Precise recognition and three-dimensional reconstruction characterization method for cracks of PHC pipe pile

Abstract

The invention provides an accurate recognition and three-dimensional reconstruction characterization method for cracks of a PHC tubular pile. The method comprises the following steps: acquiring a two-dimensional sequence image of an internal structure of a PHC tubular pile sample, and performing three-dimensional reconstruction by using Avizo software; preprocessing the original two-dimensional image; the method comprises the steps that firstly, a Trainable Weka Segment deep learning plug-in in Image-J software is activated on the basis of the expected crack shape and gray value, a model used for automatically recognizing and segmenting cracks is designed and trained, and after training is finished, a corresponding algorithm model file is generated; importing all the preprocessed two-dimensional sequence images of the pipe pile, loading the obtained algorithm model file, and then carrying out batch automatic identification and segmentation on cracks of the PHC pipe pile; and utilizing Avizo software to generate a three-dimensional reconstruction schematic diagram of the pipe pile crack. According to the method, the nondestructive testing advantage of XCT scanning and the powerful learning ability of a deep learning model are combined, automatic processing of large-batch image data is achieved, and cracks within the micrometer-to-centimeter level range are effectively recognized and reconstructed.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete structure detection, and in particular to a method for accurately identifying and characterizing cracks in a PHC pipe pile through three-dimensional reconstruction, and specifically to a method for accurately identifying and characterizing cracks in a PHC pipe pile through three-dimensional reconstruction based on XCT scanning and deep learning. Background Art

[0002] Prestressed high-strength concrete (PHC) piles are a commonly used deep foundation form and play an important role in engineering structures. However, due to material defects, construction damage, environmental factors and other reasons, PHC piles will inevitably have damage cracks during use, and the cracks are characterized by multi-scale distribution, and their width and size range affect their bearing capacity and service life. According to relevant statistics, structural failure accidents caused by cracks in piles account for more than 30% of all engineering accidents, posing a great safety hazard to the safety and reliability of underground hidden engineering pile foundation structures.

[0003] Therefore, developing an efficient and accurate PHC pile crack detection technology is of great significance to ensure engineering safety.

[0004] At present, the existing crack detection methods for PHC piles mainly include manual inspection, ultrasonic testing, eddy current testing, etc. These methods often have problems such as low accuracy and low efficiency, and mainly focus on the characterization of crack length and width. The existing ultrasonic crack depth detection method has the problems of low accuracy and difficulty in achieving measurement under complex conditions; in short, the existing crack detection and characterization methods cannot achieve accurate identification and three-dimensional characterization of cracks of different scales.

[0005] In recent years, with the rapid development of digital image correlation technology and X-ray computed tomography (XCT) technology, crack image recognition and three-dimensional reconstruction technology have received widespread attention in the field of concrete structure crack detection.

[0006] However, existing image recognition methods still have problems such as inaccurate crack recognition and incomplete three-dimensional characterization when dealing with PHC pile crack detection. In particular, the application of XCT technology in concrete structure crack detection has problems such as low image resolution and poor image quality in identifying pile cracks, and conventional crack segmentation methods cannot meet the requirements for accurate segmentation of multi-scale cracks. Summary of the invention

[0007] In order to solve the above problems, the purpose of the present invention is to provide a method for accurate identification and three-dimensional reconstruction of PHC pipe pile cracks. This technology combines the non-destructive detection advantages of XCT scanning and the powerful learning ability of deep learning models. Based on the Trainable Weka Segmentation deep learning pixel classification algorithm proposed by Arganda-Carreras et al., the characteristic morphology of cracks can be accurately identified through example learning, and the automatic processing of large quantities of image data can be realized. It can effectively identify and reconstruct cracks ranging from micrometers to centimeters, and then further count important parameters such as crack length, width, depth and orientation based on this.

[0008] The present invention is achieved through the following technical solutions:

[0009] A method for accurately identifying and three-dimensionally reconstructing cracks in a PHC pile comprises the following steps:

[0010] Step 1: Use X-ray computed tomography (XCT) technology to obtain two-dimensional sequence images of the internal structure of the PHC pile sample, and use Avizo software to perform three-dimensional reconstruction;

[0011] Step 2: Use Avizo software to preprocess the original two-dimensional image, including image position correction, noise reduction, and brightness adjustment operations;

[0012] Step 3: Based on the expected crack morphology and grayscale value, activate the Trainable WekaSegmentation deep learning plug-in in the Image-J software, design and train a model for automatic crack recognition and segmentation, and generate the corresponding algorithm model file after the training is completed;

[0013] Step 4: Import all the preprocessed 2D sequence images of pipe piles in step 2, load the algorithm model file obtained in step 3, and then perform batch automatic recognition and segmentation of PHC pipe pile cracks;

[0014] Step 5: Based on the identified and segmented two-dimensional crack sequence images, the Avizo software is used to generate a three-dimensional reconstruction diagram of the pipe pile crack; further statistics are collected on important parameters such as the length, width, and depth of the crack.

[0015] Furthermore, the specific steps of step 2 include the following:

[0016] Step 2-1 Image position correction: Use the Transform editor command in Avizo to correct the position of the original two-dimensional sequence images so that the images are arranged in a direction perpendicular to the circular section of the pile (Z direction);

[0017] Step 2-2 Image denoising: Use the Median Filter command in Avizo to denoise the two-dimensional sequence image after position correction; eliminate some image noise points generated during the shooting process;

[0018] Step 2-3 Image brightness adjustment: Use the Background Detection Correction command in Avizo to automatically adjust the brightness of the image to make the image brightness uniform.

[0019] Furthermore, the specific steps of step 3 include the following:

[0020] Step 3-1 Load the image and activate the plug-in: Load a two-dimensional slice image of the PHC pile in the Image-J software for subsequent model training, and activate the Trainable Weka Segmentation deep learning plug-in;

[0021] Step 3-2 Manually outline cracks and classify physical phases: Select the line drawing mode in the navigation bar to manually outline the cracks in the image and mark the cracks as the category Crack (red lines in the image); then switch to a circle and mark the larger steel bars, cement matrix and black background in the image as Others (green);

[0022] Step 3-3 Model training: After completing the manual segmentation and classification definition, activate the Train classifier function. The software will train the model based on the manually labeled samples so that the model can learn and identify the characteristics of cracks;

[0023] Step 3-4 Model generation: After multiple trainings, the model's crack recognition and segmentation accuracy gradually improves, and it can accurately segment unlabeled cracks in the image; at the same time, the software will generate a crack segmentation model file (file suffix .model) and a corresponding data set (file suffix .arff).

[0024] Furthermore, in this embodiment, step 4 is specifically implemented as follows:

[0025] Import all the preprocessed 2D sequence images of pipe piles in step 2, load the model file and data set trained in step 3, activate the Train classifier function again, and the model will automatically identify and segment cracks in batches; after the image processing is completed, export the segmented 2D sequence images of cracks.

[0026] Furthermore, the step 5 is specifically implemented as follows:

[0027] Step 5-1 3D reconstruction: Based on the segmented 2D crack sequence images, 3D reconstruction is performed using Avizo software, and pores and cracks are distinguished according to the shape factor to generate a 3D distribution map of the cracks;

[0028] Step 5-2 Crack parameter statistics: Based on the three-dimensional reconstructed crack distribution map, use the Label analysis command to count the length, width, and depth of the crack.

[0029] The beneficial effects of the present invention are:

[0030] 1) The present invention combines advanced XCT scanning technology, deep learning model and machine learning image segmentation method to provide an efficient and accurate PHC pile crack detection technology; based on the Trainable Weka Segmentation deep learning algorithm, the classifier in the Weka machine learning library is used to classify the image at the pixel level.

[0031] 2) When dealing with complex and multi-scale PHC pile cracks, the present invention can achieve accurate crack identification and three-dimensional characterization, further improve the accuracy and efficiency of crack detection, and provide more reliable data support for engineering safety assessment.

[0032] 3) Compared with traditional crack detection methods, the crack identification technology provided by the present invention has higher identification accuracy and shows significant advantages when dealing with small-scale cracks that are difficult to detect by traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow chart of an embodiment of the present invention, which is used to illustrate the specific implementation steps of the PHC pile crack recognition and three-dimensional characterization technology based on deep learning;

[0034] Figure 2 The original two-dimensional sequence image of the PHC pile in the embodiment of the present invention;

[0035] Figure 3 Schematic diagram of three-dimensional reconstruction of a PHC pile in an embodiment of the present invention;

[0036] Figure 4 This is a schematic diagram of model training based on deep learning in an embodiment of the present invention, which is used to illustrate the specific operations of model training;

[0037] Figure 5 A and Figure 5 B is a schematic diagram of crack segmentation before and after model training in an embodiment of the present invention;

[0038] Figure 6 A and Figure 6B is a schematic diagram of three-dimensional reconstruction of cracks in a PHC pile according to an embodiment of the present invention, showing the shape, position and distribution of the cracks;

[0039] Figure 7 It is a schematic diagram of the statistical results of the length, width and depth of the crack in the example of the present invention. DETAILED DESCRIPTION

[0040] To make the purpose, technical solution and advantages of the present invention more clearly understood, the technical solution provided by the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0041] The present invention provides a method for accurately identifying and characterizing cracks in a PHC pile using three-dimensional reconstruction, and the flow chart thereof is as follows: Figure 1 As shown, the following steps are included:

[0042] Step 1: Use X-ray computed tomography (XCT) technology to obtain two-dimensional sequence images of the internal structure of the PHC pile sample, and use Avizo software to perform three-dimensional reconstruction;

[0043] Step 2: Use Avizo software to preprocess the original two-dimensional image, including image position correction, noise reduction, and brightness adjustment operations;

[0044] Step 3: Based on the expected crack morphology and grayscale value, use the Trainable WekaSegmentation deep learning plug-in in the Image-J software to design and train a model for automatic crack recognition and segmentation. After the training is completed, the corresponding algorithm model file is generated;

[0045] Step 4: Import all the preprocessed 2D sequence images of pipe piles in step 2, and load the algorithm model file obtained in step 3, and then perform automatic batch recognition and segmentation of cracks;

[0046] Step 5: Based on the identified and segmented two-dimensional crack sequence images, the Avizo software is used to generate a three-dimensional reconstruction diagram of the pipe pile crack; the length, width, and depth of the crack are further counted.

[0047] In this embodiment, the PHC pile sample in step 1 is a hollow cylinder with an outer diameter of about 40 cm, an inner diameter of about 20 cm, and a height of about 10 cm. A 2432×1 pixel linear array detector is used for XCT scanning. The distance between the ray source and the sample is 951.9 mm, and the distance between the ray source and the detector is 1299.9 mm. The scanning parameters are set to a voltage of 520 kV, a current of 2000 μA, an interval of 16.67 ms, and an image resolution of 186 μm / voxel. The two-dimensional sequence images of the PHC pile obtained by XCT scanning are shown in FIG. Figure 2 shown.

[0048] Subsequently, the Avizo software was used to perform a three-dimensional reconstruction of the PHC pile, and the resulting schematic diagram is shown in the figure below. Figure 3 shown.

[0049] In this embodiment, step 2 is specifically implemented as follows:

[0050] Step 2-1 Image position correction: Use the Transform editor command in Avizo to correct the position of the original two-dimensional sequence images so that the images are arranged in a direction perpendicular to the circular section of the pile (Z direction);

[0051] Step 2-2 Image denoising: Use the Median Filter command in Avizo to denoise the two-dimensional sequence image after position correction; eliminate some image noise points generated during the shooting process to improve image quality;

[0052] Step 2-3 Image brightness adjustment: Use the Background Detection Correction command in Avizo to automatically adjust the brightness of the image to achieve uniform image brightness, laying a solid foundation for subsequent crack identification.

[0053] In this embodiment, step 3 is specifically implemented as follows:

[0054] Step 3-1 Load the image and activate the plug-in: Load a two-dimensional slice image of the PHC pile in the Image-J software for subsequent model training and activate the Trainable Weka Segmentation deep learning plug-in.

[0055] Step 3-2 Manual crack outlining and phase classification:

[0056] like Figure 4As shown, select the line drawing mode in the navigation bar to manually outline the cracks in the image and mark the cracks as the category Crack (red line in the figure); then switch to a circle and mark the larger steel bars, cement matrix and black background in the image as Others (green). This step is the key to training the model. Accurate labeling will directly affect the model's recognition and crack segmentation effects. In theory, the more labels there are, the better the effect of pile crack recognition and segmentation.

[0057] Step 3-3 Model training: After completing the manual segmentation and classification definition, activate the Train classifier function. The software will train the model based on the manually labeled samples so that the model can learn and identify the characteristics of cracks.

[0058] Figure 5 A and Figure 5 B is a schematic diagram of crack segmentation after model training, in which cracks (green lines in the figure) and other phases of the pipe pile (red areas in the figure) are distinguished.

[0059] Step 3-4 Model generation: After multiple trainings, the model's crack recognition and segmentation accuracy gradually improves, and it can accurately segment unlabeled cracks in the image. At the same time, the software will generate a crack segmentation model file (file suffix .model) and a corresponding data set (file suffix .arff).

[0060] In this embodiment, step 4 is specifically implemented as follows:

[0061] Import all the preprocessed 2D sequence images of pipe piles in step 2, load the model file and data set trained in step 3, activate the Train classifier function again, and the model will automatically identify and segment cracks in batches; after the image processing is completed, export the segmented 2D sequence images of cracks.

[0062] In this embodiment, step 5 is specifically implemented as follows:

[0063] Step 5-1 3D reconstruction: Based on the segmented 2D crack sequence images, 3D reconstruction is performed using Avizo software, and pores and cracks are distinguished according to the shape factor to generate a 3D distribution map of the cracks, such as Figure 6 A and Figure 6 B. The shape factor includes length, flatness and sphericity. Flatness is defined as the ratio of crack depth to crack width, length is defined as the ratio of crack width to crack length, and sphericity S is defined as the ratio of equivalent spherical surface area to measured surface area, and the expression is:

[0064]

[0065] In the formula, VP represents the volume of the pores, S P Represents the specific surface area of ​​pores.

[0066] Step 5-2 Crack parameter statistics: By analyzing the three-dimensional reconstructed distribution map of the crack, the Label analysis command is used to quantify the geometric parameters of the crack. The specific parameters are defined as follows: the crack length is based on the maximum value of the Feret diameter; the crack width is defined as the maximum distance between two parallel lines that touch but do not intersect the crack in a plane perpendicular to the length; the crack depth is the maximum length of the line segment between the two end points of the crack in a plane perpendicular to both the length and width.

[0067] Figure 7 The statistical results of the length, width and depth of the pipe pile cracks are shown. The statistical data show that the maximum values ​​of the crack length, width and depth are 123.81mm, 97.92mm and 26.42mm, respectively, while the minimum values ​​are 0.84mm, 0.19mm and 0.19mm, respectively. According to the different ranges of crack size (<1mm, 1-10mm, 10-100mm, 100mm-1000mm), the distribution ratio of crack length is 0.54%, 96.47%, 2.95% and 0.04%, the distribution ratio of width is 33.57%, 65.48%, 0.94% and 0%, and the distribution ratio of depth is 68.97%, 30.91%, 0.12% and 0%. This statistical result provides a quantitative basis for crack feature analysis.

[0068] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for accurate identification and three-dimensional reconstruction of PHC pile cracks, characterized in that: The following steps are involved: Step 1: Use X-ray computed tomography (XCT) technology to obtain two-dimensional sequence images of the internal structure of the PHC pile sample, and use Avizo software to perform three-dimensional reconstruction; Step 2: Use Avizo software to preprocess the original two-dimensional image, including image position correction, noise reduction, and brightness adjustment operations; Step 3: Based on the expected crack morphology and grayscale value, activate the Trainable WekaSegmentation deep learning plug-in in the Image-J software, design and train a model for automatic crack recognition and segmentation, and generate the corresponding algorithm model file after the training is completed; Step 4: Import all the preprocessed 2D sequence images of pipe piles in step 2, load the algorithm model file obtained in step 3, and then perform batch automatic recognition and segmentation of PHC pipe pile cracks; Step 5: Based on the identified and segmented two-dimensional crack sequence images, the Avizo software is used to generate a three-dimensional reconstruction diagram of the pipe pile crack; further statistics are collected on important parameters such as the length, width, and depth of the crack.

2. The method for accurate identification and three-dimensional reconstruction of PHC pile cracks according to claim 1, characterized in that: The step 2 is specifically implemented as follows: Step 2-1 Image position correction: Use the Transform editor command in Avizo to correct the position of the original two-dimensional sequence images so that the images are arranged in a direction perpendicular to the circular section of the pile (Z direction); Step 2-2 Image denoising: Use the Median Filter command in Avizo to denoise the two-dimensional sequence image after position correction; eliminate some image noise points generated during the shooting process; Step 2-3 Image brightness adjustment: Use the Background Detection Correction command in Avizo to automatically adjust the brightness of the image to make the image brightness uniform.

3. The method for accurate identification and three-dimensional reconstruction of cracks in PHC piles according to claim 1, characterized in that: The step 3 is specifically implemented as follows: Step 3-1 Load the image and activate the plug-in: Load a two-dimensional slice image of the PHC pile in the Image-J software for subsequent model training, and activate the Trainable Weka Segmentation deep learning plug-in; Step 3-2 Manually outline cracks and classify physical phases: Select the line drawing mode in the navigation bar to manually outline the cracks in the image and mark the cracks as the category Crack (red lines in the image); then switch to a circle and mark the larger steel bars, cement matrix and black background in the image as Others (green); Step 3-3 Model training: After completing the manual segmentation and classification definition, activate the Train classifier function. The software will train the model based on the manually labeled samples so that the model can learn and identify the characteristics of cracks; Step 3-4 Model generation: After multiple trainings, the model's crack recognition and segmentation accuracy gradually improves, and it can accurately segment unlabeled cracks in the image; at the same time, the software will generate a crack segmentation model file (file suffix .model) and a corresponding data set (file suffix .arff).

4. The method for accurate identification and three-dimensional reconstruction of cracks in PHC piles according to claim 1, characterized in that: The step 4 is specifically implemented as follows: Import all the preprocessed 2D sequence images of pipe piles in step 2, load the model file and data set trained in step 3, activate the Train classifier function again, and the model will automatically identify and segment cracks in batches; after the image processing is completed, export the segmented 2D sequence images of cracks.

5. The method for accurate identification and three-dimensional reconstruction of cracks in PHC piles according to claim 1, characterized in that: The step 5 is specifically implemented as follows: Step 5-1 3D reconstruction: Based on the segmented 2D crack sequence images, 3D reconstruction is performed using Avizo software, and pores and cracks are distinguished according to the shape factor to generate a 3D distribution map of the cracks; Step 5-2 Crack parameter statistics: Based on the three-dimensional reconstructed crack distribution map, use the Label analysis command to count the length, width, and depth of the crack.

Images