Water conservancy and hydropower gate three-dimensional deformation monitoring method and system based on visual inspection

Abstract

The invention provides a hydraulic and hydroelectric gate three-dimensional deformation monitoring method and system based on visual inspection, and the method comprises the steps: S1, collecting a high-frequency image sequence through a camera array installed at the periphery of a gate, and extracting a displacement data field reflecting oscillation deformation from the sequence; s2, calculating a corresponding frequency domain spectrogram according to the extracted displacement data field, and obtaining frequency component distribution of oscillation deformation; s3, identifying a dominant frequency peak value from the frequency domain spectrogram, and if the peak value exceeds a preset oscillation threshold value, determining a high-risk oscillation mode and marking a related timestamp; and S4, backtracking to the original image sequence aiming at the marked timestamp, and tracking the motion trail of the gate surface point to obtain a local deformation vector group. According to the invention, oscillation deformation and damage areas can be accurately identified, and the accuracy and real-time performance of gate safety monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent water conservancy monitoring technology, and in particular to a method and system for monitoring the three-dimensional deformation of water conservancy and hydropower gates based on visual detection. Background Technology

[0002] Hydropower projects are critical infrastructure for ensuring water resource allocation and flood control, and gates, as core components, directly impact the safety and efficiency of these projects due to their operational stability. In recent years, with the expansion of hydropower projects and the increasing complexity of operating environments, gate deformation monitoring during opening and closing has become a crucial research area for ensuring structural safety. Traditional monitoring methods, such as contact sensors or static measurement techniques, while capable of capturing gate deformation to some extent, often face limitations in adapting to complex water flow environments and dynamic oscillations. Specifically, these methods are insufficient in capturing high-frequency dynamic deformation, failing to comprehensively reflect the true deformation state of the gate under the coupled effects of water flow pulsation and mechanical vibration. Furthermore, traditional technologies are susceptible to interference in complex environments, and their data accuracy and real-time performance are insufficient to meet the demands of highly dynamic scenarios, resulting in an incomplete assessment of gate structural safety.

[0003] In this field, the oscillatory deformation of gates caused by water flow pulsation and mechanical vibration during opening and closing is a core challenge. The primary problem is the difficulty in accurately capturing the dynamic characteristics of oscillatory deformation. Water flow pulsation and mechanical vibration cause high-frequency oscillations in gates, and the complex temporal and spatial variations of these oscillations are difficult to record accurately using low-frequency or static measurements. For example, during high-flow-rate flood discharge, gates may experience minute but high-frequency deformations due to water flow impact. If these deformations are not captured in time, they may mask early signs of structural fatigue. Furthermore, due to insufficient frequency domain characteristic analysis of oscillatory deformation, existing technologies struggle to effectively correlate deformation data with gate structural safety. The lack of in-depth analysis of the frequency domain characteristics of oscillatory deformation makes it impossible to accurately determine which deformation modes may cause structural damage. For example, during the rapid opening and closing of a hydroelectric power station gate, specific frequency oscillations induced by water flow pulsation may lead to localized stress concentration, but existing monitoring systems struggle to identify the correlation between this frequency characteristic and potential crack risk.

[0004] Therefore, how to accurately capture the frequency domain characteristics of gate oscillation deformation through high-frequency sampling visual inspection technology, and establish a correlation assessment between deformation and structural safety, has become the key issue of this study. Summary of the Invention

[0005] A first aspect of the present invention provides a method for monitoring the three-dimensional deformation of hydraulic and hydroelectric gates based on visual inspection, the method comprising: S1. High-frequency image sequences are acquired using a camera array installed around the gate, and displacement data fields reflecting oscillating deformation are extracted from the sequences. S2. The corresponding frequency domain spectrum is calculated based on the extracted displacement data fields to obtain the frequency component distribution of the oscillating deformation. S3. The dominant frequency peak is identified from the frequency domain spectrum. If the peak exceeds a preset oscillation threshold, it is determined to be a high-risk oscillation mode and a relevant timestamp is marked. S4. The original image sequence is traced back to the marked timestamps to track the movement trajectory of points on the gate surface, obtaining a local deformation vector set. S5. The local deformation vector set is matched with a pre-established structural model. If the matching deviation is greater than a preset safety margin threshold, it is judged as a potential structural damage area and a coordinate set is output. S6. The associated frequency component distribution is obtained from the output coordinate set, the correspondence between deformation modes and damage types is classified, and the safety assessment level is determined. S7. A frequency domain feature report is generated based on the safety assessment level. The report is compared with historical data. If the comparison result shows an abnormal trend, the parameters of the structural model are updated to optimize subsequent monitoring.

[0006] Optionally, step S1 involves acquiring a high-frequency image sequence using a camera array installed around the gate, and extracting displacement data fields reflecting oscillating deformation from the sequence, including: Step S11: High-frequency image sequences are acquired by an array of cameras installed around the gate to obtain initial sequence data; Step S12: Based on the initial sequence data, process the high-frequency image sequence to determine the displacement change reflected by the oscillating deformation; Step S13: If the displacement change exceeds the preset displacement change threshold, perform time-series statistics on the displacement vector field for the displacement change, calculate the average displacement amplitude as the displacement data field, and obtain the deformation monitoring field that reflects the oscillation deformation.

[0007] Optionally, step S12, processing the high-frequency image sequence based on the initial sequence data to determine the displacement change reflected by the oscillation deformation, includes: processing the high-frequency image sequence using the OpenCV Farneback algorithm, wherein the OpenCV Farneback algorithm is based on the assumption of constant brightness, estimates the pixel motion in consecutive image frames, the input is the high-frequency image sequence, and the output is a displacement vector field representing the displacement change of each pixel.

[0008] Optionally, step S2, calculating the corresponding frequency domain spectrum based on the extracted displacement data field to obtain the frequency component distribution of the oscillating deformation, includes: Step S21: Obtain displacement data fields from the monitoring equipment; Step S22: Convert the displacement data field from the time domain to the frequency domain and calculate the preliminary frequency components; Step S23: For the initial frequency components, identify the spectral peak positions, obtain the amplitude energy distribution, determine the harmonic component decomposition results by calculating the power spectral density, and form the vibration mode; Step S24: For the vibration mode, perform distribution mapping based on the correspondence between frequency and amplitude to obtain the frequency component distribution characteristics.

[0009] Optionally, in step S22, the displacement data field is converted from the time domain to the frequency domain to calculate the preliminary frequency components, which include frequency values ​​and corresponding amplitude values.

[0010] Optionally, step S23, which identifies the spectral peak position and obtains the amplitude energy distribution for the preliminary frequency components, determines the harmonic component decomposition result by calculating the power spectral density, and forms a vibration mode, includes: the harmonic component decomposition result includes the extracted fundamental frequency and its harmonic components.

[0011] Optionally, in step S7, a frequency domain feature report is generated based on the security assessment level. The report is compared with historical data. If the comparison result shows an abnormal trend, the parameters of the structural model are updated to optimize subsequent monitoring, including: Step S71: Based on the safety assessment level, extract frequency domain features from the vibration signals collected by the sensors to obtain the first frequency domain report; Step S72: Compare the first frequency domain report with the pre-established historical data, and determine the trend anomaly by calculating the deviation value of the frequency domain characteristics; Step S73: If the deviation value of the frequency domain feature exceeds the preset trend threshold, an abnormal trend is displayed, and the parameters of the first structural model are updated to obtain the second structural model.

[0012] Optionally, step S72, which compares the first frequency domain report with pre-established historical data and determines trend anomalies by calculating the deviation value of the frequency domain characteristics, includes: The deviation value of the frequency domain feature is calculated using the following formula: Where D is the deviation value of the frequency domain characteristics. It is the current frequency domain characteristic. It is a historical frequency domain characteristic.

[0013] Optionally, in step S73, if the deviation value of the frequency domain feature exceeds a preset trend threshold, an abnormal trend is displayed, and the parameters of the first structural model are updated to obtain the second structural model. The first structural model refers to the finite element model, whose parameters include modal frequency and damping ratio. The parameters are adjusted and updated using the lsqnonlin function in MATLAB.

[0014] A second aspect of the present invention provides a three-dimensional deformation monitoring system for hydraulic and hydropower gates based on visual inspection. The system employs the method described above to monitor the three-dimensional deformation of hydraulic and hydropower gates. The system includes: an image acquisition module for acquiring high-frequency image sequences using a camera array installed around the gate, and extracting displacement data fields reflecting oscillating deformation from the sequences; a frequency domain analysis module for calculating the corresponding frequency domain spectrum based on the extracted displacement data fields, obtaining the frequency component distribution of the oscillating deformation; a risk identification module for identifying the dominant frequency peak from the frequency domain spectrum; if the peak exceeds a preset oscillation threshold, it is determined to be a high-risk oscillation mode and a relevant timestamp is marked; and a motion tracking module. The system is divided into four modules: a first module for tracing back to the original image sequence based on the marked timestamps, tracking the motion trajectory of points on the gate surface, and obtaining a set of local deformation vectors; a second module for matching the local deformation vector sets with a pre-established structural model, and if the matching deviation is greater than a preset safety margin threshold, it is judged as a potential structural damage area and outputs a coordinate set; a third module for obtaining the associated frequency component distribution from the output coordinate set, classifying the correspondence between deformation patterns and damage types, and determining the safety assessment level; and a fourth module for generating a frequency domain feature report based on the safety assessment level, comparing the report with historical data, and updating the parameters of the structural model to optimize subsequent monitoring if the comparison results show an abnormal trend.

[0015] The technical solution provided by this invention has the following beneficial effects: This invention discloses a three-dimensional deformation monitoring method for hydraulic and hydropower gates based on visual inspection. Addressing the potential structural damage caused by oscillating deformation during gate operation, the method acquires high-frequency image sequences using a camera array, extracts displacement data, and generates a frequency domain spectrum using Fourier transform. Dominant frequency peaks are identified to determine high-risk oscillation modes. For marked high-risk timestamps, an optical flow algorithm is used to track surface motion trajectories, generating local deformation vector sets, which are matched with a pre-built structural model to locate potential damage areas. Support vector machines are used to classify deformation modes and damage types, determine the safety assessment level, and generate a frequency domain feature report for comparison with historical data, dynamically updating the structural model parameters. This invention, through multi-dimensional data fusion and intelligent analysis, accurately identifies oscillating deformation and damage areas, improving the accuracy and real-time performance of gate safety monitoring and providing efficient technical support for engineering structure maintenance. Attached Figure Description

[0016] Figure 1 This is a flowchart of the three-dimensional deformation monitoring method for hydraulic and hydropower gates based on visual detection according to the present invention.

[0017] Figure 2 This is a schematic diagram of the three-dimensional deformation monitoring method for hydraulic and hydropower gates based on visual detection according to the present invention.

[0018] Figure 3 This is another schematic diagram of the three-dimensional deformation monitoring method for hydraulic and hydropower gates based on visual detection according to the present invention.

[0019] Figure 4 This is a schematic diagram of the structure of the three-dimensional deformation monitoring system for hydraulic and hydropower gates based on visual detection according to the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figures 1-3 As shown, in a first aspect, the present invention provides a method for monitoring the three-dimensional deformation of hydraulic and hydroelectric gates based on visual detection, the method comprising: S1, high-frequency image sequences are acquired by a camera array installed around the gate, and displacement data fields reflecting oscillation deformation are extracted from the sequences.

[0022] Optionally, this step also includes: Step S11: High-frequency image sequences are acquired by an array of cameras installed around the gate to obtain initial sequence data.

[0023] Step S12: Based on the initial sequence data, the high-frequency image sequence is processed using the OpenCV Farneback algorithm to determine the displacement changes reflected by the oscillation deformation. The OpenCV Farneback algorithm estimates the pixel motion in continuous image frames based on the assumption of constant brightness. The input is the high-frequency image sequence, and the output is a displacement vector field, which represents the displacement changes of each pixel.

[0024] Step S13: If the displacement change exceeds the preset displacement change threshold, then perform time-series statistics on the displacement vector field for the displacement change, calculate the average displacement amplitude as the displacement data field, and obtain the deformation monitoring field reflecting the oscillation deformation. The time-series statistics are achieved by calculating the time series average value of the displacement amplitude of each pixel in the displacement vector field using NumPy.

[0025] For example, when monitoring the oscillation and deformation of large gates in water conservancy projects, a high-frequency image sequence is first acquired using an array of cameras installed around the gate. These cameras typically shoot at a rate of more than 30 frames per second to ensure that the subtle movements of the gate under the impact of water flow are captured, thus obtaining initial sequence data.

[0026] Specifically, this data acquisition process involves deploying multiple high-definition cameras at different angles of the gate, such as the top, sides, and bottom, to cover a full field of view and avoid blind spots. The initial sequence data consists of a series of consecutive image frames, each recording the gate's instantaneous state. For example, under strong winds or fast-flowing conditions, the gate may sway slightly; these image sequences capture the entire process from static to dynamic changes. In this way, not only can data be acquired in real time, but it also provides high-resolution input for subsequent analysis, supporting precise deformation monitoring.

[0027] In one embodiment, based on these initial sequence data, the OpenCV Farneback algorithm is used to process the high-frequency image sequence to determine the displacement changes reflected by the oscillating deformation.

[0028] It's important to note that the Farneback algorithm is a dense optical flow estimation method. It's based on the assumption of constant brightness, meaning that the brightness of pixels in adjacent frames remains unchanged. It approximates the neighborhood motion of pixels through polynomial expansion, calculating the displacement vector for each pixel. The specific process involves first converting the image sequence to grayscale, then performing pyramid scaling on two consecutive frames to handle multi-scale motion. The algorithm iteratively calculates the displacement field. For example, in a gate image, if a pixel represents the gate edge and moves from position (x, y) to (x+dx, y+dy) under water pressure, Farneback will output the corresponding displacement vector (dx, dy). Thus, the entire output is a displacement vector field representing the motion changes of all pixels in the image, helping to identify the overall oscillation pattern of the gate.

[0029] For example, if the calculated displacement change exceeds the preset displacement change threshold, such as the average displacement exceeding 2 pixels, it indicates that the gate oscillates significantly, and then time-series statistics are performed on the displacement vector field for these displacement changes.

[0030] Specifically, time-series statistics calculate the average displacement amplitude of each pixel in the displacement vector field using the NumPy library. For example, by selecting multiple frames of displacement data within a time window, the average vector magnitude of each pixel is calculated to obtain the average displacement amplitude as the displacement data field. For instance, during monitoring, if the gate acquires 300 frames of images within 10 seconds, and the displacement vector field displays an amplitude sequence of [1.2, 1.5, 1.8, ...] for certain areas, NumPy will calculate its average, such as 1.5 pixels, thus forming a deformation monitoring field reflecting oscillating deformation. This method not only quantifies the degree of deformation but also enables early detection of potential structural risks, improving the safety of gate maintenance.

[0031] In one embodiment, the acquisition of this monitoring field can also be extended to an alarm system. For example, when the average displacement exceeds a safety threshold, the system automatically triggers a notification, and combines historical data to analyze deformation trends to ensure project stability.

[0032] S2. Based on the extracted displacement data field, the Fourier transform algorithm is used to calculate the corresponding frequency domain spectrum, and the frequency component distribution of the oscillating deformation is obtained.

[0033] Optionally, this step also includes: Step S21: Obtain the displacement data field from the monitoring device.

[0034] Step S22: The displacement data field is converted from the time domain to the frequency domain using the NumPy fft function to calculate the preliminary frequency components, which include frequency values ​​and corresponding amplitude values.

[0035] Step S23: For the initial frequency components, the find_peaks function of SciPy is used to identify the spectral peak positions and obtain the amplitude energy distribution. The harmonic component decomposition results are determined by calculating the power spectral density. The harmonic component decomposition results include the extracted fundamental frequency and its harmonic components, forming a vibration mode.

[0036] Step S24: For the vibration mode, based on the correspondence between frequency and amplitude, use Matplotlib's plot function to perform distribution mapping and obtain the frequency component distribution characteristics.

[0037] For example, in the business scenario of gate vibration monitoring, displacement data fields are obtained from sensor devices installed on the gate structure. These devices typically include laser displacement meters or accelerometers, which can capture the minute deformation data of the gate under the impact of water flow in real time.

[0038] Specifically, this acquisition process first involves connecting the device to the data acquisition system, converting the raw signal into a digital displacement sequence via serial port or wireless transmission. For example, a typical sequence may contain hundreds of sampling points per second, recording the horizontal and vertical displacement values ​​of key parts of the gate, thus providing basic input for subsequent analysis.

[0039] In one embodiment, the NumPy fft function is used to convert these displacement data fields from the time domain to the frequency domain. This function is based on the Fast Fourier Transform principle, which decomposes the time series signal into sinusoidal components of different frequencies and calculates the preliminary frequency components, including each frequency value and its corresponding amplitude value.

[0040] For example, when processing gate displacement sequences, the data is first preprocessed, such as detrending and applying window functions. Then, the data is input into the FFT function, which outputs a complex array, where the real and imaginary parts are used to calculate the amplitude. If the displacement sequence length is 1024 points and the sampling rate is 100Hz, the frequency resolution is about 0.1Hz. The resulting preliminary frequency components may show a significant amplitude of the gate at 2Hz, representing the dominant vibration period.

[0041] For example, for these initial frequency components, the SciPy find_peaks function is used to identify the positions of spectral peaks. This function detects local maxima by setting height thresholds and distance parameters, thereby obtaining the amplitude energy distribution.

[0042] Specifically, in gate monitoring, the amplitude spectrum output by the input FFT function will be used to output peak indices, such as identifying peaks at 1Hz, 2Hz, and 4Hz. These peaks correspond to the locations where amplitude energy is concentrated, helping to analyze the main distribution of vibration energy and avoid noise interference.

[0043] In one embodiment, the harmonic component decomposition result is determined by calculating the power spectral density, which is essentially the square of the amplitude spectrum divided by the sequence length, and is used to quantify the energy contribution of each frequency to form vibrational modes, including the extracted fundamental frequency and its harmonic components.

[0044] For example, in gate operations, the fundamental frequency may be 1Hz to represent the basic cycle of water flow, while harmonic frequencies such as 2Hz and 3Hz represent harmonic effects. The calculation process involves taking the square of the modulus of the FFT result and normalizing it. The decomposition result shows that the fundamental frequency energy accounts for 70%, which helps to identify the resonance risk of the gate structure.

[0045] For example, for vibration modes, the distribution mapping is performed using Matplotlib's plot function based on the correspondence between frequency and amplitude. This function visualizes the distribution characteristics of frequency components by plotting curves of the frequency axis and amplitude axis.

[0046] Specifically, in implementation, the frequency array J and amplitude array L are first input, and plot is called to generate a spectrum, such as showing the concentrated distribution of gate vibration in the low frequency band and the attenuation in the high frequency band. This mapping can intuitively reveal the mode characteristics, support engineers in assessing structural stability, and enable early warning in business to prevent fatigue damage.

[0047] S3. Identify the dominant frequency peak from the frequency domain spectrum. If the peak exceeds a preset oscillation threshold, it is determined to be a high-risk oscillation mode and the relevant timestamp is marked.

[0048] Optionally, this step also includes: Step S31: Obtain the frequency domain spectrum from the time domain signal, and use Fast Fourier Transform to calculate the spectral amplitude of the time domain signal to obtain the frequency domain spectrum.

[0049] Step S32: From the frequency domain spectrum, use the SciPy find_peaks function to detect the dominant frequency peak.

[0050] Step S33: If the amplitude of the dominant frequency peak exceeds a preset oscillation threshold, it is determined to be a high-risk oscillation mode.

[0051] Step S34: Mark the relevant timestamps in the original time domain signal according to the high-risk oscillation mode.

[0052] For example, in the business scenario of bridge structural health monitoring, the time-domain signals collected from sensors typically represent the displacement changes of the bridge under wind or traffic loads. These signals exist in time series form, containing the superposition of various frequency components. The Fast Fourier Transform (FFT), as an efficient algorithm, can convert these time-domain signals into a frequency-domain representation. The specific process involves discretizing the signal and then using a butterfly operation to decompose and calculate the complex amplitude at each frequency point, thereby obtaining the spectral amplitude values ​​and forming a complete frequency spectrum. The principle of this transformation lies in decomposing the signal into sinusoidal components of different frequencies, helping to identify periodic features hidden in the time domain. In practical applications, assuming the collected time-domain signal is a displacement sequence of 1024 points with a sampling frequency of 100 Hz, after calculation using the FFT, a spectral amplitude distribution from 0 to 50 Hz can be obtained, where the amplitude values ​​reflect the energy intensity of each frequency. This obtained frequency spectrum provides a foundation for subsequent analysis, ensuring a smooth transition from the time dimension to the frequency dimension.

[0053] Specifically, based on the aforementioned frequency domain spectrum, the SciPy find_peaks function is used to detect dominant frequency peaks. This is a peak finding tool that identifies significant peaks in the spectrum by setting minimum peak height and distance parameters. For example, in bridge vibration monitoring, the function scans the spectral amplitude array to find local maximum points, which correspond to potential vibration modes.

[0054] It should be noted that the principle of the find_peaks function is based on a peak detection algorithm for one-dimensional signals. It takes into account the prominence of the peak and the comparison of adjacent points, thereby avoiding noise interference.

[0055] In one embodiment, for a spectrum amplitude array, the function may detect peaks at 2 Hz and 4 Hz, which represent the main resonant frequencies of the bridge. This detection can accurately locate key frequency components in the signal, laying a data foundation for risk assessment.

[0056] For example, the process of amplitude determination for these dominant frequency peaks involves comparing them with a preset oscillation threshold, which is typically set based on historical data or engineering standards. In bridge monitoring, for instance, an oscillation threshold of 0.5 mm is set. If the peak amplitude exceeds this threshold, it is identified as a high-risk oscillation mode. The principle behind this determination is that amplitude reflects the energy level of vibration; exceeding the oscillation threshold indicates potential structural fatigue or instability. In practical applications, assuming a detected 2 Hz peak amplitude of 0.6 mm exceeds the oscillation threshold of 0.5 mm, it is identified as a high-risk mode. This helps to provide early warning of abnormal bridge vibrations and prevent safety accidents.

[0057] Specifically, based on high-risk oscillation modes, the step of marking relevant timestamps in the original time-domain signal is achieved through inverse mapping. This involves working backward from the frequency corresponding to the frequency peak in the frequency domain to trace back the time period in which that frequency component is significant, for example, using inverse Fourier transform or window analysis. The process includes calculating the phase information of that frequency and then marking intervals, such as from second 100 to second 200, on the time-domain sequence. These markings can be visualized as prominent labels on the signal graph. In bridge monitoring operations, this marking helps engineers quickly pinpoint the time of a problem, thereby guiding on-site inspections and maintenance, and ensuring the long-term stability of the structure.

[0058] For example, in an extended embodiment, if multiple high-risk patterns exist simultaneously, these markers can be further integrated to form a timestamp sequence, supporting a more comprehensive risk assessment. This expands from core single-pattern detection to multi-pattern analysis, enriching the diversity of the solution.

[0059] S4. Based on the marked timestamps, backtrack to the original image sequence, use the optical flow algorithm to track the motion trajectory of the points on the gate surface, and obtain the local deformation vector group.

[0060] Optionally, this step also includes: Step S41: Obtain the original image sequence from the timestamp markers.

[0061] Step S42: The Farneback optical flow method of the OpenCV library is used to trace the points on the gate surface to obtain motion trajectory data for the original image sequence.

[0062] Step S43: The displacement difference is calculated using the NumPy library for the motion trajectory data. The displacement difference is determined by subtracting the position vectors of adjacent time points to determine the relative offset between surface points and obtain the local deformation vector.

[0063] Step S44: Use the K-means clustering algorithm to combine the local deformation vectors into a vector group, determine the overall deformation trend of the gate surface, and obtain the local deformation vector group.

[0064] For example, in water conservancy project monitoring, the process of obtaining the original image sequence from the timestamp can be understood as follows: first, based on pre-recorded video data or frame sequences captured by real-time cameras, these sequences have precise time tags, such as each frame corresponding to a millisecond-level timestamp, thereby ensuring the timing accuracy of subsequent analysis.

[0065] Specifically, this acquisition method helps to trace the visual state of a gate at a specific moment. For example, during flood monitoring, timestamped image sequences can be extracted from the storage system. These images capture the continuous changes in the gate surface under the impact of water flow, thus providing reliable basic data for tracking. This method avoids temporal discrepancies and ensures the continuity of the movement trajectory.

[0066] In one embodiment, the Farneback optical flow method from the OpenCV library is used to track points on the gate surface of the original image sequence to obtain motion trajectory data. The Farneback optical flow method is a dense optical flow algorithm that estimates the displacement of the object's surface by calculating the motion vector of each pixel in the image. For example, in a gate monitoring scenario, adjacent frames in the sequence are first preprocessed, such as by grayscale conversion and noise filtering. Then, the Farneback algorithm is applied to calculate the optical flow field. For instance, key points on the gate surface, such as welds or edge points, are selected. The algorithm estimates the movement vectors of these points from one frame to the next based on polynomial expansion and displacement assumptions, thereby generating trajectory curves. This trajectory data reflects the minute vibration paths of the gate surface under wind loads or water pressure. This tracking helps capture dynamic changes that are difficult to detect with the naked eye, laying the foundation for subsequent deformation analysis.

[0067] For example, the NumPy library is used to calculate displacement differences for motion trajectory data. The displacement differences are determined by subtracting the position vectors of adjacent time points to determine the relative offset between surface points, thus obtaining the local deformation vector.

[0068] Specifically, NumPy, as a numerical computing library, can efficiently handle array operations. In this step, the trajectory data is represented as a two-dimensional array, with each row corresponding to a coordinate vector of a time point, such as (x... t ,y tThen, subtraction is performed on adjacent rows to obtain the difference vector. For example, if point A is (100, 200) at time t1 and (102, 201) at time t2, the difference vector is (2, 1), which represents the relative offset. By accumulating such vectors from multiple points, the degree of deformation in a local area can be quantified. For example, in the central region of a gate, if multiple points show an offset in the same direction, this may indicate stress concentration in the structure, thus obtaining a set of local deformation vectors describing surface distortion. This calculation process is simple and efficient, and can reflect the dynamic response of the gate in real time.

[0069] In one embodiment, local deformation vectors are combined into vector groups using the K-means clustering algorithm to determine the overall deformation trend of the gate surface, resulting in a set of local deformation vectors. K-means is an unsupervised clustering method that iteratively optimizes the allocation of data points into K clusters, each representing a similar deformation pattern. For example, the number of clusters is initially set to K=3, representing normal, slight, and severe deformation. Then, all local deformation vectors are input, and the algorithm calculates the Euclidean distance of each vector to the cluster center, repeatedly adjusting until convergence. For instance, if one group of vectors shows an upward shift and clusters together, this may indicate uneven stress on the upper part of the gate, while another cluster shifts downward, indicating a downward trend. Through this clustering, the overall deformation trend can be determined, such as the entire surface tending towards bending or twisting, thus obtaining the classified vector group. This method improves the early warning capability for gate safety hazards in business operations, ensuring timely maintenance to prevent accidents.

[0070] S5 matches the local deformation vector set with the pre-established structural model. If the matching deviation is greater than the preset safety margin threshold, it is judged as a potential structural damage area and the coordinate set is output.

[0071] Optionally, this step also includes: Step S51: Collect displacement data of the structural surface using sensors to obtain a set of local deformation vectors.

[0072] Step S52: Extract the coordinates of the displacement points from the local deformation vector group as the first coordinate array.

[0073] Step S53: Extract the coordinates of the corresponding points from the pre-established structural model as the second coordinate array.

[0074] Step S54: Use the NumPy library to calculate the positional difference between the first coordinate array and the second coordinate array. Specifically, use the linalg.lstsq function to solve for the least squares fitting parameters and obtain the residual vector as the matching bias.

[0075] Step S55: If the matching deviation is greater than the preset safety margin threshold, it is determined to be a damaged area and the coordinate set is output.

[0076] For example, in the business scenario of bridge structure monitoring, laser displacement sensors deployed on the bridge deck and beams collect surface displacement data in real time. These sensors record displacement changes at a frequency of 10 times per second. The data includes vector values ​​of the x-axis, y-axis and z-axis. After preliminary noise filtering, a local deformation vector group is formed. This vector group represents the deformation trend of a specific area of ​​the bridge, such as the connection point of the main beam.

[0077] Specifically, these vector sets are aggregated from displacement signals captured by sensor arrays. For example, for the main beam of a bridge spanning a river, the sensors may detect minute bending deformations caused by vehicle loads. Each vector in the vector set corresponds to the displacement direction and magnitude of a surface point, thus providing basic data support for subsequent analysis.

[0078] In one embodiment, the process of extracting displacement point coordinates from a local deformation vector group can be understood as selecting the coordinates of key nodes in the vector group. For example, in the bridge example above, the vector group contains displacement data of 100 surface points, each with three-dimensional coordinates such as (5.2, 3.1, 0.4). These coordinates are organized into a NumPy array as the first coordinate array. The dimension of this array may be (100, 3), representing the actual observed deformed position. Through this extraction, the data is converted from vector form to coordinate form, which is convenient for comparison with model data.

[0079] For example, continuing with the example of bridge structures, when extracting the coordinates of corresponding points from a pre-established structural model, this model is usually a three-dimensional digital model built based on finite element analysis software such as ANSYS. It contains the ideal coordinates of the bridge body under no-load conditions. For example, the model coordinates corresponding to the sensor points may be (5.0, 3.0, 0.0). These are extracted into a second coordinate array with the same dimensions as the first array. This forms a correspondence between the observed data and the ideal model, laying the foundation for deviation calculation.

[0080] Specifically, when using the NumPy library to calculate the positional difference between the first and second coordinate arrays, the `linalg.lstsq` function is used to solve for the least-squares fitting parameters. This function essentially solves for the optimal solution of a system of linear equations. For example, in bridge monitoring, it can fit the transformation matrix from observed coordinates to model coordinates. The process involves constructing a design matrix A, where A contains the coordinates of the observed points and the target vector b is the model coordinates. Then, `lstsq` is used to solve for the unknown vector c, making Ac as close to b as possible. The resulting residual vector is the fitted error vector. This vector quantifies the matching deviation of each point. For example, a residual value of 0.3 indicates a slight offset. The calculation process avoids direct subtraction and instead uses least squares to consider the overall fit, thus more accurately reflecting structural deformation.

[0081] In one embodiment, if the matching deviation is greater than a preset safety margin threshold, for example, when the safety margin threshold is set to 0.5 meters, for a certain beam segment of the bridge, if the deviation of multiple points in the residual vector exceeds this value, such as 0.6 or higher, the system determines that the area is a potential damage area and outputs a coordinate set such as [(5.3,3.2,0.5), (5.4,3.3,0.6)]. This output can trigger a maintenance alarm, achieving the goal of early damage identification in business operations, thereby improving the efficiency of structural safety monitoring.

[0082] For example, to expand the diversity of solutions, in another possible implementation, for structural monitoring of high-rise buildings, the displacement data collected by sensors may focus on the relative displacement between floors. After the vector group is extracted, it is compared with the coordinates of the building's BIM model. When solving using the lstsq function, a weighting factor can be added to emphasize the deviation of key parts. If the residual vector exceeds the safety margin threshold, such as 0.2 meters, the damage coordinate set is output to support preventive maintenance. The logical progression of these steps ensures a complete chain from data acquisition to damage assessment.

[0083] S6. Obtain the associated frequency component distribution from the output coordinate set, use the support vector machine algorithm to classify the correspondence between deformation patterns and damage types, and determine the safety assessment level.

[0084] Optionally, this step also includes: Step S61: Obtain the coordinate set from the coordinate data acquisition.

[0085] Step S62: Input the coordinate set as a time series into the fft function of the SciPy library, perform discrete Fourier transform to calculate the amplitude and phase of each frequency component, and obtain the frequency component distribution.

[0086] Step S63: Using the amplitude and phase of the frequency component distribution as input feature vectors, a classifier is trained using the SVC class of the Scikit-learn library through a Gaussian kernel function, and the correspondence between deformation patterns and damage types is output.

[0087] Step S64: Determine whether the deformation pattern belongs to a preset damage type range based on the correspondence. This range is a numerical limit obtained from historical damage data. If it belongs to this range, it is marked as a high-risk pattern.

[0088] Step S65: Determine the security assessment level through the high-risk mode.

[0089] For example, in practical applications of structural health monitoring, the process of acquiring coordinate sets from coordinate data collection typically involves a network of sensors deployed on the surface of bridges or buildings. These sensors, such as laser rangefinders or GPS devices, can capture the spatial position changes of the structure at different points in time in real time.

[0090] Specifically, assuming monitoring the main girder of a suspension bridge, sensors collect data once per second, generating a time-series array of three-dimensional coordinates containing x-axis, y-axis, and z-axis coordinates. For example, collecting 600 data points over 10 minutes would form a coordinate set of shape (600, 3). This data acquisition process emphasizes high precision and continuity to ensure the reliability of subsequent analysis. In this way, the coordinate set not only reflects the static position of the structure but also captures dynamic vibration information, providing a foundation for frequency analysis.

[0091] In one embodiment, the coordinate set is used as a time series input to the FFT function of the SciPy library for Discrete Fourier Transform (DFT). This is a mathematical method that converts a time-domain signal into a frequency-domain signal. Its principle is to decompose the signal into sine and cosine components of different frequencies through Fourier series decomposition, thereby revealing hidden periodic patterns.

[0092] Specifically, for the coordinate set of the aforementioned main girder of the bridge, the displacement sequence of the z-axis is selected as the input, and the fft function will calculate the frequency bins from 0 to the sampling frequency / 2. Each bin corresponds to a complex value, where the amplitude represents the intensity of the frequency component and the phase represents its initial offset.

[0093] For example, after inputting a sequence of length N=600, the output is a complex array. By taking the modulus, the amplitude spectrum is obtained. If the amplitude is 0.2 mm at the dominant frequency of 5 Hz, it indicates that the structure vibrates significantly at this frequency. The phase may be displayed as π / 4 radians, which helps to identify the vibration phase difference. The frequency component distribution obtained in this way can intuitively show the modal characteristics of the structure, such as natural frequencies and damping ratios, laying the frequency domain foundation for damage detection.

[0094] For example, continuing with bridge monitoring, the amplitude and phase of the frequency component distribution are used as input feature vectors. The SVC class of the Scikit-learn library is used to train a classifier through the Gaussian kernel function. Here, SVC stands for Support Vector Classifier. Its principle is to find a hyperplane to maximize the margin between different classes. The Gaussian kernel function maps the data to a high-dimensional space through the radial basis function to handle nonlinear problems.

[0095] Specifically, features are extracted from historical data. For example, the amplitude peak of a normal structure is in the range of 3-7Hz, and the phase shift is less than π / 2, while damaged structures may show abnormal peaks at low frequencies. The training process involves preparing a dataset, such as 100 samples, of which 50 are labeled "undamaged," with a feature vector dimension of 20 (10 amplitude + 10 phase). The model is trained using the fit method. After learning, the model can output the relationship between deformation patterns, such as "low-frequency amplification" and "crack damage." This correspondence is cross-validated to ensure an accuracy of over 90%, thus reliably classifying the current monitoring data.

[0096] In one embodiment, the deformation mode is determined to belong to a preset damage type range based on the correspondence. This range is a numerical limit obtained from historical damage data. For example, historical data shows that the amplitude of crack damage is greater than 0.15 mm and the phase offset is greater than π / 3. If the current mode matches this limit, it is marked as a high-risk mode.

[0097] Specifically, in the bridge case, if the classifier outputs a "low-frequency amplification" mode with an amplitude of 0.18 mm and a phase of π / 4, it checks whether this falls within the preset ranges [0.15, 0.3] and [π / 3, π / 2]. If so, it is marked as high-risk, which helps to provide early warning of potential collapses. The safety assessment level is determined by the high-risk mode; for example, high risk can be divided into levels 1-5, based on the severity of the mode. For instance, level 3 indicates that immediate inspection is required, thereby enabling preventative maintenance and reducing the risk of structural failure.

[0098] For example, in an extended approach, this method can be applied to the monitoring of wind turbine blades. Coordinate sets are collected from sensors on the blade surface, FFT analysis reveals the rotational frequency components, SVC classification distinguishes fatigue damage modes, the judgment range is based on historical wind load data, and the final assessment level guides maintenance scheduling to ensure operational safety.

[0099] S7. Generate a frequency domain characteristic report based on the security assessment level, compare the report with historical data, and if the comparison results show an abnormal trend, update the parameters of the structural model to optimize subsequent monitoring.

[0100] Optionally, this step also includes: Step S71: Based on the safety assessment level, extract frequency domain features from the vibration signals collected by the sensors using fast Fourier transform to obtain the first frequency domain report.

[0101] Step S72: Compare the first frequency domain report with the pre-established historical data, and determine the trend anomaly by calculating the deviation value of the frequency domain characteristics.

[0102] Preferably, the deviation value of the frequency domain feature is calculated using the following formula: Where D is the deviation value of the frequency domain characteristics. It is the current frequency domain characteristic. It is a historical frequency domain characteristic.

[0103] Step S73: If the deviation value of the frequency domain feature exceeds the preset trend threshold, an abnormal trend is displayed, and the parameters of the first structural model are updated. The first structural model refers to the finite element model, whose parameters include modal frequency and damping ratio. The parameters are adjusted and updated using the lsqnonlin function in MATLAB to obtain the second structural model.

[0104] For example, in bridge structure safety monitoring, based on the determined safety assessment level, vibration signals are first collected from accelerometers deployed on the bridge piers. These signals are typically recorded in time-series format, documenting the bridge's dynamic response under vehicle loads or wind forces. Fast Fourier Transform (FFT), as a highly efficient frequency domain analysis tool, converts these time-domain signals into a frequency-domain representation, extracting features such as the dominant frequency amplitude and phase, thereby generating a first frequency domain report. This report details the bridge's inherent frequency distribution. For instance, in a specific case, a vibration signal sequence collected by the sensors has a length of 1024 points. After calculation using FFT, the dominant frequency is approximately 2.5 Hz, corresponding to an amplitude value of 0.15g, reflecting the bridge's modal characteristics.

[0105] Specifically, when comparing the first frequency domain report with pre-established historical data, the historical data comes from a long-term monitoring database since the bridge was built, including frequency domain characteristic records under the same load conditions over the past year. The process of determining trend anomalies by calculating deviation values ​​involves quantitative comparisons of individual features, such as taking the current dominant frequency. It is 2.5Hz, while the historical average is... If the frequency is 2.7Hz, then the deviation value D of the frequency domain characteristic is calculated as |2.5-2.7| / 2.7≈0.074. If the preset trend threshold is 0.1 and D does not exceed it, it is considered normal; however, in another example, When the frequency domain characteristic deviation value D drops to 2.2Hz, it is approximately 0.185, exceeding the trend threshold and indicating an abnormal trend. This may be an early sign of fatigue damage to the bridge. This comparison helps to identify potential risks early, enabling preventative maintenance and reducing the probability of sudden accidents.

[0106] In one embodiment, if the deviation value of the frequency domain feature exceeds a preset trend threshold, indicating an abnormal trend, the parameters of the first structural model need to be updated. Here, the first structural model specifically refers to the finite element model, which is a numerical simulation method that simulates the dynamic behavior of a bridge structure by dividing it into meshes and defining material properties. Parameters such as modal frequency and damping ratio represent the vibration period and energy dissipation capacity of the structure, respectively.

[0107] For example, in a bridge monitoring project, the initial finite element model's modal frequency was set to 2.7Hz and the damping ratio to 0.05. The MATLAB function `lsqnonlin` was used for nonlinear least-squares optimization. This function iteratively minimizes the residual between the current measured data and the model's predictions. For instance, the actual measured frequency deviation was used as the objective function, and the parameters were gradually adjusted until convergence, resulting in a second structural model. In this model, the updated modal frequency might become 2.4Hz, and the damping ratio adjusted to 0.06. This update process ensures the model better reflects the actual conditions, improving the accuracy of subsequent damage predictions in practical applications.

[0108] For example, extending to practical applications, once the second structural model is obtained, it can be used to simulate the response under different damage scenarios. For instance, assuming cracks appear in the bridge beam, the influence of crack depth on frequency can be simulated by adjusting the model parameters, thereby providing a basis for maintenance decisions.

[0109] In one embodiment, this method is used in the monitoring of high-speed railway bridges. Historical data shows abnormal deviations, and the model is updated accordingly, ultimately avoiding potential derailment risks. This demonstrates the practical value of the technology in improving structural safety and operational efficiency.

[0110] Specifically, the entire process, from signal acquisition to model update, ensures the continuity and reliability of monitoring. For example, in the deviation judgment stage, if the trend threshold is set too low, it may lead to false positives. Therefore, in business practice, the threshold is often dynamically adjusted in combination with expert experience to balance sensitivity and stability.

[0111] like Figure 4As shown, in a second aspect, this invention provides a three-dimensional deformation monitoring system for hydraulic and hydropower gates based on visual detection. The system employs the method described above to monitor the three-dimensional deformation of hydraulic and hydropower gates. The system mainly includes: an image acquisition module, used to acquire high-frequency image sequences through a camera array installed around the gate, and extract displacement data fields reflecting oscillating deformation from the sequences; a frequency domain analysis module, used to calculate the corresponding frequency domain spectrum based on the extracted displacement data fields using a Fourier transform algorithm, obtaining the frequency component distribution of the oscillating deformation; a risk identification module, used to identify the dominant frequency peak from the frequency domain spectrum; if the peak exceeds a preset oscillation threshold, it is determined to be a high-risk oscillation mode and a relevant timestamp is marked; and a motion tracking module. The system is divided into several modules: a local deformation vector group and a damage detection module. The local deformation vector group is used to backtrack to the original image sequence based on the marked timestamps. If the matching deviation exceeds a preset safety margin, it is identified as a potential structural damage area, and a coordinate set is output. A pattern classification module is used to obtain the associated frequency component distribution from the output coordinate set, and uses a support vector machine algorithm to classify the correspondence between deformation patterns and damage types to determine the safety assessment level. A model optimization module is used to generate a frequency domain feature report based on the safety assessment level, compare the report with historical data, and update the structural model parameters to optimize subsequent monitoring if the comparison results show an abnormal trend. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for monitoring the three-dimensional deformation of hydraulic and hydropower gates based on visual inspection, characterized in that, The method includes: S1. High-frequency image sequences are acquired using a camera array installed around the gate, and displacement data fields reflecting oscillating deformation are extracted from the sequences. S2. The corresponding frequency domain spectrum is calculated based on the extracted displacement data fields to obtain the frequency component distribution of the oscillating deformation. S3. The dominant frequency peak is identified from the frequency domain spectrum. If the peak exceeds a preset oscillation threshold, it is determined to be a high-risk oscillation mode and a relevant timestamp is marked. S4. The original image sequence is traced back to the marked timestamps to track the movement trajectory of points on the gate surface, obtaining a local deformation vector set. S5. The local deformation vector set is matched with a pre-established structural model. If the matching deviation is greater than a preset safety margin threshold, it is judged as a potential structural damage area and a coordinate set is output. S6. The associated frequency component distribution is obtained from the output coordinate set, the correspondence between deformation modes and damage types is classified, and the safety assessment level is determined. S7. A frequency domain feature report is generated based on the safety assessment level. The report is compared with historical data. If the comparison result shows an abnormal trend, the parameters of the structural model are updated to optimize subsequent monitoring.

2. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 1, characterized in that, Step S1 involves acquiring high-frequency image sequences using a camera array installed around the gate, and extracting displacement data fields reflecting oscillating deformation from the sequences, including: Step S11: High-frequency image sequences are acquired by an array of cameras installed around the gate to obtain initial sequence data; Step S12: Based on the initial sequence data, process the high-frequency image sequence to determine the displacement change reflected by the oscillating deformation; Step S13: If the displacement change exceeds the preset displacement change threshold, perform time-series statistics on the displacement vector field for the displacement change, calculate the average displacement amplitude as the displacement data field, and obtain the deformation monitoring field that reflects the oscillation deformation.

3. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 2, characterized in that, Step S12, based on the initial sequence data, processes the high-frequency image sequence to determine the displacement change reflected by the oscillation deformation, including: processing the high-frequency image sequence using the OpenCV Farneback algorithm, wherein the OpenCV Farneback algorithm is based on the assumption of constant brightness, estimates the pixel motion in consecutive image frames, the input is the high-frequency image sequence, and the output is a displacement vector field, representing the displacement change of each pixel.

4. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 1, characterized in that, Step S2 involves calculating the corresponding frequency domain spectrum based on the extracted displacement data fields to obtain the frequency component distribution of the oscillating deformation, including: Step S21: Obtain displacement data fields from the monitoring equipment; Step S22: Convert the displacement data field from the time domain to the frequency domain and calculate the preliminary frequency components; Step S23: For the initial frequency components, identify the spectral peak positions, obtain the amplitude energy distribution, determine the harmonic component decomposition results by calculating the power spectral density, and form the vibration mode; Step S24: For the vibration mode, perform distribution mapping based on the correspondence between frequency and amplitude to obtain the frequency component distribution characteristics.

5. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 4, characterized in that, In step S22, the displacement data field is converted from the time domain to the frequency domain to calculate the preliminary frequency components, which include frequency values ​​and corresponding amplitude values.

6. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 5, characterized in that, Step S23 involves identifying the spectral peak positions and obtaining the amplitude energy distribution for the preliminary frequency components, determining the harmonic component decomposition results by calculating the power spectral density, and forming a vibration mode. The harmonic component decomposition results include the extracted fundamental frequency and its harmonic components.

7. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 1, characterized in that, Step S7 involves generating a frequency domain characteristic report based on the security assessment level, comparing the report with historical data, and updating the structural model parameters to optimize subsequent monitoring if the comparison results show an abnormal trend. This includes: Step S71: Based on the safety assessment level, extract frequency domain features from the vibration signals collected by the sensors to obtain the first frequency domain report; Step S72: Compare the first frequency domain report with the pre-established historical data, and determine the trend anomaly by calculating the deviation value of the frequency domain characteristics; Step S73: If the deviation value of the frequency domain feature exceeds the preset trend threshold, an abnormal trend is displayed, and the parameters of the first structural model are updated to obtain the second structural model.

8. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 1, characterized in that, Step S72, which compares the first frequency domain report with pre-established historical data and determines trend anomalies by calculating the deviation value of the frequency domain characteristics, includes: The deviation value of the frequency domain feature is calculated using the following formula: Where D is the deviation value of the frequency domain characteristics. It is the current frequency domain characteristic. It is a historical frequency domain characteristic.

9. The method for monitoring three-dimensional deformation of hydraulic and hydropower gates based on visual detection according to claim 8, characterized in that, In step S73, if the deviation value of the frequency domain feature exceeds the preset trend threshold, an abnormal trend is displayed, and the parameters of the first structural model are updated to obtain the second structural model. The first structural model refers to the finite element model, whose parameters include modal frequency and damping ratio. The parameters are adjusted and updated using the lsqnonlin function in MATLAB.

10. A three-dimensional deformation monitoring system for hydraulic and hydropower gates based on visual inspection, characterized in that, The system employs the method described in any one of claims 1-9 for three-dimensional deformation monitoring of hydraulic and hydropower gates. The system comprises: an image acquisition module for acquiring high-frequency image sequences using a camera array installed around the gate, and extracting displacement data fields reflecting oscillating deformation from the sequences; a frequency domain analysis module for calculating the corresponding frequency domain spectrum based on the extracted displacement data fields, obtaining the frequency component distribution of the oscillating deformation; a risk identification module for identifying the dominant frequency peak from the frequency domain spectrum, and if the peak exceeds a preset oscillation threshold, determining it as a high-risk oscillation mode and marking a relevant timestamp; and a motion tracking module for backtracking based on the marked timestamps. The system retrieves the original image sequence, tracks the motion trajectory of points on the gate surface, and obtains a set of local deformation vectors. A damage detection module matches the local deformation vectors with a pre-established structural model. If the matching deviation exceeds a preset safety margin threshold, it identifies a potential structural damage area and outputs a coordinate set. A pattern classification module obtains the associated frequency component distribution from the output coordinate set, classifies the correspondence between deformation patterns and damage types, and determines the safety assessment level. A model optimization module generates a frequency domain feature report based on the safety assessment level, compares the report with historical data, and updates the structural model parameters to optimize subsequent monitoring if the comparison results show an abnormal trend.

Images