A steel bar straight thread qualified detection method based on laser scanning and F-pointnet network

Abstract

This invention discloses a method for inspecting the quality of straight threads in reinforcing bars based on laser scanning and an F-PointNet network, belonging to the field of threaded steel bar inspection technology. The method includes the following steps: identifying the reinforcing bar to be inspected and generating a 2D image of it; using a high-precision line laser scanner combined with an F-PointNet network model to perform non-contact data acquisition of the reinforcing bar; processing the 2D image; adjusting the key parameters of the F-PointNet network using a loss function to obtain an adjusted model; calculating key geometric parameters using the adjusted model; comparing these parameters with standard parameters to achieve rapid, objective, and accurate determination of thread quality. This invention has advantages such as high inspection efficiency, strong objectivity, and high accuracy, providing strong technical support for the precise quality control of key components in building engineering.

Description

Technical Field

[0001] This invention relates to the field of rebar testing technology, and in particular to a method for detecting the conformity of straight threads in rebar based on laser scanning and F-PointNet network. Background Technology

[0002] The use of straight threaded rebar connection technology in construction engineering can simplify procedures, shorten construction time, and improve project quality. However, poor quality control of the straight thread ends can lead to a series of adverse consequences. Therefore, quality control of straight threaded rebar sleeves is particularly important. During construction, the quality of the straight thread ends of the rebar sleeves must be strictly controlled to ensure project quality. Among these technologies, the rolled straight thread technology for rebar is a commonly used and practical technique in current hydropower station construction projects, holding a crucial position in both construction and quality assurance. Straight thread technology can reduce construction costs and labor intensity, and also has significant advantages such as low learning curve and ease of operation.

[0003] However, due to the relatively short development time of straight thread technology, there are many shortcomings, such as improper thread end processing and connection, insufficient thread end length, and inadequate technical support for thread ends. Traditional thread measurement methods include comprehensive inspection and single-parameter testing. The comprehensive inspection method verifies thread conformity by manually engaging a gauge. This method is simple to use but suffers from low accuracy and easy wear on the thread surface. The single-parameter method measures a single thread parameter using measuring instruments such as external thread pitch measuring instruments and external thread height measuring instruments. Pitch diameter measurement often employs the three-wire method and the double-ball method in conjunction with a length measuring instrument. The optimal three-wire or probe ball diameter is selected based on parameters such as theoretical pitch, flank angle, and helix angle to obtain the single pitch diameter value of the thread. These two methods offer high measurement accuracy and are often used for thread gauge calibration, but they suffer from low measurement efficiency and the inability to perform multi-parameter thread measurements. In the current context, various testing technologies and measuring devices have emerged to meet different application needs. Testing methods based on different measurement principles have, to some extent, greatly improved measurement accuracy and simultaneously shortened the measurement process time.

[0004] Laser inspection technology, due to its high accuracy and versatility, has been widely used in various fields. Research on multi-parameter measurement of threads based on contour scanning is increasing, mainly divided into contact and non-contact methods. Non-contact measurement methods employ image measurement, fiber optic measurement, and laser triangulation to acquire 2D or 3D contour data of threads. Image measurement is fast, but limited by the camera's field of view, it is difficult to measure parameters of large-diameter external threads. Point laser measurement offers high accuracy, but the problems of long acquisition time and low measurement efficiency for two-dimensional or three-dimensional topographic data remain unresolved.

[0005] Therefore, providing a method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network to overcome the difficulties in the existing technology is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network, which has the advantages of high detection efficiency, strong objectivity and high accuracy, and provides strong technical support for the precise control of the quality of key components in building engineering.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for inspecting the straight thread quality of reinforcing bars based on laser scanning and F-PointNet network includes the following steps: Identify the reinforcing bars to be inspected and generate a 2D image of the reinforcing bars to be inspected; A high-precision line laser scanner was combined with an F-PointNet network model to perform non-contact data acquisition of the steel bars to be inspected. The data was then processed in conjunction with 2D images, and the key parameters of the F-PointNet network were adjusted using a loss function to obtain the adjusted model. By using the adjusted model to calculate key geometric parameters and comparing them with parameter standards, the quality of the yarn ends can be quickly, objectively, and accurately determined.

[0008] Optional, non-contact data acquisition includes: Acquire 2D image data of the actual rebar wire ends; use object detection algorithms to obtain the bounding box of the actual rebar wire ends, extend along the direction of the bounding box with the camera as the origin to form a cone, and obtain the initial point cloud.

[0009] Optionally, it also includes rotating the camera coordinate axes based on rotation invariance, so that the Z-axis of the camera coordinate axes coincides with the axis of the cone.

[0010] Optionally, obtaining the initial point cloud includes: A mature object detector is used to generate 2D bounding boxes and output one-hot classification vectors; By combining the 2D bounding box with depth information, the planes closest to and farthest from the bounding box in the depth information are found to define the 3D frustum region. All 3D points in the frustum region are collected to form the frustum point cloud.

[0011] Optionally, the F-PointNet network includes point cloud instance segmentation within the cone and point cloud object bounding box regression; Point cloud instance segmentation within a cone includes: extracting the object to be detected within the cone, and aligning the camera origin with the segmented object origin based on translation invariance; The regression of the bounding box of a point cloud object includes: adjusting the centroid of the object using T-Net, and obtaining key parameters by regressing the bounding box through the centroid.

[0012] Optionally, the loss function expression for the PointNet network model is: , in, , and These are the object detection loss, T-Net loss, and bounding box residual loss, respectively.

[0013] As can be seen from the above technical solution, compared with the prior art, the present invention provides a method for detecting the quality of straight threads in steel bars based on laser scanning and F-PointNet network, which has the following beneficial effects: The present invention combines high-precision three-dimensional sensing technology with powerful deep learning point cloud analysis capabilities, successfully solves many pain points of traditional detection methods, effectively overcomes the limitations of manual detection, and has the advantages of high detection efficiency, strong objectivity and high accuracy, providing strong technical support for the precise control of the quality of key components in building engineering. Attached Figure Description

[0014] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a flowchart of the method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network disclosed in this invention. Figure 2 This is a schematic diagram of the F-PointNet network disclosed in this invention; Figure 3 This is a schematic diagram of the regression principle of the internal boundary box of the cone disclosed in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Reference Figure 1 and Figure 2 As shown, this invention discloses a method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet networks, comprising the following steps: Identify the reinforcing bars to be inspected and generate a 2D image of the reinforcing bars to be inspected; A high-precision line laser scanner was combined with an F-PointNet network model to perform non-contact data acquisition of the steel bars to be inspected. The data was then processed in conjunction with 2D images, and the key parameters of the F-PointNet network were adjusted using a loss function to obtain the adjusted model. By using the adjusted model to calculate key geometric parameters and comparing them with parameter standards, the quality of the yarn ends can be quickly, objectively, and accurately determined.

[0018] Specifically, F-PointNet includes a 2D detection box frustum proposal module, a 3D entity segmentation module, and a 3D box regression module. Before performing point cloud processing, it uses image information to obtain some prior search ranges, which can improve both efficiency and accuracy.

[0019] Furthermore, contactless data acquisition includes: Acquire 2D image data of the actual rebar wire ends; use object detection algorithms to obtain the bounding box of the actual rebar wire ends, extend along the direction of the bounding box with the camera as the origin to form a cone, and obtain the initial point cloud.

[0020] Furthermore, refer to Figure 3 As shown, since the detected target is not necessarily at the exact center of the image, forming a cone also involves rotating the camera coordinate axes based on rotation invariance, so that the Z-axis of the camera coordinate axes coincides with the axis of the cone.

[0021] Furthermore, obtaining the initial point cloud includes: A mature object detector is used to generate 2D bounding boxes and output one-hot classification vectors; By combining the 2D bounding box with depth information, the planes closest to and farthest from the bounding box in the depth information are found to define the 3D frustum region. All 3D points in the frustum region are collected to form the frustum point cloud.

[0022] Furthermore, the F-PointNet network includes point cloud instance segmentation within the cone and point cloud object bounding box regression; Point cloud instance segmentation within a cone includes: extracting the object to be detected within the cone, and aligning the camera origin with the segmented object origin based on translation invariance; The regression of the bounding box of a point cloud object includes: adjusting the centroid of the object using T-Net, and obtaining key parameters by regressing the bounding box through the centroid.

[0023] Furthermore, the bounding box parameters include: center point: Length, width, and height: and angle .

[0024] Furthermore, the loss function expression for the PointNet network model is: , in, , and These are the object detection loss, T-Net loss, and bounding box residual loss, respectively.

[0025] In one specific embodiment, the internal thread of the mouth of a projectile of specification M48×2-6gT was tested 10 times at different times using this application, and the obtained data was analyzed to verify the stability of the system. The measurement data results are shown in Table 1.

[0026] Table 1. Inspection results of the internal threads at the muzzle of the M48×2-6gT projectile.

[0027] This shows that the test results obtained from measuring the same type of straight threaded steel bar sleeve on different days have almost no deviation, proving that the straight threaded steel bar sleeve testing system has good stability and that the measurement results can accurately reflect the true value of the qualification standard of the tested straight threaded steel bar sleeve.

[0028] Repeatable experiments were conducted using this application, with the parameters of the steel bar straight thread sleeve verified by the Metrology Institute as the true values, as shown in Table 2.

[0029] Table 2. Repeatability test results of the detection system

[0030] It can be seen that the laser scanning detection system for straight threads of steel bars has good repeatability when the turntable speed is 1 rad / s, and the standard deviations of the basic major diameter, basic pitch diameter, basic minor diameter, pitch, and thread length are all within a reasonable range, proving that the detection system can accurately reflect the various parameters of straight threads of steel bars.

[0031] Table 3 shows the results of comparisons with recent methods QueryInst, PatchDCTFastIns, and TurboInst. Compared to the best-performing TurboInst, our model improves detection accuracy (precision, recall, and F1 score) by 0.20, 0.13, and 0.17 percentage points, respectively; and segmentation accuracy (moU and mAP) by 1.24 and 1.21 percentage points, respectively. This further confirms that the network reconstructed using the MaskloU branch in DeepLabv3+ is highly effective for wildfire segmentation.

[0032] Table 3 Model performance test results

[0033] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for inspecting the straight thread quality of reinforcing bars based on laser scanning and F-PointNet network, characterized in that, Includes the following steps: Identify the reinforcing bars to be inspected and generate a 2D image of the reinforcing bars to be inspected; A high-precision line laser scanner was combined with an F-PointNet network model to perform non-contact data acquisition of the steel bars to be inspected. The data was then processed in conjunction with 2D images, and the key parameters of the F-PointNet network were adjusted using a loss function to obtain the adjusted model. By using the adjusted model to calculate key geometric parameters and comparing them with parameter standards, the quality of the yarn ends can be quickly, objectively, and accurately determined.

2. The method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network according to claim 1, characterized in that, Non-contact data acquisition includes: Acquire 2D image data of the actual rebar wire ends; use object detection algorithms to obtain the bounding box of the actual rebar wire ends, extend along the direction of the bounding box with the camera as the origin to form a cone, and obtain the initial point cloud.

3. The method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network according to claim 2, characterized in that, It also includes rotating the camera coordinate axes based on rotation invariance, so that the Z-axis of the camera coordinate axes coincides with the axis of the cone.

4. The method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network according to claim 2, characterized in that, Obtaining the initial point cloud includes: A mature object detector is used to generate 2D bounding boxes and output one-hot classification vectors; By combining the 2D bounding box with depth information, the planes closest to and farthest from the bounding box in the depth information are found to define the 3D frustum region. All 3D points in the frustum region are collected to form the frustum point cloud.

5. The method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network according to claim 2, characterized in that, The F-PointNet network includes point cloud instance segmentation within a cone and point cloud object bounding box regression; Point cloud instance segmentation within a cone includes: extracting the object to be detected within the cone, and aligning the camera origin with the segmented object origin based on translation invariance; The regression of the bounding box of a point cloud object includes: adjusting the centroid of the object using T-Net, and obtaining key parameters by regressing the bounding box through the centroid.

6. The method for detecting the quality of straight threads in reinforcing bars based on laser scanning and F-PointNet network according to claim 1, characterized in that, The loss function expression for the F-PointNet network model is: ， in, , and These are the object detection loss, T-Net loss, and bounding box residual loss, respectively.

Images