Membrane-based method for the production of a polymer

Abstract

This application provides a MEMS array-based system and method for monitoring the performance of water-retaining sand in extreme environments, relating to the field of material performance monitoring technology. The method includes: deploying a MEMS array within the water-retaining sand material to be tested; applying an excitation signal to a first location and acquiring a response signal at a second location, while collecting physical and chemical field characteristic parameters at both locations; retrieving the water transport characteristics in the material based on the response signal; establishing the spatial distribution of the material's physical and chemical fields using a three-dimensional reconstruction method; combining residual analysis with mechanistic constraints to jointly evaluate the transport characteristics and physical field distribution, identifying and classifying abnormal regions, and finally outputting the monitoring signal. This invention addresses the water-retention, water-impermeability, and structural integrity of Martian water-retaining sand under simulated extreme environments, achieving in-situ, distributed monitoring with multi-physics coupling. It can effectively identify anomalies such as membrane damage and water leakage, providing reliable technical support for the verification and application of new materials.

Description

Technical Field

[0001] This application relates to the field of material performance monitoring technology, and more specifically, to a water-retaining sand extreme environment performance monitoring system and method based on MEMS array. Background Technology

[0002] Currently, with the deepening of human exploration of extraterrestrial bodies such as Mars and the Moon, materials capable of adapting to extreme temperature differences, low air pressure, strong radiation, and highly chemically active soil environments have become a research hotspot. Among them, new materials such as "Mars water-retaining sand," which aim to achieve extraterrestrial water and soil management and ecological construction, have a significant impact on the stable operation of interstellar survival systems due to their performance reliability.

[0003] Traditional methods for monitoring material properties often rely on offline laboratory analysis or single-point physical sensors. For example, for water-retaining and water-resistant materials, weighing methods, permeation experiments, or single temperature and moisture sensors are commonly used for measurement. These methods cannot achieve online, in-situ, long-term distributed monitoring when dealing with materials like Martian water-retaining sand, which have micro-nano membrane structures, operate in extreme environments, and require consideration of multi-physics coupling effects. Furthermore, single measurements sever the inherent coupling relationships between multiple physical fields such as temperature, moisture, stress, and conductivity, making it difficult to reveal the fundamental mechanisms of material failure.

[0004] In recent years, with the development of microelectromechanical systems (MEMS) technology, multi-parameter, miniaturized sensor arrays have begun to be applied to the field of materials health monitoring. Some studies utilize distributed optical fiber or resistivity tomography to monitor soil moisture changes, while others attempt to invert transport characteristics in porous media using acoustic waves or thermal pulses. These methods have improved the information dimension of monitoring to some extent. However, existing technologies still have key shortcomings: First, most of them are designed for normal atmospheric pressure and temperature environments on Earth, lacking sensor calibration and noise suppression mechanisms for extreme environments such as the Martian surface with temperature variations from -90°C to 250°C and low atmospheric pressure. Second, existing methods fail to deeply integrate the inverted moisture transport characteristics, such as paths and diffusion coefficients, with the reconstruction process of three-dimensional physical field temperature and stress, resulting in vague and delayed localization of anomalous areas, such as localized coating damage and leaching channels for harmful substances. Third, they lack the ability to intelligently classify anomaly types, failing to distinguish between different failure modes such as physical damage, chemical corrosion, or coupling instability, thus making it difficult to provide clear guidance for material optimization and early warning.

[0005] Therefore, there is an urgent need for a water-retention and sediment-holding performance monitoring method that can adapt to extreme environments, integrate multi-physical field information, and achieve accurate anomaly location and classification to solve the above problems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, embodiments of this application provide a system and method for monitoring the performance of water-retaining sand in extreme environments based on MEMS arrays.

[0007] In a first aspect, embodiments of this application provide a method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays, including:

[0008] A MEMS array is placed in the material under test. An excitation signal is applied to a first position in the MEMS array, and a response signal is obtained through a second position. Characteristic parameters at the first and second positions are obtained, including physical field parameters and chemical field parameters.

[0009] Based on the response signal, the transmission characteristics of the preset medium in the material under test are obtained by inversion.

[0010] Based on the aforementioned feature parameters, the physical field spatial distribution and chemical field spatial distribution of the material under test are formed using a preset three-dimensional reconstruction method.

[0011] Based on mechanism-constrained residual analysis, the transmission characteristics and the spatial distribution of the physical field are jointly evaluated to identify abnormal regions.

[0012] The abnormal regions are classified into anomalies to obtain anomaly classification results, and monitoring signals are output.

[0013] Further, the step of applying an excitation signal to a first position in the MEMS array and obtaining a response signal through a second position includes:

[0014] The first position and the second position are spatially different, and the second position is located in a region with the first position as the center and within a preset radius.

[0015] The arrival time and signal strength of the response signal at each second position are extracted, and the attenuation coefficient of the response signal as the spatial distance changes is calculated based on the spatial distance between the first position and the second position.

[0016] Further, obtaining the feature parameters at the first position and the second position includes:

[0017] The physical field parameters include at least temperature parameters, moisture parameters, and stress parameters, and the chemical field parameters include at least electrical conductivity parameters.

[0018] The temperature parameter is corrected using a preset temperature drift compensation method; the moisture parameter is corrected using a preset air pressure correction method; the stress parameter is corrected using a preset thermal stress separation method; and the conductivity parameter is corrected using a preset radiation noise suppression method, thus obtaining the preprocessed characteristic parameters.

[0019] Furthermore, the step of inverting the transmission characteristics of the preset medium in the material under test based on the response signal includes:

[0020] Based on the arrival time and signal strength of the response signal at the second location and the attenuation coefficient, the transmission characteristics of the preset medium in the material under test are reconstructed by a preset inversion algorithm;

[0021] The preset medium includes water, and the transmission characteristics include the transmission path, transmission rate, and diffusion coefficient of the preset medium in the material under test.

[0022] Furthermore, the step of forming the physical field spatial distribution and chemical field spatial distribution of the material under test through a preset three-dimensional reconstruction method includes:

[0023] Based on the preprocessed feature parameters at the first and second positions, a discrete point set of each feature parameter in the spatial dimension is established.

[0024] Using a preset spatial interpolation algorithm, the transmission path is used as a spatial topological constraint to spatially reconstruct the moisture parameters in the discrete point set, thereby obtaining a three-dimensional moisture field distribution.

[0025] Based on the temperature parameters, stress parameters, and conductivity parameters in the discrete point set, spatial interpolation reconstruction without transmission path constraints is performed to obtain the three-dimensional temperature field distribution, the three-dimensional stress field distribution, and the three-dimensional conductivity field distribution. The three-dimensional conductivity field distribution is then used as the spatial distribution of the chemical field.

[0026] Furthermore, the residual analysis based on mechanistic constraints, which jointly evaluates the transmission characteristics and the spatial distribution of the physical field, includes:

[0027] A physical information neural network is constructed, and the transmission path is used as a boundary condition to train the physical information neural network to fit the three-dimensional moisture field distribution.

[0028] Calculate the physical residual of the physical information neural network at each spatial location to form a residual field distribution;

[0029] Spatial regions in the residual field distribution where the local residual energy exceeds the first preset residual threshold are marked as the first abnormal region.

[0030] Furthermore, the anomaly classification of the abnormal region includes:

[0031] Within the first abnormal region, a physical-guided generative adversarial network is constructed, which includes a generator and a discriminator.

[0032] The generator's inputs include random noise and the three-dimensional temperature field distribution, three-dimensional moisture field distribution, and three-dimensional stress field distribution, and its output is the reconstructed physical field spatial distribution;

[0033] The loss function of the generator incorporates the moisture transport equation and the thermoelastic constitutive relation as physical constraint residuals.

[0034] The input to the discriminator is the reconstructed physical field spatial distribution output by the generator and the measured physical field spatial distribution, and the output is the anomaly classification result.

[0035] Furthermore, the output of the discriminator, the anomaly classification result, includes:

[0036] The discriminator receives the three-dimensional conductivity field distribution as auxiliary input and outputs the anomaly classification result.

[0037] The reconstruction residual between the reconstructed physical field spatial distribution output by the generator and the measured physical field spatial distribution is weighted and fused with the probability corresponding to the film damage state in the anomaly classification result to obtain the anomaly score for each monitoring location.

[0038] If the anomaly score exceeds the preset anomaly threshold, the corresponding monitoring location is marked as a film damage area, which is then used as the second anomaly area, and the damage location information is output.

[0039] Furthermore, the output anomaly classification results include:

[0040] The anomaly classification results include: abnormal conductivity state, abnormal multi-physics field coupling state, and film damage state.

[0041] The abnormal conductivity state is determined based on the residual between the measured conductivity value and the reconstructed conductivity value exceeding a preset conductivity threshold and the physical field residual being lower than a preset physical field threshold.

[0042] The physical field residuals include physical constraint residuals and reconstruction residuals.

[0043] Furthermore, the output monitoring signal includes:

[0044] Anomaly information is extracted from the second anomaly region, including: the average anomaly score within the second anomaly region, the area of ​​the second anomaly region, and the anomaly classification result;

[0045] The abnormal information is input into a preset signal mapping function, and the corresponding monitoring signal is output.

[0046] The monitoring signals include at least one of the following: damage alarm signal, coupling failure early warning signal, corrosion diffusion warning signal, and general abnormality indication signal.

[0047] Secondly, embodiments of this application also provide a water-retention and sand-holding extreme environment performance monitoring system based on MEMS array, including: a data acquisition module, an inversion module, a reconstruction module, an identification module, and an output module;

[0048] Acquisition module: used to deploy a MEMS array in the material under test, apply an excitation signal to a first position in the MEMS array, acquire a response signal through a second position, and acquire characteristic parameters at the first and second positions, the characteristic parameters including physical field parameters and chemical field parameters;

[0049] Inversion module: used to invert the transmission characteristics of the preset medium in the material under test based on the response signal;

[0050] Reconstruction module: used to form the physical field spatial distribution and chemical field spatial distribution of the material under test based on the feature parameters and through a preset three-dimensional reconstruction method;

[0051] Identification module: used for residual analysis based on mechanism constraints, to jointly evaluate the transmission characteristics and the spatial distribution of the physical field, and to identify abnormal regions;

[0052] Output module: Classifies the abnormal area, obtains the abnormal classification result, and outputs the monitoring signal.

[0053] Compared with the prior art, the effective effects achieved by the present invention are as follows:

[0054] This application combines physical and chemical field parameters acquired simultaneously by a MEMS array. Through spatial interpolation and transmission path constraints, it achieves joint reconstruction of three-dimensional moisture, temperature, stress, and conductivity fields, solving the problems of isolated parameters and low spatial resolution in traditional methods. It can intuitively and comprehensively present the "water-thermal-mechanical-chemical" coupling state inside water-retaining sand, providing a data foundation for failure analysis. By embedding the moisture transport equation and thermoelastic constitutive relation as physical constraints into the network, the residual field is calculated to accurately locate abnormal areas of membrane damage or leakage. The discriminator is used in conjunction with conductivity distribution to classify anomalies, improving the robustness and accuracy of anomaly detection. Through preset correction methods such as temperature drift compensation, air pressure correction, thermal stress separation, and radiation noise suppression, the reliability of data from MEMS sensors in simulated deep space extreme environments is ensured. The final output of multi-level monitoring signals, such as damage alarm, coupling failure warning, and corrosion diffusion warning, realizes a closed loop from "data acquisition" to "intelligent diagnosis," providing a real-time, accurate, and intelligent performance monitoring tool for the design iteration, ground verification, and future in-situ interplanetary applications of new materials such as "Mars water-retaining sand." Attached Figure Description

[0055] Figure 1A flowchart of a method for monitoring the performance of water-retaining sand in extreme environments based on a MEMS array, provided in an embodiment of this application;

[0056] Figure 2 A flowchart of the three-dimensional multiphysics and chemical field reconstruction method provided in the embodiments of this application;

[0057] Figure 3 A flowchart illustrating anomaly identification and classification based on mechanistic constraints provided in this application embodiment;

[0058] Figure 4 A schematic diagram of a MEMS array-based water-retention sand extreme environment performance monitoring system provided in this application embodiment. Detailed Implementation

[0059] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0060] See Figure 1 This is a flowchart of the water-retention and sand-holding extreme environment performance monitoring method based on MEMS array provided in this embodiment. The method includes steps S101 to S105, wherein:

[0061] S101: A MEMS array is placed in the material under test. An excitation signal is applied through a first position in the MEMS array, and a response signal is obtained through a second position. Characteristic parameters at the first and second positions are obtained. The characteristic parameters include physical field parameters and chemical field parameters.

[0062] S102: Based on the response signal, the transmission characteristics of the preset medium in the material under test are obtained by inversion;

[0063] S103: Based on the feature parameters, the physical field spatial distribution and chemical field spatial distribution of the material under test are formed by a preset three-dimensional reconstruction method;

[0064] S104: Based on mechanism-constrained residual analysis, the transmission characteristics and the spatial distribution of the physical field are jointly evaluated to identify abnormal regions;

[0065] S105: Classify the abnormal area to obtain the abnormal classification result and output the monitoring signal.

[0066] It is well known that the atmospheric pressure on the surface of Mars is about 1% of that on Earth, the temperature difference between day and night exceeds 100°C, and the lowest temperature can reach -90°C.

[0067] The particle size of Martian regolith sand is mainly distributed between 60μm and 200μm, and contains fine dust of 2μm to 4μm. The sand also contains active substances such as perchlorate and hydrogen peroxide, with a total mass fraction of not less than 0.5%.

[0068] The water-retaining sand is composed of simulated Martian weathered layer sand and surface micro-nano coating. The surface micro-nano coating is coated onto the surface of the sand particles through a turbulent process under high temperature and high pressure conditions, forming a hydrophobic layer with covalent bonds.

[0069] In practical implementation, the water-retaining and water-impermeable effects of the water-retaining sand depend on the integrity and hydrophobic properties of the surface micro-nano coating. For example, when the surface micro-nano coating is intact, liquid water rolls in a spherical shape on the sand surface without penetrating, and the moisture is confined to a designated area.

[0070] Under extreme temperature cycling, radiation, and mechanical stress, the surface micro-nano coating may experience local damage or peeling. Once the surface micro-nano coating is damaged, moisture will seep into the sand grains along the damage point and come into contact with the perchlorate in the sand grains to form a solution containing perchlorate ions. The solution migrates outward, causing moisture loss on the one hand, introducing chemical pollution on the other hand, and changing the local conductivity distribution and three-dimensional moisture field distribution.

[0071] In practice, it is necessary to monitor the integrity of the water-retaining sand mulch and changes in moisture and chemical fields.

[0072] Regarding step S101:

[0073] As is generally known, MEMS refers to microelectromechanical systems, which are miniature sensors or actuators, typically ranging in size from micrometers to millimeters.

[0074] The MEMS array consists of multiple MEMS sensor nodes, each integrating a temperature-sensitive unit, a moisture-sensitive unit, a stress-strain-sensitive unit, and a conductivity-sensitive unit. The sensor nodes are arranged in a three-dimensional mesh topology within the water-retaining sand material to be tested.

[0075] In specific implementation, the MEMS array is composed of MEMS sensor nodes arranged in a three-dimensional orthogonal grid. The node spacing is set based on the following: the node spacing is not greater than 1 / 2 of the length of the water diffusion characteristic in the water-retaining sand sample.

[0076] As an optional implementation method, the moisture diffusion characteristic length is determined by selecting a defect-free water-retaining sand sample with the same material and process as the sample to be tested, applying deionized water to the surface of the sample, and measuring the radius of horizontal diffusion of moisture inside the sample. If the diffusion radius does not change by more than 5% within 3 consecutive hours, the radius is recorded as the moisture diffusion characteristic length.

[0077] For example, the water-retaining sand sample to be tested is a cuboid with a length of 30 cm, a width of 20 cm, and a height of 10 cm. The packaged side length of the MEMS sensor node is 5 mm. The node spacing is set to 5 cm.

[0078] Along the length direction, starting at 2.5 cm, a node is placed every 5 cm, for a total of 6 nodes. Similarly, along the width direction, a total of 4 nodes are placed. Along the height direction, a total of 2 nodes are placed. The total number of nodes in the MEMS array is 6 x 4 x 2, which is 48.

[0079] By applying an excitation signal to a first position in the MEMS array and obtaining a response signal through a second position, in a specific implementation, the first position and the second position are spatially different, and the second position is located in a region with the first position as the center and within a preset radius.

[0080] As an optional implementation, in the MEMS array, a node is selected sequentially as the first position, and the node is configured with an excitation generating unit, which is a thin-film heating resistor or a piezoelectric ceramic sheet.

[0081] When a thin-film heating resistor is used, the applied excitation signal is a step thermal pulse with adjustable duration; when a piezoelectric ceramic sheet is used, the applied excitation signal is a single-frequency or multi-frequency sinusoidal vibration signal.

[0082] The preset radius is set based on the following: in a defect-free water-retaining sand sample, the response signal intensity of each node is measured successively from the first position outwards. The distance at which the signal intensity first falls below three times the root mean square value of the background noise before excitation is multiplied by 0.8 and used as the preset radius.

[0083] The arrival time and signal strength of the response signal at each second position are extracted, and the attenuation coefficient of the response signal as the spatial distance changes is calculated based on the spatial distance between the first position and the second position.

[0084] In practice, when each first position is selected, a current pulse with a duration of milliseconds is applied to generate thermal pulse excitation, and at the same time, all second position nodes located within a preset radius of the first position begin data acquisition.

[0085] Each second location node acquires the temperature response signal within a time window before and after the excitation.

[0086] After the data acquisition is completed, the root mean square value of the temperature response signal within the time window before excitation is calculated for the temperature response curve of each second location node, and multiplied by 3 is used as the trigger threshold.

[0087] As an optional implementation, the search proceeds point by point from the excitation start time. When the temperature response signal first exceeds the trigger threshold, the difference between that time and the excitation start time is recorded as the arrival time.

[0088] The signal strength is obtained by subtracting the baseline temperature before excitation from the peak temperature of the temperature response curve after excitation.

[0089] The spatial distance between the first position and each second position is directly calculated using the node coordinates. The natural logarithm of each spatial distance and the signal strength is used to form a data pair. The least squares method is used to fit a straight line, and the absolute value of the slope of this line is recorded as the attenuation coefficient.

[0090] In practice, if the number of second positions involved in the fitting is less than 30, the preset radius will be automatically expanded, such as by 0.2 length units each time, until the number of second positions involved in the fitting is no less than 30.

[0091] After completing the excitation-response acquisition and attenuation coefficient calculation at the current first position, switch to the next first position and repeat the above process until all nodes have been traversed.

[0092] The characteristic parameters at the first and second positions are obtained, including physical field parameters and chemical field parameters. The physical field parameters include at least temperature, moisture, and stress parameters, and the chemical field parameters include at least electrical conductivity.

[0093] The temperature parameters are corrected by using a preset temperature drift compensation method to obtain the corrected temperature parameters.

[0094] As an optional implementation, temperature drift compensation is performed using a reference channel-based differential measurement and polynomial fitting method.

[0095] For example, at least one reference channel is provided in the MEMS array, which is placed in a constant temperature environment with the temperature kept constant at a preset value.

[0096] The differential signal is obtained by subtracting the output of all measurement channels from the output of the reference channel.

[0097] The relationship between the differential signal and the actual temperature difference is established by fitting a high-order polynomial. The polynomial coefficients are obtained by placing the entire MEMS array in an adjustable temperature chamber, applying a temperature point, recording the actual temperature of each temperature point with an external standard thermometer, collecting the differential output between each measurement channel and the reference channel, and fitting the polynomial coefficients using the least squares method.

[0098] In a specific embodiment, the constant temperature of the reference channel is set to 0 degrees Celsius, and the polynomial order is third. Twenty temperature points are applied during fitting, covering the lower to upper limits of the working temperature range. During actual measurement, the difference output values ​​between the measurement channel and the reference channel are substituted into the polynomial to calculate the corrected temperature parameters.

[0099] The moisture parameters are corrected using a preset air pressure correction method to obtain the corrected moisture parameters.

[0100] As an optional implementation method, a two-parameter compensation method based on a water activity model is used for pressure correction. The water activity model refers to a thermodynamic model that describes the relationship between the chemical potential of water in porous media and temperature and pressure. Its core parameter, water activity, is defined as the ratio of the current water vapor partial pressure to the saturated vapor pressure at the same temperature.

[0101] For example, the raw output of the moisture parameter is a physical quantity related to the dielectric constant.

[0102] In practice, the saturated vapor pressure of water decreases under low-pressure conditions, which leads to a change in the effective dielectric constant of liquid water. The Clausius-Clapeyron equation in thermodynamics is used to describe the relationship between saturated vapor pressure and temperature. Combined with the dielectric constant mixing model, a functional relationship between pressure, temperature and water output is established.

[0103] The parameters to be determined in the functional relationship are calibrated through preliminary experiments: under a combination of air pressure and temperature, the moisture output of a standard sample with known water content is measured, and the model parameters are optimized in reverse. During actual measurement, the current air pressure value and the corrected temperature parameters are substituted into the functional relationship to calculate the corrected moisture parameters.

[0104] In one specific embodiment, the dielectric constant mixing model adopts a logarithmic mixing formula for two components, such as a solid matrix and water. The calibration experiment selects 6 pressure levels and 8 temperature levels. Under each combination, three standard samples with different water contents are measured, for a total of 144 calibration points. The nonlinear least squares method is used to fit the model parameters.

[0105] The stress parameters are corrected by using a preset thermal stress separation method to obtain the corrected stress parameters.

[0106] As an optional implementation method, thermal stress separation is performed using the strain tensor decomposition method.

[0107] For example, the raw output of the stress parameters is related to the strain of the material; under temperature variation, the total strain includes both mechanical strain and thermal strain.

[0108] Thermal strain equals the coefficient of thermal expansion multiplied by the temperature change. By measuring the strain components in different directions at the same location and combining them with the generalized Hooke's law for isotropic materials, a set of equations is established, and thermal strain is solved simultaneously with mechanical strain as an unknown quantity.

[0109] For example, by substituting strain measurements from at least three different directions, such as top, bottom, and left, into the strain transformation equation under plane stress, three independent mechanical strain components in the plane are obtained, including two normal strains and one shear strain.

[0110] Thermal strain is calculated using the temperature change at the same location and the pre-calibrated coefficient of thermal expansion.

[0111] Subtracting the thermal strain from the total strain yields the pure mechanical strain, which is then multiplied by the elastic modulus to obtain the corrected stress parameters. It is known that the strain transformation equation uses strain measurements in three known directions to inversely deduce the complete strain state.

[0112] In one specific embodiment, the coefficient of thermal expansion is calibrated by an isobaric variable temperature experiment: under constant air pressure, a temperature cycle is applied to a sample without external load, the change of strain output with temperature is recorded, and the coefficient of thermal expansion is obtained by linear regression.

[0113] The elastic modulus is calibrated through a uniaxial compression test. In actual measurement, the original strain values ​​in three directions and the corrected temperature parameters are first read, the equations are solved to obtain the mechanical strain components, and then multiplied by the elastic modulus to output the corrected stress parameters.

[0114] The conductivity parameters are corrected using a preset radiated noise suppression method to obtain preprocessed conductivity parameters.

[0115] As an optional implementation, an adaptive noise cancellation and Kalman filter cascade method is used for radiated noise suppression.

[0116] The adaptive noise canceller includes a main input channel and a reference input channel; the main input channel is the original conductivity signal; the reference input channel is an independent radiation dose rate signal provided by a radiation detection channel set in the MEMS array.

[0117] The adaptive noise canceller incorporates a finite impulse response filter whose coefficients are adaptively adjusted using a least mean square algorithm. It takes the radiation dose rate signal as input and outputs an estimated radiation noise signal.

[0118] Using a reference signal as the noise reference, the coefficients of a finite impulse response filter are adaptively adjusted. The estimated radiated noise signal is subtracted from the original conductivity measurement signal to obtain the canceled signal. The canceled signal is then input into a one-dimensional Kalman filter, where the state variable of the Kalman filter is the true conductivity value, and the process noise covariance is dynamically adjusted based on the integral value of the radiation dose rate.

[0119] In one specific embodiment, the effective impulse response filter is set to order 32, and the step size factor of the least mean square algorithm is a normalized value. The radiation dose rate signal is integrated every five minutes, and the integrated value is mapped to three levels, each level corresponding to a set of preset process noise covariance parameters.

[0120] The measurement noise covariance of the Kalman filter was calibrated by static measurement in a non-radiative environment; the preprocessed conductivity parameter is the output value of the Kalman filter.

[0121] Regarding step S102:

[0122] Based on the arrival time and signal strength of the response signal at the second location and the attenuation coefficient, the transmission characteristics of the preset medium in the test material are reconstructed by a preset inversion algorithm; wherein, the preset medium includes water, and the transmission characteristics include the transmission path, transmission rate and diffusion coefficient of the preset medium in the test material.

[0123] As an optional implementation method, the inversion algorithm can employ a combination of travel-time tomography and attenuation tomography, with the solution algorithm being an algebraic reconstruction method.

[0124] In practice, the area to be measured is discretized into a three-dimensional mesh. Taking the arrival time as input, assuming an initial uniform slowness, such as the reciprocal of the transmission rate, the residual between the theoretical travel time and the measured arrival time is calculated. If the ray path is initially a straight line, it is updated to a curved path according to Fermat's principle during iteration. The residual is distributed to the mesh cells according to the path length, and the slowness value is updated.

[0125] Iterate until the model change is less than the preset convergence threshold, and output the transmission rate field, where each grid cell corresponds to a slowness value.

[0126] In practical implementation, the convergence threshold can be determined through static testing. For example, a static test is performed on the central node out of 48 nodes. In a simulated Mars environment without any excitation, arrival times are continuously collected 100 times, and the standard deviation of the arrival times is calculated, which is then converted into the uncertainty of the slowness value. When the calculated uncertainty of the slowness value is 1.5%, its half-value, 0.75%, can be taken as the convergence threshold. In actual iteration, iteration stops when the relative change in the slowness value of all grid cells between two adjacent iterations is less than 0.75%.

[0127] The diffusion coefficient inversion takes signal strength and attenuation coefficient as input, establishes an initial uniform field of attenuation coefficient, calculates the residual between theoretical attenuation and measured attenuation, distributes the residual along the same ray path, and updates the attenuation coefficient field.

[0128] After the iteration converges, the decay coefficient is converted into the diffusion coefficient by power function fitting, and the diffusion coefficient field is output.

[0129] Transmission path reconstruction treats the transmission rate field as a convection velocity field and the diffusion coefficient field as a random disturbance.

[0130] For example, the fourth-order Runge-Kutta method is used to perform streamline integration starting from each first position, while adding Langevin random terms to generate multiple trajectories. The connected path with the highest probability density is selected as the main transmission path, and a set of three-dimensional transmission path curves is output.

[0131] In this implementation, assuming there are 48 nodes with a spacing of 5 cm, all nodes take turns as the first position. An algebraic reconstruction method is used with a relaxation factor of 0.1. After each iteration, the 48 first positions are traversed, and the relative change in slowness value for each grid cell is calculated. The process stops when the relative change in slowness value for all grid cells is less than 0.75%. In this embodiment, convergence is achieved after 12 iterations, outputting the slowness values ​​corresponding to the 48 grid cells of the transmission rate field.

[0132] Regarding step S103:

[0133] Figure 2 A flowchart of a three-dimensional multiphysics and chemical field reconstruction method based on MEMS arrays provided in the embodiments of this application includes:

[0134] S201: Based on the preprocessed feature parameters at the first and second positions, establish a discrete point set of each feature parameter in the spatial dimension;

[0135] S202: Using a preset spatial interpolation algorithm, the transmission path is used as a spatial topological constraint to spatially reconstruct the moisture parameters in the discrete point set, thereby obtaining a three-dimensional moisture field distribution.

[0136] S203: Based on the temperature parameters, stress parameters and conductivity parameters in the discrete point set, perform spatial interpolation reconstruction without transmission path constraints to obtain the three-dimensional temperature field distribution, the three-dimensional stress field distribution and the three-dimensional conductivity field distribution, and use the three-dimensional conductivity field distribution as the spatial distribution of the chemical field.

[0137] In practice, the 3D reconstruction method establishes a discrete point set based on the feature parameters and uses a spatial interpolation algorithm to extend the feature parameters from discrete points to a continuous spatial distribution.

[0138] The three-dimensional spatial coordinates of each MEMS sensor node are associated with the preprocessed temperature, moisture, stress, and conductivity parameters at that node, respectively, to form temperature point sets, moisture point sets, stress point sets, and conductivity point sets; each data point in each point set contains three-dimensional spatial coordinates and a parameter value.

[0139] The three-dimensional water field is reconstructed using the radial basis function interpolation method, and the inverted water transport path is introduced as a spatial topological constraint. The radial basis function is selected as a thin plate spline.

[0140] For example, spatial points are extracted along each moisture transport path at equal intervals with a preset virtual point extraction step size as additional constraint points. The virtual point extraction step size must be less than the spacing between MEMS sensor nodes, typically half of the node spacing, and confirmed through convergence testing: the virtual point extraction step size is gradually reduced, and when the change in the moisture field interpolation result is less than one-tenth of the sensor measurement resolution, the current virtual point extraction step size is adopted.

[0141] An angular constraint is applied at each virtual point, requiring that the angle between the spatial gradient direction of the moisture parameter and the tangent direction of the transmission path does not exceed a preset angle threshold.

[0142] The angle threshold is determined through a preliminary experiment: a water-retaining sand sample with a known single straight damaged channel is prepared, and the angle between the direction of the water gradient and the tangent direction of the transport path is measured point by point along the channel under steady-state moisture distribution. The upper quartile of all measured values ​​is then used as the angle threshold.

[0143] The measured data points are merged with the virtual points with angle constraints. The Lagrange multiplier method is used to add the angle constraints as a penalty term to the linear equation system of radial basis function interpolation, and the weight coefficients are solved.

[0144] Regular grid points are generated in the sample space with a preset moisture field grid spacing. The grid spacing is set based on the fact that it is no greater than the node spacing and is usually equal to the node spacing. The grid range covers the entire sample.

[0145] By substituting the coordinates of each grid point into the constrained radial basis function interpolation formula, the moisture parameter value of that point is calculated, thus forming a three-dimensional moisture field distribution.

[0146] For the reconstruction of the three-dimensional temperature field, stress field, and conductivity field, ordinary Kriging interpolation method is used respectively, without introducing transmission path constraints.

[0147] In practice, the experimental variability function is calculated, and the data point pairs are grouped by distance. The grouping interval is usually the spacing between MEMS sensor nodes, and the maximum distance is half the length of the sample diagonal.

[0148] For all data point pairs within each group, calculate the square of the difference between the parameter values ​​of the two data points, take the average and divide by 2 to obtain the experimental variability function value corresponding to that distance; the function values ​​corresponding to all distance groups constitute the experimental variability function curve.

[0149] The experimental variability function is used to describe the structural and random nature of spatial data as distance changes; that is, the closer two points are, the more similar their parameter values ​​are.

[0150] A theoretical variogram model is fitted, with selectable models including spherical, exponential, and Gaussian models. The experimental variogram curve is fitted using weighted least squares, and the sum of squared residuals of each model is compared. The model with the smallest residual is selected as the theoretical variogram model. The nugget value, sill value, and range are extracted from the fitting results. The initial estimate of the nugget value is obtained by repeatedly measuring the variance of the parameter values ​​at the same node. The initial estimate of the sill value is given by the variance of the parameter values ​​for all data point pairs. The initial estimate of the range is given by the distance at which the experimental variogram curve first approaches stationarity.

[0151] Solve the Kriging equations at each regular grid point to be interpolated to obtain the weights of each measured point, and then sum the weighted values ​​to obtain the parameter values ​​at that point.

[0152] In practice, the output grid spacing of the temperature field, stress field, and conductivity field is consistent with the grid spacing of the moisture field.

[0153] For example, based on the aforementioned 48 nodes, node spacing of 5 cm, sample size of 30 cm × 20 cm × 10 cm, and 48 water transport paths obtained through inversion, a discrete point set is first established, with each point set containing 48 data points.

[0154] For water field reconstruction, the virtual point extraction step size is half of the node spacing, i.e., 2.5 cm. The average length of the 48 transmission paths is about 15 cm, and about 6 virtual points are extracted from each path, for a total of about 288 virtual points.

[0155] The angle threshold was calibrated through a preliminary experiment: dyed water was injected into water-retaining sand samples of the same size with a 2 mm wide through-break line. After stabilization, the gradient and tangential angle were measured every 2.5 cm. The upper quartile of the 20 measuring points was 28 degrees, so the angle threshold was set to 28 degrees.

[0156] The angular constraints of 48 measured points and 288 virtual points were combined, and a constrained system of equations was constructed using the radial basis function of thin plate splines to solve for the weight coefficients.

[0157] The output grid spacing of the moisture field is set to 5 cm. A total of 105 grid points are formed within the range of 0 to 30 cm in the length direction (7 points with a step size of 5 cm), 0 to 20 cm in the width direction (5 points), and 0 to 10 cm in the height direction (3 points). The moisture value of each point is calculated to obtain the three-dimensional moisture field.

[0158] For the temperature field, the distance between groups is 5 cm, and the maximum distance is half the diagonal length of the sample, which is about 18 cm.

[0159] After calculating the experimental variogram, the data were fitted using spherical, exponential, and Gaussian models, respectively. The sum of squared residuals were 0.023, 0.031, and 0.028, respectively. The spherical model was selected, and the fitted values ​​were 0.12 for nugget, 0.85 for sill, and 12 cm for range.

[0160] Ordinary kriging interpolation was performed on the same 105 grid points to obtain the three-dimensional temperature field.

[0161] For the stress field, spherical, exponential, and Gaussian models were used for fitting, and the sum of squared residuals were 0.045, 0.021, and 0.038, respectively. When the exponential model was selected, the nugget value was 0.08, the sill value was 0.62, and the range was 10 cm. Similarly, the three-dimensional stress field was obtained by interpolation at 105 grid points.

[0162] For the electrical conductivity field, spherical, exponential, and Gaussian models were used for fitting, and the sum of squared residuals were 0.034, 0.029, and 0.015, respectively. The Gaussian model was selected, and the nugget value was 0.05, the sill value was 0.44, and the range was 14 cm. The three-dimensional electrical conductivity field was obtained by interpolation and used as the spatial distribution of the chemical field.

[0163] The final output consists of 105 regular grid points, each containing three-dimensional coordinates and four field values: moisture, temperature, stress, and electrical conductivity.

[0164] Regarding step S104:

[0165] Figure 3 The flowchart for anomaly identification and classification based on mechanism constraints provided in the embodiments of this application includes:

[0166] A physical information neural network is constructed, and the transmission path is used as a boundary condition to train the physical information neural network to fit the three-dimensional moisture field distribution.

[0167] The physical information neural network adopts a fully connected feedforward neural network structure and consists of an input layer, four hidden layers, and an output layer.

[0168] In practice, the input layer contains three neurons, each corresponding to one of the three components of spatial coordinates: the horizontal coordinate, the vertical coordinate, and the vertical axis.

[0169] Each hidden layer contains fifty neurons, and adjacent layers are fully connected, meaning that each neuron receives the outputs of all neurons in the previous layer, sums them in a weighted manner, and then outputs the result through an activation function.

[0170] The activation function of the hidden layer is the hyperbolic tangent function, and the output layer contains one neuron that outputs the predicted water parameter value. The output layer has a linear output.

[0171] Calculate the physical residual of the physical information neural network at each spatial location to form a residual field distribution;

[0172] Spatial regions in the residual field distribution where the local residual energy exceeds the first preset residual threshold are marked as the first abnormal region.

[0173] In specific implementation, the loss function of the physical information neural network is composed of a weighted sum of a data fitting term, a physical constraint term, and a path direction constraint term. The data fitting term uses the mean square error, which is obtained by calculating the average of the squared errors between the moisture value predicted by the network and the measured moisture value at that location.

[0174] The physical constraint terms are embedded in the Richards equation as residuals. The partial derivatives of the network output with respect to the input coordinates are calculated using automatic differentiation techniques. These partial derivatives are substituted into the Richards equation, and the square of the difference between the value on the left side and the value on the right side of the equation is calculated as the physical constraint residuals.

[0175] The path direction constraint term is obtained by forcibly applying gradient direction constraints to points on the water transport path, i.e., virtual points on the inverted path, and calculating the constraint residual. The constraint residual is the square of the angle between the water gradient direction and the path tangent direction multiplied by the penalty coefficient.

[0176] As an optional implementation, the penalty coefficient can be determined using a Bayesian optimization algorithm.

[0177] For example, the search space is set to a closed interval [1,100]; the objective function is the average angle error, which is the arithmetic mean of the angles between the direction of the moisture gradient and the direction of the path tangent at all virtual points on the verification sample; the initial number of sampling points is set to 10, and the maximum number of iterations is 50.

[0178] In each iteration, a candidate penalty coefficient is selected, the physical information neural network is trained, and the corresponding average angle error is calculated. The result is fed back to the Bayesian optimizer for updating. After the iteration ends, the penalty coefficient that minimizes the average angle error is selected as the final value.

[0179] The total loss function is a weighted sum of the data fitting term, the physical constraint term, and the path direction constraint term. The weight coefficients of the three terms are determined through pre-experiments with defect-free samples.

[0180] In practice, an adaptive moment estimation optimizer is used for training. The initial learning rate is set to 0.001, and is multiplied by a decay factor of 0.9 after every 500 training rounds. The training termination condition is that the change in the total loss function is less than 0.1% over 100 consecutive iterations.

[0181] After training, each spatial coordinate point in the test area is input into the physical information neural network to obtain the predicted moisture parameter value, and the physical constraint residual of that point, i.e. the absolute value of the Richards equation residual, is calculated. The physical constraint residual values ​​of all spatial points constitute the residual field.

[0182] The first preset residual threshold is set based on the following: in a defect-free water-retaining sand sample, the physical constraint residual of each spatial coordinate point is calculated to obtain the residual distribution.

[0183] As an optional implementation, the mean square root of the squares of all physical constraint residuals is calculated within a preset radius neighborhood centered on each spatial coordinate point to obtain the local residual root mean square value of that point; the 95th percentile of the local residual root mean square values ​​of all spatial coordinate points is taken as the first preset residual threshold.

[0184] In specific implementation, local residual energy is defined as the root mean square value of the local residual obtained by taking the square root of the mean of the squares of all physical constraint residual values ​​within a preset radius neighborhood centered at a certain spatial coordinate point. Spatial regions where the local residual energy exceeds the first preset residual threshold are marked as the first abnormal region.

[0185] The preset radius can be set as the spacing between MEMS sensor nodes.

[0186] As an optional implementation, a physical-guided generative adversarial network is constructed within the first abnormal region, the physical-guided generative adversarial network including a generator and a discriminator;

[0187] The generator's inputs include random noise and the three-dimensional temperature field distribution, three-dimensional moisture field distribution, and three-dimensional stress field distribution, and its output is the reconstructed physical field spatial distribution;

[0188] The loss function of the generator incorporates the moisture transport equation and the thermoelastic constitutive relation as physical constraint residuals.

[0189] The input to the discriminator is the reconstructed physical field spatial distribution output by the generator and the measured physical field spatial distribution, and the output is the anomaly classification result.

[0190] In practice, the physical-guided generative adversarial network consists of two subnetworks: a generator and a discriminator, which are trained in an adversarial manner.

[0191] The generator adopts an encoder-decoder structure such as the U-Net architecture. The encoder part consists of four downsampling blocks, each of which contains a convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer in sequence.

[0192] The number of input channels in the first convolutional layer is the number of channels in the conditional input, including three channels: temperature field, moisture field, and stress field. The number of grid points in the input region for each field determines the actual number of channels. For example, if the input region contains 9 grid points, then each field corresponds to 9 channels, for a total of 27 channels, and the number of output channels is 32.

[0193] The second convolutional layer has 32 input channels and 64 output channels. The third convolutional layer has 64 input channels and 128 output channels. The fourth convolutional layer has 128 input channels and 256 output channels. All convolutional layers have a kernel size of 3×3, a stride of 1, and the same padding method; the activation function is a modified linear unit; the pooling window size of the max-pooling layer in each downsampling block is 2×2, with a stride of 2.

[0194] The decoder consists of four upsampling blocks, each of which contains a transposed convolutional layer, a batch normalization layer, and an activation layer, and is connected to the output of the corresponding layer of the encoder in a skip connection.

[0195] The first upsampling block has 256 input channels and 128 output channels. The second upsampling block has 128 input channels; with the skip connections from the third layer of the encoder, the actual input is 256 channels and the output is 64 channels. The third upsampling block has 64 input channels; with the skip connections from the second layer of the encoder, the actual input is 128 channels and the output is 32 channels. The fourth upsampling block has 32 input channels; with the skip connections from the first layer of the encoder, the actual input is 64 channels and the output is 3 channels (corresponding to the reconstructed moisture, temperature, and stress fields).

[0196] The transposed convolutional layer has a kernel size of 3×3, a stride of 2, and output padding on one side.

[0197] The last upsampling block is followed by a convolutional layer with a 1×1 kernel, which maps the number of channels to the number of output channels 3. The generator output is the reconstructed physical field spatial distribution, namely the moisture field, temperature field, and stress field.

[0198] The generator's input consists of two parts: a random noise vector and a conditional input. The random noise vector has a length of 100, is sampled from a standard Gaussian distribution, and is mapped through a fully connected layer to a feature map of the same size as the encoder input. This feature map is then concatenated with the conditional input channel and fed into the encoder.

[0199] The input conditions are regular grid data of temperature field, moisture field and stress field in the first anomaly region, which are then normalized.

[0200] The generator's loss function is a weighted sum of three parts: adversarial loss, reconstruction loss, and physical constraint loss. The weight coefficients of the three losses are determined through preliminary experiments.

[0201] The adversarial loss adopts the form of binary cross-entropy, with the goal of making the discriminator judge the generated field as the real field.

[0202] The reconstruction loss uses mean square error to calculate the difference between the physical field output by the generator and the original physical field obtained by reconstruction in step S103.

[0203] The physical constraint loss includes the residuals of the moisture transport equation and the thermoelastic constitutive relation. The residuals of the moisture transport equation are calculated in the same way as PINN, that is, the spatial partial derivatives of the generator's output moisture field are calculated and substituted into the Richards equation.

[0204] The residual of the thermoelastic constitutive relation is calculated as follows: using the temperature field and stress field output by the generator, combined with the thermal expansion coefficient and elastic modulus of the material, the theoretical thermal stress is calculated by multiplying the thermal expansion coefficient by the temperature change by the elastic modulus, and then compared with the thermal stress component in the stress field output by the generator, and the mean square error is taken.

[0205] The discriminator employs a convolutional neural network structure, consisting of five convolutional blocks and one fully connected classification layer. Each convolutional block contains, in sequence, a convolutional layer, a batch normalization layer, and a leak-correction linear unit activation layer.

[0206] The number of input channels for the first convolutional block is the number of channels for the input physical field, including three fields: moisture, temperature, and stress. If the input region contains 9 grid points, then each field has 9 channels, for a total of 27 channels. Adding the number of channels for the auxiliary input three-dimensional conductivity field, such as 9 channels, the total number of channels is 36.

[0207] The first convolutional block outputs 64 channels with a 4×4 kernel and a stride of 2. The second convolutional block outputs 128 channels with a 4×4 kernel and a stride of 2. The third convolutional block outputs 256 channels with a 4×4 kernel and a stride of 2. The fourth convolutional block outputs 512 channels with a 4×4 kernel and a stride of 2. The fifth convolutional block outputs 512 channels with a 3×3 kernel and a stride of 1.

[0208] In practice, the negative slope coefficient of the activation function of the leakage correction linear unit in each convolutional block is set to 0.2. The output of the last convolutional block is then fed into a fully connected layer after global average pooling. The fully connected layer outputs three nodes, corresponding to three types of abnormal states: abnormal conductivity, abnormal multiphysics coupling, and damaged coating. Finally, the probability distribution is output through a flexible maximum function.

[0209] An abnormal conductivity state is defined as the residual between the measured conductivity value and the reconstructed conductivity value generated by the generator exceeding a preset conductivity threshold, and the physical field residual being lower than a preset physical field threshold. The physical field residual is the weighted sum of the physical constraint residual and the reconstructed residual.

[0210] In practice, the reconstruction residual is obtained by averaging the absolute differences between the generator output values ​​of the moisture field, temperature field, and stress field and the corresponding values ​​of the original field at all grid points.

[0211] As an optional implementation method, the weighting coefficients are determined as follows: collect 30 standard samples containing known defect types and severity, traverse the coefficient combinations between 0 and 1 with a step size of 0.1, calculate the abnormal region identification accuracy under each set of coefficients, and select the coefficient combination corresponding to the highest accuracy as the fusion weight.

[0212] The multiphysics coupling anomaly is when the physical field residual exceeds the preset physical field threshold and the conductivity residual is lower than the preset conductivity threshold.

[0213] The coating damage state is defined as the residual conductivity and the residual physical field both exceeding their respective thresholds; when neither the residual conductivity nor the residual physical field exceeds their respective thresholds, there is no abnormality classification.

[0214] Both the preset conductivity threshold and the preset physical field threshold are determined by the 95th percentile of the statistical residual distribution in the defect-free sample.

[0215] In practice, the generator and discriminator are trained alternately. In each training round, the generator parameters are first fixed, and the discriminator parameters are updated using a batch of measured physical field samples and a batch of generated field samples output by the generator.

[0216] Then, with the discriminator parameters fixed, the generator outputs samples, and the generator parameters are updated using the generator loss function. This process is repeated for several iterations, for example, 2000 iterations, with the discriminator and generator updated once in each iteration. During training, the learning rate is initially set to 0.0002, decreasing by half every 500 iterations. The optimizer uses adaptive moment estimation.

[0217] The reconstruction residual between the reconstructed physical field output by the generator and the measured physical field is weighted and fused with the probability corresponding to the coating damage state output by the discriminator to obtain the anomaly score for each monitoring location.

[0218] The probability corresponding to the damaged state of the coating is calculated by the output layer of the discriminator through the flexible maximum value function, and the value ranges from 0 to 1.

[0219] The fusion weights were determined through preliminary experiments, and the preset anomaly threshold was determined through receiver operating characteristic (ROC) curve analysis. When the anomaly score exceeds the preset anomaly threshold, the corresponding monitoring location is marked as a damaged coating area, serving as a second anomaly area, and the damage location information is output. For example, a physical information neural network is constructed with an input layer of 3 neurons, 4 hidden layers each with 50 neurons, and an output layer of 1 neuron. The training data consists of the moisture values ​​of 48 measured nodes out of 105 grid points, with the remaining measured node moisture values ​​used for validation.

[0220] The physical constraint terms are embedded in the Richards equation, in which the saturated hydraulic conductivity is 0.05 cm / s and the porosity is 0.4, both of which are calibrated through independent soil column experiments.

[0221] The path direction constraint term applies gradient direction constraints to 288 virtual points along 48 paths, with an angle threshold of 28 degrees and a penalty coefficient of 10. In practice, the constraint residual of each virtual point is calculated, and the constraint residual values ​​of all virtual points are summed and added to the total loss function as the path direction constraint term.

[0222] The weight ratio of the loss terms for data fitting, physical constraints, and path direction constraints was set to 1:0.5:0.2 in preliminary experiments. The Adam optimizer was used for training with a learning rate of 0.001, which decayed by 0.9 every 500 epochs, for a total of 3000 epochs. Training was stopped when the loss change was less than 0.1% for 100 consecutive epochs.

[0223] After training, the physical residuals at 105 grid points are calculated. The 95th percentile of the residual distribution of defect-free samples is 0.05, therefore the first preset residual threshold is set to 0.05.

[0224] The local residual energy calculation radius is set to 10 cm. A region with a diameter of about 6 cm near the center of the sample is calculated. The local residual energy of the region containing 9 grid points is 0.08, which is greater than 0.05. It is marked as the first abnormal region.

[0225] A physical-guided generative adversarial network is constructed within the first anomalous region. The generator U-Net encoder has 27 input channels, which is 9 points × 3 fields. The output channels of the four convolutional layers are 32, 64, 128, and 256, respectively. The decoder output has 3 channels and random noise length is 100.

[0226] The discriminator has 36 input channels, including 27 physical field channels and 9 conductivity channels. The output channels of the 5 convolutional blocks are 64, 128, 256, 512 and 512 respectively. The fully connected layer outputs 3 types of abnormal states.

[0227] The training data consisted of the measured physical and electrical conductance fields of the nine grid points. The weights of the adversarial loss, reconstruction loss, and physical constraint loss in the generator's loss function were 1:10:5. The training ran for 2000 epochs, with the discriminator and generator each updated once per epoch. The learning rate was 0.0002, and it decayed by half every 500 epochs.

[0228] The 95th percentile of the electrical conductivity residual in the defect-free sample is 0.02, and the 95th percentile of the physical field residual is 0.03. Therefore, the preset electrical conductivity threshold is 0.02, and the preset physical field threshold is 0.03.

[0229] Among the 9 grid points, the conductivity residual at the center point coordinates (15 cm, 10 cm, 5 cm) is 0.08, and the physical field residual is 0.07. Both exceed the corresponding thresholds, and the coating is judged to be damaged.

[0230] The conductivity residual at a certain point on the edge is 0.09, and the physical field residual is 0.02, which is determined to be an abnormal conductivity state.

[0231] Another point is that the conductivity residual is 0.01 and the physical field residual is 0.04, which is determined to be an abnormal state of multi-physics coupling.

[0232] The weights of the reconstruction residual and the probability corresponding to the coating damage state output by the discriminator are 0.4 and 0.6, respectively. The reconstruction residual corresponding to the center point is 0.12, and the probability corresponding to the coating damage state output by the discriminator is 0.9. The anomaly score = 0.4 × 0.12 + 0.6 × 0.9 = 0.588.

[0233] The preset anomaly threshold was determined to be 0.55 through ROC curve analysis of 30 defective samples.

[0234] The anomaly score corresponding to the center point is 0.588, which is greater than 0.55, and it is marked as the second anomaly area.

[0235] The output damage location information is a spherical area with a radius of approximately 3 cm centered at (15 cm, 10 cm, 5 cm). All other points have anomaly scores below the anomaly threshold of 0.55 and are not marked.

[0236] Regarding step S105:

[0237] Anomaly information is extracted from the second anomaly region, including: the average anomaly score within the second anomaly region, the area of ​​the second anomaly region, and the anomaly classification result.

[0238] The average anomaly score is the arithmetic mean of the anomaly scores of all grid points within the second anomaly region.

[0239] The area of ​​the second abnormal region is the surface area of ​​the second abnormal region in three-dimensional space. For thin-layer damage, the unfolded area of ​​the damaged region is taken; for volumetric damage, the outer surface area is taken.

[0240] As an optional implementation, the area calculation employs a three-dimensional convex hull algorithm or a mesh surface reconstruction method, such as the traveling cube algorithm, to perform surface reconstruction on the boundary of the second abnormal region before calculation.

[0241] For example, based on the anomaly score, and using a preset anomaly threshold as the isosurface value, the isosurface mesh corresponding to the threshold is extracted. Then, for each extracted triangular facet, the area is calculated based on the coordinates of its three vertices, and the total surface area is obtained by summing them up.

[0242] If the second abnormal region is the union of multiple disconnected regions, then the area of ​​each connected region is calculated separately and then summed.

[0243] The abnormal information is input into a preset signal mapping function, and the corresponding monitoring signal is output.

[0244] The monitoring signals include at least one of the following: damage alarm signal, coupling failure early warning signal, corrosion diffusion warning signal, and general abnormality indication signal.

[0245] In practical implementation, the signal mapping function can adopt a rule-based piecewise mapping method, taking the average anomaly score and area as input, and the anomaly classification result as condition, to output one of four monitoring signals. The mapping rules are as follows:

[0246] Damage alarm signal: When the abnormality classification result is that the film is damaged, and the average abnormality score exceeds the high threshold, or the area exceeds the large area threshold, a damage alarm signal is output.

[0247] Coupling failure warning signal: When the anomaly classification result is a multi-physics coupling anomaly state, and the average anomaly score exceeds the medium threshold, or the area of ​​the region exceeds the medium area threshold, a coupling failure warning signal is output.

[0248] Corrosion diffusion warning signal: When the anomaly classification result is an abnormal conductivity state, and the average anomaly score exceeds the low threshold, or the area exceeds the small area threshold, a corrosion diffusion warning signal is output.

[0249] General anomaly warning signal: When none of the above conditions are met, but the anomaly classification result is not "no anomaly classification", a general anomaly warning signal will be output.

[0250] For example, the high, medium, and low thresholds can be determined by conducting destructive testing on standard samples with known defect types and severity.

[0251] In the specific implementation, 30 standard water-retaining sand samples were prepared, including 10 defect-free samples, 10 samples with minor damage, such as samples with a damage diameter of 1 mm to 5 mm, and 10 samples with moderate damage, such as samples with a damage diameter of 6 mm to 15 mm.

[0252] A complete monitoring process was performed on each sample to obtain an average anomaly score for each sample.

[0253] As an optional implementation method, for each type of defect, the distribution of its average abnormal score is statistically analyzed, and the 50th percentile is taken as the low threshold, the 70th percentile as the medium threshold, and the 90th percentile as the high threshold.

[0254] The threshold can be different for different anomaly categories. In this embodiment, the above experiment shows that:

[0255] Under the condition of film damage, the low threshold is 0.60, the medium threshold is 0.75, and the high threshold is 0.85;

[0256] Under multiphysics coupling anomaly conditions, the low threshold is 0.55, the medium threshold is 0.70, and the high threshold is 0.80.

[0257] Under abnormal conductivity conditions, the low threshold is 0.50, the medium threshold is 0.65, and the high threshold is 0.75.

[0258] As an optional implementation method, the small area threshold, medium area threshold, and large area threshold can be set according to the total surface area of ​​the water-retaining sand sample and the safety requirements of the application scenario.

[0259] For example, the small area threshold corresponds to the minimum damaged area that may cause localized minor leakage, which is 0.2% of the total surface area of ​​the sample.

[0260] The medium area threshold corresponds to the minimum damaged area that may affect the function of a local area but does not endanger the whole, and is taken as 1% of the total surface area of ​​the sample.

[0261] The large area threshold corresponds to the minimum damaged area that may cause system failure, which is 5% of the total surface area of ​​the sample.

[0262] In this embodiment, the sample size is 30 cm × 20 cm × 10 cm, and the total surface area is calculated as 2 × (30 × 20 + 30 × 10 + 20 × 10) = 2200 square centimeters. Therefore:

[0263] The threshold for small areas is 4.4 square centimeters, the threshold for medium areas is 22 square centimeters, and the threshold for large areas is 110 square centimeters.

[0264] Based on the same inventive concept, this application also provides a MEMS array-based water and sand extreme environment performance monitoring system corresponding to the MEMS array-based water and sand extreme environment performance monitoring method.

[0265] See Figure 4 The diagram shown is a schematic of a MEMS array-based water-retention and sand-holding extreme environment performance monitoring system provided in this application embodiment. The system includes: a data acquisition module 10, an inversion module 20, a reconstruction module 30, an identification module 40, and an output module 50, wherein:

[0266] Acquisition module 10: used to deploy a MEMS array in the material under test, apply an excitation signal to a first position in the MEMS array, acquire a response signal through a second position, and acquire characteristic parameters at the first position and the second position, the characteristic parameters including: physical field parameters and chemical field parameters;

[0267] Inversion module 20: used to invert the transmission characteristics of the preset medium in the material under test based on the response signal;

[0268] Reconstruction module 30: used to form the physical field spatial distribution and chemical field spatial distribution of the material under test based on the feature parameters and through a preset three-dimensional reconstruction method;

[0269] Identification module 40: used for residual analysis based on mechanism constraints, to jointly evaluate the transmission characteristics and the spatial distribution of the physical field, and to identify abnormal regions;

[0270] Output module 50: Classifies the abnormal area, obtains the abnormal classification result, and outputs the monitoring signal.

[0271] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0272] In the description of this specification, the terms "exemplary," "for example," "specifically," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A method for monitoring the performance of water-retaining sand in extreme environments based on a MEMS array, characterized by, The method includes: A MEMS array is placed in the material under test. An excitation signal is applied to a first position in the MEMS array, and a response signal is obtained through a second position. Characteristic parameters at the first and second positions are obtained, including physical field parameters and chemical field parameters. Based on the response signal, the transmission characteristics of the preset medium in the material under test are obtained by inversion. Based on the aforementioned feature parameters, the physical field spatial distribution and chemical field spatial distribution of the material under test are formed using a preset three-dimensional reconstruction method. Based on mechanism-constrained residual analysis, the transmission characteristics and the spatial distribution of the physical field are jointly evaluated to identify abnormal regions. The abnormal regions are classified into anomalies to obtain anomaly classification results, and monitoring signals are output.

2. The MEMS array based water retention sand extreme environment performance monitoring method according to claim 1, wherein, The step of applying an excitation signal at a first position in the MEMS array and obtaining a response signal at a second position includes: The first position and the second position are spatially different, and the second position is located in a region with the first position as the center and within a preset radius. The arrival time and signal strength of the response signal at each second position are extracted, and the attenuation coefficient of the response signal as the spatial distance changes is calculated based on the spatial distance between the first position and the second position.

3. The MEMS array based water retention sand extreme environment performance monitoring method of claim 1, wherein, The step of obtaining the feature parameters at the first position and the second position includes: The physical field parameters include at least temperature parameters, moisture parameters, and stress parameters, and the chemical field parameters include at least electrical conductivity parameters. The temperature parameter is corrected using a preset temperature drift compensation method; the moisture parameter is corrected using a preset air pressure correction method; the stress parameter is corrected using a preset thermal stress separation method; and the conductivity parameter is corrected using a preset radiation noise suppression method, thus obtaining the preprocessed characteristic parameters.

4. The MEMS array based water retention sand extreme environment performance monitoring method of claim 2, wherein, The process of inverting the transmission characteristics of the preset medium in the material under test based on the response signal includes: Based on the arrival time and signal strength of the response signal at the second location and the attenuation coefficient, the transmission characteristics of the preset medium in the material under test are reconstructed by a preset inversion algorithm; The preset medium includes water, and the transmission characteristics include the transmission path, transmission rate, and diffusion coefficient of the preset medium in the material under test.

5. The method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays according to claim 4, characterized in that, The process of forming the physical field spatial distribution and chemical field spatial distribution of the material under test through a preset three-dimensional reconstruction method includes: Based on the preprocessed feature parameters at the first and second positions, a discrete point set of each feature parameter in the spatial dimension is established. Using a preset spatial interpolation algorithm, the transmission path is used as a spatial topological constraint to spatially reconstruct the moisture parameters in the discrete point set, thereby obtaining a three-dimensional moisture field distribution. Based on the temperature parameters, stress parameters, and conductivity parameters in the discrete point set, spatial interpolation reconstruction without transmission path constraints is performed to obtain the three-dimensional temperature field distribution, the three-dimensional stress field distribution, and the three-dimensional conductivity field distribution. The three-dimensional conductivity field distribution is then used as the spatial distribution of the chemical field.

6. The method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays according to claim 5, characterized in that, The residual analysis based on mechanistic constraints, which jointly evaluates the transmission characteristics and the spatial distribution of the physical field, includes: A physical information neural network is constructed, and the transmission path is used as a boundary condition to train the physical information neural network to fit the three-dimensional moisture field distribution. Calculate the physical residual of the physical information neural network at each spatial location to form a residual field distribution; Spatial regions in the residual field distribution where the local residual energy exceeds the first preset residual threshold are marked as the first abnormal region.

7. The method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays according to claim 6, characterized in that, The abnormal region classification includes: Within the first abnormal region, a physical-guided generative adversarial network is constructed, which includes a generator and a discriminator. The generator's inputs include random noise and the three-dimensional temperature field distribution, three-dimensional moisture field distribution, and three-dimensional stress field distribution, and its output is the reconstructed physical field spatial distribution; The loss function of the generator incorporates the moisture transport equation and the thermoelastic constitutive relation as physical constraint residuals. The input to the discriminator is the reconstructed physical field spatial distribution output by the generator and the measured physical field spatial distribution, and the output is the anomaly classification result.

8. The method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays according to claim 7, characterized in that, The output of the discriminator, which is the anomaly classification result, includes: The discriminator receives the three-dimensional conductivity field distribution as auxiliary input and outputs the anomaly classification result. The reconstruction residual between the reconstructed physical field spatial distribution output by the generator and the measured physical field spatial distribution is weighted and fused with the probability corresponding to the film damage state in the anomaly classification result to obtain the anomaly score for each monitoring location. If the anomaly score exceeds the preset anomaly threshold, the corresponding monitoring location is marked as a film damage area, which is then used as the second anomaly area, and the damage location information is output.

9. The method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays according to claim 8, characterized in that, The output anomaly classification results include: The anomaly classification results include: abnormal conductivity state, abnormal multi-physics field coupling state, and film damage state. The abnormal conductivity state is determined based on the residual between the measured conductivity value and the reconstructed conductivity value exceeding a preset conductivity threshold and the physical field residual being lower than a preset physical field threshold. The physical field residuals include physical constraint residuals and reconstruction residuals.

10. The method for monitoring the water-retention and sand-holding performance in extreme environments based on MEMS arrays according to claim 9, characterized in that, The output monitoring signal includes: Anomaly information is extracted from the second anomaly region, including: the average anomaly score within the second anomaly region, the area of ​​the second anomaly region, and the anomaly classification result; The abnormal information is input into a preset signal mapping function, and the corresponding monitoring signal is output. The monitoring signals include at least one of the following: damage alarm signal, coupling failure early warning signal, corrosion diffusion warning signal, and general abnormality indication signal.

11. A MEMS array-based water and sand retention extreme environment performance monitoring system, used to implement the MEMS array-based water and sand retention extreme environment performance monitoring method according to any one of claims 1-10, characterized in that, The system includes: Acquisition module: used to deploy MEMS array in the material under test, apply an excitation signal to a first position in the MEMS array, acquire a response signal through a second position, and acquire characteristic parameters at the first and second positions, the characteristic parameters including physical field parameters and chemical field parameters; Inversion module: used to invert the transmission characteristics of the preset medium in the material under test based on the response signal; Reconstruction module: used to form the physical field spatial distribution and chemical field spatial distribution of the material under test based on the feature parameters and through a preset three-dimensional reconstruction method; Identification module: used for residual analysis based on mechanism constraints, to jointly evaluate the transmission characteristics and the spatial distribution of the physical field, and to identify abnormal regions; Output module: Classifies the abnormal area, obtains the abnormal classification result, and outputs the monitoring signal.

Images