Crack identification method and system based on fusion of acoustic emission and scanning electron microscope pictures

Abstract

This invention proposes a crack identification method and system based on the fusion of acoustic emission and scanning electron microscopy (SEM) images, relating to the field of image processing technology. The method includes: performing splitting and shearing experiments on a rock sample; acquiring acoustic emission time-domain waveform signals; separating the acoustic emission time-domain waveform signals to determine the waveform parameters and spatial distribution of each acoustic emission event; obtaining the spatial domain corresponding to the SEM scan image of the rock sample; determining the crack type corresponding to the acoustic emission event based on the waveform parameters when the real coordinates are located in the spatial domain; and using the crack type as the category label of the SEM scan image; training a preset machine learning model using the SEM scan image with the added category label to obtain a target machine learning model; and using the target machine learning model to identify the SEM image under test to determine the crack type corresponding to the SEM image under test.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a crack identification method and system based on the fusion of acoustic emission and scanning electron microscopy images. Background Technology

[0002] In the field of rock mechanics engineering, rocks are affected by excavation disturbances or in-situ stress, leading to the initiation, propagation, and penetration of internal cracks. Scanning electron microscopy (SEM) is commonly used to analyze mineral grain size distribution and identify microcracks within rocks. Its judgment is primarily based on the physical morphology of the rock after failure, making it an effective tool for studying rock damage and failure. Acoustic emission (AE) technology, as a non-destructive testing method, can not only record the acoustic emission signals generated by strain energy release during crack formation in real time but also accurately capture the location of acoustic emission events. Therefore, it is widely used in crack evolution and damage assessment studies of brittle materials.

[0003] However, the effectiveness of scanning electron microscopy (SEM) technology is highly dependent on researchers' practical experience, and the analysis of experimental results is subject to significant subjectivity and individual differences. Due to the lack of unified, quantitative analytical standards, different researchers interpret SEM observations based on their own experience, which can easily lead to biases due to differences in judgment angles and cognitive levels, resulting in insufficient consistency and objectivity in the analytical results and making it difficult to reach unified conclusions. Acoustic emission technology is limited to preliminary explorations of acoustic emission characteristic parameters and simple comparisons of average frequency (AF) and rise angle (RA) values, lacking in-depth systematic analysis. Crucially, there is currently no unified consensus in the industry regarding the slope k value of the AF / RA ratio boundary; related research is still in the exploratory stage. This results in a lack of clear core criteria for the accurate identification of crack types, affecting the accuracy and reliability of the identification results. Summary of the Invention

[0004] In view of this, the present invention proposes a crack identification method and system based on the fusion of acoustic emission and scanning electron microscopy images.

[0005] The technical solution of this invention is implemented as follows: The first aspect of this invention provides a crack identification method based on the fusion of acoustic emission and scanning electron microscopy images, comprising: Splitting and shearing tests were performed on the rock sample to collect acoustic emission time-domain waveform signals. The acoustic emission time-domain waveform signals were then separated to determine the waveform parameters and spatial distribution of each acoustic emission event. The waveform parameters include the average frequency and rise angle of a single acoustic emission event. The spatial distribution includes the actual coordinates of the acoustic emission event within the rock sample. The spatial domain corresponding to the SEM scan image of the rock sample is obtained. When the true coordinates are located in the spatial domain, the crack type corresponding to the acoustic emission event is determined based on the waveform parameters, and the crack type is used as the category label of the SEM scan image. The crack type includes tensile cracks and shear cracks. A preset machine learning model is trained using the SEM scan image with added category labels to obtain a target machine learning model. The target machine learning model is then used to identify the SEM image under test to determine the crack type corresponding to the SEM image under test.

[0006] Based on the above technical solutions, preferably, the step of separating the acoustic emission time-domain waveform signal to determine the waveform parameters and spatial distribution of each acoustic emission event includes: The acoustic emission time-domain waveform signal is separated to obtain ring count parameters, duration parameters, peak amplitude parameters, and rise time parameters. The ring count parameter represents the number of oscillations of the signal exceeding a preset threshold. The duration parameter represents the total time from the first time the signal exceeds the preset threshold to the last time it falls below the preset threshold. The peak amplitude parameter represents the peak amplitude of the signal waveform within the HDT time, starting from the first time the preset threshold is exceeded. The rise time parameter represents the time interval from the first time the signal exceeds the preset threshold to the time interval from when the signal first exceeds the preset threshold to when the peak amplitude is reached. The average frequency is determined based on the ratio of the ring count parameter to the duration parameter, and the rise angle is determined based on the ratio of the rise time parameter to the peak amplitude parameter.

[0007] Based on the above technical solutions, preferably, the step of separating the acoustic emission time-domain waveform signal to determine the waveform parameters and spatial distribution of each acoustic emission event includes: Based on the time difference or phase difference of the acoustic emission time-domain waveform signal arriving at different sensors, combined with the wave velocity and the geometric relationship of the sensor array, the initial coordinates of each acoustic emission event in the rock sample are determined. The initial coordinates are corrected based on internal damage and internal temperature to obtain the true coordinates of the acoustic emission event.

[0008] Based on the above technical solution, preferably, before obtaining the spatial domain corresponding to the SEM scan image of the rock sample, the method further includes: Based on the same reference point, the fracture surface of the rock sample was scanned by SEM partitioning to obtain multiple frames of SEM scan sub-images; The SEM scan sub-images are stitched together based on the reference point to obtain the SEM scan image of the rock sample.

[0009] Based on the above technical solutions, preferably, the step of acquiring the spatial domain corresponding to the SEM scan image of the rock sample, and determining the crack type corresponding to the acoustic emission event based on the waveform parameters when the real coordinates are located in the spatial domain, includes: The corresponding spatial domain is determined based on the two-dimensional coordinate extrema of the SEM scan image of the rock sample; the two-dimensional coordinate extrema include the maximum value of the horizontal coordinate, the minimum value of the horizontal coordinate, the maximum value of the vertical coordinate, and the minimum value of the vertical coordinate. The real coordinates are projected and transformed onto the plane where the SEM scan image is located. If the projected coordinates are located in the spatial domain, the acoustic emission event is determined to be a valid event; if the projected coordinates are not located in the spatial domain, the acoustic emission event is determined to be an invalid event. If the acoustic emission event is a valid event, the crack type corresponding to the acoustic emission event is determined based on the waveform parameters.

[0010] Based on the above technical solutions, preferably, when the acoustic emission event is a valid event, determining the crack type corresponding to the acoustic emission event based on the waveform parameters includes: If the acoustic emission event is a valid event, obtain a first ratio of the average frequency and rise angle corresponding to each acoustic emission event, and a second ratio corresponding to the maximum average frequency and maximum rise angle among all acoustic emission events; If the first ratio is greater than the second ratio, the crack in the acoustic emission event corresponding to the first ratio is determined to be a tensile crack; If the first ratio is less than the second ratio, the crack in the acoustic emission event corresponding to the first ratio is determined to be a shear crack.

[0011] Based on the above technical solutions, preferably, the preset machine learning model includes an identity branch, a spatial branch, and a channel branch; the step of training the preset machine learning model using the SEM scanned image with added category labels to obtain the target machine learning model includes: The SEM scan image with added category labels is input into the identity branch, the spatial branch, and the channel branch respectively to generate the original feature, spatial enhancement feature, and channel enhancement feature respectively; The original features, spatial enhancement features, and channel enhancement features are adaptively weighted until the generated target features meet the discrimination requirements, and the target machine learning model is determined based on the corresponding weight parameters.

[0012] Furthermore, a second aspect of the present invention provides a crack identification system based on the fusion of acoustic emission and scanning electron microscopy images, comprising: a parameter acquisition module, a type determination module, and a training and identification module; wherein, The parameter acquisition module is configured to perform splitting and shearing tests on the rock sample, acquire acoustic emission time-domain waveform signals, and separate the acoustic emission time-domain waveform signals to determine the waveform parameters and spatial distribution of each acoustic emission event; the waveform parameters include the average frequency and rise angle of a single acoustic emission event; the spatial distribution includes the true coordinates of the acoustic emission event in the rock sample; The type determination module is configured to acquire the spatial domain corresponding to the SEM scan image of the rock sample, and, when the real coordinates are located in the spatial domain, determine the crack type corresponding to the acoustic emission event based on the waveform parameters, and use the crack type as the category label of the SEM scan image; the crack type includes tensile cracks and shear cracks. The training and recognition module is configured to train a preset machine learning model using the SEM scan image with added category labels to obtain a target machine learning model, and to use the target machine learning model to recognize the SEM image to be tested to determine the crack type corresponding to the SEM image to be tested.

[0013] More preferably, a third aspect of the present invention provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the crack identification method based on the fusion of acoustic emission and scanning electron microscopy images described in the first aspect.

[0014] More preferably, a fourth aspect of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the crack identification method based on the fusion of acoustic emission and scanning electron microscope images described in the first aspect.

[0015] The crack identification method and system based on the fusion of acoustic emission and scanning electron microscopy images of the present invention have the following advantages over the prior art: 1. By utilizing acoustic emission technology to accurately capture dynamic signals generated by internal rock cracks and locate the true coordinates of events, this method overcomes the limitations of SEM scanning, which can only observe surface and localized cracks and cannot correlate the dynamic generation process of internal cracks. Furthermore, the intuitiveness of SEM images provides visual evidence for crack type determination based on acoustic emission parameters, improving the accuracy and reliability of crack identification results. Based on this, a machine learning model is trained using SEM images with category labels, resulting in a target machine learning model. This enables efficient batch recognition of a large number of test images, solving the problems of low efficiency and poor consistency in traditional manual recognition, and providing a feasible path for large-scale rapid detection of rock crack types.

[0016] 2. Based on rock splitting and shearing tests, acoustic emission time-domain waveform signals were collected and individual events were separated. The crack type was determined by the average frequency and rise angle. At the same time, by combining the spatial domain of SEM scan images with the real coordinates of acoustic emission events, dual verification of acoustic emission parameter quantification and SEM visual observation was achieved. This method got rid of the limitations of relying on researchers' experience and subjective judgment, and greatly improved the accuracy and objectivity of crack type determination.

[0017] 3. Based on the principles of time difference and phase difference, the initial coordinates are calculated by utilizing the temporal differences of signals received by multiple sensors. This captures the dynamic location information at the moment of acoustic emission events. Two key physical factors, internal damage and internal temperature, are introduced for correction. By quantifying the impact of the degree of damage to the internal rock structure and the temperature field distribution on wave propagation, the initial coordinates are dynamically calibrated, which significantly improves the theoretical accuracy and practical reliability of coordinate calculations. This makes the positioning results closer to the actual location of internal rock damage, ensuring the consistency and effectiveness of data throughout the entire process. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a crack identification method based on the fusion of acoustic emission and scanning electron microscopy images, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the acoustic emission waveform separation result provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a crack identification system based on the fusion of acoustic emission and scanning electron microscope images provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] In some embodiments, such as Figure 1 As shown, Figure 1 This is a flowchart illustrating a crack identification method based on the fusion of acoustic emission and scanning electron microscopy images provided in an embodiment of the present invention. The crack identification method based on the fusion of acoustic emission and scanning electron microscopy images provided by the present invention includes: S110 involves performing splitting and shearing tests on rock samples, acquiring acoustic emission time-domain waveform signals, separating the acoustic emission time-domain waveform signals, and determining the waveform parameters and spatial distribution of each acoustic emission event. The waveform parameters include the average frequency and rise angle of a single acoustic emission event; the spatial distribution includes the true coordinates of the acoustic emission event in the rock sample.

[0022] S120: Obtain the spatial domain corresponding to the SEM scan image of the rock sample. With the true coordinates located in the spatial domain, determine the crack type corresponding to the acoustic emission event based on the waveform parameters, and use the crack type as the category label of the SEM scan image; the crack types include tensile cracks and shear cracks.

[0023] S130: The preset machine learning model is trained using SEM scan images with added category labels to obtain the target machine learning model. The target machine learning model is then used to identify the SEM image to be tested and determine the crack type corresponding to the SEM image to be tested.

[0024] In this embodiment, splitting and shearing tests were conducted on the rock samples. During the tests, when damage behaviors such as microcracks and crack propagation occur inside the rock, elastic stress waves, i.e., acoustic emission signals, are radiated outward. Acoustic emission sensors placed on the surface of the rock samples were used to collect the acoustic emission time-domain waveform signals throughout the process. Since multiple acoustic emission events are generated during the test, each event corresponds to the generation or propagation of a microcrack. The signals of these events are discretely distributed in the continuous waveform. Based on preset acquisition and judgment rules, such as amplitude threshold, event duration, and impact interval, each independent acoustic emission event can be segmented from the continuous waveform, thereby determining the waveform parameters and spatial distribution of each acoustic emission event. Among them, the average frequency reflects the oscillation speed of the acoustic emission signal, and the rise angle characterizes the steepness of the signal rise edge.

[0025] The system determines whether the true coordinates of each acoustic emission event fall within the spatial domain of the SEM scan image. If the true coordinates are within this domain, the crack corresponding to the acoustic emission event is indeed the microscopic crack observed in the SEM image. Based on the waveform parameters, the crack type corresponding to the acoustic emission event can then be determined. The determined crack type is used as the category label for the SEM scan image, completing the accurate annotation of the SEM image. The SEM scan images with added category labels are then used as a training dataset and input into a pre-defined machine learning model. The model continuously optimizes its parameters by learning the correlation between the microscopic features of a large number of labeled images and the category labels, ultimately training a target machine learning model capable of accurately identifying crack types. Subsequently, the SEM scan image of the rock to be tested is input into the trained target machine learning model. The model automatically extracts the microscopic features of the image and matches them with the feature patterns learned during training, quickly outputting the crack type corresponding to the SEM image, thus achieving automated and accurate identification of rock crack types.

[0026] In some embodiments, the acoustic emission time-domain waveform signal is separated to determine the waveform parameters and spatial distribution of each acoustic emission event, including: The acoustic emission time-domain waveform signal is separated to obtain the ring count parameter, duration parameter, peak amplitude parameter, and rise time parameter. The ring count parameter represents the number of oscillations of the signal exceeding a preset threshold. The duration parameter represents the total time from the first time the signal exceeds the preset threshold to the last time it falls below the preset threshold. The peak amplitude parameter represents the peak amplitude of the signal waveform within the HDT time starting from the first time the preset threshold is exceeded. The rise time parameter represents the time interval from the first time the signal exceeds the preset threshold to the time interval from the first time the signal exceeds the preset threshold to the peak amplitude being reached. The average frequency is determined based on the ratio of the ring count parameter to the duration parameter, and the rise angle is determined based on the ratio of the rise time parameter to the peak amplitude parameter.

[0027] In this embodiment, the acoustic emission waveform separation results are shown below. Figure 2 The average frequency AF and the rise angle RA are defined as follows: ; in, For ringing count parameters, For duration parameter, For peak amplitude parameters, This is the rise time parameter.

[0028] In some embodiments, the acoustic emission time-domain waveform signal is separated to determine the waveform parameters and spatial distribution of each acoustic emission event, including: Based on the time difference or phase difference of the acoustic emission time-domain waveform signal arriving at different sensors, combined with the wave velocity and the geometric relationship of the sensor array, the initial coordinates of each acoustic emission event in the rock sample are determined. The initial coordinates are corrected based on internal damage and internal temperature to obtain the true coordinates of the acoustic emission event.

[0029] In this embodiment, the principle of three-dimensional positioning can be expressed as: ; For the coordinates of the sound source, The coordinates of the i-th sensor (i=1, 2, 3, 4); Characterizing the propagation speed of elastic waves in a material, It represents the time it takes for the signal to reach the i-th sensor.

[0030] Based on internal damage The above equation is corrected using the internal temperature T: ; Wave speed correction function It can be represented as: ; In the formula, Wave velocity under standard conditions Characterizing the temperature-wave velocity sensitivity coefficient (GFRP empirical value: ), Characterizing the damage attenuation index, This is the time correction for time difference synchronization.

[0031] ; ; in, The time it takes for sensor i to receive the k-th reference signal is represented by M, where M is the total number of sensors. This is a time correction parameter.

[0032] Internal damage satisfy: ; Right now ; in, Characterizes the energy of the current AE event ( ), The reference energy is used to characterize the lossless state.

[0033] In some embodiments, before acquiring the spatial domain corresponding to the SEM scan image of the rock sample, the method further includes: Based on the same reference point, the fracture surface of the rock sample was scanned by SEM partitioning to obtain multiple frames of SEM scan sub-images; The SEM scan sub-images are stitched together based on the reference points to obtain the SEM scan image of the rock sample.

[0034] In this embodiment, the fracture surface of the rock sample is cut into thin sections suitable for scanning electron microscopy (SEM) observation, and then scanned using the SEM. The scanning electron microscope emits a high-energy electron beam through an electron gun, which is focused by an electromagnetic lens to form a nanoscale electron probe that scans the sample surface point by point. The interaction between the electron beam and the sample excites secondary electrons and backscattered electrons, which are received by a detector and converted into electrical signals, ultimately generating a high-resolution grayscale image simultaneously. Since the field of view of each SEM scan is limited, a reference point is selected to divide the scanning area into sections for scanning. The scanned sections of SEM images are then stitched together based on the reference point to ensure the maximum possible reconstruction of a complete SEM image.

[0035] In some embodiments, the spatial domain corresponding to the SEM scan image of the rock sample is obtained. When the true coordinates are located in the spatial domain, the crack type corresponding to the acoustic emission event is determined based on waveform parameters, including: The spatial domain is determined by the two-dimensional coordinate extrema of the SEM scan images of rock samples; the two-dimensional coordinate extrema include the maximum, minimum, maximum and minimum values ​​of the abscissa, and the maximum and minimum values ​​of the ordinate. The real coordinates are projected and transformed onto the plane where the SEM scan image is located. If the projected coordinates are located in the spatial domain, the acoustic emission event is determined to be a valid event; if the projected coordinates are not located in the spatial domain, the acoustic emission event is determined to be an invalid event. When the acoustic emission event is a valid event, the crack type corresponding to the acoustic emission event is determined based on the waveform parameters.

[0036] In this embodiment, the coordinates of the three-dimensional acoustic emission event are set as follows: SEM image coordinates are The projection transformation function is: ; in, and Scale factor (unit: pixels / μm). This represents the translation offset.

[0037] The criteria for determining the validity of an event are as follows: if the acoustic emission event belongs to the spatial domain of the SEM image, it is determined to be a valid acoustic emission event; otherwise, it is determined to be invalid and will not be included in the SEM image discrimination.

[0038] The spatial domain of a SEM image is defined as: ; Validity conditions: ; Among them, the extended boundary domain , Use an error radius disk to ensure that valid events are within the boundary. , For projection error, This represents the upper limit of the projection error.

[0039] In some embodiments, when the acoustic emission event is a valid event, determining the crack type corresponding to the acoustic emission event based on waveform parameters includes: If the acoustic emission event is a valid event, obtain the first ratio of the average frequency and rise angle corresponding to each acoustic emission event, and the second ratio corresponding to the maximum average frequency and maximum rise angle among all acoustic emission events; If the first ratio is greater than the second ratio, the crack in the acoustic emission event corresponding to the first ratio is identified as a tensile crack. If the first ratio is less than the second ratio, the crack in the acoustic emission event corresponding to the first ratio is identified as a shear crack.

[0040] In this embodiment, the AF / RA value of a single event and the AF / RA value of all acoustic emission events are compared. By comparison, the crack type represented by each acoustic emission event can be classified. If the AF / RA value of the event is greater than k, it is judged as a tensile crack; if the AF / RA value of the event is less than k, it is judged as a shear crack.

[0041] In some embodiments, the preset machine learning model includes an identity branch, a spatial branch, and a channel branch; training the preset machine learning model using SEM scan images with added class labels yields a target machine learning model, including: The SEM scan images with added category labels are input into the identity branch, spatial branch, and channel branch respectively to generate the original features, spatial enhancement features, and channel enhancement features respectively; Adaptive weighting is applied to the original features, spatial enhancement features, and channel enhancement features until the generated target features meet the discrimination requirements. The target machine learning model is then determined based on the corresponding weight parameters.

[0042] In this embodiment, the machine learning model can be a ResNet-50+Adaptive CBAM model. This model includes an adaptive attention module, an identity branch, a spatial branch, and a channel branch. The identity branch directly passes the original feature map, the channel branch generates a channel attention map, and the spatial branch generates a spatial attention map.

[0043] Global Average Pooling (GAP): Calculates the global average value for each branch. : ; Global Max Pooling (GMP): Captures the maximum activation value for each channel. : ; in, The spatial location of the c-th channel of the input feature map The activation value at the location, where H and W are the height and width of the feature map (in pixels), and C is the number of channels in the feature map.

[0044] The results of GAP and GMP are respectively input into an MLP (Multilayer Perceptron) with shared parameters: ; (Dimensionality reduced to C / r, usually r=16). (Restore original number of channels) The channel weights are obtained by summing the outputs of the two paths and applying Sigmoid activation. : ; in, For the Sigmoid function, the output range is... ; Adaptive weighting is applied to the original features, spatial enhancement features, and channel enhancement features: ; in, This indicates channel-by-channel multiplication.

[0045] The goal of spatial branching is to focus on key regions in the spatial dimension, such as crack tips and pore edges, while weakening uniform background areas. The specific implementation steps are as follows: Channel aggregation performs both average pooling and max pooling along the channel dimension, generating two... Feature map: ; The two feature maps are concatenated. The tensors are used to generate spatial weights through a 7×7 convolutional layer. : ; Feature weighting: .

[0046] Core of dynamic weight fusion of channel attention and spatial attention: ; In the formula, and , These are trainable parameters. They adapt to different feature regions: flat regions correspond to... Value increases; color sensitive area Value increases; complex texture areas Value increases The weighted output features are obtained as follows: ; The pre-set machine learning model learns the correlation between the output features of a large number of labeled images and the category labels, continuously optimizes the model parameters, and finally trains a target machine learning model that can accurately identify crack types.

[0047] In one optional embodiment, a rock mechanics testing machine is used to conduct Brazilian splitting and direct shear tests. The loading rate is 0.05 mm / min. Before the test, the wave velocity of the rock is measured using a wave velocity detector. A reference point is set on the test surface, and four acoustic emission piezoelectric sensors are attached to the sample surface. The acoustic emission acquisition parameters are set as follows: threshold: 100 mV, HDT (duration time) 50 microseconds, PDT (amplitude definition time) 200 microseconds, HLT (lock-in time) 300 microseconds. Taking the acquisition of the SEM image set of the Brazilian splitting test as an example, the acquisition of the image set of the direct shear test is the same. After installing the sensors and adjusting the parameters, the test is started and continues until the sample fails. After failure, the waveform file acquired by the acoustic emission device is exported. The event location point information is obtained from the waveform file, and the cross-section of the sample after failure is obtained. The sample is cut into 1 mm thick slices, and the cross-section of the slices is divided into sections for electron microscopy scanning. The SEM images of the section scans are obtained and stitched together. After exporting the waveform files, Python was used to separate the waveform image of each event and obtain the waveform parameters: rise time, duration, ring count, amplitude, etc. Then, the crack type represented by each acoustic emission event was determined by AF and RA. The results of the acoustic emission event parameter separation are shown in Table 1.

[0048] Table 1. Parameter separation results for a single acoustic emission event

[0049] Based on the occurrence time and location information of each acoustic emission event, the segmented SEM images are categorized using the acoustic emission parameter determination results, resulting in the final SEM image set. To ensure accurate spatial positioning of the acoustic emission signals relative to the SEM images, the relative positional relationship between two reference points must be recorded, or the same reference point can be selected. To obtain as much dataset as possible, segmentation is based on individual acoustic emissions, following the principles below: The spatial location of the scanned SEM image is defined as a spatial domain. First, it is determined whether a particular acoustic emission event belongs to this spatial domain. Since acoustic emission events do not all originate from the fracture surface, but may also originate from the compression of spatial fissures in the rock, this step serves to eliminate invalid acoustic emission events. When an acoustic emission event belongs to the spatial domain defined by the SEM image, a circle with a diameter of 0.5 mm is drawn with the acoustic emission event as its center, and the cut area is the circumscribed square of this circle.

[0050] The SEM image set obtained from the above steps is preprocessed. Randomly combined geometric transformations are applied to the SEM images, including horizontal / vertical flipping, rotation, and offset. Then, the SEM images are uniformly scaled to 224×224 pixels for size standardization to fit the input specifications of subsequent convolutional networks. The CLAHE algorithm (contrast-limited adaptive histogram equalization) is used to enhance the contrast of the input images. The CLAHE module (preprocessing stage) divides the complete SEM image into multiple 32×32 pixel blocks. Histogram equalization is performed independently on each block to enhance the local contrast of microstructural details, such as grain boundaries and cracks. All blocks are smoothly stitched together using bilinear interpolation to output the enhanced complete image. Initial feature extraction is performed using a 7×7 convolutional kernel (stride=2, padding=3), outputting a 112×112×64 dimensional feature map. The activation function used is ReLU non-linearization. A 3×3 max-pooling layer (stride=2) is then applied to compress the spatial dimensions to 56×56×64, preserving significant texture features. Local enhancement involves adjusting the contrast of only small regions to avoid global overexposure. Detail preservation is achieved through small block segmentation to ensure that details such as micro-cracks are not blurred.

[0051] The enhanced image set is imported into the machine learning model ResNet-50+Adaptive CBAM to extract and identify crack morphology features from SEM images. Residual block group 1 extracts shallow features from the data: it performs three consecutive residual operations (ResBlock1×3), each block containing: two 3×3 convolutional layers (maintaining 64 channels), batch normalization, and an identity mapping for skip connections. A convolution with a stride of 2 is used in the first ResBlock1 to achieve downsampling to 28×28×128. Residual block group 2 (ResBlock2×4): 128 channels, output size 28×28×128; Residual block group 3 (ResBlock3×6): 256 channels, output size 14×14×256; Residual block group 4 (ResBlock4×3): 512 channels, output size 7×7×512. Each residual block group is followed by a CBAM module to achieve multi-granularity feature optimization. Channel weights are generated by global average / max pooling to an MLP, suppressing irrelevant filter responses. Features are aggregated along the channel dimension to generate spatial weights through a 7×7 convolution, focusing on key regions while maintaining an output dimension of 56×56×64. A global average pooling layer compresses the 7×7×512 feature map into a 1×1×512 dimensional vector. A fully connected layer (FC layer) maps the 512-dimensional vector to the number of classes, such as two classes: shear crack / tension crack. The Softmax function is activated, probabilities are normalized, and the final class probability distribution is output. The specific dimensional changes of the ResNet-50+CBAM model are shown in Table 2.

[0052] Table 2. Dimensional Changes in the ResNet-50+CBAM Model

[0053] In some embodiments, please refer to Figure 3 , Figure 3 This is a schematic diagram of a crack identification system based on the fusion of acoustic emission and scanning electron microscopy images, provided in an embodiment of the present invention. The present invention provides a crack identification system 300 based on the fusion of acoustic emission and scanning electron microscopy images, comprising: a parameter acquisition module 310, a type determination module 320, and a training and identification module 330; wherein, The parameter acquisition module 310 is configured to perform splitting and shearing tests on rock samples, acquire acoustic emission time-domain waveform signals, separate the acoustic emission time-domain waveform signals, and determine the waveform parameters and spatial distribution of each acoustic emission event; the waveform parameters include the average frequency and rise angle of a single acoustic emission event; the spatial distribution includes the true coordinates of the acoustic emission event in the rock sample; The type determination module 320 is configured to acquire the spatial domain corresponding to the SEM scan image of the rock sample, and, when the real coordinates are located in the spatial domain, determine the crack type corresponding to the acoustic emission event based on the waveform parameters, and use the crack type as the category label of the SEM scan image; the crack type includes tensile cracks and shear cracks. The training and recognition module 330 is configured to train a preset machine learning model using SEM scan images with added category labels to obtain a target machine learning model, and to use the target machine learning model to recognize the SEM image to be tested and determine the crack type corresponding to the SEM image to be tested.

[0054] In some embodiments, the parameter acquisition module 310 is specifically configured as follows: The acoustic emission time-domain waveform signal is separated to obtain the ring count parameter, duration parameter, peak amplitude parameter, and rise time parameter. The ring count parameter represents the number of oscillations of the signal exceeding a preset threshold. The duration parameter represents the total time from the first time the signal exceeds the preset threshold to the last time it falls below the preset threshold. The peak amplitude parameter represents the peak amplitude of the signal waveform within the HDT time starting from the first time the preset threshold is exceeded. The rise time parameter represents the time interval from the first time the signal exceeds the preset threshold to the time interval from the first time the signal exceeds the preset threshold to the peak amplitude being reached. The average frequency is determined based on the ratio of the ring count parameter to the duration parameter, and the rise angle is determined based on the ratio of the rise time parameter to the peak amplitude parameter.

[0055] In some embodiments, the parameter acquisition module 310 is specifically configured as follows: Based on the time difference or phase difference of the acoustic emission time-domain waveform signal arriving at different sensors, combined with the wave velocity and the geometric relationship of the sensor array, the initial coordinates of each acoustic emission event in the rock sample are determined. The initial coordinates are corrected based on internal damage and internal temperature to obtain the true coordinates of the acoustic emission event.

[0056] In some embodiments, the crack identification system 300 based on the fusion of acoustic emission and scanning electron microscopy images further includes an image acquisition module; the image acquisition module is specifically configured as follows: Based on the same reference point, the fracture surface of the rock sample was scanned by SEM partitioning to obtain multiple frames of SEM scan sub-images; The SEM scan sub-images are stitched together based on the reference points to obtain the SEM scan image of the rock sample.

[0057] In some embodiments, the type determination module 320 is specifically configured as follows: The spatial domain is determined by the two-dimensional coordinate extrema of the SEM scan images of rock samples; the two-dimensional coordinate extrema include the maximum, minimum, maximum and minimum values ​​of the abscissa, and the maximum and minimum values ​​of the ordinate. The real coordinates are projected and transformed onto the plane where the SEM scan image is located. If the projected coordinates are located in the spatial domain, the acoustic emission event is determined to be a valid event; if the projected coordinates are not located in the spatial domain, the acoustic emission event is determined to be an invalid event. When the acoustic emission event is a valid event, the crack type corresponding to the acoustic emission event is determined based on the waveform parameters.

[0058] In some embodiments, the type determination module 320 is specifically configured as follows: If the acoustic emission event is a valid event, obtain the first ratio of the average frequency and rise angle corresponding to each acoustic emission event, and the second ratio corresponding to the maximum average frequency and maximum rise angle among all acoustic emission events; If the first ratio is greater than the second ratio, the crack in the acoustic emission event corresponding to the first ratio is identified as a tensile crack. If the first ratio is less than the second ratio, the crack in the acoustic emission event corresponding to the first ratio is identified as a shear crack.

[0059] In some embodiments, the preset machine learning model includes an identity branch, a spatial branch, and a channel branch; the training and recognition module 330 is specifically configured as follows: The SEM scan images with added category labels are input into the identity branch, spatial branch, and channel branch respectively to generate the original features, spatial enhancement features, and channel enhancement features respectively; Adaptive weighting is applied to the original features, spatial enhancement features, and channel enhancement features until the generated target features meet the discrimination requirements. The target machine learning model is then determined based on the corresponding weight parameters.

[0060] It should be noted that the crack identification system based on acoustic emission and scanning electron microscope image fusion provided in this application embodiment and the crack identification method based on acoustic emission and scanning electron microscope image fusion provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned crack identification method based on acoustic emission and scanning electron microscope image fusion, and the repeated parts will not be described again.

[0061] In some embodiments, please refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 400 provided in this application includes a processor 410 and a memory 420; the memory 420 stores a computer program, wherein the computer program, when executed by the processor, implements the aforementioned crack identification method based on the fusion of acoustic emission and scanning electron microscopy images.

[0062] Specifically, processor 410 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. Processor 410 may also include onboard memory for caching purposes. Processor 410 may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0063] Memory 420 may be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory 420 may include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory 420 include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and may also be random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0064] This application also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, this program implements the aforementioned crack identification method based on the fusion of acoustic emission and scanning electron microscopy images. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0065] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0066] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A crack identification method based on the fusion of acoustic emission and scanning electron microscopy images, characterized in that, include: Rock samples were subjected to splitting and shearing tests to collect acoustic emission time-domain waveform signals. The acoustic emission time-domain waveform signals were then separated to determine the waveform parameters and spatial distribution of each acoustic emission event. The waveform parameters included the average frequency and rise angle of a single acoustic emission event. The spatial distribution includes the actual coordinates of the acoustic emission event in the rock sample; The process involves obtaining the spatial domain corresponding to the SEM scan image of the rock sample, and determining the crack type corresponding to the acoustic emission event based on the waveform parameters when the real coordinates are located within the spatial domain. This includes: projecting and transforming the real coordinates onto the plane of the SEM scan image; if the projected coordinates are located within the spatial domain, determining the acoustic emission event as a valid event; if the projected coordinates are not located within the spatial domain, determining the acoustic emission event as an invalid event; when the acoustic emission event is a valid event, obtaining a first ratio of the average frequency to the rise angle corresponding to each acoustic emission event, and a second ratio corresponding to the maximum average frequency and maximum rise angle among all acoustic emission events; if the first ratio is greater than the second ratio, determining the crack of the acoustic emission event corresponding to the first ratio as a tensile crack; if the first ratio is less than the second ratio, determining the crack of the acoustic emission event corresponding to the first ratio as a shear crack; and using the crack type as the category label of the SEM scan image. A preset machine learning model is trained using the SEM scan image with added category labels to obtain a target machine learning model. The target machine learning model is then used to identify the SEM image under test to determine the crack type corresponding to the SEM image under test.

2. The crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in claim 1, characterized in that, The separation of the acoustic emission time-domain waveform signal to determine the waveform parameters and spatial distribution of each acoustic emission event includes: The acoustic emission time-domain waveform signal is separated to obtain ring count parameters, duration parameters, peak amplitude parameters, and rise time parameters. The ring count parameter represents the number of oscillations of the signal exceeding a preset threshold. The duration parameter represents the total time from the first time the signal exceeds the preset threshold to the last time it falls below the preset threshold. The peak amplitude parameter represents the peak amplitude of the signal waveform within the HDT time, starting from the first time the preset threshold is exceeded. The rise time parameter represents the time interval from the first time the signal exceeds the preset threshold to the time interval from when the signal first exceeds the preset threshold to when the peak amplitude is reached. The average frequency is determined based on the ratio of the ring count parameter to the duration parameter, and the rise angle is determined based on the ratio of the rise time parameter to the peak amplitude parameter.

3. The crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in claim 1, characterized in that, The separation of the acoustic emission time-domain waveform signal to determine the waveform parameters and spatial distribution of each acoustic emission event includes: Based on the time difference or phase difference of the acoustic emission time-domain waveform signal arriving at different sensors, combined with the wave velocity and the geometric relationship of the sensor array, the initial coordinates of each acoustic emission event in the rock sample are determined. The initial coordinates are corrected based on internal damage and internal temperature to obtain the true coordinates of the acoustic emission event.

4. The crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in claim 1, characterized in that, Before obtaining the spatial domain corresponding to the SEM scan image of the rock sample, the process also includes: Based on the same reference point, the fracture surface of the rock sample was scanned by SEM partitioning to obtain multiple frames of SEM scan sub-images; The SEM scan sub-images are stitched together based on the reference point to obtain the SEM scan image of the rock sample.

5. The crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in claim 1, characterized in that, The step of acquiring the spatial domain corresponding to the SEM scan image of the rock sample, and determining the crack type corresponding to the acoustic emission event based on the waveform parameters when the true coordinates are located in the spatial domain, includes: The spatial domain is determined based on the two-dimensional coordinate extrema of the SEM scan image of the rock sample; the two-dimensional coordinate extrema include the maximum value of the horizontal coordinate, the minimum value of the horizontal coordinate, the maximum value of the vertical coordinate, and the minimum value of the vertical coordinate.

6. The crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in claim 1, characterized in that, The preset machine learning model includes identity branch, spatial branch, and channel branch; The step of training a preset machine learning model using the SEM scanned image with added category labels to obtain a target machine learning model includes: The SEM scan image with added category labels is input into the identity branch, the spatial branch, and the channel branch respectively to generate the original feature, spatial enhancement feature, and channel enhancement feature respectively; The original features, spatial enhancement features, and channel enhancement features are adaptively weighted until the generated target features meet the discrimination requirements, and the target machine learning model is determined based on the corresponding weight parameters.

7. A crack identification system based on the fusion of acoustic emission and scanning electron microscopy images, characterized in that, include: The module consists of a parameter acquisition module, a type determination module, and a training and recognition module; among which, The parameter acquisition module is configured to perform splitting and shearing tests on the rock sample, acquire acoustic emission time-domain waveform signals, and separate the acoustic emission time-domain waveform signals to determine the waveform parameters and spatial distribution of each acoustic emission event; the waveform parameters include the average frequency and rise angle of a single acoustic emission event; the spatial distribution includes the true coordinates of the acoustic emission event in the rock sample; The type determination module is configured to acquire the spatial domain corresponding to the SEM scan image of the rock sample, and, when the real coordinates are located in the spatial domain, determine the crack type corresponding to the acoustic emission event based on the waveform parameters, including: projecting and transforming the real coordinates to the plane where the SEM scan image is located; if the projected coordinates are located in the spatial domain, determining the acoustic emission event as a valid event; if the projected coordinates are not located in the spatial domain, determining the acoustic emission event as an invalid event; when the acoustic emission event is a valid event, acquiring a first ratio of the average frequency and rise angle corresponding to each acoustic emission event, and a second ratio corresponding to the maximum average frequency and maximum rise angle among all acoustic emission events; if the first ratio is greater than the second ratio, determining the crack of the acoustic emission event corresponding to the first ratio as a tensile crack; if the first ratio is less than the second ratio, determining the crack of the acoustic emission event corresponding to the first ratio as a shear crack; and using the crack type as a category label for the SEM scan image; The training and recognition module is configured to train a preset machine learning model using the SEM scan image with added category labels to obtain a target machine learning model, and to use the target machine learning model to recognize the SEM image to be tested to determine the crack type corresponding to the SEM image to be tested.

8. An electronic device, characterized in that, It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, characterized in that, It stores a computer program, wherein when the computer program is executed by a processor, it implements the crack identification method based on the fusion of acoustic emission and scanning electron microscopy images as described in any one of claims 1 to 6.

Images