Composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning

Abstract

The invention relates to the technical field of composite material damage detection, and particularly discloses a composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning, and the method comprises the steps: S1, data collection, S2, ultrasonic data processing, S3, simulation model construction, S4, PVT-RCNN model processing, S5, RI CP processing, and S6, damage model generation. According to the method, an ultrasonic point cloud data acquisition platform and a data processing method are designed, a sufficient composite laminate ultrasonic point cloud data set with obvious features is constructed, a PVT-RCNN model which is based on PV-RCNN and is improved by Swi n Transformer is designed, the feature extraction capability of the model on point cloud is improved, three-dimensional space information of damage is intelligently obtained, and the accuracy of the damage is improved. An RI CP registration algorithm and a damage model generation method are designed, a finite element model with damage is directly generated according to damage point cloud, and the material performance can be obtained through calculation.

Description

Technical Field

[0001] The present invention belongs to the technical field of composite material damage detection, and specifically relates to an intelligent detection and modeling method for composite material damage based on ultrasonic point cloud and deep learning. Background Art

[0002] Advanced fiber-reinforced composite materials are increasingly used in high-performance structural applications, such as aircraft main structural components, wind turbine blades, etc., due to their high specific stiffness and specific strength, corrosion resistance and fatigue resistance. Composite materials are subject to alternating loads, foreign object impacts, etc. during use, which can easily cause damage such as debonding and delamination, affecting the performance and life of the material. For this reason, composite materials need to be regularly inspected and monitored to identify damage, evaluate and treat it before the damage develops to a level where the structural strength is below the acceptable limit.

[0003] In order to inspect composite materials, there are several common nondestructive testing methods: visual inspection, ultrasonic testing, infrared thermal imaging, terahertz testing, X-ray tomography (CT), etc. The most widely used in nondestructive testing of composite materials is ultrasonic testing technology, which has the advantages of high resolution, real-time dynamic imaging, high safety, convenience and speed. Common ultrasonic testing results include A scan, B scan, C scan, etc., which can intuitively distinguish between damage and normal areas, but cannot directly obtain the three-dimensional spatial information of damage. Therefore, many studies have been carried out on ultrasonic three-dimensional imaging in recent years. Benjamin et al. used a high-resolution ultrasonic C-scan system to detect low-speed impact damage of composite materials, and presented the internal 3D contour of the damage through an unbiased layer-by-layer technique. McKee et al. used a two-dimensional phased array probe to experimentally image artificial damage in a hyperbolic test piece under immersion. The experiment successfully obtained a three-dimensional imaging of the inside of the test piece with a root mean square error of only 0.04 mm. Ohara et al. used a Phased-Array system based on a piezoelectric and laser ultrasonic system (PLUS) to create high-resolution 3D images of internal damage. Wang et al. constructed a scanning system based on phased array ultrasound to detect low-speed impact damage of composite thin plates, and achieved three-dimensional size quantification of damage. Jiang et al. used in-situ computed tomography and digital volume correlation technology to detect composite materials and constructed a three-dimensional model. Three-dimensional imaging methods mainly include point clouds and voxels. Ultrasonic three-dimensional imaging based on the principle of ultrasonic reflection is a point cloud imaging method. Ultrasonic point clouds contain not only material surface information, but also material internal information, and can directly obtain three-dimensional spatial information of damage.

[0004] The performance evaluation methods for composite materials after damage are: obtaining empirical parameter evaluation through experiments, which is a commonly used method in industry; or using finite element modeling for numerical simulation. There are a lot of studies on finite element analysis of composite material damage. Zhou et al. used a progressive damage model to study the mechanical response and damage development of laminates under low-speed impact. Higuchi et al. designed a microscale simulation scheme to evaluate the in-situ damage and strength characteristics of laminates and effectively predicted the propagation of transverse cracks. The 3D finite element-progressive damage method (FE-PDM) proposed by Liu et al. can accurately predict the complex mechanical behavior and damage evolution of composite materials. However, constructing a composite material damage model is complicated. Therefore, it is of great significance to quickly generate a composite material damage model using the results of ultrasonic testing and conduct finite element analysis experiments.

[0005] In response to this, the inventors proposed an intelligent detection and modeling method for composite material damage based on ultrasonic point cloud and deep learning to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent detection and modeling method for composite material damage based on ultrasonic point cloud and deep learning to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The intelligent detection and modeling method of composite material damage based on ultrasonic point cloud and deep learning includes the following steps:

[0009] S1. Data acquisition: build an experimental platform, use a robotic arm to control the probe to scan the material along the specified path, and obtain ultrasonic data, including A-scan signal, B-scan image, C-scan image and ultrasonic point cloud;

[0010] S2, ultrasonic data processing, integrating the ultrasonic signal with the coordinates of the end of the robotic arm to obtain the original ultrasonic point cloud data, performing attenuation compensation, body mass center downsampling, half-wave height method and data enhancement processing to obtain the required ultrasonic data features;

[0011] S3, build a simulation model, build a damage-free material model according to the characteristics of the composite material specimen, mesh the node point cloud according to the damage-free material model, extract the coordinates corresponding to each node in the inp file, and finally form a point cloud;

[0012] S4, PVT-RCNN model processing, including using SwinTransformer feature extraction network for voxel feature extraction, region proposal network to generate preliminary 3D bounding box, obtain key points mapped to voxel feature maps of different scales of the backbone network, obtain fused key point features by splicing these key point features and sending them to the key point prediction module, and send the key point features and preliminary predicted 3D bounding box together to the RoI-grid Pooling Module to obtain the final predicted bounding box and confidence;

[0013] S5, RICP processing, point cloud registration work is carried out, the damage corresponding to the damage point cloud identified by the PVT-RCNN model is generated into the finite element model, and point cloud registration work is carried out;

[0014] S6. Generate a damage model, complete the coordinate transformation of the ultrasonic point cloud and damage extraction, and then reflect the information of the damage point cloud on the model.

[0015] Preferably, the attenuation expression of the plane wave in the attenuation compensation in step S2 is:

[0016] P x =P0e -αx

[0017] Where P0 is the initial sound pressure of the wave source, P x is the sound pressure at a distance x from the wave source, e is the natural logarithm, α is the medium attenuation coefficient, unit is dB / mm, and x is the distance from the wave source;

[0018] The compensation formula based on the ultrasonic attenuation formula is:

[0019] I n =I0e -αx

[0020] Where I0 is the initial signal strength; I n is the signal strength after compensation; x is the distance from the collected signal to the upper surface of the material.

[0021] Preferably, the centroid downsampling in step S2 divides the point cloud data space into a series of small three-dimensional grids, which are called voxels. The centroid of all points inside each voxel is calculated, and all points in the voxel are replaced by the centroid. In this way, the original point cloud data is replaced by a group of evenly distributed and small number of points, thus realizing downsampling. The centroid calculation expression is:

[0022]

[0023] By comparing the two types of data, we can see that physical fitness downsampling effectively reduces the amount of data, and the characteristics of the data can also be preserved.

[0024] Preferably, the half-wave height method, i.e., the 6db-drop method, processes the damage point cloud. After the network identifies the damage point cloud, it finds the maximum signal strength therein, sets half of its signal strength as the threshold, removes points less than the threshold, and realizes further processing of the damage point cloud, thereby obtaining more accurate damage information.

[0025] Preferably, the data enhancement processing in step S2 is used to collect data from 40 composite materials with different pre-embedded damages. In order to obtain more experimental data, a total of 160 groups of ultrasonic point cloud data are obtained by selecting different ultrasonic gains and front and back side collection methods. Random flipping, random rotation, random scaling and noise addition are used to finally obtain 800 groups of data, of which the training set and test set are 600 groups and 200 groups respectively. The ultrasonic point cloud of the damaged area is annotated using "CloudCompare".

[0026] Preferably, in step S4, the Swin Transformer feature extraction network is used to extract voxel features, specifically: the ultrasonic point cloud is voxelized, and then Patch Partition processing is input to obtain multiple patches of the same size, and then Linear Enbedding and two Swin Transformer Blocks are input to obtain the first layer of feature maps, and then multiple Patch Merging and Swin Transformer Blocks are performed to obtain multi-scale feature maps with resolutions of 1 / 2, 1 / 4 and 1 / 8 of the original input 2D voxel feature map.

[0027] Preferably, the Swin Transformer block consists of a shifted window based MSA module followed by 2 layers of MLP with a GELU nonlinearity in between, a LayerNorm (LN) layer is applied before each MSA module and each MLP, and a residual connection is applied after each module;

[0028] Layer 1 adopts a conventional window partitioning scheme, which is used to divide the voxel feature map to generate a local window. Focusing the self-attention in the local window can effectively improve the computational efficiency and avoid the interference caused by irrelevant distant pixels;

[0029] The window partition in Layer l+1 is shifted to generate a new window, which is achieved by moving the window position on the X and Y axes with the same step length. The self-attention calculation in the new window spans the boundaries of the previous window in the previous layer, thus providing a connection between them.

[0030] Preferably, the Swin Transformer Block is constructed by replacing the standard multi-head self-attention MSA module in the Transformer block with a shift window-based module, and the other layers remain unchanged. The calculation process of the two Swin TransformerBlocks is expressed as:

[0031]

[0032] in, and z l They represent the SW-MSA module output features and the MLP module output features of the lth block respectively.

[0033] Preferably, the RICP processing in step S5 is based on the RICP algorithm of the Rodrigues rotation matrix to prevent the registration from falling into the local optimum, and effectively replace the coarse registration step. The formula of the Rodrigues rotation matrix is:

[0034]

[0035] The ultrasonic point cloud is multiplied by a uniformly decreasing rotation matrix, which causes the point cloud to rotate around the Z axis. The value uniformly decreases from 90° to 0° with the number of iterations. Corresponding points are searched in the rotated ultrasonic point cloud and the node point cloud. The rotation matrix R and translation matrix t are obtained by calculation. The ultrasonic point cloud coordinates are updated according to R and t, and the mean square error E(R, t) between the ultrasonic point cloud and the node point cloud is calculated. Finally, the error value is compared with the set iteration threshold. If it is greater than or equal to the iteration threshold, it is iterated again. If it is less than or equal to the iteration threshold, the iteration ends and the point cloud registration is completed.

[0036] Preferably, in step S1, the experimental platform includes a robotic arm, a water tank, an ultrasonic probe and a probe fixture.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] (1) The present invention designs an ultrasonic point cloud data acquisition platform and data processing method, constructs a sufficient ultrasonic point cloud data set of composite laminates with obvious characteristics, designs a PVT-RCNN model based on PV-RCNN and improved by Swin Transformer, improves the model's feature extraction capability for point clouds, intelligently obtains three-dimensional spatial information of damage, designs a RICP registration algorithm and a damage model generation method, directly generates a damaged finite element model based on the damage point cloud, and obtains material properties through calculation.

[0039] (2) In the present invention, the robotic arm controls the ultrasonic probe to scan the composite material to obtain raw data, and then obtains ultrasonic point cloud data with obvious characteristics after data processing. At the same time, a damage-free composite material model is constructed according to the characteristics of the composite material, and its node coordinates are extracted to obtain the node point cloud. Then, the PVT-RCNN intelligently detects the damage point cloud in the ultrasonic point cloud, and obtains the spatial position and size of the damage through the 6dB-drop method. The ultrasonic point cloud and the node point cloud are aligned by the RICP algorithm to obtain the transformation matrix (R, t). The results of the two can be fused to obtain the damage point cloud that matches the composite material model. Finally, the damage represented by the damage point cloud is generated into the composite material model through the damage model generation module to obtain a composite material model with damage. Finite element analysis of the model can be performed to evaluate the performance of the composite material being tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Ultrasound data graph obtained for the present invention;

[0041] Figure 2 A flow chart for generating a damage model based on a damage point cloud of the present invention;

[0042] Figure 3 This is a rendering of the physical fitness center downsampling of the present invention;

[0043] Figure 4 It is the effect diagram of the straight-through filter of the present invention;

[0044] Figure 5 This is a set of enhanced effect diagrams of ultrasonic point cloud data of the present invention;

[0045] Figure 6 The schematic diagram of the MSA module based on the shift window of the present invention;

[0046] Figure 7 This is a strip damage detection result diagram of the present invention;

[0047] Figure 8 This is a square damage detection result diagram of the present invention;

[0048] Fig. 9 This is a circular damage detection result diagram of the present invention;

[0049] Fig.10 This is a diagram of the registration experiment results of the ultrasonic point cloud and the node point cloud of the present invention;

[0050] Fig.11 A damage model diagram generated based on a circular damage point cloud according to the present invention;

[0051] Fig.12 A damage model diagram generated based on a square damage point cloud according to the present invention;

[0052] Fig.13 It is the overall framework diagram of the experiment of the present invention;

[0053] Fig.14 Schematic diagram of the experimental platform of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Embodiment 1:

[0056] See also Figures 1 to 14 As shown, the intelligent detection and modeling method of composite material damage based on ultrasonic point cloud and deep learning includes the following steps:

[0057] S1. Data collection and building of experimental platform, such as Fig.14 As shown in the figure, the experimental platform includes a robotic arm, a water tank, an ultrasonic probe and a probe fixture. The robotic arm is used to control the probe to scan the material along the specified path. The ultrasonic data obtained include A-scan signal, B-scan image, C-scan image and ultrasonic point cloud; the robotic arm is a JAKA Zu 7 collaborative robot, the ultrasonic A-scan device is a UFD BOX, and the ultrasonic probe is a 5MHz straight probe. The experimental material is a prefabricated carbon fiber laminate with internal damage, with a size of 150mm×100mm×5mm, a material sound velocity of 2900m / s, and a collection range of 7mm. The dotted line in the figure represents the probe movement path planning, and the robotic arm controls the probe to scan the material along the S-shaped path in the figure. The data sampling range is: X=170mm, Y=120mm. Figure 1 The ultrasonic data obtained in the experiment of the present invention include A-scan signal, B-scan image, C-scan image and ultrasonic point cloud, where T represents the upper surface signal of the specimen, B represents the lower surface signal of the specimen, and D represents the damage signal. It can be seen from the figure that A, B, and C scans can all identify damage, but cannot obtain the spatial information of the damage. The ultrasonic point cloud contains information such as the spatial position and size of the damage, based on which a finite element model of a composite material containing damage can be established;

[0058] S2, ultrasonic data processing, integrating the ultrasonic signal with the coordinates of the end of the robotic arm to obtain the original ultrasonic point cloud data, which contains 4 sets of data (x, y, z, intensity), where x and y are provided by the robotic arm, z is the depth corresponding to the ultrasonic signal, and intensity is the intensity of the ultrasonic signal. The required ultrasonic data features are obtained by performing attenuation compensation, physical centroid downsampling, pass-through filter and data enhancement processing;

[0059] S3. Construct a simulation model based on the characteristics of the composite material specimen. Fig.13 The damage-free material model in the overall experimental framework diagram has a size of 150mm*100mm*5mm and consists of 40 layers. The layers are cohesive units with 0 thickness. The ply sequence is: [45 / 0 / -45 / 0 / 90 / 0 / 45 / 0 / -45 / 0 / 45 / 0 / -45 / 0 / 90 / 0 / 45 / 0 / -45 / 0]s. In order to balance the calculation efficiency and accuracy, the grid density is 1mm×1mm in the middle 70mm square area, and 1mm×2mm and 2mm×2mm in the surrounding areas. Fig.13 The node point cloud in the overall framework diagram of the experiment is the point cloud finally formed by extracting the coordinates corresponding to each node in the inp file after meshing the damage-free material model;

[0060] S4, PVT-RCNN model processing, including using Swin Transformer feature extraction network for voxel feature extraction, region proposal network to generate preliminary 3D bounding box, obtain key points mapped to different scales of voxel feature maps of the backbone network, obtain fused key point features by splicing these key point features and sending them to the key point prediction module, and send the key point features and preliminary predicted 3D bounding box together to the RoI-grid Pooling Module to obtain the final predicted bounding box and confidence;

[0061] S5, RICP processing, point cloud registration work is carried out, the damage corresponding to the damage point cloud identified by the PVT-RCNN model is generated into the finite element model, and point cloud registration work is carried out;

[0062] S6. Generate a damage model. After completing the coordinate transformation and damage extraction of the ultrasonic point cloud, reflect the information of the damage point cloud on the model. Calculate the area of ​​the unit through the coordinates of the unit node in the inp file, and use a loop to determine whether all the units in the middle layer contain the damage site in the scan result. If it contains one or more damage sites, the unit will be included in a collection and written to a new inp file, and deleted in abaqusCAE. Because only layered damage is considered, only the cohesive layer units of the model are judged and deleted.

[0063] From the above, it can be seen that by designing an ultrasonic point cloud data acquisition platform and data processing method, a sufficient ultrasonic point cloud data set of composite laminates with obvious characteristics was constructed, and a PVT-RCNN model based on PV-RCNN and improved by SwinTransformer was designed to enhance the model's feature extraction capability for point clouds and intelligently obtain three-dimensional spatial information of damage. The RICP registration algorithm and damage model generation method were designed, and a damaged finite element model was directly generated based on the damage point cloud, and material properties can be obtained through calculation.

[0064] Embodiment 2:

[0065] See also Figures 1 to 14 As shown in Figure 2, the attenuation expression of the plane wave in attenuation compensation is:

[0066] P x =P0e -αx

[0067] Where P0 is the initial sound pressure of the wave source, P x is the sound pressure at a distance x from the wave source, e is the natural logarithm, α is the medium attenuation coefficient, unit is dB / mm, and x is the distance from the wave source;

[0068] The compensation formula based on the ultrasonic attenuation formula is:

[0069] I n =I0e -αx

[0070] Where I0 is the initial signal strength; I n is the signal strength after compensation; x is the distance from the collected signal to the upper surface of the material;

[0071] For this material, the compensation formula is obtained based on the ultrasonic attenuation formula: I n =I0e -0.002x

[0072] like Figure 2 The figure below is the effect diagram of attenuation compensation. The upper figure is the original ultrasonic A-scan signal, and the lower figure is the compensated ultrasonic A-scan signal. From the results, we can see that attenuation compensation can effectively correct the energy loss of the signal during the transmission process and ensure the accuracy and reliability of the signal.

[0073] The centroid downsampling divides the point cloud data space into a series of small three-dimensional grids, which are called voxels. The centroid of all points inside each voxel is calculated, and all points in the voxel are replaced by the centroid. In this way, the original point cloud data is replaced by a group of evenly distributed and small number of points, thus realizing downsampling. The centroid calculation expression is:

[0074]

[0075] like Figure 3 This is an effect diagram of body mass centroid downsampling. In the present invention, the voxel is a cube with a side length of 0.1 mm. By comparing the two types of data, it can be seen that body mass centroid downsampling effectively reduces the amount of data, and the characteristics of the data can also be preserved.

[0076] The 6db-drop method sets a pass filter with a minimum strength threshold of 1000 to remove interference signals, such as Figure 4 This is the effect diagram of the straight-through filter. The top is the point cloud data before filtering, which is square and has no obvious features. The bottom is the point cloud data after filtering, which can directly distinguish the upper surface, lower surface and damage of the material, and process the damaged point cloud. After the network recognizes the damaged point cloud, it finds the maximum signal strength, sets half of its signal strength as the threshold, removes points less than the threshold, and realizes further processing of the damaged point cloud, thereby obtaining more accurate damage information.

[0077] Data enhancement processing was used to collect data from 40 composite materials with different pre-embedded damages. In order to obtain more experimental data, a total of 160 sets of ultrasonic point cloud data were obtained by selecting different ultrasonic gains and front and back acquisition methods. Random flipping, random rotation, random scaling and noise addition were used to finally obtain 800 sets of data, of which the training set and test set were 600 and 200 sets respectively. "CloudCompare" was used to annotate the ultrasonic point cloud of the damaged area, as shown in the figure. Figure 5 This is a set of enhanced renderings of ultrasonic point cloud data.

[0078] The Swin Transformer feature extraction network is used for voxel feature extraction as follows: the ultrasonic point cloud is voxelized, then input into Patch Partition processing to obtain multiple patches of the same size, and then input into LinearEnbedding and two Swin Transformer Blocks to obtain the first layer of feature maps. After that, multiple PatchMerging and Swin Transformer Blocks are performed to obtain multi-scale feature maps with resolutions of 1 / 2, 1 / 4 and 1 / 8 of the original input 2D voxel feature map.

[0079] The Swin Transformer block consists of a shifted window-based MSA module followed by 2 layers of MLP with a GELU nonlinearity in between. A LayerNorm (LN) layer is applied before each MSA module and each MLP, and a residual connection is applied after each module.

[0080] like Figure 6Schematic diagram of the MSA module based on shifted windows. Layer 1 adopts a conventional window partitioning scheme, which is used to partition the voxel feature map to generate local windows. Focusing the self-attention in the local window can effectively improve the computational efficiency and avoid the interference caused by irrelevant distant pixels.

[0081] The window partition in Layer l+1 is shifted to generate a new window, which is achieved by moving the window position on the X and Y axes with the same step length. The self-attention calculation in the new window spans the boundaries of the previous window in the previous layer, thus providing a connection between them.

[0082] The Swin Transformer Block is constructed by replacing the standard multi-head self-attention MSA module in the Transformer block with a shift window-based module, and the other layers remain unchanged. The calculation process of the two Swin Transformer Blocks is expressed as:

[0083]

[0084] in, and z l They represent the SW-MSA module output features and the MLP module output features of the lth block respectively.

[0085] In step S5, the RICP processing based on the RICP algorithm of the Rodrigues rotation matrix prevents the registration from falling into the local optimum and effectively replaces the coarse registration step. The formula of the Rodrigues rotation matrix is:

[0086]

[0087] The ultrasonic point cloud is multiplied by a uniformly decreasing rotation matrix, which causes the point cloud to rotate around the Z axis. The value uniformly decreases from 90° to 0° with the number of iterations. Corresponding points are searched in the rotated ultrasonic point cloud and the node point cloud. The rotation matrix R and translation matrix t are obtained by calculation. The ultrasonic point cloud coordinates are updated according to R and t, and the mean square error E(R, t) between the ultrasonic point cloud and the node point cloud is calculated. Finally, the error value is compared with the set iteration threshold. If it is greater than or equal to the iteration threshold, it is iterated again. If it is less than or equal to the iteration threshold, the iteration ends and the point cloud registration is completed.

[0088] As can be seen from the above, the robotic arm controls the ultrasonic probe to scan the composite material to obtain the original data, and then obtains the ultrasonic point cloud data with obvious characteristics after data processing. At the same time, a damage-free composite material model is constructed according to the characteristics of the composite material, and its node coordinates are extracted to obtain the node point cloud. Then, the PVT-RCNN intelligently detects the damaged point cloud in the ultrasonic point cloud, and obtains the spatial position and size of the damage through the 6dB-drop method. The RICP algorithm aligns the ultrasonic point cloud with the node point cloud to obtain the transformation matrix (R, t). The fusion of the two results can obtain the damaged point cloud that matches the composite material model. Finally, the damage represented by the damage point cloud is generated into the composite material model through the damage model generation module to obtain a composite material model with damage. The finite element analysis of the model can realize the performance evaluation of the tested composite material.

[0089] Embodiment three:

[0090] See also Figures 1 to 14 As shown, the hardware platform used in the present invention is an Intel(R) Core(R) 10700k CPU, an NVIDIA RTX 3090 graphics card, an operating system of Windows 10, a development language of Python, a PyTorch version of 1.7, a CUDA version of 11.1, and a simulation software of ABAQUS 2019;

[0091] 1) Ultrasonic point cloud intelligent detection experimental results:

[0092] The training process of the ultrasonic point cloud intelligent detection experiment uses the Adam optimizer to update parameters, the initial learning rate is set to 0.01, the attenuation factor is 0.8, the training epoch is set to 80, and the batch_size is set to 2. Because the detection object is only the type of damaged point cloud, the present invention uses 3D AP (representing the accuracy of the 3D detection box) as the evaluation index, and the IoU threshold is 0.7.

[0093] Figure 7-9 This is the result diagram of ultrasonic point cloud damage intelligent detection. Figure 7 is the strip damage detection result, Figure 8 is the square damage detection result, Fig. 9 The first row is the result of point cloud detection, and the second row is the damage point cloud obtained after 6dB drop method processing. The x, y, z in the figure represent the center coordinates of the detection frame, and l, w, h represent the length, width and height of the detection frame, in mm.

[0094] Depend on Figure 7It is known that the PVT-RCNN model accurately detected the strip damage point cloud in the ultrasonic point cloud, and the center point coordinates of the detection box are (64.50, 81.51, 2.58) and the size is (9.00, 55.22, 1.21). After the damage point cloud is processed by the 6dB drop method, the center point coordinates become (64.50, 81.63, 2.44), the size becomes (7.00, 53.81, 0.85), the length is reduced by 2, the width is reduced by 1.41, and the height is reduced by 0.36.

[0095] Figure 8 The coordinates of the center point of the square lesion are (67.50, 82.21, 2.50) and the size is (33.00, 66.62, 0.82).

[0096] Fig. 9 The coordinates of the center point of the circular lesion are (67.50, 84.67, 2.53) and the size is (61.00, 63.99, 0.76).

[0097] In summary, the method of the present invention can accurately identify the damage point cloud in the ultrasonic point cloud and obtain the three-dimensional spatial information of the damage point cloud. The 6dB drop method can effectively remove the interference signal and obtain more accurate damage point cloud information.

[0098] 2) Experimental results of RICP registration algorithm:

[0099] In this experiment, the ultrasonic point cloud and the node point cloud are registered, and the convergence time of the algorithm, the root mean square error (RMSE) of the point cloud distance and the registration effect are used as evaluation indicators.

[0100] Table 2 below shows the results of different registration algorithms:

[0101]

[0102] Table 2

[0103] Fig.10 This is the experimental result of the registration of ultrasonic point cloud and node point cloud.

[0104] The ultrasonic point cloud in Point cloud 1 contains a circular damage point cloud. It takes 5.8 seconds to process the ICP, and the RMSE is 0.0880mm. It can be seen from the figure that the two point clouds are 90° different in direction after registration, and are obviously not aligned. RICP takes 7.0 seconds, and the RMSE is 0.0103mm, which is 1.2 seconds longer than ICP, and the RMSE is 0.0777mm less. It can be seen from the figure that the two point clouds are basically aligned, and the effect is significantly better than ICP.

[0105] The ultrasonic point cloud in Point cloud 2 contains a square damage point cloud. It takes 5.5s for ICP and the RMSE is 0.0815mm. It can be seen from the figure that the two point clouds are 90° different in direction after registration, and are obviously not aligned. RICP takes 6.4s and the RMSE is 0.0136mm, which is 0.9s longer than ICP and 0.0679mm less than RICP. It can be seen from the figure that the two point clouds are basically aligned, and the effect is significantly better than ICP. Therefore, the RICP algorithm can accurately register the ultrasonic point cloud and the node point cloud, and has application value.

[0106] 3) Damage model generation and evaluation experimental results:

[0107] Fig.11 is a damage model generated based on a circular damage point cloud. Fig.12 This is a damage model generated based on a square damage point cloud. This figure reflects the cohesive units in the 20th and 21st plies of the model. It can be seen from the figure that the finite element model with damage can be accurately generated using the detected damage point cloud.

[0108] It can be seen from the above that the method of the present invention can accurately identify the damage point cloud in the ultrasonic point cloud and obtain the three-dimensional spatial information of the damage point cloud. The 6dB drop method can effectively remove the interference signal and obtain more accurate damage point cloud information. The RICP algorithm can accurately align the ultrasonic point cloud with the node point cloud. The detected damage point cloud can accurately generate a finite element model with damage, so the above method has a wide range of application value.

[0109] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0110] In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0111] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Intelligent detection and modeling method of composite material damage based on ultrasonic point cloud and deep learning, characterized in that: The following steps are involved: S1. Data acquisition: build an experimental platform, use a robotic arm to control the probe to scan the material along the specified path, and obtain ultrasonic data, including A-scan signal, B-scan image, C-scan image and ultrasonic point cloud; S2, ultrasonic data processing, integrating the ultrasonic signal with the coordinates of the end of the robotic arm to obtain the original ultrasonic point cloud data, performing attenuation compensation, body mass center downsampling, half-wave height method and data enhancement processing to obtain the required ultrasonic data features; S3, build a simulation model, build a damage-free material model according to the characteristics of the composite material specimen, mesh the node point cloud according to the damage-free material model, extract the coordinates corresponding to each node in the inp file, and finally form a point cloud; S4, PVT-RCNN model processing, including using Swin Transformer feature extraction network for voxel feature extraction, region proposal network to generate preliminary 3D bounding box, obtain key points mapped to different scales of voxel feature maps of the backbone network, obtain fused key point features by splicing these key point features and sending them to the key point prediction module, and send the key point features and preliminary predicted 3D bounding box together to the RoI-grid Pooling Module to obtain the final predicted bounding box and confidence; S5, RICP processing, point cloud registration work is carried out, the damage corresponding to the damage point cloud identified by the PVT-RCNN model is generated into the finite element model, and point cloud registration work is carried out; S6. Generate a damage model, complete the coordinate transformation of the ultrasonic point cloud and damage extraction, and then reflect the information of the damage point cloud on the model.

2. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1 is characterized in that: The attenuation expression of the plane wave in the attenuation compensation in step S2 is: P x =P0e -αx Where P0 is the initial sound pressure of the wave source, P x is the sound pressure at a distance x from the wave source, e is the natural logarithm, α is the medium attenuation coefficient, unit is dB / mm, and x is the distance from the wave source; The compensation formula based on the ultrasonic attenuation formula is: I n =I0e -αx Where I0 is the initial signal strength; I n is the signal strength after compensation; x is the distance from the collected signal to the upper surface of the material.

3. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1 is characterized in that: In step S2, the centroid downsampling divides the point cloud data space into a series of small three-dimensional grids, which are called voxels. The centroid of all points inside each voxel is calculated, and all points in the voxel are replaced by the centroid. In this way, the original point cloud data is replaced by a group of evenly distributed and small number of points, thus achieving downsampling. The centroid calculation expression is: By comparing the two types of data, we can see that physical fitness downsampling effectively reduces the amount of data, and the characteristics of the data can also be preserved.

4. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1 is characterized in that: The half-wave height method, namely the 6db-drop method, processes the damage point cloud. After the network recognizes the damage point cloud, it finds the maximum signal strength therein, sets half of its signal strength as the threshold, removes points less than the threshold, and realizes further processing of the damage point cloud, thereby obtaining more accurate damage information.

5. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1 is characterized in that: In step S2, data enhancement processing is performed on 40 composite materials with different pre-embedded damages. In order to obtain more experimental data, a total of 160 groups of ultrasonic point cloud data are obtained by selecting different ultrasonic gains and front and back acquisition methods. Random flipping, random rotation, random scaling and noise addition are used to finally obtain 800 groups of data, of which the training set and test set are 600 and 200 groups respectively. The ultrasonic point cloud of the damaged area is annotated using "CloudCompare".

6. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1 is characterized in that: In step S4, the Swin Transformer feature extraction network is used to extract voxel features. Specifically, the ultrasonic point cloud is voxelized, and then input into Patch Partition processing to obtain multiple patches of the same size, and then input into Linear Enbedding and two Swin Transformer Blocks to obtain the first layer of feature maps. After that, multiple Patch Merging and Swin Transformer Blocks are performed to obtain multi-scale feature maps with resolutions of 1 / 2, 1 / 4 and 1 / 8 of the original input 2D voxel feature map.

7. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 6 is characterized in that: The Swin Transformer block consists of a shifted window-based MSA module followed by 2 layers of MLP with a GELU nonlinearity in between. A LayerNorm (LN) layer is applied before each MSA module and each MLP, and a residual connection is applied after each module. Layer 1 adopts a conventional window partitioning scheme, which is used to divide the voxel feature map to generate a local window. Focusing the self-attention in the local window can effectively improve the computational efficiency and avoid the interference caused by irrelevant distant pixels; The window partition in Layer l+1 is shifted to generate a new window, which is achieved by moving the window position on the X and Y axes with the same step length. The self-attention calculation in the new window spans the boundaries of the previous window in the previous layer, thus providing a connection between them.

8. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 7 is characterized in that: The Swin Transformer Block is constructed by replacing the standard multi-head self-attention MSA module in the Transformer block with a shift window-based module, and the other layers remain unchanged. The calculation process of the two Swin TransformerBlocks is expressed as: in, and z l They represent the SW-MSA module output features and the MLP module output features of the lth block respectively.

9. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1 is characterized in that: In step S5, the RICP processing based on the RICP algorithm of the Rodrigues rotation matrix prevents the registration from falling into the local optimum and effectively replaces the coarse registration step. The formula of the Rodrigues rotation matrix is: The ultrasonic point cloud is multiplied by a uniformly decreasing rotation matrix, which causes the point cloud to rotate around the Z axis. The value uniformly decreases from 90° to 0° with the number of iterations. Corresponding points are searched in the rotated ultrasonic point cloud and the node point cloud. The rotation matrix R and translation matrix t are obtained by calculation. The ultrasonic point cloud coordinates are updated according to R and t, and the mean square error E(R, t) between the ultrasonic point cloud and the node point cloud is calculated. Finally, the error value is compared with the set iteration threshold. If it is greater than or equal to the iteration threshold, it is iterated again. If it is less than or equal to the iteration threshold, the iteration ends and the point cloud registration is completed.

10. The composite material damage intelligent detection and modeling method based on ultrasonic point cloud and deep learning according to claim 1, characterized in that: In step S1, the experimental platform includes a robotic arm, a water tank, an ultrasonic probe and a probe fixture.

Images