A method and related device for real-time and accurate monitoring of displacement field of multi-branch paths of concrete crack tip

Abstract

The application discloses a concrete crack tip multi-branch path displacement field real-time accurate monitoring method and related equipment, and relates to the technical field of concrete structure health monitoring, and specifically discloses the following technical scheme: according to linear elastic fracture mechanics, a cohesive force model and computer graphics, a virtual data set is generated; the virtual data set is input into an initial depth displacement calculation network, concrete micro-feature extraction and motion feature enhancement operations are performed on the virtual data set, and an enhanced feature map is obtained; through a deformable convolution layer, non-continuous displacement matching operations are performed on the enhanced feature map, and a coarse displacement field is obtained; sub-pixel level refinement processing is performed on the coarse displacement field, and a fine displacement field is obtained; according to the real displacement field and the fine displacement field, the initial depth displacement calculation network is optimized, and a target depth displacement calculation network is obtained; a material surface speckle image is input into the target depth displacement calculation network, and a crack tip deformation displacement field is obtained. The application can improve the accuracy and robustness of displacement field acquisition, and can be widely applied to the technical field of concrete structure health monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of concrete structure health monitoring, and in particular to a method for real-time and accurate monitoring of displacement fields of multiple branched paths of concrete crack tips and related equipment. BACKGROUND

[0002] In the entire life cycle of concrete structures such as bridges and dams, high-precision, real-time and robust monitoring of the dynamic development, expansion path and microscopic displacement of the tip of internal or surface cracks is the cornerstone of ensuring structural safety, warning failure risks and guiding maintenance decisions. However, the existing technology faces significant and widespread shortcomings in achieving this goal: 1) Existing crack monitoring methods either rely on costly and complex physical experiments or use general image datasets for deep learning training. However, physical experiments are difficult to stably reproduce extreme and critical complex fracture behaviors such as crack tip multi-branching and high strain concentration; while general image datasets cannot effectively represent the unique mesoscopic heterogeneity of concrete materials, resulting in the model failing to fully learn the visual features of real concrete cracks during training, which limits the accuracy and robustness of monitoring from the source. 2) Traditional image processing methods, such as digital image correlation (DIC), often require extremely high computational resources and time when pursuing sub-pixel level accuracy of crack tip micro-displacement, making it impossible to meet the real-time monitoring needs of structures under dynamic loading conditions. Although existing deep learning algorithms have advantages in speed, they are easily disturbed by background noise when facing complex textures on the concrete surface and low signal-to-noise ratio areas at the crack tip, leading to a decrease in prediction accuracy, especially making it difficult to accurately analyze micron-level displacement. SUMMARY

[0003] Therefore, the embodiments of the present application mainly aim to provide a method for real-time and accurate monitoring of displacement fields of multiple branched paths of concrete crack tips and related equipment, in order to solve at least one of the problems in the prior art, and the present application can improve the accuracy and robustness of displacement field acquisition.

[0004] To achieve the above-mentioned purpose, one aspect of the embodiments of the present application provides a method for real-time and accurate monitoring of displacement fields of multiple branched paths of concrete crack tips, which comprises:

[0005] Generating a virtual dataset according to linear elastic fracture mechanics, cohesive zone model and computer graphics;

[0006] Inputting the virtual dataset into an initial deep displacement calculation network, performing concrete mesoscopic feature extraction and motion feature enhancement operations on the virtual dataset to obtain an enhanced feature map;

[0007] Performing a non-continuous displacement matching operation on the enhanced feature map through a deformable convolution layer to obtain a coarse displacement field;

[0008] performing sub-pixel level refinement on the coarse displacement field to obtain a fine displacement field;

[0009] optimizing the initial depth displacement resolving network according to the real displacement field and the fine displacement field to obtain a target depth displacement resolving network;

[0010] inputting a material surface speckle image into the target depth displacement resolving network to obtain a crack tip deformation displacement field.

[0011] In some embodiments, a virtual data set is generated according to linear elastic fracture mechanics, a cohesive zone model, and computer graphics, including the following steps:

[0012] constructing a physical displacement field by fusing the linear elastic fracture mechanics and the cohesive zone model;

[0013] generating an artificial speckle image with concrete meso-features by the computer graphics;

[0014] generating the virtual data set according to the physical displacement field and the artificial speckle image.

[0015] In some embodiments, the concrete meso-feature extraction and motion feature enhancement operation on the virtual data set to obtain an enhanced feature map includes the following steps:

[0016] inputting the virtual data set into a multi-scale feature extractor to perform feature extraction operation on the virtual data set to obtain a composite feature map;

[0017] inputting the composite feature map into a space-channel collaborative attention module to perform secondary weighted optimization on the composite feature map to obtain the enhanced feature map.

[0018] In some embodiments, the inputting the virtual data set into a multi-scale feature extractor to perform feature extraction operation on the virtual data set to obtain a composite feature map includes the following steps:

[0019] extracting low-frequency structural features of the virtual data set by a large-scale convolution kernel of the multi-scale feature extractor;

[0020] extracting high-frequency detail features of the virtual data set by a small-scale convolution kernel of the multi-scale feature extractor;

[0021] obtaining the composite feature map according to the low-frequency structural features and the high-frequency detail features.

[0022] In some embodiments, the inputting the composite feature map into the space-channel collaborative attention module, performing secondary weighting optimization on the composite feature map, and obtaining the enhanced feature map, comprises the following steps:

[0023] inputting the composite feature map into a channel attention submodule, enhancing the motion feature channel of the composite feature map through channel attention weight, and suppressing the static feature channel of the composite feature map, to obtain a channel weighted feature map;

[0024] inputting the channel weighted feature map into a space attention submodule, enhancing the high gradient strain region weight of the channel weighted feature map through space attention weight, and suppressing the static region weight of the channel weighted feature map, to generate the enhanced feature map;

[0025] The space-channel collaborative attention module is formed in series by the channel attention submodule and the space attention submodule.

[0026] In some embodiments, the optimizing the initial depth displacement calculation network according to the real displacement field and the fine displacement field, and obtaining a target depth displacement calculation network, comprises the following steps:

[0027] constructing a multi-scale optical flow supervision loss according to the real displacement field and the fine displacement field;

[0028] constructing a fracture mechanics regularization loss according to the fine displacement field;

[0029] constructing a hybrid loss function according to the multi-scale optical flow supervision loss and the fracture mechanics regularization loss;

[0030] optimizing the initial depth displacement calculation network according to the hybrid loss function, and obtaining the target depth displacement calculation network.

[0031] In some embodiments, the formula used for optimizing the initial depth displacement calculation network according to the real displacement field and the fine displacement field, and obtaining a target depth displacement calculation network, comprises:

[0032] ;

[0033] ;

[0034] ;

[0035] In the formula, represents a hybrid loss function; represents a multi-scale optical flow supervision loss; represents a fracture mechanics regularization loss; a multi-scale optical flow supervision loss weight is represented; a fracture mechanics regularization loss weight is represented; a scale index is represented, ; a weight coefficient of each scale is represented; a predicted displacement field of the th scale is represented; a real displacement field of the th scale is represented; a spatial position is represented; a crack tip region is represented; a minimum displacement gradient of an initial depth displacement solving network in the crack tip region is represented; a predicted displacement gradient at a spatial position is represented; a Euclidean norm is represented; a Frobenius norm is represented.

[0036] To achieve the above object, another aspect of the embodiment of the present application provides a device for real-time accurate monitoring of a concrete crack tip multi-branch path displacement field, the device comprising:

[0037] a data generation module configured to generate a virtual data set according to linear elastic fracture mechanics, a cohesive force model and computer graphics;

[0038] a feature extraction and enhancement module configured to input the virtual data set into an initial depth displacement solving network, and perform concrete micro-feature extraction and motion feature enhancement operations on the virtual data set to obtain an enhanced feature map;

[0039] a non-continuous displacement matching module configured to perform a non-continuous displacement matching operation on the enhanced feature map through a deformable convolution layer to obtain a coarse displacement field;

[0040] a fine processing module configured to perform sub-pixel level fine processing on the coarse displacement field to obtain a fine displacement field;

[0041] a model training and optimization module configured to optimize the initial depth displacement solving network according to a real displacement field and the fine displacement field to obtain a target depth displacement solving network;

[0042] a real-time monitoring module configured to input a material surface speckle image into the target depth displacement solving network to obtain a crack tip deformation displacement field.

[0043] To achieve the above object, another aspect of the embodiment of the present application provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the method described above when executing the computer program.

[0044] To achieve the above object, another aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method described above.

[0045] To achieve the above object, another aspect of the embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the method described above.

[0046] The embodiment of the present application at least has the following beneficial effects: the present application provides a concrete crack tip multi-branch path displacement field real-time accurate monitoring method and related equipment, which generates a virtual data set according to linear elastic fracture mechanics, a cohesive force model and computer graphics, and provides a data basis for downstream tasks; inputs the virtual data set into an initial depth displacement solving network, performs concrete micro-feature extraction and motion feature enhancement operations on the virtual data set, obtains an enhanced feature map, fully extracts features, and reduces noise interference; performs non-continuous displacement matching operations on the enhanced feature map through a deformable convolution layer, realizes high robustness tracking of irregular crack paths, and obtains a coarse displacement field; performs sub-pixel level refinement processing on the coarse displacement field, reduces local analytical errors in the displacement field, and obtains a fine displacement field; optimizes the initial depth displacement solving network according to the real displacement field and the fine displacement field, improves the robustness of the network model, and obtains a target depth displacement solving network; inputs a material surface speckle image into the target depth displacement solving network, obtains a crack tip deformation displacement field, and improves the accuracy and robustness of displacement field acquisition. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0048] Figure 1 is a flow chart of the concrete crack tip multi-branch path displacement field real-time accurate monitoring method provided by the embodiment of the present application;

[0049] Figure 2 is a schematic diagram of the concrete crack monitoring system provided by the embodiment of the present application;

[0050] Figure 3is a schematic diagram of a deep optical flow network architecture provided by an embodiment of the present application.

[0051] Figure 4 is a schematic diagram of a horizontal displacement field comparison result provided by an embodiment of the present application.

[0052] Figure 5 is a schematic diagram of a hardware structure of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application, but are only examples of devices and methods consistent with some aspects of the embodiments of the present application as described in the appended claims.

[0054] It should be noted that although the functional modules are divided in the system schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a manner different from the module division in the system or the order in the flowchart. The terms "first / S100", "second / S200" in the specification and claims and the above-described drawings can be used in this paper to describe various concepts, but unless specifically stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "when" or "in response to determining".

[0055] The terms "at least one", "multiple", "each", "any" and the like used in the present application include one, two or more than two, multiple includes two or more than two, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0057] Before the embodiments of the present application are described in detail, first, some nouns and terms involved in the embodiments of the present application are described, and the nouns and terms involved in the embodiments of the present application are applicable to the following explanations.

[0058] Linear Elastic Fracture Mechanics (LEFM) is a branch of linear elastic mechanics, mainly used for analyzing the fracture problems of brittle materials. The theory assumes that the material is in an elastic state throughout the loading process, and describes the stress and strain fields at the crack tip through stress intensity factors and other methods.

[0059] Crack tip singular field refers to the stress and strain fields near the crack tip in an object containing a crack, which have singular distribution characteristics. This singular characteristic is mainly due to the very sharp geometry of the crack tip, which causes the stress and strain to change sharply near the tip.

[0060] Artificial speckle image refers to a random pattern generated on the surface of a test specimen by spraying, printing or other methods. These patterns are used to track the deformation and displacement of objects in digital image correlation technology.

[0061] Computer Graphics (CG) is a science that uses mathematical algorithms to convert two-dimensional or three-dimensional graphics into raster form on a computer display. Its main research content includes the principles and algorithms of representation, calculation, processing and display of graphics.

[0062] Existing crack tip area displacement field monitoring methods lack high-fidelity data sets that can accurately characterize the complex crack development characteristics of concrete, limiting the training effect of deep learning models; under the premise of ensuring the measurement accuracy of small and high gradient displacement at the crack tip, the method has large calculation amount and cannot realize real-time monitoring under dynamic loading conditions; and the existing deep learning model design does not fully incorporate the fracture mechanics theory and the mesoscopic characteristics of concrete materials, making it difficult to effectively handle the discontinuity of the crack path and the noise interference of the background texture, affecting the robustness of the prediction.

[0063] Therefore, in the embodiments of the present application, a concrete crack tip multi-branch path displacement field real-time accurate monitoring method and related equipment are provided, a high-fidelity virtual data set driven by physical information is constructed, the virtual data set is used as a training basis, and a deep displacement calculation network is designed for concrete crack displacement field analysis, as shown in Figure 3 On the basis of the deep optical flow network framework, the mesoscopic feature extraction module, the motion feature enhancement module, the discontinuous displacement matching module and the sub-pixel level displacement field refinement module are integrated. Through efficient calculation, high-precision and high-robustness displacement field real-time acquisition at the crack tip area is realized, forming a closed-loop optimization process from high-fidelity data construction, special network design to real-time monitoring application.

[0064] Figure 1is an optional flowchart of a method for real-time accurate monitoring of concrete crack tip multi-branch path displacement field provided by an embodiment of the present application, Figure 1 The method in the method can include but is not limited to steps S100 to S600:

[0065] Step S100, according to linear elastic fracture mechanics, cohesive force model and computer graphics, a virtual data set is generated;

[0066] Step S200, input the virtual data set into the initial depth displacement calculation network, and perform concrete micro feature extraction and motion feature enhancement operation on the virtual data set to obtain an enhanced feature map;

[0067] Step S300, through the deformable convolution layer, the enhanced feature map is subjected to non-continuous displacement matching operation to obtain a coarse displacement field;

[0068] Step S400, the coarse displacement field is subjected to sub-pixel level refinement processing to obtain a fine displacement field;

[0069] Step S500, according to the real displacement field and the fine displacement field, the initial depth displacement calculation network is optimized to obtain a target depth displacement calculation network;

[0070] Step S600, input the material surface speckle image into the target depth displacement calculation network to obtain a crack tip deformation displacement field.

[0071] In step S100 of some embodiments, by deeply fusing physical theory and advanced numerical simulation technology, a data set with high fidelity, accurate physical semantics and rich key failure modes is generated. Illustratively, the physical displacement field is generated by combining linear elastic fracture mechanics and cohesive force model, the artificial speckle image matching the concrete microstructure is created by using computer graphics, the virtual data set containing high-precision displacement true value label and being able to represent complex crack branching and other complex crack scenarios is formed. The virtual data set construction of the embodiment of the present application avoids the traditional experience limitations of high acquisition cost, difficulty in quantification and inability to stably reproduce the complex crack behavior of concrete of the real experimental data, and lays a unique data foundation for the subsequent deep displacement calculation network to realize high-precision, high-robustness real-time monitoring of the concrete crack tip deformation displacement field.

[0072] In some embodiments, step S100 can include but is not limited to steps S110 to S130:

[0073] Step S110, by fusing linear elastic fracture mechanics and cohesive force model, a physical displacement field is constructed;

[0074] Step S120, by computer graphics, an artificial speckle image with concrete micro features is generated;

[0075] Step S130, according to the physical displacement field and the artificial speckle image, a virtual data set is generated.

[0076] In step S110 of some embodiments, by deeply coupling linear elastic fracture mechanics and cohesive zone model, the crack propagation process of concrete material is simulated, the crack tip region is obtained, and the displacement field (data matrix) conforming to the physical characteristics of the crack tip singularity field can be generated. A hierarchical meshing strategy is adopted to perform ultra-high density grid encryption on the crack tip region to obtain micron-level displacement data and high-gradient strain data containing the physical characteristics of the crack tip singularity field. The micron-level displacement data and high-gradient strain data are output as pixel-level accurate true value labels to obtain an accurate physical displacement field generated based on physical theory, ensuring the physical authenticity and accuracy of the data set.

[0077] In step S120 of some embodiments, a computer graphics technology is used to generate an artificial speckle image through parameterized control. The artificial speckle image simulates the mesoscopic heterogeneous characteristics of concrete, so that the generated artificial speckle image is highly consistent with real concrete in terms of macro and micro texture, thereby effectively avoiding the decline in generalization ability of the deep displacement solving network model due to data distribution differences in actual application. Optionally, the mesoscopic heterogeneous characteristics include random distribution of concrete aggregates, pore structure, and cementitious material matrix, etc.

[0078] In step S130 of some embodiments, according to the generated physical displacement field and the artificial speckle image, a virtual data set for training the deep displacement solving network model is formed. Illustratively, the generated physical displacement field and the artificial speckle image are obtained, and the artificial speckle image is geometrically deformed using the physical displacement field to simulate the real deformation effect of the object surface under physical load and generate a deformed image. The physical displacement field usually contains horizontal displacement and vertical displacement The artificial speckle image is used as a reference image In the process of applying the physical displacement field to generate the deformed image, for each target pixel point on the deformed image the following processing is performed: according to the displacement field, the original position of the target pixel point in the reference image is found Since the original position is usually not an integer coordinate, an interpolation algorithm is needed to calculate the gray value of the position from the known pixel points of the original image. A virtual data set is generated by grouping the physical displacement field, artificial speckle image and deformed image into a data triple. Optionally, the interpolation algorithm uses a bicubic interpolation method, which can calculate a smooth gray value from the surrounding 16 pixel points, effectively reducing the jagged phenomenon caused by interpolation and generating a high-quality, smooth deformed image.

[0079] In step S200 of some embodiments, based on the deep optical flow network framework, a module integrating microscopic feature extraction and motion feature enhancement is added, and an attention mechanism is further integrated, and the deep optical flow network framework is improved. Illustratively, based on the improved deep optical flow network framework, an initial deep displacement calculation network is formed. Through the initial deep displacement calculation network, the input virtual data set is subjected to multi-scale microscopic feature guidance and motion feature enhancement driven by the attention mechanism, to generate an enhanced feature map. Through the cooperation of multi-scale microscopic feature perception and adaptive attention mechanism, the features in the concrete scene can be fully extracted, and noise interference can be reduced.

[0080] In some embodiments, step S200 can include but is not limited to steps S210 to S220:

[0081] In step S210, the virtual data set is input into a multi-scale feature extractor to perform feature extraction operation on the virtual data set, to obtain a composite feature map;

[0082] In step S220, the composite feature map is input into a spatial-channel collaborative attention module to perform secondary weighting optimization on the composite feature map, to obtain an enhanced feature map.

[0083] In step S210 of some embodiments, the virtual data set is input into the multi-scale feature extractor of the initial deep displacement calculation network. A multi-scale convolution kernel combination is used in the multi-scale feature extractor to construct a feature pyramid. Key information is captured by the large-scale convolution kernel and the small-scale convolution kernel in the multi-scale convolution kernel combination. Through the design of the multi-scale convolution kernel combination, it is ensured that the deep displacement calculation network can understand both the macrostructure and the microscopic details of the concrete from the bottom layer, laying a solid data foundation for subsequent displacement analysis.

[0084] In some embodiments, step S210 can include but is not limited to steps S211 to S213:

[0085] In step S211, the low-frequency structural features of the virtual data set are extracted by the large-scale convolution kernel of the multi-scale feature extractor;

[0086] Step S212, high-frequency detail features of the virtual data set are extracted by a small-scale convolution kernel in the multi-scale feature extractor;

[0087] Step S213, a composite feature map is obtained according to the low-frequency structural features and the high-frequency detail features.

[0088] In step S211 of some embodiments, a large-scale convolution kernel in the multi-scale feature extractor is used to capture low-frequency information in a large range in the virtual data set, which can include but is not limited to low-frequency structural features such as aggregate spatial distribution, cement mortar matrix, and macroscopic pores.

[0089] In step S212 of some embodiments, a small-scale convolution kernel in the multi-scale feature extractor is used to focus on high-frequency detail features of the virtual data set, which can include but is not limited to high-frequency details such as micro-crack paths, fine aggregate edges, and pore profiles.

[0090] In step S213 of some embodiments, according to the low-frequency structural features and the high-frequency detail features extracted by the multi-scale feature extractor, a composite feature map can be output.

[0091] In step S220 of some embodiments, a space-channel collaborative attention module is integrated after the multi-scale feature extractor and before the encoder of the feature map entering the main body of the deep displacement calculation network, so that the deep displacement calculation network can actively focus on key areas with displacement and suppress the interference of static backgrounds. A series combination mechanism of channel attention and spatial attention is adopted to form a space-channel collaborative attention module composed of a channel attention submodule and a spatial attention submodule. The composite feature map is input into the space-channel collaborative attention module, and the composite feature map is weighted and optimized twice to obtain an enhanced feature map with the same size as the composite feature map but with significantly higher information.

[0092] In some embodiments, step S220 can include but is not limited to steps S221 to S222:

[0093] Step S221, the composite feature map is input into the channel attention submodule, the motion feature channels of the composite feature map are enhanced and the static feature channels of the composite feature map are suppressed by channel attention weights, to obtain a channel weighted feature map;

[0094] Step S222, the channel weighted feature map is input into the spatial attention submodule, the high gradient strain region weights of the channel weighted feature map are enhanced and the static region weights of the channel weighted feature map are suppressed by spatial attention weights, to generate an enhanced feature map;

[0095] In step S221 of some embodiments, the composite feature map is sent to a channel attention sub-module, which learns the importance weights of each feature channel of the composite feature map through global average pooling and a fully connected layer, and performs first weighted optimization on the composite feature map, thereby enhancing the motion feature channels related to semantics such as “motion” and “deformation”, while suppressing the static feature channels expressing “static texture”, to obtain a channel weighted feature map.

[0096] In step S222 of some embodiments, the channel weighted feature map generated after optimization by the channel attention sub-module is sent to a spatial attention sub-module. The sub-module generates a two-dimensional spatial attention map, i.e., an enhanced feature map, by performing maximum pooling and average pooling in the channel dimension and processing through a convolution layer. Illustratively, in the spatial attention sub-module, each pixel position on the channel weighted feature map is assigned a spatial attention weight, so that high gradient strain regions with significant displacement changes such as crack tips and expansion paths obtain high weights, while the weights of other static regions in the channel weighted feature map are effectively suppressed to generate the enhanced feature map. The enhanced feature map will be the input of all subsequent network levels, guiding the entire network to focus computing resources on key deformation regions.

[0097] In step S300 of some embodiments, a deformable convolution layer is also integrated in the initial depth displacement calculation network. By dynamically adjusting the sampling position of the convolution kernel, local non-continuous displacement caused by complex paths such as multiple branching cracks and tortuous extensions of concrete is adaptively captured, so that the initial depth displacement calculation network can accurately match the upper displacement field according to the actual morphology and deformation characteristics of the cracks in the enhanced feature map, generate a coarse displacement field, and achieve high robustness in tracking irregular crack paths.

[0098] In step S400 of some embodiments, a sub-pixel level residual correction unit is also built into the initial depth displacement calculation network for the purpose of micro displacement analysis. Through the sub-pixel level residual correction unit, the coarse displacement field is refined at the sub-pixel level, and a lightweight convolution unit is used to learn the micro-level residual displacement distribution pattern to correct local analysis errors in the coarse displacement field, ensuring that the full-field displacement reaches sub-pixel accuracy, generating a fine displacement field, and achieving accurate measurement of micro-level displacement.

[0099] In step S500 of some embodiments, a hybrid error integrating multi-scale optical flow supervision and fracture mechanics theory is used as a hybrid loss function, and the initial depth displacement calculation network model is trained and optimized end-to-end through the hybrid loss function, and related mechanisms are introduced to improve the model generalization ability, so that the model not only learns the pixel-level visual matching, but also understands the underlying fracture mechanics behavior.

[0100] In some embodiments, step S500 can include but is not limited to steps S510-S540:

[0101] Step S510, constructing a multi-scale optical flow supervision loss according to the real displacement field and the fine displacement field;

[0102] Step S520, constructing a fracture mechanics regularization loss according to the fine displacement field;

[0103] Step S530, constructing a hybrid loss function according to the multi-scale optical flow supervision loss and the fracture mechanics regularization loss;

[0104] Step S540, optimizing the initial depth displacement calculation network according to the hybrid loss function to obtain a target depth displacement calculation network.

[0105] In step S510 of some embodiments, the fine displacement field is taken as a predicted displacement field, and a multi-scale optical flow supervision loss is constructed according to the real displacement field and the predicted displacement field. The real displacement field is generated by numerical simulation, and optionally, a physical displacement field constructed by fusing linear elastic fracture mechanics and cohesive force model is taken as the real displacement field. The multi-scale optical flow supervision loss is a data-driven term, which ensures that the model can approximate the real displacement at each scale from macro to micro. Exemplarily, the expression of the multi-scale optical flow supervision loss is as follows:

[0106]

[0107] In the formula, represents the multi-scale optical flow supervision loss; represents the scale index, represents the weight coefficient of each scale, which is used to balance the importance of different scales; represents the predicted displacement field of the i-th scale; represents the real displacement field of the i-th scale; represents the Euclidean norm.

[0108] ​​​​In step S520 of some embodiments, a fine displacement field is taken as a predicted displacement field, and a fracture mechanics regularization loss is constructed according to the fine displacement field. The fracture mechanics regularization term converts the fracture mechanics theory into a soft constraint on the output of the deep displacement solver network, guiding the network to learn a physically reasonable displacement field. Illustratively, by fusing the linear elastic fracture mechanics and the cohesive zone model to construct a physical displacement field (i.e., a ground truth label), the position information of the physical displacement field is used to demarcate the crack tip region, and during the training process of the deep displacement solver network, the fracture mechanics regularization term calculates the spatial gradient of the network predicted displacement field, i.e., the deep displacement solver network calculates the fracture mechanics regularization loss in the specific crack tip region. According to the linear elastic fracture mechanics theory, there is stress singularity at the crack tip, which must be accompanied by extremely high displacement gradient. Therefore, the fracture mechanics regularization term will impose a significant penalty on the solution of the predicted displacement gradient in the crack tip region. The expression of the fracture mechanics regularization loss is:

[0109] ;

[0110] In the formula, denotes the fracture mechanics regularization loss; denotes the spatial position; denotes the crack tip region; denotes the minimum displacement gradient that the initial deep displacement solver network should reach in the crack tip region, which is determined by the actual magnitude of the displacement gradient in the data set; denotes the predicted displacement gradient at the spatial position ; denotes the Frobenius norm.

[0111] In step S530 of some embodiments, a hybrid loss function is constructed by a weighted combination of the multi-scale optical flow supervision loss and the fracture mechanics regularization loss. The formula used includes:

[0112] ;

[0113] In the formula, denotes the hybrid loss function; denotes the multi-scale optical flow supervision loss weight; denotes the fracture mechanics regularization loss weight.

[0114] In step S540 of some embodiments, the initial deep displacement solver network is iteratively trained and optimized by the hybrid loss function, so that the target deep displacement solver network can make more accurate and robust predictions even when it faces new crack patterns that have never been seen in real-world scenarios.

[0115] In step S600 of some embodiments, the target depth displacement solving network is deployed, and by collecting the material surface speckle image on site and inputting it into the target depth displacement solving network, the crack tip deformation displacement field can be quickly and accurately output, realizing real-time high-precision monitoring of crack dynamic expansion. Exemplarily, as shown in Figure 2 The target depth displacement solving network is deployed to the loading device 701, the material surface speckle image on site is collected by the image acquisition system 702, and the crack tip deformation displacement field 704 is generated relying on the GPU acceleration capability of the deep learning end-to-end parallel computing and GPU computing unit 703.

[0116] The embodiments of the present application realize the dual breakthrough of precision and real-time in the field of concrete crack tip displacement field monitoring, effectively overcoming the application bottleneck of traditional digital image correlation technology in high gradient non-continuous deformation area. Compared with the calculation divergence or smoothing distortion problem caused by fine grid division in the key area such as crack tip in the traditional method, the embodiments of the present application generate virtual data set by physical model driving, and design depth displacement solving network architecture specially for non-continuous field, which significantly improves the monitoring precision and robustness of displacement field. As shown in Figure 4 The crack tip deformation displacement field output by the embodiments of the present application as shown in part (a) of Figure 4 The crack tip deformation displacement field output by the DIC method as shown in part (b) of The displacement cloud map output by the embodiments of the present application presents clear and smooth gradient characteristics in the crack tip, which accurately reflects the stress concentration and expansion behavior. In terms of real-time, relying on the deep learning end-to-end parallel computing and GPU acceleration capability, the present application shortens the thousand-frame image processing time from 126 minutes of the traditional method to 10 minutes in the same hardware environment, realizing more than 12 times efficiency improvement. This order of magnitude breakthrough makes it possible to realize millisecond-level response high-frequency online monitoring of structure under dynamic load.

[0117] The embodiments of the present application also provide a concrete crack tip multi-branch path displacement field real-time accurate monitoring device, which can realize the above-mentioned concrete crack tip multi-branch path displacement field real-time accurate monitoring method. The device comprises:

[0118] A data generation module is configured to generate a virtual data set according to linear elastic fracture mechanics, cohesive force model and computer graphics;

[0119] A feature extraction and enhancement module is configured to input the virtual data set into an initial depth displacement solving network, perform concrete micro-feature extraction and motion feature enhancement operations on the virtual data set, and obtain an enhanced feature map;

[0120] The non-continuous displacement matching module is configured to perform a non-continuous displacement matching operation on the enhanced feature map through a deformable convolution layer to obtain a coarse displacement field.

[0121] The fine processing module is configured to perform sub-pixel level fine processing on the coarse displacement field to obtain a fine displacement field.

[0122] The model training optimization module is configured to optimize the initial depth displacement calculation network according to the real displacement field and the fine displacement field to obtain a target depth displacement calculation network.

[0123] The real-time monitoring module is configured to input the material surface speckle image into the target depth displacement calculation network to obtain a crack tip deformation displacement field.

[0124] It can be understood that the content in the above method embodiments is applicable to the device embodiments, the device embodiments specifically implement the functions of the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0125] The embodiment of the application further provides an electronic device, which comprises a processor and a memory, the memory stores a computer program, and the processor implements the above method when executing the computer program. The electronic device can be any intelligent terminal, such as a tablet computer or a vehicle-mounted computer.

[0126] It can be understood that the content in the above method embodiments is applicable to the device embodiments, the device embodiments specifically implement the functions of the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0127] Reference Figure 5 , Figure 5 The electronic device of another embodiment is illustrated, and the electronic device comprises:

[0128] The processor 801 can be implemented in a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, and is configured to execute a related program to implement the technical solutions provided by the embodiments of the application.

[0129] The memory 802 can be implemented in the form of read only memory (ROM), static storage device, dynamic storage device or random access memory (RAM), etc. The memory 802 can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present specification are implemented by software or firmware, the related program codes are stored in the memory 802 and are called and executed by the processor 801 to execute the method of the embodiments of the present application;

[0130] The input / output interface 803 is used to realize information input and output;

[0131] The communication interface 804 is used to realize the communication interaction between the device and other devices, and the communication can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0132] The bus 805 transmits information between various components (such as the processor 801, the memory 802, the input / output interface 803 and the communication interface 804) of the device;

[0133] The processor 801, the memory 802, the input / output interface 803 and the communication interface 804 are connected to each other through the bus 805 for internal communication connection in the device.

[0134] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the above method.

[0135] It can be understood that the contents in the above method embodiments are all applicable to the present storage medium embodiment, the functions specifically realized by the present storage medium embodiment are the same as those of the above method embodiments, and the beneficial effects achieved by the present storage medium embodiment are also the same as those of the above method embodiments.

[0136] The embodiment of the present application also provides a computer program product or computer program, which comprises computer instructions stored in a computer readable storage medium. The processor of the computer device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the above method.

[0137] In summary, the concrete crack tip multi-branch path displacement field real-time accurate monitoring method and related device of the embodiment of the present application has the following advantages:

[0138] 1、The embodiment of the present application is based on a high-fidelity virtual data set construction method driven by physical theory, which changes the passive situation of traditional deep learning in concrete crack monitoring, which is limited by the difficulty of obtaining real test data and low data fidelity. By accurately reproducing the mesoscopic characteristics and complex fracture mechanics behavior of concrete, this virtual data set lays a solid foundation for the robustness and accuracy of the deep displacement calculation network model from the source, solving the problem that general data sets cannot effectively represent the physical properties of concrete.

[0139] 2、The embodiment of the present application breaks through the limitation of the existing general optical flow model in the aspect of deep displacement calculation network architecture, which is insufficient in processing the discontinuous and high gradient deformation field of crack tip. Through the mesoscopic feature extraction module and the motion feature enhancement module, the key information in the concrete scene is fully extracted, and the attention mechanism is further integrated to focus on the key area where displacement occurs and suppress the interference of static background. Combined with the deformable convolution, the discontinuous displacement of multi-branch cracks is adaptively captured, and the sub-pixel level residual correction unit is designed to realize micron-level displacement accurate analysis, which balances the real-time and sub-pixel accuracy of displacement calculation.

[0140] 3、The embodiment of the present application forms an end-to-end closed loop from virtual data set generation to real-time monitoring, which provides reliable technical support for the structural health monitoring of major infrastructure such as bridges and dams, and provides a solid technical support for prolonging the service life of concrete structures.

[0141] In some alternative embodiments, the functions / operations mentioned in the block diagram can not occur in the order mentioned in the operation diagram. For example, depending on the functions / operations involved, two blocks shown in succession can actually be executed substantially simultaneously or the blocks can sometimes be executed in reverse order. In addition, the embodiments presented and described in the flowcharts of the present application are provided by way of example, with the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are independently executed.

[0142] Furthermore, although the present application is described in the context of functional modules, it is to be understood that one or more of the described functions and / or features can be integrated in a single physical device and / or software module, or one or more functions and / or features can be implemented in separate physical devices or software modules. It will also be appreciated that detailed discussion of the actual implementation of each module is not necessary to an understanding of the application. Rather, the actual implementation of the modules, in combination with their attributes, functions, and internal relationships, are to be understood within the context of the devices disclosed herein. Thus, those skilled in the art with access to patents, scientific journals, and other public sources known by those skilled in the art will be able, using ordinary skill, to practice the application as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative and that the scope of the present application is to be determined by the entire scope of the claims, along with all equivalents of the claims and their equivalents.

[0143] If the functions are implemented in software, the functions can be stored in or implemented as one or more computer program products, which can be incorporated into a computer- readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this document, a "computer-readable medium" can be any means that can store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0144] The logic and / or steps represented in the flowcharts and / or otherwise described herein, for example, can be embodied in non-transitory computer-readable media, which can be executed by an instruction execution system, apparatus, or device such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this document, a "computer-readable medium" can be any means that can store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0145] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0146] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques can be used to implement the hardware used in the described embodiments: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so forth.

[0147] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. Illustrative expressions of the above terms in the specification do not necessarily refer to the same embodiment or example. Also, the specific feature, structure, material or characteristic described can be combined in any appropriate manner in one or more embodiments or examples.

[0148] Although embodiments of the application have been shown and described, it would be recognized by those of ordinary skill in the art that various changes, modifications, substitutions, and alterations can be made hereto without departing from the spirit and scope of the application, which is defined by the following claims and their equivalents.

[0149] The above is a specific description of the preferred embodiments of the present application, but the present application is not limited to the described embodiments, and those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the present application, and these equivalent modifications or replacements are all included in the scope defined by the claims of the present application.

Claims

1. A method for real-time and accurate monitoring of displacement field along multiple bifurcation paths at the tip of a concrete crack, characterized in that, Includes the following steps: Based on linear elastic fracture mechanics, cohesive force models, and computer graphics, a virtual dataset is generated, including: A physical displacement field is constructed by integrating the linear elastic fracture mechanics with the cohesive force model; an artificial speckle image with the microscopic characteristics of concrete is generated using computer graphics; and the virtual dataset is generated based on the physical displacement field and the artificial speckle image. The virtual dataset is input into the initial depth displacement calculation network, and concrete microstructure features are extracted and motion features are enhanced to obtain an enhanced feature map. By performing discontinuous displacement matching on the enhanced feature map through deformable convolutional layers, a coarse displacement field is obtained. The coarse displacement field is refined at the subpixel level to obtain the fine displacement field; Based on the actual displacement field and the refined displacement field, the initial depth displacement calculation network is optimized to obtain the target depth displacement calculation network, including: Based on the true displacement field and the fine displacement field, a multi-scale optical flow supervision loss is constructed; based on the fine displacement field, a fracture mechanics regularization loss is constructed; based on the multi-scale optical flow supervision loss and the fracture mechanics regularization loss, a hybrid loss function is constructed; based on the hybrid loss function, the initial depth displacement calculation network is optimized to obtain the target depth displacement calculation network. The speckle image of the material surface is input into the target depth displacement calculation network to obtain the deformation displacement field at the crack tip.

2. The method according to claim 1, characterized in that, The step of performing concrete microstructure feature extraction and motion feature enhancement operations on the virtual dataset to obtain an enhanced feature map includes the following steps: The virtual dataset is input into a multi-scale feature extractor to perform feature extraction operations on the virtual dataset, resulting in a composite feature map; The composite feature map is input into the spatial-channel collaborative attention module, and the composite feature map is subjected to secondary weighted optimization to obtain the enhanced feature map.

3. The method according to claim 2, characterized in that, The step of inputting the virtual dataset into a multi-scale feature extractor to perform feature extraction on the virtual dataset and obtain a composite feature map includes the following steps: The low-frequency structural features of the virtual dataset are extracted using the large-scale convolution kernel of the multi-scale feature extractor. The high-frequency detail features of the virtual dataset are extracted using the small-scale convolution kernel of the multi-scale feature extractor. The composite feature map is obtained based on the low-frequency structural features and the high-frequency detail features.

4. The method according to claim 2, characterized in that, The step of inputting the composite feature map into the spatial-channel collaborative attention module and performing secondary weighted optimization on the composite feature map to obtain the enhanced feature map includes the following steps: The composite feature map is input into the channel attention submodule. Through channel attention weights, the motion feature channels of the composite feature map are enhanced, and the static feature channels of the composite feature map are suppressed, resulting in a channel-weighted feature map. The channel-weighted feature map is input into the spatial attention submodule. Through spatial attention weights, the weights of high-gradient strain regions in the channel-weighted feature map are enhanced, and the weights of static regions in the channel-weighted feature map are suppressed, thereby generating the enhanced feature map. The spatial-channel collaborative attention module is formed by connecting the channel attention submodule and the spatial attention submodule in series.

5. The method according to claim 1, characterized in that, The initial depth displacement calculation network is optimized based on the real displacement field and the refined displacement field to obtain the target depth displacement calculation network. The formulas used include: ； ； ； In the formula, Represents the mixed loss function; This represents the loss under multi-scale optical flow supervision; This represents the fracture mechanics regularization loss. Indicates the weights of the multi-scale optical flow supervision loss; This represents the weight of the fracture mechanics regularization loss. Indicates scale index. ; This represents the weighting coefficient for each scale; Indicates the first Predicted displacement fields at various scales; Indicates the first A true displacement field at a scale; Indicates spatial location; Indicates the region at the tip of the crack; This represents the minimum displacement gradient of the initial depth displacement calculation network in the crack tip region; Indicates spatial location Predicted displacement gradient at the location; Denotes the Euclidean norm; This represents the Frobenius norm.

6. A real-time and precise monitoring device for the displacement field of multi-branching paths at the tip of a concrete crack, characterized in that, include: The data generation module is used to generate a virtual dataset based on linear elastic fracture mechanics, cohesion model, and computer graphics. Specifically, the data generation module is used to: construct a physical displacement field by fusing the linear elastic fracture mechanics and cohesion model; generate artificial speckle images with microscopic features of concrete using computer graphics; and generate the virtual dataset based on the physical displacement field and the artificial speckle images. The feature extraction and enhancement module is used to input the virtual dataset into the initial depth displacement calculation network, and perform concrete micro-feature extraction and motion feature enhancement operations on the virtual dataset to obtain an enhanced feature map. The discontinuous displacement matching module is used to perform discontinuous displacement matching operation on the enhanced feature map through a deformable convolutional layer to obtain a coarse displacement field. The fine-processing module is used to perform sub-pixel-level fine-processing on the coarse displacement field to obtain the fine displacement field. The model training optimization module is used to optimize the initial depth displacement calculation network based on the real displacement field and the fine displacement field to obtain the target depth displacement calculation network. Specifically, the model training optimization module is used to: construct a multi-scale optical flow supervision loss based on the real displacement field and the fine displacement field; construct a fracture mechanics regularization loss based on the fine displacement field; construct a hybrid loss function based on the multi-scale optical flow supervision loss and the fracture mechanics regularization loss; and optimize the initial depth displacement calculation network based on the hybrid loss function to obtain the target depth displacement calculation network. The real-time monitoring module is used to input the speckle image of the material surface into the target depth displacement calculation network to obtain the deformation displacement field at the crack tip.

7. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images