Settlement data analysis and early warning method based on cloud platform

Abstract

The invention relates to a settlement data analysis and early warning method based on a cloud platform, and the method comprises the steps: collecting settlement data of a basic part of a main transformer in real time through a high-precision electronic settlement observation device array, obtaining deformation stress data through data preprocessing and coordinate calculation, and uploading the deformation stress data to the cloud platform; transmitting an ultrasonic pulse signal to the stress concentration position of the local area through ultrasonic detection equipment, receiving reflected wave data, and identifying the depth and range of the crack initiation position according to the reflected wave data; a sensor array is arranged on the basic surface of a crack initiation position, sound wave signals released in the crack propagation process are captured, and a time sequence of the crack propagation direction and speed is determined; and detecting crack depth data of the repair priority region through ultrasonic waves, calculating an adjusted basic health state value, comparing the adjusted basic health state value with a preset safety threshold, and generating a corresponding safety early warning signal.

Description

Technical Field

[0001] The present application relates to the field of electrical digital data technology, and specifically to a settlement data analysis and early warning method based on a cloud platform. Background Art

[0002] Settlement data analysis and early warning technologies are crucial for infrastructure safety, especially for structures supporting heavy equipment, such as substation main transformer foundations, whose stability and safety are directly linked to the reliable operation of power systems. Uneven settlement can trigger abnormal stresses within foundations, inducing crack initiation and propagation, and thus threatening the overall load-bearing capacity of the structure. Therefore, developing efficient settlement monitoring and crack identification methods is not only a pressing need for engineering safety but also a key direction for advancing intelligent monitoring technologies. Currently, traditional settlement monitoring relies on single measurement methods, such as levels or conventional sensors, which struggle to comprehensively capture strain changes and crack dynamics within foundations. Even when combined with some structural health monitoring technologies, data processing still primarily relies on offline analysis, lacking real-time and systematic capabilities, making it difficult to accurately predict crack initiation locations and propagation trends. This approach is limited by its inability to effectively integrate multi-source data and its inability to cope with the complex stress distribution changes under long-term heavy loads. The main challenge facing research lies in accurately identifying the initiation and propagation processes of cracks within foundations. Specifically, while arrays of high-precision settlement observation devices can provide abundant displacement data, converting this data into reliable indicators of internal strain remains a technical challenge. Furthermore, when detecting cracks, technologies such as ultrasonic testing and acoustic emission monitoring are limited by signal processing capabilities and environmental interference, making it difficult to accurately determine the direction and speed of crack expansion. These unresolved technical factors lead to significant uncertainty in the assessment of the impact of cracks on the foundation's bearing capacity, creating a difficult problem that urgently needs to be overcome. Therefore, how to integrate real-time data from an array of high-precision electronic settlement observation devices on a cloud platform, calculate the strain distribution of various parts of the foundation, and combine ultrasonic testing and acoustic emission monitoring technologies to accurately identify the initiation location, expansion direction, and speed of cracks within the main transformer foundation, and then assess their impact on the overall bearing capacity, has become a key issue that needs to be addressed in this study. Summary of the Invention

[0003] In order to solve the problems existing in the above-mentioned prior art, the purpose of this application is to provide a settlement data analysis and early warning method based on a cloud platform.

[0004] The cloud platform-based settlement data analysis and early warning method described in this application includes the following steps:

[0005] S101. Using a high-precision electronic settlement observation device array, collect settlement data of the main transformer foundation in real time, obtain the real-time deformation distribution status of the foundation, and upload it to the cloud platform;

[0006] S102: Calculate the strain values ​​of various parts of the main transformer foundation based on the real-time deformation distribution state and determine the stress concentration location. Specifically, obtain stress distribution data of the foundation structure and identify local high stress areas through meshing, strain calculation, and stress analysis of the three-dimensional deformation data.

[0007] S103, using ultrasonic testing equipment to detect cracks at stress concentration locations on the main transformer foundation, emitting pulse signals and analyzing reflected wave data to identify the initiation location, depth, and range of the cracks;

[0008] S104, placing an acoustic emission sensor array at the location where the cracks initiation occurs in the main transformer foundation to capture the acoustic wave signals during crack expansion, and determining the time series of the crack expansion direction and speed through signal processing and time series analysis;

[0009] S105. Performing an update calculation based on the time series of crack expansion direction and velocity combined with the strain distribution matrix to analyze the impact of crack expansion on the strain distribution of the main transformer foundation and determine the trend of local strain aggravation;

[0010] S106. Predict the crack initiation location, expansion direction, and speed based on the local strain intensification trend, and calculate the dynamic changes in the overall bearing capacity of the main transformer foundation through a structural mechanics simulation model to provide data support for safety assessment;

[0011] S107. If the dynamic change value of the main transformer foundation bearing capacity is lower than the safety threshold, the settlement observation and acoustic emission monitoring data are integrated through the cloud platform to generate a three-dimensional visualization path of the crack extension and determine the repair priority area, providing a decision-making basis for engineering maintenance;

[0012] S108. Repair the crack depth in the priority area through ultrasonic detection and calculate the adjusted basic health status value of the main transformer. After comparing it with the preset safety threshold, generate a corresponding safety warning signal to provide real-time feedback for project safety management.

[0013] Preferably, the step S101 includes:

[0014] Collecting triaxial displacement data from the high-precision electronic sedimentation observation device array according to a preset sampling period through a distributed sensor gateway, performing linear correction on the original displacement data according to the sensor calibration curve, and using Kalman filtering to eliminate random error interference in the corrected displacement data to obtain filtered displacement data;

[0015] For the filtered displacement data, reference coordinate point information is obtained from a pre-calibrated reference pile, a coordinate transformation matrix is ​​established using the least squares method, and three-dimensional spatial coordinates are calculated for the filtered displacement data according to the coordinate transformation matrix;

[0016] The three-dimensional spatial coordinates after the solution are optimized by using a least squares adjustment algorithm, and the optimized three-dimensional spatial coordinates are normalized according to the reference coordinate point information to obtain normalized coordinate data;

[0017] For the normalized coordinate data, stress and strain parameters are obtained from preset monitoring points, and the stress and strain distribution state is calculated using an elastic mechanics model to obtain deformation stress data;

[0018] Establishing a deformation trend curve based on the deformation stress data, and predicting the deformation trend using a long short-term memory network to obtain predicted deformation data;

[0019] Constructing a basic deformation distribution map based on the predicted deformation data, obtaining corresponding monitoring point thresholds from a pre-set deformation threshold library, and comparing the predicted deformation data with the thresholds;

[0020] If the predicted deformation data of the monitoring point exceeds the corresponding threshold, an early warning signal is triggered, and the early warning signal and the predicted deformation data are encrypted by the data encryption module and transmitted to the cloud platform database for storage.

[0021] Preferably, the step S102 includes:

[0022] The three-dimensional deformation data of the foundation parts are obtained from the deformation monitoring database. An orthogonal grid division scheme is established based on the deformation data. The foundation structure is divided into hexahedral grid units using an adaptive mesher, and the strain measurement positions are calibrated at the grid nodes.

[0023] For the grid nodes, the elastic modulus parameters of the foundation concrete are obtained from a pre-established material library, the initial strain value of the node is calculated using Hooke's law, and the degree of strain abnormality is determined by comparing the strain value with the preset reference strain;

[0024] Based on the initial strain value of the node, the finite element strain energy density calculation formula is used to solve the isotropic strain components, and the strain energy density equation is numerically integrated by the fourth-order Runge-Kutta method to obtain the strain distribution cloud map;

[0025] According to the strain distribution cloud map, the node strain gradient is calculated using the Gaussian integral method, and the principal value and principal direction of the strain are solved by the principal strain calculation formula to obtain the principal direction distribution data of the foundation structure strain;

[0026] Based on the strain principal direction distribution data, the principal stress distribution is calculated using the elastic mechanics strain-stress conversion equation, and the maximum shear stress surface is determined using the Mohr circle criterion to obtain the foundation structure stress distribution data;

[0027] According to the stress distribution data, the stress intensity calculation formula is used to solve the node stress intensity value, and the stress concentration area is determined by judging the stress intensity threshold to obtain the coordinate data of the local stress concentration position of the foundation structure.

[0028] Preferably, the step S103 includes:

[0029] According to the coordinates of the stress concentration location, an array ultrasonic probe is used to generate a longitudinal wave pulse signal with a frequency range of 2MHz to 5MHz. The transmission waveform parameters are set through the probe vibrator controller, and the concrete acoustic characteristic parameters are obtained from the ultrasonic parameter database. The ultrasonic pulse signal is matched with the acoustic impedance to obtain the matched transmission wave signal.

[0030] For the matched transmitted wave signal, an ultrasonic receiver is used to collect the reflected wave signal, a bandpass filter is used to eliminate the concrete medium noise, and a Hilbert transform is performed on the filtered reflected wave signal to extract the signal envelope to obtain envelope feature data;

[0031] According to the envelope characteristic data, an attenuation compensation coefficient is obtained from an acoustic wave compensation database, and exponential attenuation compensation is performed on the intensity of the reflected wave signal to obtain a compensated reflected wave signal;

[0032] For the compensated reflected wave signal, discrete wavelet transform is used to extract time-frequency features, characteristic coefficients are obtained through multi-scale decomposition, and a time-frequency spectrum of the reflected wave is obtained;

[0033] According to the reflected wave time-frequency spectrum, a crack standard feature vector is obtained from a defect feature database, and a feature matching degree is calculated using a random forest identifier to obtain a crack type discrimination result;

[0034] Based on the crack type identification results, the waveform peak time difference method is used to calculate the crack depth parameters, and the phase spectrum analysis method is used to calculate the crack propagation direction to obtain the crack spatial distribution characteristic data.

[0035] Preferably, the step S104 includes:

[0036] Based on the crack initiation location, an acoustic emission sensor array was arranged in a hexagonal topology, with a spacing of 30 cm between adjacent sensors. The sampling frequency and trigger voltage were set using a sensor matrix controller, and waveform parameter thresholds were obtained from the acoustic emission calibration database. The sensitivity of the sensor array was calibrated to obtain the calibrated acquisition parameters.

[0037] Based on the calibrated acquisition parameters, a data collector is used to collect acoustic emission signals at a preset frequency, environmental noise is eliminated through a dynamic threshold filter, and envelope extraction is performed on the filtered signal to obtain acoustic wave event data;

[0038] According to the acoustic wave event data, Fourier transform is used to extract spectrum features, characteristic frequency band signals are extracted through frequency segmentation filtering, and time-frequency analysis is performed on the characteristic frequency band signals to obtain acoustic emission characteristic maps;

[0039] Based on the acoustic emission characteristic spectrum, the triangulation positioning algorithm is used to calculate the position coordinates of the acoustic wave source, the sound source direction angle is calculated by the time difference of the sound wave arrival, and the sound source coordinates are arranged in time sequence to obtain the crack expansion trajectory data;

[0040] Based on the crack propagation trajectory data, a time series predictor is established using a deep recurrent neural network, the crack propagation rate is calculated by acoustic emission energy density, and the propagation rate is accumulated over time to obtain a crack propagation velocity sequence;

[0041] For the crack propagation velocity sequence, a Kalman filter is used to smooth the velocity data, the propagation direction is calculated by vector synthesis, and a time series is constructed by combining the propagation direction and velocity to obtain the crack propagation state parameters.

[0042] Preferably, the step S104 further includes:

[0043] According to the original sound wave signal collected by the sensor array, the frequency band range parameters are obtained from the sound wave acquisition parameter library, the frequency is filtered through a bandpass filter, and the filtered signal is denoised using an adaptive threshold to obtain a noise-reduced sound wave signal;

[0044] For the noise-reduced sound wave signal, extract the signal envelope curve using Hilbert transform, calculate the amplitude sequence through the envelope curve, calculate the spectrum sequence through Fourier transform, and obtain the sound wave feature data;

[0045] Based on the acoustic wave characteristic data, a spectrum peak identifier is used to extract the characteristic frequency band, and the amplitude change trend is obtained through wavelet decomposition. If the amplitude change exceeds the preset monitoring threshold, the characteristic triggering moment is recorded to obtain characteristic triggering data;

[0046] Based on the characteristic trigger data, the cross-correlation function is used to calculate the time difference sequence of the sensor signal, the array layout position is obtained from the sensor coordinate library, and the sound source coordinates are calculated by the acoustic wave positioning algorithm to obtain the crack extension coordinates;

[0047] According to the crack propagation coordinates, the propagation displacement vector is calculated by using the coordinate difference of adjacent moments, the propagation direction is determined by the direction angle of the displacement vector, and the propagation rate is calculated by taking the quotient of the propagation displacement and the time interval to obtain the crack propagation parameter;

[0048] According to the crack propagation parameters, a velocity prediction model is constructed using a support vector regression predictor. Continuous prediction results are obtained by sliding a time window. The propagation direction and velocity data are synchronized in time to obtain a crack propagation time series.

[0049] Preferably, the step S105 includes:

[0050] According to the crack propagation time series, the strain redistribution calculator is used to update the basic strain distribution data. The measured strain values ​​are extracted from the strain monitoring database. The boundary of the propagation affected area is determined by strain difference calculation to obtain the real-time strain distribution matrix.

[0051] Based on the real-time strain distribution matrix, a tensor mapping method is used to calculate the strain change around the crack propagation trajectory, the strain field reconstruction result is obtained from the finite element calculator, and the strain distribution update data is established through the grid interpolation method to obtain the strain influence tensor;

[0052] According to the strain influence tensor, a recursive neural network is used to establish a strain transfer function, the influence of the extended parameter on the strain field is calculated from the strain transfer function, the influence coefficient of each monitoring point is determined by strain coupling calculation, and the strain transfer matrix is ​​obtained;

[0053] For the strain transfer matrix, the strain change rate of the high stress area is extracted using the strain time series, the mutation point position is identified from the strain change rate curve, the range of the strain intensification area is determined by local strain gradient calculation, and the local strain intensification trend is obtained.

[0054] Preferably, the step S106 includes:

[0055] According to the local strain aggravation trend data, the strain gradient field distribution is calculated using the five-point difference format. The strain growth rate is extracted from the strain gradient field data. The mutation point is identified through the strain rate curve to obtain the strain mutation position data.

[0056] For the strain mutation position data, the strain contour line is calculated by least square fitting, the crack initiation position is determined from the change in the curvature of the contour line, and the expansion trend angle is calculated through the main strain direction to obtain the crack space parameters;

[0057] According to the crack spatial parameters, a deep neural network is used to predict the expansion path, the expansion velocity is calculated from the strain growth curve, and the expansion path and velocity are mapped in time and space to obtain crack expansion prediction data;

[0058] Based on the crack propagation prediction data, the concrete elastic modulus, Poisson's ratio, and compressive strength parameters are read from the basic parameter library, and the structural stiffness degradation amount is calculated through the elastic-plastic damage function to obtain stiffness damage data;

[0059] Based on the stiffness damage data, a stress redistribution algorithm is used to calculate the stress distribution of the foundation structure, the principal stress direction is extracted from the stress distribution cloud map, the bearing level is calculated by the ratio of the principal stress to the allowable stress, and the bearing capacity change data is obtained;

[0060] For the bearing capacity change data, a multi-point regression algorithm is used to fit the bearing capacity evolution curve, the bearing values ​​at key moments are extracted from the evolution curve, the overall bearing capacity is determined by bearing ratio calculation, and the bearing capacity assessment result of the foundation structure is obtained.

[0061] Preferably, the step S107 includes:

[0062] According to the dynamic change value of the foundation bearing capacity, the structural bearing capacity safety threshold is obtained from the safety monitoring database. The threshold is judged based on the bearing capacity change value. If it is lower than the safety threshold, the settlement displacement data and acoustic emission data are extracted from the online monitoring platform to obtain monitoring fusion data.

[0063] For the monitoring fusion data, a time series matching algorithm is used to perform time alignment of the monitoring data, and a mapping relationship between the settlement monitoring points and the acoustic emission source positions is established through spatial coordinate transformation to obtain calibration coordinate data;

[0064] Based on the calibration coordinate data, a deep convolutional neural network is used to extract the spatial features of the cracks, the deformation trend is calculated from the surface displacement monitoring data, and a crack extension mapping function is established by combining the features to obtain the spatial data of the crack extension;

[0065] Based on the spatial data of crack propagation, a three-dimensional mesher is used to construct a crack propagation reference surface, and depth layered data is extracted from the reference surface. The point cloud is fitted using an iterative closest point algorithm to obtain a three-dimensional contour of the crack propagation.

[0066] Based on the three-dimensional crack extension profile, a bearing capacity loss calculator is used to establish a regional damage distribution map, hazard levels are extracted from the damage distribution map, and the hazard levels are sorted using a hierarchical analysis method to obtain a structural hazard classification;

[0067] According to the structural hazard classification, a priority sorting algorithm is used to determine the order of repair areas, the repair urgency is calculated from the degree of regional damage, and the regional classification identification is generated through the repair level matrix to obtain the repair priority data.

[0068] Preferably, the step S108 includes:

[0069] According to the coordinates of the repair priority area, an ultrasonic detection array is used to transmit detection waves. The ultrasonic frequency and amplitude parameters are extracted from the standard waveform library. The echo signal is extracted through a bandpass filter. The filtered signal is subjected to Hilbert transform to obtain the ultrasonic envelope curve.

[0070] Based on the ultrasonic envelope curve, a peak recognition algorithm is used to extract the peak position of the reflected wave, the crack depth parameter is calculated from the time difference of the reflected wave, and the actual crack depth value is calculated through acoustic wave attenuation compensation to obtain crack depth distribution data;

[0071] Based on the crack depth distribution data, a stress redistribution function is constructed using an elastic-plastic mechanics calculator, elastic modulus and strength parameters are extracted from a material parameter library, and the structural bearing capacity is calculated using a stress equilibrium equation to obtain foundation bearing state data;

[0072] Based on the foundation bearing state data, a deep feedback neural network is used to establish a state assessment function, the displacement field and strain field features are extracted from the monitoring data, and the comprehensive score of the foundation structure is calculated through feature fusion to obtain the foundation health index;

[0073] According to the basic health index, a fuzzy judgment matrix is ​​used to establish a warning classification standard, and the safety threshold parameter is obtained from the warning grade library. If the health index is lower than the safety threshold, a corresponding level warning mark is generated to obtain the basic structure safety status;

[0074] According to the safety status of the infrastructure, an early warning signal is constructed using an early warning information generator, sound and light warning parameters are extracted from a signal feature library, the type of early warning signal is determined through an early warning level matrix, and basic safety early warning data is obtained.

[0075] The cloud-based settlement data analysis and early warning method described in this application has the advantages of real-time collection of basic settlement data through an array of high-precision electronic settlement observation devices, determination of the strain distribution matrix in combination with finite element analysis and material mechanics models, identification of stress concentration locations, positioning of crack initiation locations using ultrasonic detection technology, and capture of acoustic wave signals during crack expansion using a sensor array.

[0076] The present invention predicts crack expansion trends by analyzing the relationship between crack propagation parameters and the strain distribution matrix, and inputs the predicted trends into a structural mechanics simulation model to calculate the dynamic changes in the foundation's bearing capacity. When the bearing capacity is lower than a safety threshold, the present invention generates a three-dimensional visualization path for crack propagation, determines the repair priority area, and generates a safety warning signal based on the ultrasonic detection results.

[0077] This method achieves early warning and precise repair guidance of cracks in the main transformer foundation, effectively ensuring the safe operation of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is the process of a cloud platform-based settlement data analysis and early warning method described in this application Figure 1 ;

[0079] Figure 2 This is the process of a cloud platform-based settlement data analysis and early warning method described in this application Figure 2 . DETAILED DESCRIPTION

[0080] like Figure 1-Figure 2As shown, the cloud platform-based settlement data analysis and early warning method described in this application includes the following steps:

[0081] S101. Using a high-precision electronic settlement observation device array, collect settlement data of the main transformer foundation in real time, obtain the real-time deformation distribution status of the foundation, and upload it to the cloud platform;

[0082] S102: Calculate the strain values ​​of various parts of the main transformer foundation based on the real-time deformation distribution state and determine the stress concentration location. Specifically, obtain stress distribution data of the foundation structure and identify local high stress areas through meshing, strain calculation, and stress analysis of the three-dimensional deformation data.

[0083] S103, using ultrasonic testing equipment to detect cracks at stress concentration locations on the main transformer foundation, emitting pulse signals and analyzing reflected wave data to identify the initiation location, depth, and range of the cracks;

[0084] S104, placing an acoustic emission sensor array at the location where the cracks initiation occurs in the main transformer foundation to capture the acoustic wave signals during crack expansion, and determining the time series of the crack expansion direction and speed through signal processing and time series analysis;

[0085] S105. Performing an update calculation based on the time series of crack expansion direction and velocity combined with the strain distribution matrix to analyze the impact of crack expansion on the strain distribution of the main transformer foundation and determine the trend of local strain aggravation;

[0086] S106. Predict the crack initiation location, expansion direction, and speed based on the local strain intensification trend, and calculate the dynamic changes in the overall bearing capacity of the main transformer foundation through a structural mechanics simulation model to provide data support for safety assessment;

[0087] S107. If the dynamic change value of the main transformer foundation bearing capacity is lower than the safety threshold, the settlement observation and acoustic emission monitoring data are integrated through the cloud platform to generate a three-dimensional visualization path of the crack extension and determine the repair priority area, providing a decision-making basis for engineering maintenance;

[0088] S108. Repair the crack depth in the priority area through ultrasonic detection and calculate the adjusted basic health status value of the main transformer. After comparing it with the preset safety threshold, generate a corresponding safety warning signal to provide real-time feedback for project safety management.

[0089] like Figure 1-Figure 2 As shown, in step S101, the settlement data of the main transformer foundation is collected in real time by an array of high-precision electronic settlement observation devices, the real-time deformation distribution state of the foundation is obtained, and uploaded to the cloud platform;

[0090] Furthermore, in step S101, S1011, displacement data in three axes is collected from an array of high-precision electronic sedimentation observation devices through a distributed sensor gateway according to a preset sampling period, and after linear correction is performed on the raw data based on the sensor calibration curve, Kalman filtering technology is used to eliminate random error interference to obtain smooth filtered displacement data;

[0091] Then, the reference coordinate point information is obtained from the pre-calibrated reference pile, and the coordinate transformation matrix is ​​constructed using the least squares method. The three-dimensional spatial coordinates of the filtered displacement data are solved and normalized to generate normalized coordinate data.

[0092] The normalized coordinate data is then uploaded to the cloud platform to provide basic data support for analysis. The sampling period can be set according to actual needs, for example, it can be set to 100Hz to achieve millimeter-level accuracy. The sensor gateway is deployed in a star topology, and each gateway connects multiple sensors to form a collaborative observation network.

[0093] S1012. After obtaining normalized coordinate data, extract stress and strain parameters from preset monitoring points. Calculate the stress and strain distribution of each part of the foundation using an elastic mechanics model to generate deformation stress data. Construct a deformation trend curve based on the deformation stress data. Use a long-short-term memory network to predict the deformation trend over a period of time. Obtain predicted deformation data and construct a foundation deformation distribution map. Obtain the threshold value for each monitoring point from a preset deformation threshold library and compare it with the predicted deformation data. If the threshold value is exceeded, trigger an early warning signal. The early warning signal and predicted deformation data are encrypted and transmitted to the cloud platform database for storage via a data encryption module. Monitoring points can be divided into multiple units. For example, the foundation can be divided into 25 regions, each equipped with stress and strain sensors to cover key load-bearing areas.

[0094] In the embodiment of the present invention, the deployment of an array of high-precision electronic settlement observation devices fully considers the multi-point monitoring requirements of the main transformer foundation. The distributed sensor gateway connects 6 sensors via a star topology, and the calibration curve uses a piecewise linear fitting method to ensure that the error is less than 0.01mm within the 0-50mm range. The Kalman filter effectively removes Gaussian white noise through the prediction and update stages, and the smoothness of the filtered data is significantly improved. The reference coordinate points are selected from the four corners and the center of the foundation, and are fixed with reinforced concrete piles with a depth of 15 meters to ensure that the coordinate system conversion accuracy is better than 0.1mm.

[0095] For stress-strain analysis, the elastic mechanics model comprehensively considers the foundation's bearing capacity and load distribution to generate stress field distribution data. Deformation trend curves calculate displacement, velocity, and acceleration characteristics over a 48-hour sliding window. The long-short-term memory network utilizes a three-layer structure with 128 hidden neurons, predicting deformation trends over the next 24 hours with an error of less than 0.5mm. The deformation threshold library sets a 5mm threshold for load-bearing areas and an 8mm threshold for non-load-bearing areas. Data transmission utilizes a 2048-bit asymmetric encryption algorithm to ensure security.

[0096] In practical applications, such as monitoring the foundation of a main transformer at a 380kV substation, the system detected a cumulative settlement of 4.2mm at the northwest corner over 48 hours, with a settlement rate of 0.175mm / hour. The predicted settlement after 48 hours would be close to 7.8mm. The cloud platform issued a timely warning signal, and after engineers implemented reinforcement measures, the settlement rate dropped to below 0.02mm / hour, validating the effectiveness of the method.

[0097] It is understood that the embodiments of the present invention do not impose specific limitations on the sampling frequency, number of sensors, or network structure. These parameters can be adjusted by technical personnel based on actual scenarios to accommodate different monitoring needs. Subsequent steps will further analyze strain and crack states based on this real-time deformation data. The specific implementation methods will be detailed in the following embodiments.

[0098] like Figure 1-Figure 2 As shown, in step S102, the strain values ​​of various parts of the main transformer foundation are calculated according to the real-time deformation distribution state and the stress concentration location is determined. Specifically, the stress distribution data of the foundation structure is obtained and the local high stress area is identified through meshing, strain calculation and stress analysis of the three-dimensional deformation data.

[0099] Further, in step S102, S1021, the three-dimensional deformation data of the main transformer foundation is extracted from the deformation monitoring database stored in the cloud platform, the foundation structure is divided into hexahedral grid units using an adaptive mesher and grid node coordinate data is generated, and then the elastic modulus and Poisson's ratio parameters of the concrete are read from a pre-built material property library, and the initial strain value of each grid node is calculated using Hooke's law and compared with a preset baseline strain threshold to determine whether there is an abnormal strain area;

[0100] For abnormal strain areas, the finite element analysis method is used to calculate the strain energy density, and the energy equation is numerically integrated using the fourth-order Runge-Kutta method to generate cloud map data reflecting the strain distribution. When dividing the mesh, the foundation as a whole adopts an orthogonal grid with a side length of 200mm, while the key parts such as supports and edges are encrypted to a side length of 100mm to improve the calculation resolution of the stress concentration area. Strain energy density is the core indicator for measuring the deformation capacity of materials. Its calculation process takes into account the accumulation of elastic potential energy of concrete under loading. The typical value range is 0.1 to 0.5kJ / m 3 The integration step is set to 0.01s to ensure accuracy.

[0101] S1022. Based on the generated strain distribution cloud map data, the strain gradient of each grid node is calculated using the Gaussian integral method and the gradient change rate is extracted. The principal strain value and its direction are solved using the principal strain calculation formula combined with the three-dimensional characteristic equation. The principal stress distribution is then calculated using the stress-strain conversion relationship in elastic mechanics to generate stress distribution data for the foundation structure.

[0102] Subsequently, the maximum shear stress surface and its angle with the horizontal plane were analyzed using the Mohr circle criterion, and the stress intensity value of each node was evaluated in combination with the stress intensity calculation formula. If the stress intensity of a node exceeds twice the average value, it is determined to be a stress concentration area and its coordinate position is recorded. A stiffness matrix correction is introduced into the principal strain calculation to reflect the anisotropic properties of concrete in different directions, and the stress distribution data is derived using the generalized Hooke's law to ensure that the results are consistent with the actual stress state.

[0103] In this embodiment of the present invention, the acquisition frequency of 3D deformation data is set to 10 Hz, enabling real-time recording of minute displacement changes on the foundation surface. The concrete parameters stored in the material property library include an elastic modulus of 30 GPa, a Poisson's ratio of 0.2, and a compressive strength of 30 MPa. These parameters are applicable to stabilized concrete older than 90 days, and strain calculations are performed based on a linear elastic constitutive model. A baseline strain threshold of 200 microstrains serves as a reference for abnormality determination.

[0104] The strain distribution cloud map is displayed in the form of colored contour lines, graded in 100-microstrain increments from 0 to 1000 microstrain, making it easy to intuitively identify high-strain areas. The fourth-order Runge-Kutta method improves calculation stability through multiple iterations in numerical integration, and the error is controlled within 0.1%. When calculating the principal stress distribution, special attention is paid to the area below the load application point, as it is often the starting point of stress concentration. A measured case shows that the maximum principal strain value of the main transformer foundation of a 500kV substation is approximately 400 microstrain, located directly below the support, with an accompanying stress intensity of 12MPa, close to 40% of the design strength.

[0105] Based on the stress distribution data, S1023 further analyzes the characteristics of stress concentration areas. Using the stress intensity ratio method, it determines the distribution of local high stress points and annotates them in the basic 3D model. For example, within a 300mm radius around the transformer support, an elliptical stress concentration area is identified, with the maximum shear stress plane pointing toward the edge at a 45-degree angle to the horizontal. These coordinates provide a precise spatial reference for predicting crack initiation locations. Identifying stress concentration areas also helps optimize the placement of monitoring points. The stress intensity calculation formula comprehensively considers the combined effects of principal and shear stresses, ensuring the reliability of the results.

[0106] In an embodiment of the present invention, the use of an adaptive mesher improves computational efficiency. Its meshing scheme dynamically adjusts based on the spatial distribution of deformation data, avoiding the inaccuracy of traditional uniform meshes in high-strain regions. The application of the Mohr circle criterion provides theoretical support for directional analysis of shear stress surfaces, making the determination of stress concentration areas more scientific. During actual monitoring, the coordinate data of stress concentration locations can be directly imported into the cloud platform's 3D visualization module, enabling engineers to quickly locate and implement targeted measures.

[0107] It is understood that this invention does not impose any specific restrictions on the grid size, calculation step size, or material parameter values. These can be adjusted by technical personnel based on actual engineering needs to accommodate different infrastructure analysis scenarios. Subsequent steps will involve crack detection and propagation analysis based on these stress concentration locations.

[0108] like Figure 1-Figure 2 As shown, in step S103, crack detection is performed on the stress concentration position of the main transformer foundation by ultrasonic detection equipment, pulse signals are emitted and reflected wave data are analyzed to identify the initiation position, depth and range of the cracks.

[0109] Furthermore, in step S103, in an embodiment of the present invention, an array ultrasonic probe is used to generate a longitudinal wave pulse signal with a frequency of 2 MHz to 5 MHz for the coordinate data of the stress concentration position, and the transmission waveform parameters are set by the probe vibrator controller and the acoustic impedance matching is performed in combination with the acoustic characteristics of concrete to generate an optimized transmission wave signal; then an ultrasonic receiver is used to collect the reflected wave data, the noise is removed by a bandpass filter, and the envelope characteristics are extracted by Hilbert transform, and then the spatial distribution characteristics of the cracks are obtained by combining attenuation compensation and time-frequency analysis. The array probe uses 8 probes arranged linearly with a spacing of 25 mm and a coverage length of 200 mm. The transmission frequency is preferably 3 MHz to adapt to the wavelength of 1.2 mm in concrete, and the sound wave propagation speed is 3600 m / s. The acoustic impedance matching is optimized by coupling agent, and the acoustic impedance of concrete is 8×10 6 kg / m 2 s, the acoustic impedance of the coupling agent is 4×10 6 kg / m2 s, ensuring efficient energy transfer.

[0110] S1031. After the transmission wave signal is generated, a Gaussian modulated pulse is used with a pulse width of 1 μs and a repetition frequency of 1 kHz. The reflected wave signal is captured by an ultrasonic receiver and a bandpass filter with a frequency range of 2.5 MHz to 3.5 MHz is used to filter out low-frequency aggregate scattering noise and high-frequency electromagnetic interference. The attenuation band rejection ratio is greater than 40 dB. The filtered reflected wave signal is then subjected to a Hilbert transform to extract time-domain envelope feature data to highlight the echo peak distribution. The correlation between peak amplitude and crack size provides a basis for subsequent analysis. Based on the envelope feature data, an attenuation compensation coefficient, such as 0.8 dB / cm, is read from the acoustic wave compensation database. Exponential attenuation compensation is applied to the reflected wave signal to enhance deep signal strength. For example, the amplitude at a depth of 80 cm can be increased by 12 dB, thereby enhancing the detection sensitivity of deep cracks. The Hilbert transform decomposes the signal into amplitude and phase components, making the time-domain characteristics of the crack reflection wave more prominent, facilitating subsequent feature extraction.

[0111] S1032. After obtaining the compensated reflected wave signal, discrete wavelet transform (DWT) was used to extract time-frequency features. A five-layer multiscale decomposition was performed using the db4 wavelet basis function to generate a time-frequency spectrum of the reflected wave. The dominant frequency component was concentrated between 2.8MHz and 3.2MHz, with a spectrum broadened by approximately 0.4MHz. Based on the time-frequency spectrum, standard crack feature vectors were extracted from the defect feature database, covering sample data of varying depths and inclinations. A random forest identifier was used to calculate feature matching. The training sample consisted of 1,000 data sets, with 15 feature dimensions, including time-domain waveforms, spectra, and statistical parameters. The resulting crack type identification results were then output. The waveform peak time difference method was used to calculate crack depth, with a travel time resolution of 0.1μs and a depth accuracy better than 2mm. Phase spectrum analysis was used to determine the crack propagation direction, with a phase sensitivity of 0.5° / degree. This generated characteristic data for the spatial distribution of cracks. The multiscale decomposition of the wavelet transform can separate noise from valid signals. The time-frequency spectrum intuitively reflects the frequency distribution characteristics of the crack reflection wave. The random forest algorithm improves recognition robustness through multi-decision tree voting.

[0112] In this embodiment of the present invention, ultrasonic testing is implemented to address the inaccuracy of traditional methods in identifying deep cracks. Acoustic impedance matching improves detection sensitivity by reducing acoustic wave reflection losses, while a bandpass filter effectively isolates complex interference within the concrete medium. The application of attenuation compensation restores deep reflection wave signals, ensuring the ability to detect penetrating cracks.

[0113] During an actual inspection of the main transformer foundation of a 500kV substation, a through-crack was discovered with an initiation depth of 15cm, a 35-degree inclination, and an extension length of 45cm. The reflected wave amplitude was 6dB higher than that of the intact area, the spectrum was widened by 0.6MHz, and the phase difference varied by more than 15 degrees. Using the time difference method and phase analysis, the crack location and trend were accurately demarcated, providing critical data support for reinforcement measures. The crack feature extraction and matching process fully demonstrated the accuracy of the method, and the detection results can be directly used for subsequent analysis on the cloud platform.

[0114] It is understood that the present invention does not impose any specific restrictions on the number of probes, frequency range, or filter parameters, which can be adjusted by technicians based on the actual characteristics of concrete. For example, if the object being tested is high-density concrete, the transmission frequency can be appropriately increased to shorten the wavelength and improve resolution.

[0115] like Figure 1-Figure 2 As shown, in step S104, an acoustic emission sensor array is arranged at the crack initiation location of the main transformer foundation to capture the acoustic wave signal when the crack is expanding, and the time series of the crack expansion direction and speed is determined through signal processing and time series analysis.

[0116] Furthermore, in step S104, in an embodiment of the present invention, an acoustic emission sensor array is arranged in a hexagonal topology according to the crack initiation location. The central sensor is placed directly above the crack initiation point, and the surrounding six sensors are distributed in a regular hexagon with a spacing of 300 mm. The sampling frequency is set to 1 MHz and the trigger voltage is set to 50 mV by the sensor matrix controller. Standard waveform parameters are obtained from the acoustic emission calibration database for sensitivity calibration to generate calibrated acquisition parameters. The acoustic emission signal is collected using a data collector, and the 100 kHz to 300 kHz frequency band signal is filtered by a bandpass filter. The ambient noise is removed by an adaptive threshold filter with the threshold set to 2.5 times the root mean square value of the background noise to obtain clear acoustic wave event data. The dynamic adjustment of the adaptive threshold can optimize the filtering effect in real time according to the on-site noise level to ensure signal quality. The hexagonal topology improves the coverage and accuracy of sound source localization through multi-point collaborative monitoring.

[0117] S1041. For the collected acoustic event data, Fourier transforms were used to calculate spectral features, extracting the main frequency band signal from 150kHz to 250kHz. This was then decomposed into an 8-level time-frequency spectrum using the db4 wavelet basis function, clearly demonstrating the frequency distribution and temporal variations of the acoustic emission events. A triangulation algorithm was then used to calculate the sound source location based on the time difference of arrival of the sound waves, achieving a positioning accuracy of better than 5mm and an angular resolution of 2 degrees. Crack propagation trajectory data was generated by arranging the source coordinates in time series. A deep recurrent neural network was further used to construct a time series prediction model. The network consists of three layers, with 128 time series nodes in the input layer and 256 neurons in the hidden layer. An acoustic emission energy density calculator was used to analyze the crack propagation rate, using a 10ms window width and a 2ms sliding step size, and outputting crack propagation state parameters. The triangulation algorithm restores the spatial location of the acoustic source by analyzing the time difference between multiple sensors, while the deep recurrent neural network predicts future propagation trends by learning from historical data, enhancing the predictive power of the analysis.

[0118] S1042. After acquiring crack propagation trajectory data, the Hilbert transform is used to extract the envelope curve of the acoustic signal to reflect the energy release process. The time difference series between adjacent sensors is calculated using the cross-correlation function. The sound source coordinates are calculated and the propagation displacement vector is determined by combining the sensor coordinate library. The propagation direction is calculated based on the azimuth of the displacement vector, and the propagation rate is calculated by the ratio of the displacement to the time interval. A velocity prediction model is constructed using a support vector regression predictor. The kernel function uses the radial basis function, the parameter γ is 0.1, the penalty factor C is 100, and the training sample is the most recent 100 event data. Continuous propagation velocity prediction results are generated. The velocity data is smoothed using a Kalman filter with a process noise covariance of 0.01 and a measurement noise covariance of 0.1. The propagation direction is calculated using vector synthesis, and a crack propagation time series containing coordinates, velocity, and direction is generated. Support vector regression optimizes the accuracy of velocity prediction through nonlinear mapping, while the Kalman filter effectively eliminates random fluctuations in the data, ensuring the smoothness and reliability of the time series.

[0119] In an embodiment of the present invention, the acquisition bandwidth of the acoustic emission signal is set to 100kHz to 400kHz, the pre-amplification factor is 40dB, and the single-channel cache depth is 32MB to meet the real-time storage requirements of high-frequency signals. Measured data show that the amplitude of the acoustic emission signal during crack expansion fluctuates between 100mV and 500mV, the main frequency is concentrated around 150kHz, the spectrum is widened by about 50kHz, and the duration is 0.5ms to 2ms. Monitoring of the main transformer foundation of a 500kV substation showed that the crack expanded at an angle of 45 degrees under lateral load. 458 acoustic emission events were recorded within 24 hours, of which 80% were in the rapid expansion stage, and the event density within 50mm of the crack initiation point reached 12 / cm. 2The maximum expansion speed is 2mm / s, the average rate is about 0.5mm / s, and the total length is 45mm.

[0120] It is understood that this invention does not impose strict limits on the specific values ​​of sensor spacing, sampling frequency, or filtering parameters. These can be adjusted based on the scale of the crack and the monitoring environment. For example, in high-noise environments, the threshold multiplier can be appropriately increased to enhance interference resistance. Subsequent steps will further evaluate the impact of the cracks on the foundation based on this time series.

[0121] like Figure 1-Figure 2 As shown, in step S105, the time series of crack extension direction and speed is combined with the strain distribution matrix to perform update calculations, analyze the impact of crack extension on the strain distribution of the main transformer foundation, and determine the local strain aggravation trend.

[0122] Furthermore, in step S105, in an embodiment of the present invention, measured strain values ​​are extracted from the strain monitoring database and updated using a strain redistribution calculator. The acquisition frequency is set to 10 Hz, and strain data is acquired using fiber Bragg grating sensors with a measurement range of ±5000 microstrain and a resolution of 1 microstrain. The foundation is divided into 200 100 mm × 100 mm monitoring units to record strain changes in real time. Strain differences are calculated based on the crack propagation time series to determine the boundaries of the affected area and generate a real-time strain distribution matrix. Subsequently, a tensor mapping method is used to calculate the strain change around the crack propagation trajectory. Nine-node quadrilateral elements are used for spatial discretization, and the strain field is reconstructed using a bilinear interpolation function. Combined with the finite element calculator results, radial basis function interpolation is used to establish the strain distribution update data. The interpolation radius is set to 100 mm, and the weight decays exponentially with distance. A strain influence tensor is generated. Tensor mapping transforms the local effects of crack propagation into global strain field changes through discretization, ensuring the spatial continuity of the calculation results.

[0123] S1051. A recurrent neural network (RNN) was used to construct a strain transfer function for the strain influence tensor to quantify the effect of crack propagation on the strain field. The network consisted of 128 time series nodes, with long short-term memory units (LSTMs) used in the hidden layer. Input features included crack propagation velocity, azimuth, and strain increment. The training data consisted of 1,000 sets of 60-second strain field evolution sequences. Strain coupling calculations were used to determine the strain influence coefficients at each monitoring point and generate a strain transfer matrix. The strain time series of high-stress areas were then extracted from the strain transfer matrix. The strain rate of change was calculated using a 10-second sliding window and a 1-second step size. A wavelet transform was used to identify breakpoints, where the wavelet coefficient amplitude increased. Local strain gradients were calculated using a central difference scheme with a 50-mm grid spacing. The extent of strain exacerbation regions was determined and strain concentration trend data were generated. By capturing temporal dependencies, the RNN effectively simulated the dynamic process of strain transfer, while the wavelet transform enhanced the sensitivity of breakpoint detection.

[0124] In the embodiments of the present invention, the impact of crack expansion on the foundation strain distribution exhibits significant spatial heterogeneity. Measured data show that the strain gradient within 50 mm in front of the crack tip rises sharply, with the maximum strain value increasing from 300 microstrain to 800 microstrain, with an exponential increase, reflecting the amplification effect of crack expansion on the local stress field. Monitoring of the main transformer foundation of a 500 kV substation showed that crack expansion caused significant changes in the strain field within 80 mm. The influence coefficient decayed from the center of the crack outward, dropping to 20% of the initial value at the boundary. For every 0.1 mm / s increase in expansion speed, the affected area expanded by approximately 5 mm.

[0125] S1052. In strain concentration trend analysis, the strain rate in high-stress areas rapidly increased from an initial 5 microstrain / s to 20 microstrain / s. The strain-intensified region exhibited an elliptical distribution, with the major axis aligned with the crack propagation direction and the minor axis perpendicular to it. Within 24 hours, the area expanded 2.5 times, with a maximum strain gradient reaching 15 microstrain / mm, most pronounced at the crack intersection. Strain transfer matrix analysis revealed a positive correlation between crack propagation velocity and the strain transfer coefficient. This correlation provides data support for predicting local strain intensification. The generation of strain concentration trend data also helps identify potential high-risk areas. Strain gradient calculations quantify the spatial rate of strain change through a differential method, providing a precise basis for dynamically tracking the crack's impact range.

[0126] It is understood that this invention does not impose strict restrictions on the specific settings of monitoring unit division, interpolation radius, or network parameters. These can be flexibly adjusted based on the foundation scale and crack characteristics. For example, in complex stress areas, the monitoring unit density can be increased to improve resolution. Subsequent steps will further evaluate the foundation's bearing capacity based on this strain trend data.

[0127] like Figure 1-Figure 2 As shown, in step S106, the initiation location, expansion direction and speed of the crack are predicted according to the local strain aggravation trend, and the dynamic change of the overall bearing capacity of the main transformer foundation is calculated through the structural mechanics simulation model to provide data support for safety assessment.

[0128] Furthermore, in step S106, in an embodiment of the present invention, a five-point difference format is used to calculate the strain gradient field distribution using local strain exacerbation trend data. The grid spacing is set to 50 mm, and is increased to 25 mm in high-strain areas to improve resolution. The strain growth rate curve is extracted from the strain gradient field, and the strain mutation location data is obtained by identifying the curve mutation point. The strain rate at the mutation point can jump from 15 microstrain / s to 35 microstrain / s within 0.1 s. The least squares method is used to fit the mutation location data. The contour line spacing is 0.5 mm, and the spacing is reduced to 0.1 mm in the crack initiation area. The crack initiation location is determined by the area with a curvature radius less than 100 mm. The expansion trend angle is calculated in combination with the principal strain direction to generate the crack space parameters. The five-point difference format improves the accuracy of gradient calculation through multi-point numerical approximation, while the contour line curvature analysis intuitively reflects the local anomalies of the strain field, providing a reliable basis for crack location.

[0129] S1061. Based on the spatial parameters of cracks, a deep neural network was used to predict the crack propagation path and velocity. The network consists of three layers: the input layer integrates 15 features, including strain, stress, and displacement fields; the hidden layer is configured with 256 neurons; and the output layer generates propagation path and velocity predictions. The propagation velocity was calculated from the strain growth curve, and the path and velocity were combined through spatiotemporal mapping to generate crack propagation prediction data. The prediction showed that the initial propagation direction was approximately 45 degrees from the maximum principal stress, and the velocity could reach 0.8 mm / s. Concrete material parameters were then read from a basic parameter library, including an elastic modulus of 30 GPa, a Poisson's ratio of 0.2, a compressive strength of 30 MPa, and a tensile strength of 2.5 MPa. The stiffness degradation was calculated using an elastic-plastic damage function. The damage factor was increased from 0 to 0.35, and the degradation region was fan-shaped with a 60-degree vertex angle and an impact range of twice the crack length. The deep neural network optimized the path prediction accuracy through multidimensional feature learning, while the elastic-plastic damage function quantified the weakening effect of the crack on the material stiffness.

[0130] S1062. After obtaining crack propagation prediction data, a stress redistribution algorithm was used to calculate the stress distribution of the foundation structure. The principal stress directions were extracted from the stress distribution cloud map, and the ratio of principal stress to allowable stress was calculated to assess the load-bearing capacity. This generated data on the change in bearing capacity after stiffness damage. A multi-point regression algorithm was used to fit the bearing capacity evolution curve. 50 characteristic points on the foundation surface were selected for calculation. The bearing values ​​at critical moments were extracted, and the overall bearing capacity was determined using the bearing ratio. The foundation was divided into 25 calculation units, and the dynamic change in bearing capacity was assessed using the segmented accumulation method. The stress redistribution algorithm simulated the stress transfer process caused by cracks through iterative calculations, while the multi-point regression algorithm captured the nonlinear downward trend of bearing capacity through curve fitting.

[0131] In an embodiment of the present invention, monitoring of the foundation of a main transformer in a 500kV substation revealed that crack expansion significantly increased stress within a 50mm area, shifting the principal stress direction by 25 degrees and forming a new stress concentration zone. The load-bearing capacity decreased in a stepwise manner, decreasing by approximately 2% for every 15mm of expansion, totaling 12% over 24 hours, with 80% of the decrease occurring during the rapid expansion phase. The maximum drop in the load-bearing capacity of a local unit reached 35%, while the drop in adjacent units ranged from 15% to 25%. Ultimately, the bending capacity decreased by 18%, and the shear capacity decreased by 22%.

[0132] It is understood that this invention does not impose fixed restrictions on the grid spacing, number of neural network layers, or specific values ​​of material parameters. These can be adjusted based on actual project needs. For example, in areas with dense cracks, the grid can be further refined to improve calculation accuracy. Subsequent steps will generate early warning signals based on this bearing capacity data.

[0133] like Figure 1-Figure 2 As shown, in step S107, if the dynamic change value of the main transformer foundation bearing capacity is lower than the safety threshold, the settlement observation and acoustic emission monitoring data are integrated through the cloud platform to generate a three-dimensional visualization path of crack expansion and determine the repair priority area, providing a decision-making basis for engineering maintenance.

[0134] Furthermore, in step S107, in an embodiment of the present invention, a structural bearing capacity safety threshold is extracted from the safety monitoring database. This threshold is typically set at 80% of the design bearing capacity. For example, for a 500kV substation foundation with a design bearing capacity of 24MPa, the safety threshold is 19.2MPa. When the measured value drops to 17.8MPa, an early warning is triggered. For situations where the value falls below the threshold, settlement displacement data and acoustic emission data are acquired from the online monitoring platform. The settlement monitoring frequency is 10Hz, and the acoustic emission monitoring frequency is 1MHz. Time alignment is performed using a timing matching algorithm with a 1s sliding window and a 0.1s step size, achieving an alignment accuracy of better than 0.1ms. The least squares method is then used to perform spatial coordinate transformation, establishing a mapping relationship between settlement monitoring points and acoustic emission sources with a registration accuracy of 2mm, and generating calibrated coordinate data. Timing matching ensures the synchronization of multi-source data, while coordinate transformation achieves precise spatial position correspondence through matrix calculation.

[0135] S1071. Based on the calibrated coordinate data, a deep convolutional neural network was used to extract spatial features of the crack. The network was designed as a five-layer structure. The input features included displacement fields, strain fields, and acoustic emission signal features, and the output was a three-dimensional feature vector of the crack. Deformation trends were calculated from the surface displacement data, revealing a maximum settlement of 12 mm. The settlement basin was elliptical, with the major axis aligned with the fracture direction. A crack propagation mapping function was constructed by combining these features, linking surface deformation with internal cracks. The mapping accuracy reached 85%, generating spatial data for crack propagation. A three-dimensional mesher was then used to construct a crack propagation reference plane using 50 mm hexahedral elements. This was then refined to 25 mm within the fractured area. Depth-layered data was extracted from the reference plane. Point cloud fitting was performed using an iterative closest point algorithm combined with a KD tree accelerated search. After 50 iterations and a convergence threshold of 0.1 mm, a 3D crack propagation contour was generated, showing a depth of 180 mm, a width of 0.8 mm, and a 45-degree diagonal extension. The deep convolutional neural network extracted the spatial pattern of the crack through multiple layers of convolution and pooling, while the iterative closest point algorithm optimized the efficiency and accuracy of the point cloud fitting.

[0136] S1072. After generating a three-dimensional crack expansion profile, a bearing capacity loss calculator was used to construct a regional damage distribution map. Damage levels were indicated using red, yellow, and blue colors: greater than 0.6 (red), 0.3 to 0.6 (yellow), and less than 0.3 (blue). The hazard level was ranked using the Analytic Hierarchy Process (AHP), taking into account crack depth, width, and expansion rate, with weights of 0.5, 0.3, and 0.2, respectively. Three hazardous areas were identified, with the area below the load-bearing support having a hazard level of 0.82. A comprehensive scoring method was then used for prioritization, combining damage severity, location importance, and expansion trend. Repair urgency was calculated based on the damage growth rate, with growth rates exceeding 5% considered urgent. The foundation was divided into 25 zones using a repair level matrix. Repair priority data was generated based on damage severity, location coefficient, and load factor. Four areas surrounding the support, with a damage level of 0.75, a daily growth rate exceeding 8%, and a load factor of 1.5, were classified as Level 1 repair areas. The AHP method improved the scientific nature of the ranking by using quantitative indicators, while the repair level matrix provided a systematic assessment framework for zone division.

[0137] In the embodiment of the present invention, the data fusion and visualization capabilities of the cloud platform significantly improve the intuitiveness of the crack expansion path and the pertinence of the repair decision. The monitoring data shows the high risk of the load-bearing area, providing a key basis for timely intervention.

[0138] It is understood that this invention does not impose strict restrictions on grid size, number of neural network layers, or weight distribution. These can be adjusted based on actual monitoring needs. For example, in high-risk areas, the grid can be further refined or input features can be added to improve accuracy. Subsequent steps will involve verification and early warning based on the repair priority data.

[0139] like Figure 1-Figure 2 As shown, in step S108, the crack depth in the repair priority area is detected by ultrasonic detection and the adjusted basic health status value of the main transformer is calculated. After comparison with the preset safety threshold, a corresponding safety warning signal is generated to provide real-time feedback for project safety management.

[0140] Furthermore, in step S108, in an embodiment of the present invention, an 8-channel ultrasonic detection array is used to emit Gaussian modulated pulse detection waves at the coordinate location of the repair priority area, with a center frequency of 3MHz, a pulse width of 1μs, a probe spacing of 25mm, and a coverage length of 200mm. Frequency and amplitude parameters are extracted from a standard waveform library, and the signal is enhanced by a 40dB preamplifier. A bandpass filter is used to filter the 2.5MHz to 3.5MHz frequency range to remove concrete aggregate scattering noise. The filtered signal is Hilbert transformed to generate an envelope curve that clearly reflects the echo amplitude distribution. A peak recognition algorithm is used to extract the peak position of the reflected wave with a threshold of 3 times the noise root mean square value. The crack depth is calculated through time difference analysis. The actual depth is quantified by combining the sound wave propagation velocity of 3600m / s and the attenuation coefficient of 0.8dB / cm to generate crack depth distribution data. The measurement error is controlled within 5mm, and the time difference accuracy is better than 0.1μs. Ultrasonic detection ensures the reliability of deep crack detection by combining high-frequency pulses with precise filtering.

[0141] S1081. Based on crack depth distribution data, an elastic-plastic mechanics calculator was used to construct a stress redistribution function. Concrete elastic modulus (30 GPa), compressive strength (30 MPa), and tensile strength (2.5 MPa) were extracted from the material parameter library. Taking into account the nonlinear properties of the material, the structural bearing capacity was calculated using the stress equilibrium equation, generating foundation bearing state data. The calculations showed a 35% increase in stress within a 50 mm radius around the crack, with a corresponding decrease in local bearing capacity. A deep feedback neural network was then used to establish a state assessment function. The input features included 15 items, including displacement, strain, and crack parameters. Long short-term memory units were used in the hidden layer to capture temporal dependencies. Features were fused using a weighted average method, and weights were optimized using a backpropagation algorithm. The output was a basic health index. In one monitoring case, the health index decreased from 0.95 to 0.68. The elastic-plastic calculation quantified the weakening effect of cracks on bearing capacity, while the neural network enhanced the comprehensiveness of the state assessment through multidimensional feature analysis.

[0142] S1082. After obtaining the basic health index, a fuzzy evaluation matrix was used to categorize warning levels. Based on Monte Carlo simulation, a safety threshold of 0.75, a yellow warning threshold of 0.65, an orange warning threshold of 0.55, and a red warning threshold of 0.45 were set. If the index falls below 0.75, a warning is triggered. For example, a yellow warning is generated at an index of 0.68. A warning information generator extracts acoustic and optical parameters from a signal feature library. Yellow warnings are issued once an hour, with a sound frequency of 500Hz to 2kHz and a low LED flashing frequency. Orange warnings are issued every 30 minutes, and red warnings are issued more frequently, every 10 minutes. Basic safety warning data with level identification is generated. During continuous monitoring, 32 warnings were recorded: 25 yellow, 5 orange, and 2 red, with an accuracy rate of 95%. Fuzzy evaluation optimizes threshold setting through probability distribution, and the graded warning mechanism ensures timely and differentiated responses.

[0143] In an embodiment of the present invention, the foundation is divided into 25 calculation units for load-bearing status assessment. The health index of the crack area decreases significantly. The generation of early warning signals effectively indicates the repair effect and potential risks, providing an intuitive basis for subsequent maintenance.

[0144] It is understandable that the present invention does not impose fixed restrictions on the number of probes, filtering frequency bands or warning frequencies, and can be adjusted according to the characteristics of the repair area. For example, in a high-noise environment, the filtering range can be appropriately widened to improve signal quality.

[0145] Those skilled in the art can make various other corresponding changes and deformations based on the technical solutions and concepts described above, and all of these changes and deformations should fall within the scope of protection of the claims of this application.

Claims

1. A settlement data analysis and early warning method based on a cloud platform, characterized in that: include: The distributed sensor gateway collects the three-axis displacement data of the electronic settlement observation device array, performs three-dimensional spatial coordinate solution, calculates deformation stress data, and uploads it to the cloud platform; Establish hexahedral mesh elements based on deformation stress data, calculate node strain energy density, and determine stress concentration locations of the foundation structure through strain gradient analysis; Ultrasonic pulse signals are emitted to the stress concentration location to identify the crack initiation location. An acoustic emission sensor array is arranged in the crack area to capture the extended acoustic wave signal and construct a time series of the crack propagation direction and speed. Establish a dynamic correlation model between the strain distribution matrix and crack propagation parameters to determine the local strain aggravation trend; The local strain intensification trend is input into the structural mechanics simulation model to calculate the dynamic change value of the foundation bearing capacity. When the dynamic change value is lower than the safety threshold, a three-dimensional visualization path of the crack is generated and the repair priority area is determined; Conduct ultrasonic depth testing on priority repair areas, calculate the structural bearing capacity and generate graded warning signals.

2. The cloud platform-based settlement data analysis and early warning method according to claim 1, characterized in that: The method collects the three-axis displacement data of the electronic sedimentation observation device array through the distributed sensor gateway, performs three-dimensional spatial coordinate solution and calculates the deformation stress data and uploads it to the cloud platform, including: Displacement data is collected through the sensor gateway and Kalman filtering and smoothing are applied; the least squares method is used to solve the coordinates and generate normalized coordinate data; the coordinate data is uploaded to the cloud platform; stress and strain parameters are extracted and deformation stress data is calculated using an elastic mechanics model; a deformation trend curve is constructed and predicted using a long short-term memory network; a warning is triggered by comparing with the preset threshold and the data is encrypted and transmitted to the cloud platform.

3. The cloud platform-based settlement data analysis and early warning method according to claim 1, characterized in that: The method of establishing hexahedral grid units based on deformation stress data, calculating node strain energy density, and determining stress concentration locations of the foundation structure through strain gradient analysis includes: Three-dimensional deformation data is extracted from the cloud platform, adaptive meshing is performed, and initial strain values ​​are calculated. Finite element analysis and the Runge-Kutta method are used to calculate the strain energy density of abnormal strain areas and generate cloud maps. The strain gradient and principal strain values ​​are calculated to determine the principal stress distribution. The stress intensity is evaluated using the Mohr circle criterion and the stress intensity formula, and nodes that exceed twice the average value are identified as stress concentration locations.

4. The cloud platform-based settlement data analysis and early warning method according to claim 1, characterized in that: The method of transmitting ultrasonic pulse signals to the stress concentration position, identifying the crack initiation position, arranging an acoustic emission sensor array in the crack area to capture the extended acoustic wave signal, and constructing a time series of the crack extension direction and speed includes: An acoustic emission sensor array with a hexagonal topology is used to collect data, and an adaptive threshold filter is used to remove noise. The acoustic wave event data is processed through Fourier transform, wavelet decomposition and triangulation positioning algorithm to generate crack propagation trajectories. Deep recurrent neural network and Hilbert transform are used to analyze crack propagation state parameters. The propagation direction and velocity are calculated using cross-correlation function and support vector regression, and the data is smoothed through Kalman filter to obtain a crack propagation time series containing coordinates, velocity and direction.

5. The cloud platform-based settlement data analysis and early warning method according to claim 1, characterized in that: The establishment of a dynamic correlation model between the strain distribution matrix and the crack propagation parameters to determine the local strain aggravation trend includes: The measured strain values ​​are extracted from the strain monitoring database and the strain distribution is updated; the strain changes are recorded in real time using fiber Bragg grating sensors; the strain difference is calculated based on the crack extension time series and the boundary of the affected area is determined; the strain distribution update data is established using tensor mapping and finite element calculator results; the strain transfer function is constructed through a recursive neural network and the strain influence coefficient is calculated; the strain time series of the high stress area is extracted and the strain change rate is calculated; the wavelet transform is used to identify the mutation point and calculate the local strain gradient, determine the range of the strain intensification area, and generate strain concentration trend data.

6. The cloud platform-based settlement data analysis and early warning method according to claim 1, characterized in that: The local strain aggravation trend is input into the structural mechanics simulation model to calculate the dynamic change value of the foundation bearing capacity. When the dynamic change value is lower than the safety threshold, a three-dimensional visualization path of the crack is generated and the repair priority area is determined, including: Extract the structural bearing capacity safety threshold and trigger an early warning; obtain settlement displacement and acoustic emission data, and align them in time and space through time series matching and coordinate transformation; use a deep convolutional neural network to extract the spatial characteristics of cracks and construct a crack extension mapping function; use three-dimensional meshing and an iterative closest point algorithm to generate a three-dimensional contour of crack extension; construct a regional damage distribution map and rank the hazard levels; and use a comprehensive scoring method and a repair level matrix to generate repair priority data.

7. The cloud platform-based settlement data analysis and early warning method according to claim 1, characterized in that: The ultrasonic depth detection of the repair priority area and calculation of the structural bearing capacity to generate a graded warning signal include: An ultrasonic detection array is used to obtain crack depth distribution data; an elastic-plastic mechanics calculator is used to construct a stress redistribution function and calculate the structural bearing capacity; a deep feedback neural network is used to establish a state assessment function and generate a basic health index; a fuzzy judgment matrix is ​​used to divide the warning level and trigger the warning; and a graded warning signal is sent through a warning information generator.

Images