1. Introduction
In recent years, considerable attention has been paid to the health monitoring of advanced composite structures. Damage identification through various non-destructive testing techniques predict or diagnose the health of composite structures [
1,
2,
3]. Reactive Powder Concrete (RPC), an emerging materials, are widely concerned in railway engineering, roads, buildings, airports and nuclear industrial facilities—ultra high strength, high toughness and high durability [
4,
5,
6,
7,
8,
9]. However, there are few researches on damage identification and detection of composite structures using RPC as raw materials. It is critical to popularize intelligent damage monitoring technology of RPC structure damage.
Certain damage, which have considerable impact on structure reliability and security, are inevitably retained because of the influence of fatigue load [
10,
11,
12]. The process of causing damage caused by cyclic loading includes the combined results of the cumulative process of crack initiation, expansion and final fracture. When the RPC structure is subjected to a continuous fatigue load, initial microcracks appear on the surface of the structure. The stress generated by the matrix crack increases and continues to aggregate, expanding and eventually forming macroscopic cracks. Scholars evaluate the health status of the structure by monitoring the development of damage [
13,
14,
15]. Finding damage in time and predicting injuries is the key to avoiding temporary accidents. Therefore, it is very necessary to fully grasp the development of crack damage as well as find an intelligent method to quickly and conveniently monitor the health of the structure.
At present, the research on RPC mainly focuses on the influence of material parameters on its performance and the establishment of constitutive models. However, there is a lack of investigations on the damage characterization of materials and damage evolution mechanism [
5,
16,
17]. The stress-strain characteristics of RPC under fatigue loading are nonlinear. Therefore, it is important to use appropriate non-destructive testing methods to probe the quality of RPC structures. Accompanied by the destruction process, the structure is deformed and accompanied by local crack damage, which can be recorded by the 3D laser scanning technology (3D LST). Some scholars have found that 3D laser scanning can be used as an effective technique for fatigue damage accumulation research in RPC (see, for example, [
18,
19,
20] as well as [
21]). The 3D LST can be used to continuously monitor the development of damage or structure deformation and defects in RPC structure dynamically. In addition, we can use image processing methods to correlate the damage generation process with the feature parameters of the 3D image. The damage mode of RPC is extremely complicated, and the displacement and deformation characteristics exhibited during the fatigue loading process are obvious, which indicates that the 3D LST is an extremely effective technique for detecting the damage process. Moreover, many recent studies show that in the detection method of damage, the 3D LST, which can quickly and accurately implement structural modeling, non-destructive, non-contact, and does not require a light source and cost of wiring, has significant advantages [
22,
23]. At the same time, in the intelligent method of pattern recognition, artificial intelligence recognition technology, such as clustering technology and data mining, which can accurately and quickly identify the type of the damage, has received widespread attention. These results open the door to research the life prediction and mechanical behavior of RPC structure. More importantly, the 3D laser scanning technology affords unique advantages in the fatigue damage detection of RPC structure because it can evaluate the information contained in the damage to assess damage on-line.
However, the study of bending fatigue damage performance is extremely complicated. At present, the research object is limited to small-scale specimen experiments, the experimental data obtained is relatively simple and the versatility is insufficient. Especially for building structures, as a new type of material, RPC has just been invested in by engineering construction. Compared with regular specimen structures, extracting the characteristics of point cloud data acquired by 3D laser scanning technology is more complex and more cost. There are few reports on the damage characteristics of such complex structures under fatigue loads until now.
The purpose of this study was to investigate the evolution and development of RPC rupture process under 3D laser scanning technology under fatigue loading. The main part of damage extraction are performed by the Gray Level Co-occurrence Matrix (GLCM) method. In addition, the paper also discusses the variation of fatigue damage characteristics at different stages. The different techniques implemented in this work, that is, the extraction of 3D laser scanning events at various stages of the test, can classify and analyze crack development. It also provides an effective way to monitor damage.
3. Result and Analysis
The whole process of damage mechanism of RPC under fatigue loading are investigated using the 3D laser scanning technology as show in
Figure 4.
The overall process consists of six parts in
Figure 4. The establishment of a 3D laser scanning system. The execution of RPC fatigue loading test. Further, the GLCM is generated. Including extracted feature parameters, parameter screening and verification processes. Stack each part as the main line of research. And mark the core content of each part.
3.1. Bending Strength
Four sets of test pieces were selected for the static load bending test, and the average value of the 28-day bending strength was measured, as shown in
Table 5.
There was no significant change in the surface of the test piece before the test loading. The initial crack occurs in the test piece when the loading force reaches 70% to 80% of the ultimate bending strength. The flexural strength value did not decrease as the specimen appeared across the crack. It was not until the steel fiber in the cracked cross-section was pulled out that the test specimen was completely damaged.
3.2. Data Collection
The structure is located in the scanning sector detection area by adjusting the horizontal position of the holder body. In order to ensure complete structural modeling information, the effective detection range in the vertical direction is greater than the height of the detected object. The data acquisition process as shown in Figure 8.
Figure 5 shows the process of obtaining a structural point cloud image. Verification of scanning system accuracy based on bending fatigue load test. A total of 100 scans were performed in each group.
Figure 6 is a structural point cloud diagram.
3.3. Damage Parameter Acquisition
When the damage degree of the RPC structure is slight, the displacement and deflection of the damage source area are small. However, the location and elevation information of the damaged area is constantly changing as the degree of damage continues to develop. This change can be obtained in two ways. On the one hand, the continuous development of cracks in 2D plane can be achieved. On the other hand, the displacement of the crack and the change in height information in the 3D space.
The 2D image can be captured by the CCD camera built in the 2D laser ranging sensor. According to the principle of the independently developed 3D laser scanning system, the point cloud data for obtaining the scanned test piece is a digital matrix composed of data points having position information. This value can be calculated by the Formula (2). Therefore, the damage region of the digital matrix composed of the 2D image and the point cloud elevation information is selected as the Region of Interest (ROI), respectively. The GLCM of the ROI is constructed separately, the feature parameters are extracted, and the texture features are calculated therefrom. The angles are determined as 0°, 45°, 90°, and 135°, the image grayscale is selected as 265, and the distance between the pixel pairs is 2 pixels. Therefore, when the construction factor is d = 2, g = 28 and θ selects 0°, 45°, 90° and 135°, and six characteristic parameters of the damaged area of the test piece are extracted. In order to satisfy the image rotation invariance, the average of the 4 angles is taken as the final value function parameter.
3.4. Parameter Extraction in 2D Plane
The images of the fixed damage area are acquired with the same number of frames, as shown in
Figure 7.
The ROI of damage images were captured in
Figure 7. The number of image acquisition interval frames is 10 images. The direction of the arrow indicates the direction of damage development. As fatigue loads continue to accumulate, damage continues to expand and develop. A nonlinear texture change is formed on a 2D plane. The characteristic parameters of each image were acquired and data recorded.
3.5. Parameter Extraction in 3D Space
A transformation algorithm for transforming 3D scene data into 2D elevation image is proposed based on the principle of orthogonal projection.
Figure 8 is a schematic diagram of a conversion algorithm for 3D point cloud data projection elevation data information.
In
Figure 8, the elevation information of the3D image is represented by the Z-axis in the coordinate system. Therefore, the projection point on the projection plane is the elevation data of the image, which can be represented by a matrix Z(x, y) consisting of discrete quantities:
where x is the range of values (0, m) and the range of y is (0, n).
All points in the Z(x, y) matrix are represented by one byte of 8 bits, converted to any value in the 0–255 interval, based on the one-dimensional sampling theorem and the principle of linear quantization. Therefore, the matrix Z(x, y) is a grayscale image of 256 gradation values. The converted image is named an elevation projection. An elevation digital matrix corresponding to images of a 2D plane is obtained. Then, convert them into projected images in sequentially. Feature parameters of the projected image are extracted separately. Finally, data statistics are performed.
3.6. Damage Diagnosis
For the purpose of obtaining the damage indicators of fatigue damage model, the parameters is screening use the DFS (Digital Feature Screening) method. The DFS method establishes the selection criteria of the damage characteristic index: obtaining the crack image in a fixed area under the step of increasing the loading level, and sequentially performing image numbering. Finally, damage characteristic index of the crack image is obtained. Further, a digital feature screening analysis chart is established, as shown in
Figure 9.
Figure 9a shows the variation of the characteristic parameters as the damage grows in 3D space. After analysis, it is known that as the fatigue damage increases, the ASM characteristic parameters will continue to increase according to certain rules. The remaining characteristic parameters change irregularly with the development of the fracture, so the remaining characteristic parameters are excluded. In order to monitor the development of fractures in 3D space, ASM characteristic parameters with good positive correlation with fracture development trends were selected.
Figure 9b shows a tendency of characteristic parameters to change as the damage advances in the 2D plane. Analysis of the data changes in the graph shows that as the fatigue damage increases, the CLS characteristic parameters continue to decrease. As the crack develops, the other characteristic parameters show different trends. Therefore, it is excluded. In order to monitor the development of two-dimensional spatial crack damage, the CLS characteristic parameters which are negatively correlated with the crack development trend are selected.
Figure 9a,b respectively count the changes of the crack development of the six characteristic parameters, and obtain the monitoring basis of the 2D plane and 3D space cracks. The characteristic parameters ASM and CLS can quantify the development of cracks in different spaces and fully describe the development of cracks. Therefore, CLS and ASM characteristic parameters can be used as damage indicators for fatigue damage in RPC bending loading test to detect the development of damage.