Method and system for health diagnosis of stone structure diversion channel and storage medium

Abstract

The present application relates to the technical field of civil engineering structure health monitoring, and particularly relates to a health diagnosis method and system for a stone structure diversion channel and a storage medium. The health diagnosis method for the stone structure diversion channel comprises: generating acceleration response data of health and damage working conditions based on a finite element model, constructing time-frequency matrix data by performing fast discrete orthogonal S transformation on the acceleration signal data; calculating bilateral time-frequency power spectrum density data of each acceleration response data; calculating global transfer function matrix data of each acceleration response data through a transfer function; extracting damage features by comparing the global transfer function matrix data before and after damage, and constructing damage feature vector data; and establishing a mapping model of the acceleration response data and the damage feature vector data by using a Laplace neural operator. The present application significantly improves the accuracy and efficiency of health diagnosis of the stone structure diversion channel in view of the material and structural characteristics of the stone structure diversion channel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of civil engineering structure health monitoring, in particular to a stone structure diversion channel health diagnosis method, system, device and computer readable storage medium. BACKGROUND

[0002] As a key water conservancy infrastructure, the stone structure diversion channel continuously bears complex and changeable environmental effects and external loads in the long-term service process, including but not limited to hydraulic scouring, foundation settlement, temperature alternation, and material performance degradation (such as weathering, dissolution) and environmental erosion (such as freeze-thaw cycle, chemical corrosion). The above factors jointly cause structural cumulative damage, significantly reducing its structural safety margin, service performance and long-term operation reliability.

[0003] Compared with modern homogeneous material structures (such as reinforced concrete), the stone structure diversion channel presents unique mechanical response behavior and complex damage evolution mode (such as block loosening, misplacement, mortar loss, local collapse) due to the anisotropy, heterogeneity and discrete block connection characteristics of the masonry material. However, the traditional structure state evaluation method mainly relies on manual inspection and empirical judgment, which has inherent defects such as strong subjectivity, low efficiency, limited coverage and difficulty in quantitatively characterizing internal damage, and cannot meet the urgent needs of modern infrastructure for efficient, accurate and automated monitoring.

[0004] Especially critical is that under the coupling action of complex natural environment and random external excitation load (such as water flow pulsation, seismic motion), the dynamic response of the stone structure diversion channel presents significant time-varying nonlinear dynamics behavior, and its vibration signal exhibits typical non-stationary characteristics, and its time-frequency energy distribution has high dynamic evolution characteristics. These time-frequency characteristics containing structure state information are important physical basis for evaluating the health state. However, the existing mainstream analysis methods are mostly limited to a single time domain (such as time history analysis) or a single frequency domain (such as Fourier spectrum analysis), and have theoretical limitations in analyzing such wideband, non-stationary, nonlinear coupled signals, making it difficult to comprehensively and accurately capture the local damage initiation, damage spatial positioning and damage degree quantification information of the structure under complex working conditions. This directly leads to major technical bottlenecks in the overall and local health state perception and accurate diagnosis of the stone structure diversion channel.

[0005] Therefore, in view of the material and structural characteristics of the stone structure diversion channel and the non-stationary characteristics of its dynamic response, it is urgent to develop an innovative health monitoring and damage diagnosis technology that integrates advanced signal processing and intelligent learning algorithms to overcome the shortcomings of existing technologies, achieve high sensitivity identification, accurate positioning and quantitative evaluation of structural damage, and provide reliable technical support for the safe operation and maintenance decision of the facility. SUMMARY

[0006] Embodiments of the present application provide a stone structure diversion channel health diagnosis method, system, device and computer readable storage medium to solve the problem that the traditional method is difficult to comprehensively and accurately capture the damage positioning and damage degree quantification information of the stone structure diversion channel under complex working conditions.

[0007] To achieve the above-mentioned purpose, in one aspect, a stone structure diversion channel health diagnosis method is provided, comprising the following steps:

[0008] Based on the finite element model, acceleration response data of health and damage working conditions are generated, and time-frequency matrix data is constructed by performing fast discrete orthogonal S transform on the acceleration response data;

[0009] According to the time-frequency matrix data, bilateral time-frequency power spectrum density data of each acceleration response data is calculated;

[0010] According to the bilateral time-frequency power spectrum density data, global transfer function matrix data of each acceleration response data is calculated through a transfer function;

[0011] The global transfer function matrix data of each acceleration response data under the health working condition is calculated, the global transfer function matrix data of each acceleration response data under the damage working condition is calculated, the damage feature is extracted by comparing the global transfer function matrix data before and after damage, and damage feature vector data is constructed;

[0012] According to the damage feature vector data, a mapping model of the acceleration response data and the damage feature vector data is established by using a Laplace neural operator, and damage positioning, local damage quantification and global damage index data are obtained by inputting the measured acceleration response data into the mapping model.

[0013] In some embodiments, after the step of establishing a mapping model of the acceleration response data and the damage feature vector data by using a Laplace neural operator according to the damage feature vector data, and obtaining damage positioning, local damage quantification and global damage index data by inputting the measured acceleration response data into the mapping model, the step further comprises:

[0014] Based on the measured acceleration response data of the stone structure diversion channel, the damage positioning, local damage quantification and global damage index data are verified, and the parameters of the mapping model are fine-tuned.

[0015] In some embodiments, in the step of generating acceleration response data of health and damage working conditions based on a finite element model, and constructing time-frequency matrix data by performing fast discrete orthogonal S transform on the acceleration response data, the step comprises:

[0016] A random damage field is generated by randomly reducing the stiffness of each element of the finite element model to simulate the damage working condition;

[0017] The real environment load is simulated by applying random white noise to the finite element model.

[0018] In some embodiments, the step of generating acceleration response data of health and damage working conditions based on the finite element model, and constructing time-frequency matrix data by performing fast discrete orthogonal S transform on the acceleration response data, the calculation formula of the time-frequency matrix data is:

[0019]

[0020] In the formula, is the time-frequency matrix data, is the frequency index, indicating the central frequency currently calculated; is the frequency bandwidth parameter, indicating the frequency bandwidth of the Gaussian window used for weighting; is the time offset, indicating the translation position of the signal in the time domain; is the discrete frequency index, indicating the set of discrete frequency components; is the normalization coefficient, used to ensure that the result of the transform satisfies the energy conservation, i.e., the total energy of the input signal and the output signal remains consistent; is the representation of the effect of time translation operation on the signal, specifically the change in phase; is the representation of the signal being modulated by the Gaussian window in the time domain, and the center frequency of the window is adjusted; is the frequency domain representation of the input signal.

[0021] In some embodiments, the step of calculating bilateral time-frequency power spectral density data of each acceleration response data according to the time-frequency matrix data, the formula for calculating the bilateral time-frequency power spectral density data is:

[0022]

[0023] In the formula, is the bilateral time-frequency power spectral density data; is the absolute value of the time-frequency matrix data; t is the time; is the absolute value of the frequency; is the normalization constant, and 0.2≤ ≤0.3.

[0024] In some embodiments, the step of calculating the global transfer function matrix data of each acceleration response data from the bilateral time-frequency power spectral density data by using the transfer function includes:

[0025] Normalizing the bilateral time-frequency power spectral density data:

[0026]

[0027] wherein, denotes normalization, max denotes maximum value, is normalized bilateral time-frequency power spectral density data, is bilateral time-frequency power spectral density data, is maximum value of bilateral time-frequency power spectral density data under healthy condition;

[0028] Under each condition, bilateral time-frequency power spectral density data of the sensor closest to the excitation point is selected as a reference, and for each non-reference sensor , a transfer function between the non-reference sensor and the reference sensor is calculated:

[0029]

[0030] wherein, is the transfer function of the th non-reference sensor, is normalized bilateral time-frequency power spectral density data of the th non-reference sensor, is normalized bilateral time-frequency power spectral density data of the reference sensor;

[0031] The transfer functions of all sensors in each condition are spliced according to time sequence and frequency to form global transfer function matrix data:

[0032]

[0033] wherein, is global transfer function matrix data, is the transfer function of the th non-reference sensor, N is the number of sensors and N≥2.

[0034] In some embodiments, in the step of establishing a mapping model of acceleration response data and damage feature vector data according to the damage feature vector data, the step includes:

[0035] inputting to a preset Laplacian neural operator, is acceleration response data under each damage condition;

[0036] by a shallow neural network to raise the dimension of the input function to obtain v ; t ;

[0037] applying a Laplacian layer and a local linear transformation , and applying a pole-residue method in the Laplacian layer to calculate a system pole-based and residual transient response residual for expressing the transient response in the laplace domain while applying pole-residual method to calculate input pole based and residual steady state response residual for expressing the steady state response in the laplace domain ,

[0038] passing , and v ( t ) through a local linear transformation after transformation, add and activate, use shallow neural network project the activated function to the feature vector , establish the mapping model of acceleration response data and damage feature vector data.

[0039] In another aspect, a health diagnosis system for a stone structure diversion channel is provided, comprising:

[0040] a time-frequency matrix module for generating acceleration response data of health and damage conditions based on a finite element model, and constructing time-frequency matrix data by performing fast discrete orthogonal S transform on the acceleration response data;

[0041] a power spectral density module for calculating bilateral time-frequency power spectral density data of each acceleration response data according to the time-frequency matrix data;

[0042] a transfer function matrix module for calculating global transfer function matrix data of each acceleration response data by transfer function according to the bilateral time-frequency power spectral density data;

[0043] a damage feature vector module for calculating global transfer function matrix data of each acceleration response data under the health condition, calculating global transfer function matrix data of each acceleration response data under the damage condition, extracting damage features by comparing the global transfer function matrix data before and after damage, and constructing damage feature vector data;

[0044] a mapping model module for establishing a mapping model of acceleration response data and damage feature vector data by using a laplace neural operator according to the damage feature vector data, and obtaining damage positioning, local damage quantification and global damage index data by inputting measured acceleration response data to the mapping model.

[0045] In still another aspect, a health diagnosis device for a stone structure water conduit is provided, which includes a memory and a processor, the memory storing at least one program, and the at least one program being executed by the processor to implement the health diagnosis method for the stone structure water conduit.

[0046] In still another aspect, a computer readable storage medium is provided, in which at least one program is stored, and the at least one program is executed by a processor to implement the health diagnosis method for the stone structure water conduit.

[0047] The above technical solution has the following technical effects:

[0048] The present application realizes real-time and accurate diagnosis of damage to the stone structure water conduit by fusing fast discrete orthogonal S transform (FDOST) time-frequency feature extraction, global transfer function damage sensitivity analysis and Laplacian neural operator (LNO) physical mechanism driven mapping: FDOST effectively analyzes the dynamic time-frequency energy distribution of non-stationary vibration signals, and combines the difference of the transfer function to extract and fuse multi-dimensional features of damage position, local strength and overall degradation; LNO explicitly models the transient / steady state response physical process through the Laplace layer pole-residual mechanism, and constructs an end-to-end functional mapping from acceleration signals to damage features; under the cooperative optimization of finite element multi-damage pre-training and on-site data fine-tuning, the accuracy of damage positioning is further improved by breaking through the limitations of environmental noise and model deviation, and a full-automatic, high-robust and long-period health monitoring closed loop is provided for water conservancy facilities. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 FIG. 1 is a flowchart of a health diagnosis method for a stone structure water conduit according to an embodiment of the present application;

[0050] Figure 2 FIG. 2 is a structural diagram of a stone structure water conduit according to an embodiment of the present application;

[0051] Figure 3 FIG. 3 is a finite element structure diagram of a stone structure water conduit according to another embodiment of the present application;

[0052] Figure 4 FIG. 4 is an algorithm diagram of a Laplacian neural operator according to still another embodiment of the present application;

[0053] Figure 5 FIG. 5 is a flowchart of an algorithm for establishing a mapping model of acceleration response data and damage feature vector data according to still another embodiment of the present application;

[0054] Figure 6 FIG. 6 is a structural diagram of a health diagnosis device for a stone structure water conduit according to still another embodiment of the present application. DETAILED DESCRIPTION

[0055] For further illustration of the embodiments, the present application provides accompanying drawings. These drawings are part of the disclosure of the present application, mainly used to illustrate the embodiments, and can be used to explain the operating principles of the embodiments in conjunction with the related description of the specification. Those of ordinary skill in the art should be able to understand other possible implementations and advantages of the present application in conjunction with these. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0056] The stone structure diversion channel undertakes the core functions of regional water supply, irrigation, and flood control. Once a local damage triggers a chain collapse (such as block instability or channel seepage leading to foundation erosion), it may cause major disasters such as flood and water pollution, threatening the safety of life and property of downstream residents. Most existing stone structure diversion channels are historical engineering heritage (such as the Dujiangyan and Lingqu branch projects), and their structural stability directly affects the survival of cultural heritage. Health diagnosis is a necessary technical means for living protection. Traditional overhaul costs can reach 70% of new construction projects. Through precise damage positioning and quantitative evaluation, preventive maintenance can be achieved to avoid comprehensive renovation and reduce the life cycle cost.

[0057] Due to the large strength dispersion of natural stone materials in stone structure diversion channels (the difference can reach 300%), it is difficult to uniformly calibrate the damage threshold; internal mortar corrosion and base hollowing are difficult to visually identify, and by the time the surface appears, it has entered the accelerated destruction period; the block instability of the stone structure diversion channel has the characteristics of brittle collapse (no significant deformation warning before the critical point); a single point damage can trigger a "domino effect" continuous collapse. In addition, the time-varying spectral characteristics of the response signal are caused by excitation sources such as water flow impact and wind load, and traditional Fourier analysis fails. Stone structure diversion channels are usually in strong environmental interference in the field environment, and the signal-to-noise ratio of field environmental noise (wind and rain, traffic) is usually less than 10 dB, which can drown out the damage characteristics. Based on the damage and interference characteristics of the stone structure diversion channel, a health diagnosis method for the stone structure diversion channel with high sensitivity, precise positioning and quantitative evaluation is urgently needed.

[0058] The technical scheme of the health diagnosis method for the stone structure diversion channel of the present application has universality and can be expanded to other scenarios, for example:

[0059] Railway / highway bridges: the bridge produces transient impact response under vehicle dynamic load, and the time-frequency energy distribution evolves dynamically (similar to the non-stationary signal of the diversion channel).

[0060] Wind turbine tower: the tower is affected by wind load and mechanical vibration, producing non-stationary vibration signals (similar to the water flow impact characteristics of the diversion channel), and the present application can detect bolt preload loss or tower inclination.

[0061] Aerospace structures (wings / airframes): the aerodynamic-structural coupled vibration in flight is strongly non-stationary, and needs to be diagnosed in real time at the millisecond level (LNO end-to-end mapping meets the delay <1s requirement).

[0062] Ancient building wood structure: mortise and tenon joint nonlinear friction vibration signal is complex, and a traditional method is difficult to analyze (LNO physical embedding modeling is required).

[0063] The application will be further described in conjunction with the drawings and specific embodiments.

[0064] Referring to Figure 1 The application provides a diagnosis method for the state of a marine photovoltaic support structure, comprising the following steps:

[0065] Step S101, generating acceleration response data of health and damage working conditions based on a finite element model, and constructing time-frequency matrix data by performing fast discrete orthogonal S transform on the acceleration response data;

[0066] Step S102, calculating bilateral time-frequency power spectrum density data of each acceleration response data according to the time-frequency matrix data;

[0067] Step S103, calculating global transfer function matrix data of each acceleration response data by a transfer function according to the bilateral time-frequency power spectrum density data;

[0068] Step S104, calculating global transfer function matrix data of each acceleration response data under the health working condition, calculating global transfer function matrix data of each acceleration response data under the damage working condition, extracting damage features by comparing the global transfer function matrix data before and after damage, and constructing damage feature vector data;

[0069] Step S105, establishing a mapping model of the acceleration response data and the damage feature vector data by using a Laplace neural operator according to the damage feature vector data, and obtaining damage positioning, local damage quantification and global damage index data by inputting measured acceleration response data into the mapping model.

[0070] In the above embodiments, the Finite Element Model (FEM) is a numerical analysis tool used to discretize complex physical systems such as mechanical structures, fluids, electromagnetic fields, etc. into a number of simple small elements such as triangles, quadrilaterals, hexahedrons, etc. By solving the mathematical equations of each element, an approximate solution for the entire system is obtained. The core idea is "divide and conquer", which transforms continuous problems into discrete problems, suitable for analyzing complex geometries and nonlinear behavior. In this application, the continuous structure of the stone structure diversion channel is divided into a finite number of simple elements, and the elements are connected through nodes, thereby transforming the continuous problem into a discrete equation set. The finite element model is suitable for the structural stability evaluation of the stone structure diversion channel in this application, and its application in the structural mechanics analysis of the stone structure diversion channel can simulate the dynamic response of the stone structure diversion channel under healthy and different damage conditions.

[0071] In some embodiments, after the steps of obtaining damage localization, local damage quantification, and global damage indicator data, further comprising: based on the measured acceleration response data of the stone structure diversion channel, verifying the damage localization, local damage quantification, and global damage indicator data, and fine-tuning the parameters of the mapping model.

[0072] As shown in Figure 2 , it is a structural schematic diagram of the stone structure diversion channel. The stone structure diversion channel is affected by complex environmental conditions during actual service, such as material aging, uneven load distribution, environmental noise, etc. These conditions may be different from the idealized assumptions of the finite element model simulation data. By fine-tuning the field measurement data, the pre-trained LNO model can better adapt to the actual working conditions, and improve the accuracy and robustness of the model prediction. Although the acceleration response data generated based on the finite element model can cover a variety of damage conditions, these data may not fully reflect the real dynamic characteristics of the field stone structure diversion channel. Through fine-tuning, the model can learn more features related to the real damage mode from the field data, further bridging the gap between simulation data and actual data.

[0073] High-precision acceleration sensors (such as piezoelectric or MEMS sensors) are selected, the sensor arrangement position is the same as the arrangement position in the finite element model, and the acceleration response signal is collected for a month to fine-tune the pre-trained stone structure diversion channel damage monitoring LNO.

[0074] Referring to Figure 3SAP2000 is selected as an example of finite element model to build the finite element model of the stone aqueduct and to divide the grid; of course, other finite element models such as ABAQUS, OpenSees, etc. can also be selected. A large number of acceleration response data under different working conditions are generated through the finite element model, including healthy working conditions and various self-defined damage working conditions. For example, the finite element model of the stone aqueduct is built using SAP2000, and the aqueduct is divided into 500 finite element units, each unit representing a part of the aqueduct, ensuring that the model resolution is high enough to capture local damage characteristics, and acceleration sensors are arranged at the midspan and the end of the span.

[0075] Each unit is assigned an initial stiffness K 0, and a continuous random damage field is generated according to the full-unit distribution field of the stone aqueduct using a Gaussian random field, as shown in formula (1):

[0076] (1)

[0077] In the formula, the mean value μ is 0.15, the variance is 64, the block effect is 5, I is the unit matrix, the scaling parameter is 3, N () represents the Gaussian distribution, Δ represents the Laplacian operator, γ is a parameter that controls the smoothness, usually γ > 0.

[0078] The range of random distribution is 0% to 90% stiffness reduction to ensure the physical meaning of the obtained damage field. Therefore, the random continuous stiffness reduction of each unit generates a random damage field, i.e. a damage working condition.

[0079] Random white noise excitation is applied to the finite element model, and acceleration response signals under healthy and various damage working conditions are collected. The fast discrete orthogonal S transform is calculated according to formula (2) to construct a time-frequency matrix database of the acceleration response signals:

[0080] (2)

[0081] In the formula:

[0082] —time-frequency matrix data;

[0083] —frequency index, representing the center frequency currently calculated;

[0084] —frequency band width parameter, representing the frequency band width of the Gaussian window used for weighting;

[0085] —time offset, representing the shift position of the signal in the time domain;

[0086] —discrete frequency index, representing a set of discrete frequency components;

[0087] —normalization coefficient, used to ensure that the result of the transformation satisfies the energy conservation, i.e. the total energy of the input signal and the output signal remains consistent;

[0088] —representing the influence of the time shift operation on the signal, specifically the change in phase;

[0089] —representing the modulation of the signal in the time domain through a Gaussian window, and adjusting the center frequency of the window;

[0090] —frequency domain representation of the input signal.

[0091] On the basis of the time-frequency matrix database constructed by FDOST, according to the concept of bilateral time-frequency power spectral density, the bilateral time-frequency power spectral density under healthy and various damage conditions is calculated, and then the global transfer function matrix is calculated according to the concept of transfer function and the damage sensitivity characteristics are extracted, and the corresponding feature vector is output.

[0092] The bilateral time-frequency power spectral density function is a representation of the energy distribution of a signal in the time-frequency plane, which satisfies the energy conservation requirement to ensure that the energy distribution of the signal in the time-frequency plane is consistent with the total energy of the original signal, avoiding information loss or distortion problems. This is particularly important for the analysis of non-stationary signals, as it allows for effective bridging between the time and frequency domains.

[0093] The calculation of bilateral time-frequency power spectral density is shown in equation (3):

[0094] (3)

[0095] In the formula:

[0096] —bilateral time-frequency power spectral density data;

[0097] —absolute value of time-frequency matrix data;

[0098] t —time;

[0099] —discrete frequency index, representing a set of discrete frequency components;

[0100] absolute value of frequency;

[0101] normalization constant, and 0.2≤ ≤0.3.

[0102] In some embodiments, in the step of calculating global transfer function matrix data of each acceleration response data by transfer function according to the bilateral time-frequency power spectral density data, the step comprises:

[0103] normalizing the bilateral time-frequency power spectral density data:

[0104] (4)

[0105] wherein, ~ represents normalization, max represents maximum value, is the normalized bilateral time-frequency power spectral density data, is the bilateral time-frequency power spectral density data, is the maximum value of the bilateral time-frequency power spectral density data under healthy working condition;

[0106] In each working condition, the bilateral time-frequency power spectral density data of the sensor closest to the excitation point is selected as the reference, and for each non-reference sensor , the transfer function between the non-reference sensor and the reference sensor is calculated:

[0107] (5)

[0108] wherein, is the transfer function of the th non-reference sensor, is the normalized bilateral time-frequency power spectral density data of the th non-reference sensor, is the normalized bilateral time-frequency power spectral density data of the reference sensor;

[0109] The transfer functions of all sensors in each working condition are spliced according to time sequence and frequency to form global transfer function matrix data:

[0110] (6)

[0111] wherein, is the global transfer function matrix data, is the transfer function of the th non-reference sensor, N is the number of sensors and N≥2.

[0112] In some embodiments, the damage feature extraction is performed in combination with the global transfer function matrix data before and after the damage, and the damage feature vector data is constructed, including:

[0113] The difference of the global transfer function matrix before and after the damage is calculated:

[0114] (7)

[0115] In the formula, is the difference of the global transfer function matrix before and after the damage, represents the global transfer function matrix data under the damage condition, represents the global transfer function matrix data under the healthy condition;

[0116] Select the high frequency range f > 50 Hz, calculate the change of the total energy in the high frequency range before and after the damage:

[0117] (8)

[0118] In the formula, is the change value of the total energy in the high frequency range before and after the damage, represents the mean value of the bilateral time-frequency power spectrum density of all sensors in the high frequency range under the damage condition, represents the mean value of the bilateral time-frequency power spectrum density of all sensors in the high frequency range under the healthy condition, is the frequency, dt is the integral of the small change amount with respect to time t ;

[0119] Extract the energy below 5 Hz from the mean bilateral time-frequency power spectrum density:

[0120] (9)

[0121] In the formula, is the total energy in the low frequency range , is the mean value of the bilateral time-frequency power spectrum density of all sensors in the low frequency range , is the frequency;

[0122] Linear regression is performed on to fit the energy attenuation curve:

[0123] (10)

[0124] In the formula, is the low frequency range total energy of the low frequency range, a decay rate, b constant, o residual, t time;

[0125] solved using the least square method a

[0126] (11)

[0127] wherein, a decay rate, t time, total energy of the low frequency range total energy of the low frequency range, m denotes the total number of data points;

[0128] The features are spliced to construct the damage feature vector data:

[0129] (12)

[0130] wherein, F is the damage feature vector data, the damage feature vector data has damage localization data, the greater the difference value, the closer to the damage position, the high frequency energy change local damage quantitative data, the energy decay rate a is the global damage index data, used to measure the overall performance degradation of the structure.

[0131] Laplace neural operator (LNO) is an innovative deep learning method proposed in the field of operator learning in recent years, referring to Figure 4 is a schematic diagram of the Laplace neural operator. Unlike traditional deep learning methods based on discrete grids, LNO establishes an end-to-end mapping from continuous function space to continuous function space, showing stronger generalization ability. LNO expands the Fourier transform based on a set of complex exponential functions, raises the input function to a higher dimension through a shallow neural network, making it possible to accurately track patterns in the spatial domain and effectively learn function mapping. LNO introduces the mathematical properties of Laplace transform, showing significant advantages in solving highly nonlinear partial differential equations (PDEs) with complex geometric domains and non-standard boundary conditions.

[0132] Referring to Figure 5 In some embodiments, according to the damage feature vector data, the step of establishing a mapping model of the acceleration response data and the damage feature vector data by using the Laplace neural operator includes:

[0133] Step S1051, inputting to the preset Laplace neural operator, ​Acceleration response data for each damage condition;

[0134] Step S1052, through a shallow neural network Increase the dimensionality of the input function to obtain v ( t );

[0135] Step S1053: Apply the Laplace layer and local linear transformation. The pole-residual method is applied to the Laplace layer to calculate the system poles. and residual transient response residual Used to express transient response in the Laplace domain Simultaneously, the pole-residual method is applied to calculate the input poles. and residual steady-state response residual Used to express steady-state response in the Laplace domain ;

[0136] Step S1054, will , as well as v ( t After local linear transformation After transformation, the elements are added together and activated using a shallow neural network. The activated function is projected onto the feature vector. A mapping model between acceleration response data and damage feature vector data was established.

[0137] By extracting the bilateral time-frequency power spectral density from the time-frequency matrix constructed using FDOST, and combining it with global transferability function difference analysis, a multi-dimensional feature vector is constructed by fusing high-frequency energy changes (locating local damage) and low-frequency attenuation rate (assessing overall degradation), overcoming the limitations of single-domain analysis. LNO utilizes Laplace layer pole-residual calculations to explicitly model transient / steady-state responses, establishing an end-to-end functional mapping from acceleration signal to damage features, significantly improving damage location accuracy and quantitative assessment reliability. Based on the synergistic innovation of the Laplace neural operator (LNO) and the Fast Discrete Orthogonal S-Transform (FDOST), this invention achieves real-time and accurate diagnosis of damage in stone structure irrigation canals. Through finite element multi-damage condition pre-training and fine-tuning with field measured data, environmental noise interference and model bias are effectively overcome, maintaining high accuracy even at low signal-to-noise ratios. The diagnostic process is real-time, providing a fully automated, robust, and long-term health monitoring solution for stone structure irrigation canals.

[0138] On the other hand, the present invention also provides a health diagnosis system for a stone structure irrigation canal, comprising:

[0139] A time-frequency matrix module is configured to generate acceleration response data of healthy and damage conditions based on a finite element model, and to construct time-frequency matrix data by performing a fast discrete orthogonal S transform on the acceleration response data.

[0140] A power spectrum density module is configured to calculate bilateral time-frequency power spectrum density data of each acceleration response data based on the time-frequency matrix data.

[0141] A transfer function matrix module is configured to calculate global transfer function matrix data of each acceleration response data by a transfer function based on the bilateral time-frequency power spectrum density data.

[0142] A damage feature vector module is configured to calculate global transfer function matrix data of each acceleration response data in a healthy condition, to calculate global transfer function matrix data of each acceleration response data in a damage condition, to extract damage features by comparing the global transfer function matrix data before and after damage, and to construct damage feature vector data.

[0143] A mapping model module is configured to establish a mapping model of acceleration response data and damage feature vector data by using a Laplace neural operator based on the damage feature vector data, to input measured acceleration response data into the mapping model, and to obtain damage positioning, local damage quantification, and global damage index data.

[0144] In some embodiments, the system further includes a fine-tuning verification module configured to fine-tune model parameters based on experimental data and / or field data, and to verify and correct the corresponding relationship between the acceleration response data and the damage feature vector data in the mapping model.

[0145] The application further provides a health diagnosis device for a stone structure diversion channel, as shown in the drawings. Figure 6 The device includes a processor 601, a memory 602, a bus 603, and a computer program stored in the memory 602 and executable on the processor 601, the processor 601 includes one or more processing cores, the memory 602 is connected to the processor 601 through the bus 603, the memory 602 is used to store program instructions, and the processor executes the computer program to implement the steps in the above method embodiments.

[0146] Further, as an executable solution, the stone structure water diversion channel health diagnosis device can be a computer unit, which can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The computer unit can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above-mentioned constituent structure of the computer unit is only an example of the computer unit, and does not constitute a limitation on the computer unit, and can include more or fewer components than the above, or combine certain components, or different components. For example, the computer unit can also include an input / output device, a network access device, a bus, and the like, and the embodiments of the present application do not limit this.

[0147] Further, as an executable solution, the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like, and the processor is the control center of the computer unit, which connects various parts of the computer unit through various interfaces and lines.

[0148] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the computer unit by running or executing the computer program and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; the data storage area can store data created according to the use of the mobile phone, and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0149] In some embodiments, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements the steps of the above method according to the embodiments of the present application.

[0150] The modules / units integrated with the computer unit, if realized in the form of software functional units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, the present application can also be implemented by a computer program to instruct related hardware to complete all or part of the above-mentioned method embodiments. The computer program can be stored in a computer readable storage medium, and the computer program, when executed by a processor, can implement the steps of the above-mentioned method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM) and software distribution medium, etc. It should be noted that the computer readable medium can include appropriate contents according to the requirements of legislation and patent practice in the jurisdiction.

[0151] In some embodiments, the present application also provides a computer program product, which includes a computer program, and the computer program, when executed by a processor, implements the steps of the above method.

[0152] Although the present application is specifically shown and described in connection with the preferred embodiments, those skilled in the art should understand that various changes in form and details can be made to the present application without departing from the spirit and scope of the present application as defined by the appended claims.

Claims

1. A method for health diagnosis of a stone-structured irrigation canal, characterized in that, Includes the following steps: Acceleration response data under healthy and damaged conditions are generated based on the finite element model, and time-frequency matrix data is constructed by fast discrete orthogonal S-transformation of the acceleration response data. Calculate the two-sided time-frequency power spectral density data of each acceleration response data based on the time-frequency matrix data; Based on the bilateral time-frequency power spectral density data, the global transfer function matrix data of each acceleration response data is calculated using the transfer function. The global transitivity function matrix data of each acceleration response data under healthy working conditions is calculated, and the global transitivity function matrix data of each acceleration response data under damaged working conditions is calculated. By comparing the global transitivity function matrix data before and after the damage, damage features are extracted and damage feature vector data is constructed. Based on the damage feature vector data, a mapping model between acceleration response data and damage feature vector data is established using the Laplacian neural operator. By inputting measured acceleration response data into the mapping model, damage localization, local damage quantification, and global damage index data are obtained. The calculation formula for constructing the time-frequency matrix data by performing a fast discrete orthogonal S-transform on the acceleration response data is as follows: In the formula, This is time-frequency matrix data. This is the frequency index, representing the center frequency currently being calculated; This is a bandwidth parameter, representing the bandwidth used for the weighted Gaussian window; Time offset represents the position of the signal shift in the time domain; This is a discrete frequency index, representing a set of discrete frequency components; These are normalization coefficients used to ensure that the transformation result satisfies energy conservation, that is, the total energy of the input signal and the output signal remains the same. To represent the effect of time shifting on the signal, it is specifically manifested as a change in phase; This represents modulating the signal in the time domain using a Gaussian window and adjusting the center frequency of the window; This is the frequency domain representation of the input signal.

2. The health diagnosis method for a stone-structured irrigation canal according to claim 1, characterized in that, After the steps of establishing a mapping model between acceleration response data and damage feature vector data using the Laplacian neural operator based on damage feature vector data, and obtaining damage localization, local damage quantification, and global damage index data by inputting measured acceleration response data into the mapping model, the method further includes: Based on the measured acceleration response data of the stone structure water diversion channel, the damage location, local damage quantification, and global damage index data were verified, and the parameters of the mapping model were fine-tuned.

3. The health diagnosis method for a stone-structured irrigation canal according to claim 1, characterized in that, The steps of generating acceleration response data for healthy and damaged conditions based on the finite element model, and constructing time-frequency matrix data by performing a fast discrete orthogonal S-transform on the acceleration response data, include: By randomly reducing the stiffness of each element in the finite element model, a random damage field is generated to simulate the damage condition. Real environmental loads are simulated by applying random white noise to the finite element model.

4. The health diagnosis method for a stone-structured irrigation canal according to claim 1, characterized in that, In the step of calculating the two-sided time-frequency power spectral density data of each acceleration response data based on the time-frequency matrix data, the formula for calculating the two-sided time-frequency power spectral density data is as follows: In the formula, This is two-sided time-frequency power spectral density data; The absolute value of the time-frequency matrix data; t For time; The absolute value of the frequency; Let be a normalization constant, and 0.2 ≤ ≤0.

3.

5. The health diagnosis method for a stone-structured irrigation canal according to claim 1, characterized in that, The step of calculating the global transfer function matrix data of each acceleration response data based on the bilateral time-frequency power spectral density data includes: The two-sided time-frequency power spectral density data are normalized: In the formula, ~ represents normalization, and max represents the maximum value. The normalized two-sided time-frequency power spectral density data, This is two-sided time-frequency power spectral density data. The maximum value of the bilateral time-frequency power spectral density data under healthy operating conditions; Under each operating condition, the bilateral time-frequency power spectral density data of the sensor closest to the excitation point is selected as the benchmark. For each non-benchmark sensor... Calculate its transfer function with respect to the reference sensor: In the formula, For the first The transfer function of a non-reference sensor, For the first Normalized two-sided time-frequency power spectral density data of a non-reference sensor It is the normalized two-sided time-frequency power spectral density data of the reference sensor; The transfer functions of all sensors under each operating condition are concatenated according to time sequence and frequency to form a global transfer function matrix data: In the formula, This is global transitivity function matrix data. For the first The transfer function of a non-reference sensor, N is the number of sensors and N≥2.

6. The health diagnosis method for a stone-structured irrigation canal according to claim 1, characterized in that, The steps of calculating the global transferability function matrix data of each acceleration response data under healthy operating conditions, calculating the global transferability function matrix data of each acceleration response data under damaged operating conditions, and extracting damage features by comparing the global transferability function matrix data before and after damage to construct damage feature vector data include: Calculate the difference in the global transitivity function matrix before and after the damage: In the formula, The difference between the global transitivity function matrix before and after the damage. This represents the global transitivity function matrix data under damage conditions. Represents the global transitivity function matrix data under healthy operating conditions; Select high frequency range f >50Hz, calculate the change in total energy in the high-frequency range before and after the damage: In the formula, For the high frequency range before and after damage The change in total energy, This indicates that all sensors under damage conditions operate in the high-frequency range. The mean of the two-sided time-frequency power spectral density, This indicates that all sensors are operating in the high-frequency range under healthy conditions. The mean of the two-sided time-frequency power spectral density, For frequency, dt For time t Integrate the small changes; Extracting energy below 5Hz from the mean two-sided time-frequency power spectral density: In the formula, Low frequency range Total energy, For all sensors in the low frequency range The mean of the two-sided time-frequency power spectral density, For frequency; right Perform linear regression to fit the energy decay curve: In the formula, Low frequency range Total energy, a The attenuation rate, b It is a constant. o For residuals, t For time; Solve using the least squares method a : In the formula, a The attenuation rate, t For time, Low frequency range Total energy, m This represents the total number of data points; The features are concatenated to construct the damage feature vector data: Where F represents the damage feature vector data, and the damage feature vector data... For damage localization data, the larger the difference value, the closer to the damage location; high-frequency energy changes. The data provided are quantitative data on local damage, while the energy decay rate (σ) is a global damage index used to measure the overall performance degradation of the structure.

7. The health diagnosis method for a stone-structured irrigation canal according to claim 1, characterized in that, The step of establishing a mapping model between acceleration response data and damage feature vector data using the Laplacian neural operator based on damage feature vector data includes: Will The input is given to the preset Laplace neural operator. Acceleration response data for each damage condition; Through a shallow neural network Increase the dimensionality of the input function to obtain v ( t ); Application of Laplace layer and local linear transformation W The pole-residual method is applied to the Laplace layer to calculate the system poles. and residual transient response residual Used to express transient response in the Laplace domain Simultaneously, the pole-residual method is applied to calculate the input poles. and residual steady-state response residual Used to express steady-state response in the Laplace domain ; Will , as well as v ( t After local linear transformation The functions are then summed and activated, and a shallow neural network is used to project the activated function onto the feature vector F, thus establishing a mapping model between acceleration response data and damage feature vector data.

8. A health diagnosis system for a stone-structured aqueduct, configured to perform the health diagnosis method for a stone-structured aqueduct as described in any one of claims 1 to 7, characterized in that, include: The time-frequency matrix module is used to generate acceleration response data under healthy and damaged conditions based on the finite element model, and to construct time-frequency matrix data by performing a fast discrete orthogonal S-transform on the acceleration response data. The power spectral density module is used to calculate the two-sided time-frequency power spectral density data of each acceleration response data based on the time-frequency matrix data. The transferability function matrix module is used to calculate the global transferability function matrix data of each acceleration response data based on the bilateral time-frequency power spectral density data through transferability functions. The damage feature vector module is used to calculate the global transitivity function matrix data of each acceleration response data under healthy conditions, and to calculate the global transitivity function matrix data of each acceleration response data under damaged conditions. By comparing the global transitivity function matrix data before and after damage, damage features are extracted and damage feature vector data is constructed. The mapping model module is used to establish a mapping model between acceleration response data and damage feature vector data based on damage feature vector data using the Laplacian neural operator. By inputting measured acceleration response data into the mapping model, damage localization, local damage quantification, and global damage index data can be obtained.

9. A health diagnostic device for a stone-structured irrigation canal, characterized in that, It includes a memory and a processor, the memory storing at least one program, the at least one program being executed by the processor to implement the health diagnosis method for a stone structure irrigation canal as described in any one of claims 1 to 7.

Images