Reinforcement cage AI intelligent detection system

Abstract

A reinforcement cage AI intelligent detection system disclosed by the present invention comprises a console and a detection support cooperating with a reinforcement cage, detection cameras are arranged in the six visual directions of the detection support, and an AI detection terminal, a PLC controller and an alarm group cooperating with the detection cameras are arranged in the console. And the AI detection terminal identifies reinforcement cage information collected by the detection camera, and transmits the reinforcement cage information to the PLC controller to control the alarm group to give an alarm. The detection cameras are arranged in the six visual directions of the reinforcement cage, image information collection can be carried out on the reinforcement cage, finally the collected information is transmitted to the AI detection terminal to carry out intelligent identification and measurement based on a visual algorithm, various index parameters of the reinforcement cage can be efficiently and accurately detected, and the detection accuracy is improved. And the whole detection process is within a defined safety range, that is, the detection area is automatically carried out, so that the potential safety hazard of quality inspection personnel during detection can be greatly eliminated.

Description

Technical Field

[0001] This invention relates to the field of rebar cage inspection technology, and in particular to an AI-powered intelligent inspection system for rebar cages. Background Technology

[0002] During tunnel boring machine (TBM) construction, the tunnel segments are the main assembled components. Their assembly is interconnected, and to ensure the strength and load-bearing capacity of the segments, reinforcing cages are installed within them. The reinforcing cage workshop at the segment prefabrication base provides assurance for the performance and quality of the segments. To guarantee the quality of the reinforcing cages, quality inspection is required after fabrication. This mainly involves checking parameters such as the quantity, spacing, and diameter of the main reinforcing bars, stirrups, and uprights. However, current quality inspection of tunnel segment reinforcing cages relies primarily on manual labor. Manual inspection is prone to omissions or significant errors, resulting in low efficiency, high personnel costs, and potential safety hazards for inspectors. Summary of the Invention

[0003] In order to overcome the shortcomings of the existing technology, the technical problem to be solved by the present invention is to propose an AI intelligent inspection system for steel cages, which improves quality inspection efficiency, significantly reduces personnel input, and reduces safety hazards during segment inspection based on machine vision and artificial intelligence technology.

[0004] To achieve this objective, the present invention adopts the following technical solution: This invention provides an AI-powered intelligent inspection system for rebar cages, comprising a control console and an inspection bracket that works with the rebar cage. The inspection bracket is equipped with inspection cameras arranged in six viewing directions, all of which are binocular cameras. The control console contains an AI inspection terminal, a PLC controller, and an alarm group that work with the inspection cameras. During inspection, the AI ​​inspection terminal identifies the rebar cage information collected by the inspection cameras and transmits it to the PLC controller to control the alarm group to issue a warning.

[0005] A preferred embodiment of the present invention further includes a fence, which encloses a rectangular detection area centered on the detection bracket. The control console and the detection cameras are both located within this detection area to ensure operational safety. Furthermore, a gantry that cooperates with the detection bracket is also provided within the detection area. Three detection cameras are positioned at the top and three at the bottom of the detection bracket. The three cameras at the top are equidistantly arranged along the crossbeam of the gantry, and one camera is positioned at each of the other viewing angles. The detection cameras at the bottom are equipped with a light source to ensure clear image capture.

[0006] A preferred embodiment of the present invention is that the control console is further equipped with a display, which is connected to the AI ​​detection terminal and the PLC controller to display the recognition results and link the PLC controller to control the alarm group; the alarm group includes quantity alarms, spacing alarms, and diameter alarms, so as to distinguish the types of unqualified results through different alarms.

[0007] A preferred embodiment of the present invention is that the detection bracket is a hollow bracket, the top of the detection bracket is an arc shape that cooperates with the reinforcing cage, and the lowest point of the detection bracket is at least 90cm from the ground. This ensures that the detection bracket can properly support the reinforcing cage.

[0008] A preferred embodiment of the present invention is that the AI ​​detection terminal includes an identification module, a qualification judgment module, and a detection result display and storage module. The identification module detects the quantity of main reinforcing bars, stirrups, and vertical reinforcing bars, and detects the relative position of individual targets on the image. The qualification judgment module compares the data processed by the identification module with the defined data of the same type to determine the qualification of individual targets in terms of quantity, diameter, spacing, etc., and finally outputs the judgment result. The detection result display and storage module plots the detected data and images on the displayed image and stores the detection results data and images in a database.

[0009] The beneficial effects of this invention are as follows: (1) This case adopts an algorithm combining binocular vision and deep learning. The deep learning algorithm uses the semantic segmentation algorithm and object detection algorithm of YOLOv8 to detect the pixel coordinates and number of individual targets in the collected image, so as to calculate the pixel spacing of the individual targets. Then, the pixel spacing and pixel diameter are converted into real data in cm by the fitting curve in the above recognition module. The various surfaces of the steel cage are not planar structures and have a certain curvature. Therefore, the spacing of individual targets is further calculated by combining binocular vision. Binocular vision detection mainly uses a binocular camera. After the camera is installed and fixed, the Zhang calibration method is used to perform binocular stereo calibration and binocular stereo correction of camera distortion. The obtained parameter matrix Q is used to calculate the depth of the individual target relative to the camera. (2) By setting up detection cameras in the six directions of the steel cage, the steel cage can be collected in all directions without blind spots. Finally, the collected information is transmitted to the AI ​​detection terminal for intelligent recognition and measurement based on visual algorithms. It can efficiently and accurately detect various index parameters of the steel cage, such as the number of main bars / stirrups / upright bars, the spacing of main bars, the diameter of main bars, etc. The entire detection process is carried out automatically within the designated safety range, that is, within the detection area, which can also greatly eliminate the safety hazards of quality inspectors during the detection. (3) This case provides a cost-effective solution; the algorithm is highly scalable and can be upgraded to detect the quantity, size and other properties of more types of steel bars; the algorithm accuracy reaches 1.0mm, which can meet the needs of subsequent occasions with increasingly higher quality requirements; the detection time is reduced to less than 10s; the hardware and construction structure are simple, the operation is convenient, and it is easy to replicate to other production lines; the vision is directly captured, no motion mechanism is required, and the noise is <45db; the mechanism and optical methods are used to filter out the influence of vibration on the detection scheme. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the overall structure of the reinforcing cage provided in a specific embodiment of the present invention; Figure 2 This is a schematic diagram of the overall structure of an AI intelligent detection system for steel cages provided in a specific embodiment of the present invention; Figure 3 This is a schematic diagram of the principle of an AI intelligent detection system for steel cages provided in a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the inference test real-shot images and generated inference results in the verification of the main reinforcement position detection algorithm provided in a specific embodiment of the present invention; Figure 5 This is a schematic diagram of the real-life images of the inference test and the generated inference results during the verification of the lumbar muscle position detection algorithm provided in a specific embodiment of the present invention. Figure 6 This is a schematic diagram of the inference test images and the generated inference results during the verification of the pole reinforcement position detection algorithm provided in a specific embodiment of the present invention.

[0011] In the picture: 1. Control console; 2. Detection bracket; 3. Detection camera; 11. AI detection terminal; 12. PLC controller; 13. Alarm group; 4. Fence; 41. Detection area; 14. Monitor; 42. Gantry. Detailed Implementation

[0012] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0013] As shown in the figure, a rebar cage AI intelligent detection system includes a control console 1 and a detection bracket 2 that works with the rebar cage. The detection bracket 2 is equipped with detection cameras 3 in six viewing directions. All detection cameras 3 are binocular cameras. The control console 1 is equipped with an AI detection terminal 11, a PLC controller 12, and an alarm group 13 that work with the detection cameras 3. During detection, the AI ​​detection terminal 11 identifies the rebar cage information collected by the detection cameras 3 and transmits it to the PLC controller 12 to control the alarm group 13 to issue a warning. During inspection, the steel cage to be inspected is first hoisted onto the inspection bracket 2. Then, by setting inspection cameras 3 in six directions of the steel cage, image information from various angles of the steel cage can be fully collected, such as the quantity of main bars / stirrups / upright bars, the spacing of main bars, and the diameter of main bars. Finally, the information is transmitted to the AI ​​inspection terminal 11. The AI ​​inspection terminal 11 can quickly perform intelligent recognition and measurement based on vision and deep learning algorithms. When the inspection result is unqualified, it can link the PLC controller 12 to control the alarm group 13 to issue an alarm, reminding the staff to handle it. In order to distinguish the alarm type, the alarm group 13 includes a quantity alarm, a spacing alarm, and a diameter alarm, so as to accurately identify the type of unqualified. Furthermore, in order to facilitate the display of the inspection results, a display 14 is also set on the control console 1. The display 14 is connected to the AI ​​inspection terminal 11 and the PLC controller 12 to display the recognition effect and link the PLC controller 12 to control the alarm group 13. This vision-based AI intelligent inspection can efficiently complete the inspection of various parameters of the steel cage. Moreover, the entire inspection system platform has a simple structure and the entire inspection process is fully automated, which can reduce manual input and avoid the safety hazards of manual inspection.

[0014] Preferably, to ensure personnel safety during inspection, a fence 4 is also included. The fence 4 forms a rectangular inspection area 41 centered on the inspection bracket 2. The control console 1 and the inspection camera 3 are both located within the inspection area 41. Furthermore, to prevent the top inspection camera 3 from obstructing the hoisting of the rebar cage onto the inspection bracket 2, a gantry 42 that cooperates with the inspection bracket 2 is also provided within the inspection area 41. That is, the gantry 42 can be set on one side of the inspection bracket 2. Three inspection cameras 3 are set in the top and bottom views of the inspection bracket 2. The three top inspection cameras 3 are arranged equidistantly along the crossbeam of the gantry 42. One inspection camera 3 is set in each of the other viewing directions. This facilitates the hoisting of the rebar cage to be inspected and allows the six-view inspection cameras to operate normally at the workstation. Setting more cameras in the top and bottom views allows for accurate observation of a larger area. At the same time, to ensure image clarity, the bottom view inspection camera 3 is equipped with a light source to facilitate the acquisition of image information by the bottom view inspection camera 3. When acquiring information, the six viewing directions are front, back, left, right, up, and down. Due to limitations in the field of view of the detection camera 3 and the size of the rebar cage, each view is divided into a head region, a mid region, and a tail region. The mid region captures the middle area of ​​the rebar cage, while the head and tail regions capture the two ends of the rebar cage. Two cameras are set up in each region for binocular detection, resulting in a total of six detection cameras 3 per view. The shooting angle of all detection cameras 3 is perpendicular to the plane of the view being captured, and image data is acquired via RTSP protocol. Once installed, the position and angle of the detection cameras 3 cannot be moved. The vertical view is responsible for detecting the main and vertical reinforcement bars of the rebar cage, the front and back view is mainly for detecting the stirrups, and the left and right view is mainly for detecting the main reinforcement bars. For each type of rebar to be detected, it is jointly detected and corrected by two view directions. For example, stirrups are jointly detected by the left and right view directions. After the algorithm detects the quantity in each view direction, it compares the quantity difference between the two view directions to ensure the accuracy of the stirrup detection quantity. The camera parameters and spatial offset information are obtained through stereo calibration and stereo correction in binocular detection. The diameter and spacing are calculated using this information and the identified information to ensure the detection accuracy. Furthermore, the detection bracket 2 is a hollow bracket with an arc shape at the top that matches the steel cage. The lowest point of the detection bracket 2 is at least 90cm from the ground.

[0015] Furthermore, the AI ​​detection terminal includes an identification module, a qualification judgment module, and a detection result display and storage module. The identification module detects the quantity of main reinforcing bars, stirrups, and vertical reinforcing bars, and detects the relative position of individual targets on the image. The qualification judgment module compares the data processed by the identification module with the data of the same type as the defined data to determine the qualification of individual targets in terms of quantity, diameter, spacing, etc., and finally outputs the judgment result. The detection result display and storage module plots the detected data and the detected image on the displayed image. The detection results and images are stored in the database. The recognition module mainly uses the YOLOv8 model to detect the number of main bars, stirrups, and vertical bars, and detects the relative position of individual targets (based on pixel coordinates) on the image (e.g., when detecting stirrups, one stirrup is considered as one individual, the same below). The actual distance corresponding to each point on the horizontal axis x under pixel coordinates is determined by curve fitting (the camera cannot be moved or its angle shifted after installation, otherwise the relationship between the fitted pixels and the actual distance will be deviated, and the subsequent binocular detection will also be deviated). The contour information of individual targets is detected using the segment method of YOLOv8, and then the pixel diameter of the individual target is obtained. The pixel diameter is then converted into the actual diameter using the above-fitted curve. After YOLOv8 detects two adjacent individual targets, its pixel distance is calculated and converted into the actual distance d using the above-fitted curve. Then, the depth Z of the two adjacent individual targets relative to the camera is detected by binocular stereo vision. i Z i+1 (i represents the i-th single target), and finally, the formula is calculated using the hypotenuse of a right triangle: The actual distance is calculated. The qualification judgment module first acquires standard pre-produced images of different models of rebar cages, then delineates individual targets in the head, tail, and mid images of the same type of rebar cage from various viewpoints, serving as the basis for subsequent product qualification judgment. The data processed by the recognition module is compared with the delineated data of the same model to determine the number, diameter, and spacing of individual targets for qualification judgment, and finally, the judgment result is output. The detection result display and storage module displays the detected data and images, including diameter, spacing, and quantity, on the AI ​​intelligent terminal platform after target detection and qualification judgment are completed. The data is also plotted on the displayed image. The AI ​​intelligent terminal stores the detection results and images in the MongoDB database for traceability; simultaneously, the latest real-time detection results are displayed on the web screen of console 1.

[0016] Verification of main reinforcement position detection algorithm: (1) The location of the main reinforcement is distinguished by using the yolov8n-segment algorithm; (2) A model was trained using existing video data of steel reinforcement, and the model was used for inference testing; (3) The number of main reinforcement bars can be calculated using this algorithm; like Figure 4 As shown, the verification results of the pole reinforcement position detection algorithm are consistent with the actual results.

[0017] Validation of the lumbar muscle position detection algorithm (1) The yolov8n-segment algorithm is used to distinguish the location of the lumbar muscles; (2) A model was trained using existing video data of steel reinforcement, and the model was used for inference testing; (3) The number of lumbar ligaments can be calculated using this algorithm; like Figure 5 As shown, the verification results of the pole reinforcement position detection algorithm are consistent with the actual results.

[0018] Verification of the pole reinforcement position detection algorithm (1) The yolov8n-detect algorithm is used to distinguish the position of the upright reinforcement; (2) A model was trained using existing video data of steel reinforcement, and the model was used for inference testing; (3) The number of upright reinforcement bars can be calculated using this algorithm; like Figure 6 As shown, the verification results of the pole reinforcement position detection algorithm are consistent with the actual results.

[0019] This invention has been described through preferred embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. This invention is not limited to the specific embodiments disclosed herein; other embodiments falling within the scope of the claims are also within the protection scope of this invention.

Claims

1. A steel cage AI intelligent detection system, characterized in that: The application relates to a steel reinforcement cage AI intelligent detection system which comprises a control console (1) and a detection support (2) matched with a steel reinforcement cage, detection cameras (3) are arranged at six visual directions of the detection support (2), an AI detection terminal (11), a PLC controller (12) and an alarm group (13) matched with the detection cameras (3) are arranged in the control console (1), during detection, the AI detection terminal (11) identifies the steel reinforcement cage information collected by the detection cameras (3) and transmits the information to the PLC controller (12) to control the alarm group (13) to give a warning.

2. The steel reinforcement cage AI intelligent detection system according to claim 1, characterized in that: It further comprises a fence (4) which surrounds a rectangular detection area (41) with the detection support (2) as the center, and the control console (1) and the detection cameras (3) are arranged in the detection area (41).

3. The steel reinforcement cage AI intelligent detection system according to claim 1, characterized in that: A display (14) is further arranged on the control console (1), the display (14) is connected to the AI detection terminal (11) and the PLC controller (12) to display the identification effect and control the alarm group (13) through the PLC controller (12).

4. The steel reinforcement cage AI intelligent detection system according to claim 3, characterized in that: The alarm group (13) comprises a quantity alarm, a spacing alarm and a diameter alarm.

5. The steel reinforcement cage AI intelligent detection system according to claim 2, characterized in that: A portal (42) matched with the detection support (2) is further arranged in the detection area (41), three detection cameras (3) are arranged at the top and bottom visual angles of the detection support (2), the three detection cameras (3) arranged at the top are equidistantly arranged along the crossbeam of the portal (42), and one detection camera (3) is arranged at other visual angles.

6. The steel reinforcement cage AI intelligent detection system according to claim 5, characterized in that: The detection cameras (3) are all binocular cameras.

7. The steel reinforcement cage AI intelligent detection system according to claim 1, characterized in that: The detection cameras (3) at the bottom visual angle are provided with light sources.

8. The steel reinforcement cage AI intelligent detection system according to claim 1, characterized in that: The detection support (2) is a hollow support, the top of the detection support (2) is arc-shaped and matched with the steel reinforcement cage, and the lowest point of the detection support (2) is at least 90 cm away from the ground.

9. The steel reinforcement cage AI intelligent detection system according to claim 5, characterized in that: The AI detection terminal comprises an identification module, a qualified judgment module and a detection result display and storage module, the identification module detects the number of main reinforcement, stirrup and vertical steel and the relative position of single targets on an image, the qualified judgment module judges the number, diameter and spacing of single targets according to the data processed by the identification module and the same type of data circled, and finally outputs the judgment result. The display module displays the detected data and image on the display image, and stores the detected data and image in the database.

Images