisecond level.

[0081] By using the intrinsic and extrinsic parameters of three cameras, each camera independently extracts subpixel-level displacement time histories, and vibration components in the X (downwind), Y (crosswind), and Z (vertical) directions can be obtained respectively. Extraction results: Maximum amplitude in the downwind direction ±15cm, maximum amplitude in the crosswind direction ±8cm, and vertical amplitude ±3cm.

[0082] Parameter identification and early warning: Spectrum analysis identified the fundamental frequency of the structure as 0.18Hz, which deviates from the design value of 0.19Hz by approximately 5%. Comparison with historical data revealed that the deviation was within the normal range. No alert was triggered; only data was recorded.

[0083] Example 2: Multi-view collaborative monitoring of long-span bridges.

[0084] Application scenario: Vibration monitoring at mid-span of the main span of a suspension bridge.

[0085] like Figure 6 As shown, the deployment plan is as follows: Main view camera 1: On the land on the shore at one end of the bridge pier, the line of sight is along the bridge and aimed at the middle of the span; Main view camera 2: On the land on the other side of the bridge pier, the line of sight is along the bridge and aimed at the middle of the span; Subordinate perspective camera: The land on the shore is a certain distance from the bridge pier, and the line of sight is at a certain angle to the bridge direction, aimed at the middle of the span.

[0086] Key parameters: First displacement pixel threshold: 0.3 pixels; Linkage delay: <50ms; Three-dimensional displacement measurement accuracy: ±1mm (10m distance); Implementation results: This system successfully captured the vertical vibration (amplitude 12mm), lateral sway (amplitude 3mm), and longitudinal drift (amplitude 2mm) of the bridge during a heavy vehicle passage event, providing complete measured data for the analysis of the bridge's dynamic characteristics.

[0087] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for real-time intelligent monitoring of vibration in engineering structures, characterized in that, Includes the following steps: S1. Multiple cameras are deployed around the engineering structure to form a monitoring network for long-term monitoring in a low-power monitoring mode. Each camera continuously captures a time-series video stream of the engineering structure's surface; S2. In the time-series video stream of each camera, the displacement pixel amplitude of feature points in the monitoring area is calculated in real time. When the displacement pixel amplitude of a camera continuously exceeds the preset first displacement pixel threshold, it is determined that a valid vibration event has occurred, and the monitoring network switches from low-power guard mode to high-precision analysis mode. S3. In high-precision analysis mode, the camera that first detects vibration is used as the main perspective to track the vibration target; and one or more cameras are activated as subordinate perspectives to synchronously track the vibration target; two-dimensional displacement data of the vibration target under different perspectives are obtained; the two-dimensional displacement data of each perspective are fused to calculate the three-dimensional real displacement of the vibration target in three-dimensional space at different times and its vibration components in each direction, so as to obtain the three-dimensional vibration displacement time history. S4. Based on the three-dimensional vibration displacement time history, identify the measured modal parameters of the engineering structure, use the measured modal parameters and their preset benchmark values ​​to obtain multi-dimensional vibration parameters, and conduct health status assessment and dynamic early warning of the engineering structure based on the comparison of the multi-dimensional vibration parameters with their preset thresholds.

2. The method according to claim 1, characterized in that, In step S2, the method for determining the first displacement pixel threshold is as follows: based on long-term monitoring data of the engineering structure under normal environmental vibration, the baseline value and standard deviation of the displacement pixel amplitude are statistically obtained, and the baseline value plus 2-3 times the standard deviation is used as the first displacement pixel threshold.

3. The method according to claim 1, characterized in that, In step S3, under the high-precision analysis mode, both the main-view camera and the subordinate-view camera track the vibration target based on the high-precision feature tracking algorithm; the high-precision feature tracking algorithm extracts the displacement of the vibration target based on the sub-pixel-level feature tracking algorithm.

4. The method according to claim 3, characterized in that, In step S3, the sub-pixel level feature tracking algorithm includes the following steps: For each frame of image, a local window ROI containing the vibration target of the monitored object is selected, centered on the coordinates of the target feature point in the current frame. Within this window, the spatial gradient of image grayscale in the horizontal and vertical directions, as well as the grayscale temporal change rate between adjacent frames, are calculated, and optical flow constraint equations are established. The optical flow constraint equation is solved by the least squares method to obtain the optical flow vector within the local window ROI. The displacement increment of the feature point in the next frame is obtained, the coordinates of the target feature point are updated, and the local window is extracted again. The above calculation is repeated until the displacement increment is less than its corresponding preset threshold. The final output is the sub-pixel coordinates of the target feature points in each frame, which are 0.1 pixels in unit, thus obtaining the sub-pixel coordinate sequence of the vibrating target in continuous time sequence.

5. The method according to claim 1, characterized in that, In step S3, two-dimensional displacement data of the vibrating target from different perspectives are obtained. Specifically, based on the sub-pixel coordinate sequences of multiple cameras obtained by the sub-pixel level feature tracking algorithm, the sub-pixel coordinates of the target feature points of each camera at the same time in the different sub-pixel coordinate sequences are extracted, and the sub-pixel coordinates of multiple target feature points are used as the two-dimensional displacement data of the vibrating target.

6. The method according to claim 1, characterized in that, In step S3, the two-dimensional displacement data from various perspectives are fused: the two-dimensional displacement data from various perspectives are fused through multi-view geometric calculation; wherein, multi-view geometric calculation includes: Each camera is pre-calibrated to establish a viewpoint mapping relationship between image coordinates from different viewpoints, and a viewpoint mapping matrix is ​​obtained. The main viewpoint maps the image coordinates to preset bit commands of the subordinate viewpoint in real time through the viewpoint mapping matrix.

7. The method according to claim 6, characterized in that, The three-dimensional true displacement and its vibration components in each direction of the vibrating target in three-dimensional space at different times are calculated to obtain the three-dimensional vibration displacement time history, specifically: Based on the pinhole imaging model, the projection matrix is ​​used to represent the projection relationship between the three-dimensional point of the vibration target acquired by each camera in the world coordinate system and the sub-pixel coordinates of the imaging point in the image, and an overdetermined set of equations for the vibration target captured by multiple cameras is constructed. The intrinsic and extrinsic parameter matrices of each camera, which were pre-calibrated using Zhang's method, and the sub-pixel coordinates of the target feature points of the vibrating target at the same moment extracted from different perspectives, were used as inputs to the overdetermined equation system. The weighted least squares method was used to solve the overdetermined equation system to obtain the three-dimensional coordinates of the vibrating target in the world coordinate system at a certain moment. Substituting the subpixel coordinates at all times into the overdetermined equations, the three-dimensional coordinates of the vibrating target in the world coordinate system at different times are repeatedly calculated to obtain the three-dimensional coordinate sequence of the vibrating target in the world coordinate system. Using the initial 3D coordinates in the 3D coordinate sequence as reference coordinates, the displacement of the 3D coordinates relative to the reference coordinates at each time step is calculated as the true 3D displacement; ultimately forming a 3D true displacement sequence. The three vibration components of the three-dimensional true displacement at each moment are obtained, namely the vibration components in the rightward direction with the wind, the rightward direction with the crosswind, and the vertical upward direction, to obtain the three-dimensional vibration displacement time history.

8. The method according to claim 1, characterized in that, In step S4, the measured modal parameters of the engineering structure are identified based on the three-dimensional vibration displacement time history, including the identification of natural frequencies, damping ratios, and mode shapes; the identification of natural frequencies; The three-dimensional vibration displacement time history data is segmented, with 50% overlap between adjacent segments. Calculate the square of the fast Fourier transform amplitude of each data segment to obtain the power spectrum of that segment; The final power spectral density curve is obtained by arithmetically averaging the power spectra of all segments. In the power spectral density curve, significant peak frequencies are automatically identified using a peak search algorithm, with each peak corresponding to a natural frequency; Damping ratio identification: The damping ratio at each natural frequency is calculated using the half-power bandwidth method; Mode shape identification: At the natural frequency, the Fourier transform amplitude and phase of the displacement signals of the engineering structure at each measuring point are obtained, and after normalization, the r-th order mode shape vector is obtained. .

9. The method according to claim 1, characterized in that, Multidimensional vibration parameters, including the rate of change of frequency and mode shape correlation coefficient; based on the comparison of the multidimensional vibration parameters with their preset thresholds, the health status of the engineering structure is assessed. When the rate of change of frequency exceeds its preset threshold, it is determined that the stiffness of the engineering structure may be degraded. Modal correlation coefficient This indicates a significant change in the vibration mode of the engineering structure, suggesting possible localized damage.

10. A real-time intelligent monitoring system for vibration of engineering structures, characterized in that, For performing the method according to any one of claims 1-9: comprising: The data sensing module includes a monitoring network consisting of multiple fixed cameras, used to perform data acquisition and preprocessing of the engineering structure; Collaborative control central module: used to perform displacement pixel amplitude and first displacement pixel threshold calculation and comparison on the data collected by the monitoring network; and to control the linkage of multi-view cameras, switching from low-power monitoring mode to high-precision analysis mode. Data processing and analysis module: In high-precision analysis mode, it integrates two-dimensional displacement data from various perspectives, uses the intrinsic and extrinsic parameter matrices of each camera, and the two-dimensional displacement data to calculate the true displacement of the vibrating target at different times and its vibration components in each direction; identifies the measured modal parameters of the engineering structure, and uses the measured modal parameters and their preset benchmark values ​​to obtain multi-dimensional vibration parameters for assessing the health status of the engineering structure. Early warning module: Based on the health status assessment information of the engineering structure, it generates and publishes early warning information, records video summaries and timestamps of abnormal vibration moments, and circles the abnormal vibration area on key frames.

Images