A method for testing the dynamic compressive mechanical properties of ice structures

Abstract

This invention relates to the field of ice mechanical performance testing technology and discloses a method for testing the dynamic compression mechanical properties of ice structures. The method includes simultaneously acquiring three-dimensional coordinates, time series, acoustic emission signals, and thermal infrared images of an ice structure during dynamic compression using multiple sensors to construct a multimodal dataset; extracting mechanical indices, acoustic features, and thermodynamic parameters after spatiotemporal alignment processing to generate an integrated feature set; automatically identifying consistent and abnormal regions based on spatial distance and feature similarity; simulating the evolution of mechanical properties under different working conditions using consistent region data; calculating anomaly risk indices by combining acoustic energy distribution and infrared temperature difference changes to form a high-risk point set; and finally generating a risk warning report through failure mode identification. This method achieves multi-dimensional automated evaluation of the mechanical properties of ice structures, accurately locating defective areas and quantifying risk levels, thus improving the accuracy of testing and analysis and the ability to provide early warnings.

Description

Technical Field

[0001] This invention relates to the field of ice mechanical performance testing technology, specifically to a method for testing the dynamic compressive mechanical properties of ice structures. Background Technology

[0002] Ice structures are increasingly used in bridges, ships, polar engineering, and other fields, and their dynamic mechanical properties directly affect the safety and durability of the structures. Existing methods for testing the mechanical properties of ice mainly rely on a single type of sensor, or supplemented by acoustic emission sensors to monitor crack propagation. The data obtained by these methods are limited in scope, reflecting only the average response of the ice structure as a whole or a single local physical phenomenon.

[0003] The drawback of traditional technical solutions lies in their inability to effectively correlate the spatiotemporal relationship between mechanical response, internal damage acoustic signals, and surface thermal field distribution. Mechanical sensors struggle to accurately pinpoint the initiation location of internal microcracks; while acoustic emission technology can detect damage events, it lacks corresponding information on local mechanical states and temperature changes. Furthermore, without simultaneous corroboration from mechanical and acoustic data, it is difficult to determine whether the surface temperature field changes observed by infrared thermal imagers are caused by plastic work-thermal conversion or localized stress concentration due to defects. This fragmented data leads to an insufficiently in-depth analysis of the damage evolution mechanism of ice structures under dynamic loading.

[0004] Due to the lack of collaborative analysis of multiphysics data, existing methods struggle to objectively assess the uniformity of the internal properties of ice structures and cannot automatically identify potential defect areas or material inhomogeneities that are inconsistent with the overall mechanical behavior. This makes risk assessment often rely on threshold alarms of overall parameters or macroscopic observations of failure morphology after testing. There is a lack of a precise, quantitative, temporal and spatial criterion for determining which specific locations are failure initiation points during compression and what their risk levels are. Summary of the Invention

[0005] The purpose of this invention is to provide a method for testing the dynamic compressive mechanical properties of ice structures, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for testing the dynamic compressive mechanical properties of ice structures, the method comprising:

[0007] A multimodal test dataset was constructed by synchronously acquiring the three-dimensional spatial coordinates, time series data, acoustic emission signals, and thermal infrared image data of ice structure samples during dynamic compression using a multi-sensor array.

[0008] The multimodal test dataset is spatiotemporally aligned to extract the mechanical performance indicators, acoustic feature vectors, and thermodynamic parameters of each sampling point, generating an integrated feature set.

[0009] Based on the integrated feature set, and according to spatial distance and feature similarity, the ice structure region is automatically divided, and the boundary labels of consistent and abnormal regions are output.

[0010] By utilizing feature data within a consistent region, the evolution of mechanical properties under different compression rates and temperature gradients is simulated, generating a dynamic impact assessment map.

[0011] Anomalies in the compressed response are identified from the abnormal regions. The risk index of each anomaly is calculated by combining the energy distribution of the acoustic emission signal and the temperature difference change of the thermal infrared image, forming a set of high-risk anomalies.

[0012] Failure mode identification is performed on the set of high-risk anomalies, and a mechanical performance test and risk warning report for ice structures is generated based on preset mechanical performance thresholds and acoustic feature thresholds.

[0013] Preferably, the step of constructing the multimodal test dataset includes:

[0014] An acoustic sensor array and a thermal infrared camera were deployed around the ice structure sample to synchronously record acoustic emission signals and thermal infrared images at a fixed sampling frequency during the compression experiment.

[0015] The surface point cloud data of the ice structure sample was acquired by a 3D laser scanner and synchronized with the time encoder of the compressor to generate a time series of 3D spatial coordinates.

[0016] Wavelet transform processing is performed on the acoustic emission signal to extract frequency domain feature energy values;

[0017] Pixel-level temperature calibration is performed on thermal infrared images to generate a temperature distribution matrix;

[0018] The three-dimensional spatial coordinates, time series data, acoustic feature energy values, and temperature distribution matrix are aligned according to time points and fused into a multimodal test dataset.

[0019] Preferably, the step of generating the integrated feature set includes:

[0020] Mechanical performance indicators, including compressive modulus and yield strength, are extracted from each sampling point in the multimodal test dataset.

[0021] Calculate the short-time energy integral and spectral centroid of the acoustic emitted signal to form an acoustic feature vector;

[0022] The temperature gradient and heat flux density of each sampling point are extracted from the thermal infrared image as thermodynamic parameters.

[0023] The mechanical performance indicators, acoustic eigenvectors, and thermodynamic parameters are normalized to eliminate dimensional differences.

[0024] The normalized features are combined into a multi-dimensional feature vector according to the sampling point index to generate an integrated feature set.

[0025] Preferably, the step of labeling the boundaries of the output consistent regions and abnormal regions includes:

[0026] Calculate the spatial Euclidean distance and feature cosine similarity of each sampling point in the integrated feature set;

[0027] A density-based spatial clustering algorithm is used to group the sampling points using the minimum number of samples and the neighborhood radius as parameters;

[0028] Calculate the average silhouette coefficient for each cluster, and select high-consistency clusters with silhouette coefficients greater than a threshold as consistent regions.

[0029] The remaining low-consistency clusters are marked as anomalous regions, and polygon annotations of the region boundaries are generated;

[0030] Output the geometric center coordinates and characteristic statistics of the consistent and abnormal regions.

[0031] Preferably, the step of generating the dynamic impact assessment map includes:

[0032] Extract compression rate time series and temperature time series from the consistent region;

[0033] A partial differential equation model of compression dynamics is established, with compression rate and temperature as input variables and mechanical properties as output variables.

[0034] The mechanical property distribution under different compression rates and temperature combinations was simulated by solving partial differential equations using the finite element method.

[0035] Generate a compression rate sensitivity map to show the gradient of mechanical properties as a function of compression rate;

[0036] Generate a temperature-coupled response surface to show the mechanical response under the interaction of temperature and compression rate;

[0037] The sensitivity map and response surface are integrated into a dynamic impact assessment map.

[0038] Preferably, the step of forming the set of high-risk anomalies includes:

[0039] Extract the compressed response value of each sampling point from the abnormal region and calculate the deviation from the average value of the consistent region;

[0040] Analyze the energy distribution of acoustic emission signals to identify points where the energy peak exceeds the background noise level;

[0041] Calculate the temperature gradient from the thermal infrared image and filter out points where the rate of temperature change is greater than a threshold;

[0042] The comprehensive risk index for each point is calculated by combining the compression response deviation, acoustic energy peak, and temperature gradient.

[0043] Points with a comprehensive risk index greater than a preset risk threshold are included in the set of high-risk anomalies.

[0044] Record the three-dimensional coordinates, risk index, and timestamp of each anomaly.

[0045] Preferably, the steps for generating the risk warning report include:

[0046] Each point in the set of high-risk anomalies is classified into failure modes, including brittle fracture and plastic deformation modes;

[0047] Based on the critical values ​​of mechanical properties, determine whether each point has reached the failure condition;

[0048] The reliability of failure modes is verified by combining acoustic feature thresholds;

[0049] Generate a risk warning report, including a list of anomalies, failure mode classification, risk level, and a summary of spatial distribution;

[0050] The risk warning report is output in a machine-readable format for subsequent decision support.

[0051] Preferably, after generating the risk warning report, the following steps are also included:

[0052] Based on the set of high-risk anomalies in the risk warning report, the spatiotemporal characteristics and risk index of each anomaly are extracted, input into the adaptive control algorithm, and a compression experiment parameter adjustment instruction is generated.

[0053] According to the compression experiment parameter adjustment instructions, the loading rate of the compressor and the ambient temperature control device are adjusted in real time to make the compression process of the ice structure sample deviate from the high-risk state.

[0054] During the adjustment process, new multimodal test datasets are continuously collected, the integrated feature set is updated, and the risk index of outliers is recalculated to form a closed-loop control cycle.

[0055] Preferably, the step of generating the compression experiment parameter adjustment instruction includes:

[0056] Read the time series of risk indices from the set of high-risk outliers and calculate the risk change gradient;

[0057] Correlation analysis was performed between the risk change gradient and compression rate and temperature parameters to establish a parameter adjustment mapping table;

[0058] Using a fuzzy inference mechanism, the compression rate correction and temperature offset are output based on the current risk index and historical adjustment effects.

[0059] The compression rate correction and temperature offset are encapsulated as machine-readable instructions and sent to the compressor control system.

[0060] Preferably, the method further includes a long-term data assimilation process:

[0061] Regularly archive multimodal test datasets, integrated feature sets, and risk warning reports to build a historical experiment database;

[0062] Using spatiotemporally aligned data from a historical experimental database, a time series prediction model is trained to output the degradation trend of the mechanical properties of ice structures.

[0063] The predicted degradation trend is fed back into the compression dynamics simulation model to correct the model parameters and improve the prediction accuracy of the dynamic impact assessment map.

[0064] After each new experiment, the historical experiment database and prediction model are automatically updated, forming a continuous learning mechanism.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] By spatiotemporally aligning and fusing mechanical performance indicators, acoustic feature vectors, and thermodynamic parameters, an integrated feature set is generated, and consistent and anomalous regions are automatically divided based on spatial distance and feature similarity. This technical solution can automatically identify uniform regions with consistent mechanical behavior and localized regions with anomalous responses within ice structures from multi-dimensional data, and output their precise boundaries. Compared to methods that rely on a single parameter or manual experience to delineate regions, this solution objectively reveals the true distribution of internal material properties, allowing the analysis focus to be precisely on potential weak points, thus improving the accuracy and automation level of material inhomogeneity identification.

[0067] This approach identifies anomalous points in the compression response within anomaly regions and calculates a risk index for each anomalous point by integrating the energy distribution characteristics of acoustic emission signals with temperature difference changes from thermal infrared images. This technical solution correlates acoustic energy, characterizing the intensity of energy release, with changes in the thermal field reflecting local energy dissipation density, constructing a dynamic risk quantification model. This allows the assessment to move beyond static threshold judgments and capture the dynamic evolution of the response characteristics of anomalous points under load. The calculation of the risk index integrates acoustic precursors and thermodynamic responses to crack propagation, enabling refined ranking and early warning of the hazard level of anomalous points, providing more reliable and advanced decision-making information for targeted measures. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating the working principle of the dynamic compressive mechanical property testing method based on ice structure described in this invention.

[0069] Figure 2 A flowchart for generating an integrated feature set;

[0070] Figure 3 A flowchart for outputting boundary annotations for consistent and abnormal regions;

[0071] Figure 4 An analysis chart is generated for the set of high-risk anomalies;

[0072] Figure 5 This is a closed-loop control cyclic analysis diagram. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1 This invention provides a method for testing the dynamic compressive mechanical properties of ice structures. The method includes: deployment and data acquisition of a multi-sensor array, construction of a multimodal test dataset, spatiotemporal alignment and feature extraction, region division and boundary labeling, mechanical property evolution simulation, anomaly risk calculation, and risk warning report generation. The multi-sensor array consists of acoustic sensors, a thermal infrared camera, and a 3D laser scanner, arranged around the ice structure sample. During the compression experiment, it synchronously acquires 3D spatial coordinates, time-series data, acoustic emission signals, and thermal infrared image data at a fixed sampling frequency. After wavelet transform and temperature calibration, the acquired data is aligned according to time points to form a multimodal test dataset. Mechanical performance indicators, acoustic feature vectors, and thermodynamic parameters for each sampling point are extracted from the dataset and normalized to generate an integrated feature set. Based on spatial distance and feature similarity, a clustering algorithm is used to divide the ice structure region into consistent and anomalous regions, outputting boundary labels. Feature data within the consistent region is used to simulate the evolution of mechanical properties under different compression rates and temperature gradients, generating a dynamic impact assessment map. Sampling points in the abnormal region are used to calculate a risk index based on compression response deviation, acoustic energy distribution, and temperature difference changes, forming a set of high-risk anomalies. Failure modes are identified for these high-risk points based on mechanical performance thresholds and acoustic characteristic thresholds, generating a machine-readable risk warning report.

[0075] Example 1: See Figure 2In practical implementation, the multimodal test dataset based on the dynamic compression mechanical performance testing method of ice structures is constructed by deploying an acoustic sensor array and a thermal infrared camera around the ice structure sample. The acoustic sensor array uses piezoelectric sensor elements fixed on the experimental support in a linear or ring layout to ensure that the sensor probe maintains a constant distance from the surface of the ice structure sample. The thermal infrared camera is installed in front of or to the side of the sample, with a field of view covering the entire compression area. During the compression experiment, the acoustic emission signal and thermal infrared image are recorded synchronously at a fixed sampling frequency. The fixed sampling frequency is set according to the experimental requirements. The acoustic sampling frequency is usually in the range of 100kHz to 1MHz, and the thermal infrared image acquisition frequency is 30 frames / second to 100 frames / second. The synchronization mechanism is realized through a hardware trigger signal. All sensor devices are connected to a central synchronization controller, and the controller sends out a unified timestamp signal to ensure the time consistency of data acquisition. A 3D laser scanner is placed above or to the side of the sample. The scanner emits a laser beam to scan the surface of the ice structure sample, acquiring high-precision point cloud data at a rate of 1000 to 10000 points per second. This point cloud data contains 3D spatial coordinate information. The 3D laser scanner is synchronized with the compressor's time encoder via cable or wireless protocol. The time encoder records compression displacement and load data, generating a time-series 3D spatial coordinate data. This time-series data is stored with millisecond-level precision, and each data point is associated with an absolute timestamp. The acoustic emission signal undergoes wavelet transform processing using the Daubechies wavelet basis function, with a decomposition level of 5 to 10 layers. Frequency domain feature energy values ​​are extracted, including energy integrals from the low-frequency, mid-frequency, and high-frequency bands. The energy distribution within each time window is calculated. Pixel-level temperature calibration is performed using thermal infrared images. A blackbody radiation source is placed beside the sample, and multiple reference images are acquired. A linear interpolation algorithm is used to correct the temperature value of each pixel, generating a temperature distribution matrix. The matrix dimensions are consistent with the image resolution, and each element represents the temperature value of a specific point on the sample surface. Three-dimensional spatial coordinates, time series data, acoustic feature energy values, and temperature distribution matrices are aligned according to time points. The alignment algorithm is based on timestamp matching and uses the nearest neighbor interpolation method to handle time deviations. They are then fused into a multimodal test dataset, which is stored in a structured format and includes time index, spatial coordinate vector, acoustic energy array, and temperature matrix fields.

[0076] In some embodiments, the generation of the integrated feature set involves extracting mechanical performance indicators for each sampling point from a multimodal test dataset. These indicators include compressive modulus and yield strength. The compressive modulus is calculated using a stress-strain curve, with stress data derived from a compressor load sensor and strain data from displacement differential analysis of a 3D laser scanner. The yield strength is determined based on inflection point identification of the stress-strain curve, using derivative analysis to identify the onset of plastic deformation. The short-time energy integral of the acoustic emission signal is calculated using a Hamming window function with a window length of 10 to 50 milliseconds and an overlap rate of 50%. The spectral centroid is derived using a Fast Fourier Transform (FFT) with 1024 or 2048 points, forming an acoustic feature vector. The acoustic feature vector dimensions include the short-time energy integral value, spectral centroid frequency, spectral bandwidth, and spectral roll-off point. The temperature gradient and heat flux density for each sampling point are extracted from the thermal infrared image. The temperature gradient is calculated using the spatial difference method, with the temperature distribution matrix processed using the Sobel operator or central difference method. The heat flux density is derived based on Fourier's law of thermal conductivity, considering the thermal conductivity parameter of ice materials as a thermodynamic parameter. Mechanical performance indicators, acoustic feature vectors, and thermodynamic parameters are normalized using a min-max scaling algorithm. This algorithm linearly transforms the values ​​of each feature dimension to the [0,1] interval, eliminating dimensional differences. The normalization parameters are statistically derived from the training data. The normalized features are then combined into multi-dimensional feature vectors based on their sampling point indices, which correspond to the time series. Each feature vector includes fields such as compressibility modulus, yield strength, short-time energy integral, spectral centroid, temperature gradient, and heat flux density, generating an integrated feature set. This integrated feature set is stored in matrix form, with the number of rows corresponding to the number of sampling points and the number of columns corresponding to the number of feature dimensions, allowing for direct access by subsequent machine learning algorithms.

[0077] It is understandable that the deployment of the acoustic sensor array during the construction of the multimodal test dataset needs to consider environmental noise suppression. The acoustic sensor array is connected to the data acquisition card using shielded cables. The data acquisition card is equipped with a bandpass filter to filter out low-frequency mechanical vibrations and high-frequency electromagnetic interference. Temperature calibration of the thermal infrared camera needs to be performed regularly, with the calibration cycle being before each experiment. A standard temperature source is used to verify the calibration curve to ensure the accuracy of the temperature distribution matrix. The preprocessing of the point cloud data from the 3D laser scanner includes denoising and registration steps. The denoising algorithm uses statistical outlier removal, and the registration uses an iterative nearest-neighbor algorithm to align the point clouds at different time points to the same coordinate system. After extracting the feature energy values ​​in the frequency domain, the wavelet transform processing of the acoustic emission signal also needs to perform threshold denoising. The threshold is adaptively set based on the background noise level to retain significant event signals. Pixel-level temperature calibration of the thermal infrared image involves emissivity correction. The emissivity value is set to the range of 0.96 to 0.98 according to the ice material properties. After the temperature distribution matrix is ​​generated, spatial interpolation is performed to fill missing pixels. The interpolation method is bilinear interpolation or nearest-neighbor interpolation. During the alignment of the multimodal test dataset, the timestamp matching algorithm needs to handle clock drift between sensors and use linear regression to correct small time deviations. The fused dataset contains metadata such as experimental condition annotations and device parameter records.

[0078] Optionally, the extraction of mechanical performance indicators during the generation of the integrated feature set can be combined with real-time load data. Load data is read from the compressor's built-in sensors, with the sampling frequency synchronized with the acoustic signal. The calculation of compressive modulus and yield strength values ​​uses a moving average filter to smooth the data, with a window size of 5 to 10 sampling points. The short-time energy integral calculation of the acoustic feature vectors can be extended to multi-channel analysis. Signals from multiple channels of the acoustic sensor array are processed separately, and the average value is taken as the final feature. The calculation of the spectral centroid requires normalization of the frequency axis, expressed in Hertz. The temperature gradient calculation of thermodynamic parameters can incorporate a time dimension to calculate the instantaneous temperature change rate. The derivation of heat flux density needs to consider the thickness and boundary conditions of the ice sample, using the finite difference method to approximate the heat conduction equation. After combining the normalized feature vectors, a feature selection step is performed. The feature selection method is based on variance thresholding or correlation analysis, removing redundant features with low variance or high correlation to improve the generalization ability of the integrated feature set. The integrated feature set storage format supports multiple data interfaces, such as CSV or binary format, facilitating subsequent spatiotemporal alignment processing algorithms for reading and processing.

[0079] In some embodiments, the synchronous acquisition stage of the multimodal test dataset construction can employ a high-precision clock source, such as a GPS clock or atomic clock, to ensure that the time synchronization error of the acoustic sensor array, thermal infrared camera, and 3D laser scanner is less than 1 millisecond. Gain control is added during acoustic emission signal recording, and the gain value is dynamically adjusted according to the signal strength to avoid saturation or distortion. Thermal infrared image acquisition requires control of the influence of environmental thermal radiation. The inner wall of the experimental chamber is covered with low-emissivity materials to reduce reflection interference. After the temperature distribution matrix is ​​generated, flat field correction is performed, and reference images are used to eliminate the camera's own thermal noise. When the point cloud data of the 3D laser scanner is synchronized with the time encoder, the encoder outputs a pulse signal to trigger the scan. Time tags are added to the point cloud data, and the 3D spatial coordinates of the time series are stored in array form, with each point containing X, Y, Z coordinates and a timestamp. After extracting the acoustic feature energy values, event detection is performed. The detection algorithm is based on energy thresholds or pattern matching to identify significant acoustic events, and event parameters such as duration and energy peak are stored in the dataset. The fusion step of the multimodal test dataset includes data compression. The compression algorithm uses lossless encoding such as ZIP or LZ77 to reduce storage space while preserving the integrity of the original data.

[0080] Example 2: See Figure 3 In practical implementation, the boundary labeling output of consistent and abnormal regions is based on the spatial Euclidean distance and feature cosine similarity of each sampling point in the integrated feature set. The spatial Euclidean distance is calculated through 3D coordinate difference, and the 3D coordinates of each sampling point come from the point cloud data of the 3D laser scanner. The formula for calculating the spatial Euclidean distance is the square root of the sum of the squares of the differences between the coordinates of two sampling points. The feature cosine similarity measures the directional consistency of the feature vectors of two sampling points in multidimensional space, and is obtained by dividing the vector dot product by the product of the vector magnitudes. A density-based spatial clustering algorithm is used to group the sampling points. The density-based spatial clustering algorithm uses the minimum number of samples and the neighborhood radius as input parameters. The minimum number of samples is set to 5, representing the minimum number of points required in the neighborhood of a core point. The neighborhood radius is set to 0.1 meters, defining the neighborhood range of the sampling point. The density-based spatial clustering algorithm is implemented using the DBSCAN algorithm. The DBSCAN algorithm can identify clusters of arbitrary shapes and effectively distinguish noise points. The DBSCAN algorithm processing includes finding core points, connecting density-reachable points, and forming the final cluster. The average silhouette coefficient for each cluster is obtained by calculating the ratio of within-group distance to between-group distance. The formula for the average silhouette coefficient is:

[0081] ,

[0082] in: This represents the contour coefficient of sampling point i. This represents the average distance from sampling point i to other points in the same cluster. This represents the average distance from sampling point i to all points in the nearest heterogeneous cluster.

[0083] Clusters with a silhouette coefficient greater than 0.7 are selected as consistent regions. The silhouette coefficient threshold of 0.7 is determined based on historical experimental data analysis to ensure high consistency within each cluster. The remaining low-consistency clusters are marked as anomalous regions, which may contain boundary points or noise points. The boundaries of anomalous regions are labeled using a convex hull algorithm to generate polygonal boundaries. This algorithm employs either the Graham scan method or the Jarvis step method to calculate the convex hull shape of the outer points of the cluster, generating polygonal labels for the region boundaries. The geometric center coordinates and feature statistics of both consistent and anomalous regions are output. The geometric center coordinates are obtained by calculating the average of the coordinates of all points within the cluster. The feature statistics include the mean, variance, maximum, and minimum values ​​of the feature values, stored in a structured format.

[0084] In some embodiments, the generation of the dynamic impact assessment map extracts compression rate time series and temperature time series from a consistent region. The compression rate time series is derived from the first derivative calculation of the compressor displacement sensor, and the temperature time series is derived from the regional average temperature value of the thermal infrared image. A partial differential equation model of compression kinetics is established. This model is based on viscoelastic constitutive relations, using compression rate and temperature as input variables and mechanical properties as output variables. The general form of the compression kinetics partial differential equation includes stress balance equations and energy conservation equations. The partial differential equations are solved using the finite element method. The finite element method uses tetrahedral elements for spatial discretization and implicit Euler method for time discretization to simulate the distribution of mechanical properties under different combinations of compression rates and temperatures. The mechanical property distribution includes stress field, strain field, and damage variable field. A compression rate sensitivity map is generated, showing the gradient of mechanical properties with compression rate. The gradient is calculated using partial derivatives, and the magnitude of the gradient is visualized using color mapping. A temperature-coupled response surface is generated, showing the mechanical response under the interaction of temperature and compression rate. The surface is constructed through multivariate regression analysis, and the regression model includes quadratic interaction terms. The compression rate sensitivity map and temperature coupling response surface are integrated into a dynamic impact assessment map. The integration process uses layer overlay technology to maintain the consistency of the coordinate axes and supports three-dimensional rotation and slice viewing functions.

[0085] It is understandable that the spatial Euclidean distance calculation during the partitioning of consistent and anomalous regions requires consideration of coordinate unit unification; all three-dimensional coordinates are converted to the International System of Units (SI) meters. Before calculating the feature cosine similarity, the feature vectors need to be normalized to eliminate the influence of feature dimensions. The parameter settings of the density-based spatial clustering algorithm need to be adjusted according to the dataset size; for large datasets, the neighborhood radius can be appropriately increased to ensure cluster integrity. The calculation of the average silhouette coefficient requires a sufficient number of sampling points; small-scale clustering may produce unreliable silhouette coefficient values, therefore a minimum cluster point threshold is set. When generating polygonal boundaries using the convex hull algorithm, concave polygons may be produced, requiring subsequent processing to convert them to convex polygons to ensure the geometric rationality of the boundary labels. The geometric center coordinates of consistent and anomalous regions are output in vector format for easy integration into GIS systems, and the feature statistics include confidence interval information to improve the reliability of the results.

[0086] In some embodiments, the partial differential equation model of compression kinetics for dynamic impact assessment maps requires the determination of material parameters, which are calibrated through standard mechanical experiments, including elastic modulus, Poisson's ratio, and coefficient of thermal expansion. When solving using the finite element method, boundary conditions need to be set, determined based on the actual constraints of the experimental setup, such as fixed supports or sliding boundaries. The gradient calculation of the compression rate sensitivity map uses the central difference method to ensure numerical accuracy, and the gradient values ​​are normalized and mapped to the chromatographic range. The multiple regression analysis of the temperature-coupled response surface needs to verify the model's significance; analysis of variance is used to check the validity of the regression terms, and insignificant terms are removed. The integration of dynamic impact assessment maps requires consideration of data volume optimization, employing multi-level detail techniques to achieve real-time rendering of large-scale data, and the map output supports various interactive operations such as zooming and panning.

[0087] Optionally, the segmentation of consistent and anomalous regions can incorporate multi-scale clustering analysis, applying density-based spatial clustering algorithms at different spatial scales to detect feature regions of varying sizes. The generation of dynamic impact assessment maps can be extended to multiphysics coupling analysis, incorporating the influence of humidity or chemical fields to enrich the assessment dimensions. Boundary labeling output can add topological descriptions, recording the adjacency relationships between regions and providing more information for spatial analysis. The temperature-coupled response surface can utilize neural networks instead of multiple regression, improving the fitting accuracy of complex nonlinear relationships. The dynamic impact assessment map can integrate time-varying animation functionality to demonstrate the evolution of mechanical properties over time.

[0088] In some embodiments, density-based spatial clustering algorithms can be optimized using spatial indexing when processing large-scale data, employing kd-trees or R-trees to accelerate neighborhood queries and improve computational efficiency. Silhouette coefficient calculation can be parallelized, with the silhouette coefficient of each sampling point calculated independently and then aggregated. Anomaly region boundary labeling can increase uncertainty quantification, using Monte Carlo methods to assess the confidence interval of boundary locations. Solving the compressed dynamic partial differential equations can employ adaptive meshing techniques, refining the mesh in regions with large gradients to improve accuracy. The color mapping scheme for the dynamic impact assessment map must consider colorblindness to ensure the readability of the visualization results.

[0089] Example 3: In specific implementation, the high-risk anomaly point set is formed by extracting the compression response value of each sampling point from the anomaly region. The compression response value includes real-time stress and strain data. The real-time stress data comes from the readings of the compressor load sensor, and the real-time strain data is calculated by displacement difference calculation using a 3D laser scanner. The deviation of the compression response value of each sampling point from the average value of the consistent region is calculated. The average value of the consistent region is obtained by calculating the arithmetic mean of the compression response values ​​of all sampling points in the consistent region. The deviation is calculated as an absolute percentage, with the formula being: deviation equals the sampling point value minus the average value of the consistent region, the absolute value is divided by the average value of the consistent region, and then multiplied by 100%. The energy distribution of the acoustic emission signal is analyzed through spectrum analysis. The time-domain signal is converted to the frequency domain using a fast Fourier transform to identify points where the energy peak exceeds the background noise level. The background noise level is statistically obtained from the acoustic signal in the no-load stage of the early stage of the experiment. Usually, the average value plus three times the standard deviation is taken as the threshold. The energy peak detection uses a local maximum algorithm to filter out points that exceed the threshold. The temperature gradient is calculated from the thermal infrared image. The temperature distribution matrix is ​​processed using the spatial difference method. The temperature difference between each pixel and its neighbors is calculated using the central difference method, and then divided by the spatial distance to obtain the gradient value. Points with a temperature change rate greater than a threshold of 0.5℃ / s are selected. This threshold is based on the heat capacity and thermal conductivity parameters of the ice material. Combining the compression response deviation, acoustic energy peak, and temperature gradient, a comprehensive risk index is calculated for each point. The comprehensive risk index is calculated using a weighted summation formula:

[0090] ,

[0091] in: This represents the overall risk index. This represents the normalized value indicating the compression response deviation. The normalized value representing the peak acoustic energy. This represents the normalized value of the temperature gradient. The weighting factor for the compression response deviation is set to 0.5. The weighting factor for the peak acoustic energy is set to 0.3. The weighting coefficient for the temperature gradient is set to 0.2, and this coefficient is allocated based on expert experience to ensure the balance of the risk index. Points with a comprehensive risk index greater than a preset risk threshold of 0.8 are included in the high-risk anomaly set. The risk threshold is adjusted based on the safety standards of ice structures and historical failure data. The three-dimensional coordinates, risk index, and timestamp of each anomaly are recorded. The three-dimensional coordinates are directly obtained from the point cloud data, the risk index is stored as a floating-point number, and the timestamp is synchronized with the global clock of the compression experiment. The high-risk anomaly set is saved in tabular form, containing an index, coordinate values, risk values, and time information.

[0092] In some embodiments, the risk warning report is generated by classifying each point in the set of high-risk anomalies according to its failure mode. Failure modes include brittle fracture and plastic deformation. Brittle fracture is characterized by a sudden drop in stress and high-frequency acoustic emission signals, while plastic deformation is characterized by a stress plateau and low-frequency acoustic emission signals. The classification is based on shape analysis of the mechanical response curve and acoustic signal pattern recognition. The mechanical response curve is plotted using stress-strain data, identifying abrupt changes or plateaus in the curve. Acoustic signal patterns are distinguished by spectral characteristics such as dominant frequency and bandwidth. Each point is judged to have reached the failure condition based on critical mechanical performance values, including the compressive strength limit and the strain limit. The compressive strength limit is obtained from a material database, with a typical value being the maximum stress value of ice at a specific temperature. The strain limit is set based on the plastic deformation initiation point. The judgment logic is that if the stress or strain at a sampling point exceeds 80% of the critical value, it is marked as high-risk. The credibility of failure modes is verified by combining acoustic feature thresholds. These thresholds include peak energy thresholds and frequency thresholds. The peak energy threshold is set to five times the background noise level. The frequency threshold is set to the lower limit of the high-frequency band (e.g., 50kHz) for brittle fracture and the upper limit of the low-frequency band (e.g., 20kHz) for plastic deformation. The verification process checks whether the acoustic signal simultaneously meets the mode and threshold conditions. A risk warning report is generated, containing a list of anomalies, failure mode classifications, risk levels, and a spatial distribution summary. The anomaly list is sorted in descending order by the comprehensive risk index. Each point includes detailed information such as coordinates and a timestamp. Failure mode classifications are identified by tags. Risk levels are divided into high, medium, and low based on the comprehensive risk index. The spatial distribution summary describes the clustering of anomalies through cluster analysis. The report output is in a machine-readable format, such as XML or JSON, and includes metadata such as the experiment number and generation time for integration into decision support systems.

[0093] It is understandable that the calculation of compression response deviation during the formation of the high-risk anomaly set requires consideration of data smoothing. A moving average filter is used to preprocess the original stress-strain data, with a window size of 5 sampling points to reduce noise impact. The identification of acoustic energy peaks must avoid false peaks, employing dual verification of duration and amplitude, with a duration threshold set to at least 1 millisecond. When calculating the temperature gradient, the spatial resolution limitations of the thermal infrared image must be addressed, using bilinear interpolation to improve gradient accuracy and ensure the reliability of the rate of change calculation. The weighting coefficients of the comprehensive risk index can be adjusted according to specific application scenarios; for example, the weight of the temperature gradient can be increased in high-temperature environments, while the core weight range remains stable. The storage format of the high-risk anomaly set supports fast querying, and the index is built on coordinates and timestamps for convenient subsequent real-time analysis.

[0094] In some embodiments, failure mode classification can be aided by machine learning algorithms. The classification model is trained using labeled samples from historical data, with features including mechanical response curve parameters and acoustic spectral characteristics. However, the underlying classification remains rule-based to ensure interpretability. The setting of mechanical performance thresholds needs to consider temperature compensation, as the strength of ice varies with temperature; interpolation functions are used to adjust the thresholds based on real-time temperature. Verification of acoustic feature thresholds can incorporate multi-channel consistency checks; if multiple acoustic sensors simultaneously detect anomalous signals, credibility is increased. Spatial distribution summaries of risk warning reports can generate heatmap visualizations, drawn based on anomaly density, and integrated into the report attachments. Machine-readable report definition modes (XSD) or JSON formats ensure standardized data exchange.

[0095] Optionally, the formation of high-risk anomaly sets can be extended to multi-timescale analysis, considering not only instantaneous risk but also risk change trends. These trends are calculated using a sliding window to determine the derivative of the risk index. The failure mode classification in the risk warning report can be expanded to include subcategories, such as crack propagation stages in brittle fracture, providing a more refined risk assessment. The report output format supports real-time streaming, pushing data to the monitoring interface via the WebSocket protocol for immediate alerts.

[0096] In some embodiments, the compression response deviation can be calculated as a relative deviation rather than an absolute percentage. The relative deviation is standardized based on the standard deviation of a consistent region, improving sensitivity to outliers. The background noise level of the acoustic energy peak can be dynamically updated, with the threshold adjusted in real time according to changes in the experimental environment. The temperature gradient threshold can be set in zones, as different parts of the ice structure may have different thermal sensitivities; the zoning is based on the thermal uniformity of the consistent region. The normalization process in the comprehensive risk index formula uses the Z-score method, ensuring that each feature value has zero mean and unit variance. Records of high-risk outlier sets can be supplemented with environmental parameters such as humidity to expand the dimensions of risk assessment.

[0097] See Figure 4 This figure illustrates the spatial distribution of high-risk anomalies in an ice structure during dynamic compression. The figure uses color gradients to represent the comprehensive risk index at different locations, with red markers indicating high-risk anomaly areas exceeding safety thresholds. The figure includes the fusion analysis results of three key parameters: compression response deviation, acoustic emission energy, and temperature gradient. Each data point represents a monitoring location on the ice structure sample, with its color intensity reflecting the comprehensive risk level at that location. High-risk areas typically exhibit characteristics such as stress concentration, acoustic energy anomalies, and drastic temperature changes. This data is crucial for identifying potential failure areas in ice structures and provides quantitative evidence for engineering safety assessments. Spatial distribution pattern analysis can identify weak points and potential crack propagation paths in the structure, providing data support for preventative maintenance and structural optimization.

[0098] Example 4: In specific implementation, the closed-loop control cycle based on the risk warning report is initiated by extracting the spatiotemporal features and risk index from the set of high-risk anomalies. The set of high-risk anomalies comes from the output of the risk warning report. The spatiotemporal features of each anomaly include three-dimensional coordinates, a timestamp, and a risk value. The risk index is a floating-point value of the comprehensive risk index calculation result. These data are input into the adaptive control algorithm, which is designed using the proportional-integral-derivative (PID) controller principle. The input variable of the PID controller is the deviation between the risk index and the target risk threshold, and the output variable is the compression experiment parameter adjustment command. The compression experiment parameter adjustment command adjusts the compressor loading rate and the ambient temperature control device in real time. The loading rate adjustment is achieved through a servo motor control system. The servo motor receives digital commands to change the compression displacement speed. The ambient temperature control device uses a Peltier element or a liquid nitrogen injection system. The Peltier element controls the heating or cooling power through current, causing the compression process of the ice structure sample to deviate from the high-risk state. During the adjustment process, the multi-sensor array continuously collects new multimodal test datasets. The acoustic sensor array, thermal infrared camera, and 3D laser scanner work synchronously at a fixed sampling frequency to update the integrated feature set. The update of the integrated feature set includes recalculating mechanical performance indicators, acoustic feature vectors, and thermodynamic parameters, as well as recalculating the risk index of anomalies, forming a closed-loop control cycle. The period of the closed-loop control cycle is set to 1 second to ensure the real-time performance and stability of the control system.

[0099] Referring to Table 1, the generation of compression experiment parameter adjustment instructions involves reading the risk index time series from the high-risk anomaly point set. This risk index time series is processed using sliding window analysis, with the window size set to 10 historical data points. The risk change gradient is calculated using the first-order difference method to determine the difference between the current risk index and the previous risk index. The risk change gradient is then correlated with compression rate and temperature parameters. The correlation analysis method uses the Pearson correlation coefficient to calculate the linear correlation between the risk change gradient and compression rate, as well as the linear correlation between the risk change gradient and temperature parameters. A parameter adjustment mapping table is established. This table is a two-dimensional lookup table where the row index represents the discrete value of the risk change gradient, and the column index represents the current combination of compression rate and temperature. The table content includes the recommended compression rate correction and temperature offset. A fuzzy inference mechanism is employed to output compression rate correction and temperature offset based on the current risk index and historical adjustment effects. The input variables of the fuzzy inference mechanism include the current risk index, risk change gradient, and historical adjustment effect score. The historical adjustment effect score is calculated based on the degree of risk index reduction over a past period. The output variables of the fuzzy inference mechanism are fuzzy sets of compression rate correction and temperature offset, which are converted to precise values ​​through defuzzification. The compression rate correction and temperature offset are encapsulated into machine-readable instructions, with the instruction format conforming to the CAN bus protocol or Modbus protocol, containing target parameter values ​​and timestamp information, and sent to the compressor control system. The compressor control system parses the instructions and drives the actuators to operate.

[0100] Table 1: Parameter Adjustment Mapping Table

[0101] ,

[0102] In some embodiments, the adaptive control algorithm of the closed-loop control cycle can incorporate a feedforward control loop. This feedforward control loop predicts parameter adjustments based on the initial attributes of the ice structure sample, including ice density and bubble content. However, the core control still relies primarily on feedback from the risk index. The construction of the parameter adjustment mapping table can be optimized using machine learning methods. Reinforcement learning algorithms are used to update the mapping table content based on historical control effects, improving control accuracy. The membership function of the fuzzy inference mechanism can be a triangular function or a Gaussian function. The fuzzification interval of the input variables is determined statistically based on experimental data, and the defuzzification method for the output variables uses the centroid method. Verification mechanisms, such as cyclic redundancy check codes, can be added to the transmission of machine-readable instructions to ensure the reliability of instruction transmission. The response time of the compressor control system must be less than the closed-loop control cycle period to avoid control delay.

[0103] See Figure 5This figure presents a complete monitoring view of the closed-loop control process based on risk warning. Three curves of different colors represent the changing trends of the comprehensive risk index, compression rate, and ambient temperature, respectively. A dual Y-axis system clearly illustrates the dynamic relationships between these parameters. During the control process, when the risk index exceeds a preset high-risk threshold, the system automatically adjusts the compression rate and ambient temperature parameters to deviate the compression process of the ice structure from the high-risk state. The figure illustrates the complete closed-loop process of risk monitoring, threshold judgment, and control response, demonstrating the system's adaptive control capability. Changes in control parameters reflect the system's dynamic response mechanism to risk states. Adjustments to the compression rate directly affect the kinetic characteristics of the loading process, while temperature control influences the structural response by altering the thermodynamic properties of the material. This multi-parameter collaborative control strategy ensures the safety and stability of the testing process, providing a reliable experimental platform for the study of the mechanical properties of ice structures.

[0104] Example 5: In specific implementation, the long-term data assimilation process periodically archives multimodal test datasets, integrated feature sets, and risk warning reports to build a historical experimental database. The archiving operation is automatically triggered after each compression experiment. The multimodal test dataset contains the original time-series data collected by the sensors, the integrated feature set contains processed normalized feature vectors, and the risk warning report contains machine-readable structured evaluation results. The historical experimental database is implemented using a relational database management system. The database table structure includes fields such as experiment number, timestamp, original sensor data, feature vectors, and risk indicators. The archiving process is automatically executed through a data pipeline. After verifying the data integrity, the data pipeline performs a write operation. A time series prediction model is trained using spatiotemporally aligned data from a historical experimental database. Spatiotemporally aligned data refers to multimodal data sequences with a unified time index. The time series prediction model employs a Long Short-Term Memory (LSTM) neural network architecture. The input features of the LSM include historical compression rate sequences, temperature sequences, mechanical property sequences, and risk index sequences. The output variable is the degradation trend of the ice structure's mechanical properties at future time steps, represented by the rate of change of the elastic modulus over time. The training process uses 90% of the data from the historical experimental database as the training set and the remaining 10% as the validation set. The Adam algorithm is used for optimization, with the mean squared error as the loss function. The predicted degradation trend is fed back to the compression dynamics simulation model, a partial differential equation model based on physical laws. The feedback data is input through a parameter correction interface. The correction model parameters include the viscoelastic coefficient and thermal expansion coefficient of ice. The correction method uses the Kalman filter algorithm, which integrates the predicted degradation trend as an observation with the simulation model's predictions to update the model parameter state estimates and improve the prediction accuracy of the dynamic impact assessment map. After each new experiment, the historical experiment database and prediction model are automatically updated. The update process is implemented through a script program. The script program calls the database stored procedure to add new records and triggers the model retraining process, forming a continuous learning mechanism. The continuous learning mechanism ensures that the system continuously optimizes its performance as data accumulates.

[0105] In some embodiments, the archiving strategy for the historical experimental database can be configured with a retention strategy, such as retaining only the most recent 100 experimental data to control storage size, while key metadata is permanently stored. The long short-term memory neural network of the time series prediction model can be designed as a multi-layer encoder-decoder structure, where the encoder processes historical sequences and the decoder outputs multi-step prediction results. Parameter correction of the compression dynamics simulation model can introduce uncertainty quantification, and the Kalman filter algorithm outputs the covariance matrix of parameter estimates to evaluate the reliability of prediction accuracy. The automatic update process can be set with trigger conditions, such as updating immediately when new experimental data differs significantly from historical patterns, otherwise updating periodically according to a plan. The continuous learning mechanism can include model performance monitoring, issuing an alarm when the prediction error exceeds a threshold, prompting manual inspection.

[0106] Optionally, the long-term data assimilation process can be extended to a distributed database architecture, with historical experimental databases deployed across multiple nodes to improve data access efficiency. Time series prediction models can integrate attention mechanisms to enhance feature extraction capabilities at key time points. Parameter calibration of the compression dynamics simulation model can be combined with Bayesian inference, leveraging prior distributions to improve calibration robustness with small datasets. The automatic update process supports incremental learning modes, with long short-term memory neural networks employing online learning algorithms to avoid the computational overhead of full retraining. The continuous learning mechanism can record model version history, supporting rollback to specific versions to address performance degradation. Archiving of historical experimental databases can add data encryption functionality to protect the intellectual property rights of experimental data. Early stopping strategies can be introduced into the training of time series prediction models to prevent overfitting and improve generalization ability. Parameter calibration of the compression dynamics simulation model can be extended to multi-model ensembles, calibrating multiple candidate models and then weighted averaging the prediction results. The automatic update process can be designed for asynchronous execution, with new experimental data cached and updated by background tasks, without blocking the real-time testing process. The continuous learning mechanism provides learning curve visualization to help users monitor the system's evolution process.

[0107] Optionally, the long-term data assimilation process can interface with other experimental databases to import publicly available ice mechanics experimental data, expanding the diversity of training data. The time-series prediction model can output prediction intervals, providing not only estimates of degradation trend points but also the range of uncertainty. Parameter calibration for the compression dynamics simulation model can be performed separately for different regions of the ice structure, considering material inhomogeneity. The automatic update process can include a data quality check module to automatically filter out abnormal experimental data and prevent model contamination. A continuous learning mechanism can support transfer learning, using trained model parameters as initialization values ​​for new models to accelerate convergence.

[0108] It is understandable that the historical experimental database for the long-term data assimilation process needs to be backed up regularly to prevent data loss, and the training cycle of the time series prediction model needs to be coordinated with the experimental frequency to avoid model lag. Parameter calibration of the compression dynamics simulation model needs to ensure numerical stability, and the covariance initialization of the Kalman filter algorithm affects the convergence speed. The scripts for the automatic update process need to be fault-tolerant to handle abnormal situations such as database connection interruptions. Long-term operation of the continuous learning mechanism requires monitoring of computing resource usage to avoid memory leaks or insufficient storage space. The effectiveness of the entire long-term data assimilation process depends on the degree of matching between data quality and model structure, and joint validation needs to be performed regularly.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for testing the dynamic compressive mechanical properties of ice structures based on, characterized by, The method comprises: Synchronously collecting three-dimensional spatial coordinates, time series data, acoustic emission signals and thermal infrared image data of the ice structure sample during dynamic compression by a multi-sensor array, and constructing a multi-modal test data set; Performing space-time alignment processing on the multi-modal test data set, extracting the mechanical performance index, acoustic feature vector and thermodynamic parameter of each sampling point, and generating an integrated feature set; Based on the integrated feature set, and according to the spatial distance and feature similarity, automatically dividing the ice structure region, and outputting the boundary label of the consistent region and the abnormal region; The step of outputting the boundary label of the consistent region and the abnormal region comprises: Calculating the spatial Euclidean distance and feature cosine similarity of each sampling point in the integrated feature set; Using a density-based spatial clustering algorithm, taking the minimum sample number and neighborhood radius as parameters, grouping the sampling points; Calculating the average profile coefficient of each cluster, and screening the high-consistency cluster with a profile coefficient greater than a threshold value as a consistent region; Marking the remaining low-consistency clusters as abnormal regions, and generating a region boundary polygon label; Outputting the geometric center coordinates and feature statistics of the consistent region and the abnormal region; Using the feature data in the consistent region, simulating the evolution law of the mechanical performance under different compression rates and temperature gradients, and generating a dynamic impact evaluation atlas; The step of generating a dynamic impact evaluation atlas comprises: Extracting the compression rate time series and temperature time series from the consistent region; Establishing a compression dynamics partial differential equation model, taking the compression rate and temperature as input variables, and the mechanical performance as output variables; Solving the partial differential equation by the finite element method to simulate the mechanical performance distribution under different compression rates and temperature combinations; Generating a compression rate sensitivity map to show the change gradient of the mechanical performance with the compression rate; Generating a temperature coupling response surface to show the mechanical response under the interaction of temperature and compression rate; Integrating the sensitivity map and the response surface into a dynamic impact evaluation atlas; Identifying compression response abnormal points from the abnormal region, calculating the risk index of each abnormal point by combining the energy distribution of the acoustic emission signal and the temperature difference change of the thermal infrared image, and forming a high-risk abnormal point set; Performing failure mode identification on the high-risk abnormal point set, generating an ice structure mechanical performance test and risk warning report based on the preset mechanical performance critical value and acoustic feature threshold.

2. The method for testing the dynamic compression mechanical properties of ice structures according to claim 1, characterized in that, The step of constructing a multi-modal test data set comprises: Deploying an acoustic sensor array and a thermal infrared camera around the ice structure sample, and synchronously recording acoustic emission signals and thermal infrared images at a fixed sampling frequency during the compression experiment; Acquiring surface point cloud data of the ice structure sample by a three-dimensional laser scanner, and synchronizing with a time encoder of the compressor to generate time series three-dimensional spatial coordinates; Performing wavelet transform processing on the acoustic emission signal to extract frequency domain feature energy values; Performing pixel-level temperature calibration on the thermal infrared image to generate a temperature distribution matrix; Aligning the three-dimensional spatial coordinates, time series data, acoustic feature energy values and temperature distribution matrix by time points, and fusing into a multi-modal test data set.

3. The method for testing the dynamic compression mechanical properties of ice structures according to claim 2, characterized in that, The step of generating an integrated feature set comprises: Extracting mechanical performance indicators, including compression modulus and yield strength, from each sampling point in the multi-modal test dataset; Calculating the short-time energy integral and spectral centroid of the acoustic emission signal to form an acoustic feature vector; Extracting temperature gradient and heat flux density from each sampling point in the thermal infrared image as thermodynamic parameters; Normalizing the mechanical performance indicators, acoustic feature vectors, and thermodynamic parameters to eliminate dimensional differences; Combining the normalized features into a multi-dimensional feature vector according to the sampling point index to generate an integrated feature set.

4. The method for testing the dynamic compression mechanical properties of ice structures according to claim 1, characterized in that, The step of forming a high-risk anomaly point set includes: Extracting the compression response value of each sampling point from the anomaly region and calculating the deviation from the average value of the consistent region; Analyzing the energy distribution of the acoustic emission signal to identify points with energy peaks exceeding background noise levels; Calculating the temperature difference gradient from the thermal infrared image and screening points with temperature difference change rates greater than a threshold value; Combining the compression response deviation, acoustic energy peak, and temperature difference gradient to calculate the comprehensive risk index of each point; Including points with a comprehensive risk index greater than a preset risk threshold in the high-risk anomaly point set; Recording the three-dimensional coordinates, risk index, and timestamp of each anomaly point.

5. The method for testing the dynamic compression mechanical properties of ice structures according to claim 4, characterized in that, The step of generating the risk warning report includes: Classifying each point in the high-risk anomaly point set into failure modes, including brittle fracture and plastic deformation modes; Determining whether each point meets the failure condition based on the mechanical performance threshold value; Verifying the credibility of the failure mode based on the acoustic feature threshold value; Generating a risk warning report containing an anomaly point list, failure mode classification, risk level, and spatial distribution summary; Outputting the risk warning report in a machine-readable format for subsequent decision support.

6. The method for testing the dynamic compression mechanical properties of ice structures according to claim 5, characterized in that, After generating the risk warning report, it also includes: Based on the high-risk anomaly point set in the risk warning report, extracting the spatio-temporal features and risk index of each anomaly point and inputting them into an adaptive control algorithm to generate a compression experiment parameter adjustment instruction; According to the compression experiment parameter adjustment instruction, adjusting the loading rate of the compressor and the environmental temperature control device in real time to make the compression process of the ice structure sample deviate from the high-risk state; During the adjustment process, continuously collecting new multi-modal test datasets, updating the integrated feature set, and recalculating the risk index of the anomaly points to form a closed-loop control cycle.

7. The method for testing the dynamic compression mechanical properties of ice structures according to claim 6, characterized in that, The step of generating the compression experiment parameter adjustment instruction includes: Reading the risk index time series from the high-risk anomaly point set and calculating the risk change gradient; Correlating the risk change gradient with the compression rate and temperature parameters to establish a parameter adjustment mapping table; Using a fuzzy inference mechanism to output the compression rate correction and temperature offset based on the current risk index and historical adjustment effects; Packaging the compression rate correction and temperature offset as machine-readable instructions and sending them to the compressor control system.

8. The method for testing the dynamic compression mechanical property of ice structure according to claim 1, characterized in that, It also includes a long-term data assimilation process: Periodically archiving multi-modal test datasets, integrated feature sets, and risk warning reports to build a historical experiment database; Using the spatio-temporal aligned data in the historical experiment database to train a time series prediction model and output the mechanical performance degradation trend of ice structures; The predicted degradation trend is fed back to the compression dynamics simulation model to correct the model parameters and improve the prediction accuracy of the dynamic impact evaluation atlas; After each new experiment, the historical experiment database and the prediction model are automatically updated to form a continuous learning mechanism.

Images