Concrete-filled steel tube arch structure apparent damage and internal debonding and void detecting and positioning robot and operation method thereof

Abstract

The invention provides a concrete-filled steel tube arch structure apparent damage and internal debonding and void detecting and positioning robot and an operation method thereof. The device comprises an arch climbing robot trolley, a lengthened mechanical arm, a knocking device, a depth-of-field camera, a signal acquisition device and a control module. The robot trolley stably moves on the arch surface through the permanent magnet wheels and the electromagnets, the lengthened mechanical arm is fixed to the top of the trolley, the knocking range can cover half of the outer surface of the arch through one-time walking through the two-section connecting arm, and full-arch detection can be completed through two-time crawling. The knocking device is located at the tail end of the mechanical arm and generates audio signals. The depth-of-field camera transmits images in real time to detect apparent damage. The signal acquisition device acquires and returns audio signals, and the control module controls the motor through a chip so as to control other modules. The invention provides a use method of the robot, internal void and apparent damage can be positioned in real time in the detection period by analyzing the collected audio and video signal features, the detection efficiency is high, and the application prospect is good.

Description

Technical Field

[0001] The present invention belongs to the field of structural damage detection, and particularly relates to a robot for detecting and positioning apparent damage and internal debonding and voids in a concrete-filled steel tube arch structure and an operation method thereof. Background Art

[0002] The concrete-filled steel tube arch structure is an efficient, durable and economical bridge structure form, especially suitable for bridge projects requiring high bearing capacity and durability. However, due to construction quality problems, temperature changes, shrinkage of the concrete inside the tube, corrosion and other reasons, various diseases such as debonding, voids, and apparent coating damage may occur. Timely damage detection of the concrete-filled steel tube arch structure is beneficial to accurately grasp its service status.

[0003] At present, there are problems in non-destructive testing of concrete-filled steel tube arch bridges, such as difficult data collection, low efficiency, and low accuracy. The non-destructive testing methods for concrete-filled steel tubes mainly include infrared thermal imaging method, ultrasonic method, impact echo method, etc. These methods require ultrasonic instruments or sensors to be attached to the surface of the concrete-filled steel tube or embedded inside the concrete-filled steel tube. The operation is relatively complex, the cost is relatively high, and the efficiency is relatively low, resulting in limited application in the detection of concrete-filled steel tube structures. The manual tapping acoustic method is expected to be widely promoted and applied due to its advantages of good economy and high efficiency. However, the manual tapping method still has problems that need to be solved urgently, such as high danger of manual climbing and tapping on the arch, poor automation, and low accuracy due to the dependence on subjective judgment for defect identification.

[0004] The present invention uses a climbing robot to replace manual tapping. The climbing robot taps on the outer surface of the arch, and then completes the collection of sound and video through wireless signal transmission. The sound signal is transmitted back to a remote computer for debonding and void judgment through an algorithm. At the same time, an apparent damage is assisted in detection through a depth camera, and it is expected to be well applied to the actual project of the concrete-filled steel tube arch structure. Summary of the Invention

[0005] The present invention provides a robot for detecting and positioning apparent damage and internal debonding and voids in a concrete-filled steel tube arch structure and an operation method thereof. The robot significantly improves the detection efficiency by equipping with an extended robotic arm, and locates defects such as debonding, voids, and apparent damage. By analyzing the collected audio and video data, the detection accuracy is improved. At the same time, an operation method of the robot is provided, and transfer learning is carried out by using the detection data set of the first bridge to improve the detection efficiency and accuracy, and reduce the data requirements and training costs.

[0006] In order to achieve the above invention purpose, the present invention provides the following technical solutions:

[0007] A robot for detecting and locating apparent damage and internal debonding and voids in a concrete-filled steel tube arch structure, comprising a climbing arch robot trolley (1), an extended robotic arm (2), a knocking device (3), a depth camera (4), a signal acquisition device (5) and a control module (6);

[0008] The climbing arch robot trolley (1) includes a carbon fiber chassis (11), permanent magnet adsorption wheels (12), an electromagnet (13), a battery (14) and a wireless signal transmitter (15); the permanent magnet adsorption wheels (12) are connected to the four corners of the carbon fiber chassis (11), and each side is provided with a motor, and each wheel is independently controlled by the motor; the electromagnet (13) is placed at the center of the bottom of the carbon fiber chassis (11); the battery (14) is placed on the top of the carbon fiber chassis (11); the permanent magnet adsorption wheels (12), the electromagnet (13) and the motor work together to control the movement of the trolley on the arch; by controlling the magnitude of the current, the adsorption force of the electromagnet (13) is controlled to ensure the safety and stability of the trolley on the arch;

[0009] The extended robotic arm (2) is fixed to the center of the top of the carbon fiber chassis (11) and includes a horizontal connecting arm (21) and a vertical connecting arm (22); when the climbing arch robot trolley (1) is fixed on the arch surface, the knocking range of its extended robotic arm (2) can cover 1 / 2 of the outer circle of the arch cross-section; a pan-tilt with a camera is mounted at the end of the extended robotic arm (2) and is fixed on one side of the knocking device (3);

[0010] The knocking device (3) is fixed to the end of the extended robotic arm (2) and includes a knocking hammer (31), a spring drive device (32) and a connecting rod (33), all of which are controlled by a motor;

[0011] The depth camera (4) is fixed to the top of the carbon fiber chassis (11) on the side where the climbing arch robot trolley (1) advances; the depth camera (4) and the pan-tilt with a camera are connected to the wireless signal transmitter (15) through wires to transmit image information;

[0012] The signal acquisition device (5) is fixed to the bottom of the carbon fiber chassis (11) on the side where the climbing arch robot trolley (1) advances and is composed of a wireless audio transmission device (51) and a microphone (52), which are fixed at the end of the robotic arm, on the side close to the body of the robot trolley;

[0013] The control module (6) is fixed to the top of the carbon fiber chassis (11); the control module (6) includes a printed circuit board (61), a chip (62), an ultrasonic ranging sensor (63) and a motor driver (64); the motor driver (64) is connected to the chip (62) and the motor on both sides through wires to control the movement of the climbing arch robot trolley (1).

[0014] An operation method for a detection and positioning robot for apparent damage and internal debonding and voids in a concrete-filled steel tube arch structure, comprising the following steps:

[0015] S1. Place the detection and positioning robot on the outer surface of the arch, and the detector sends instructions through a wireless remote control;

[0016] S2. The lengthened robotic arm (2) starts to rotate from one side of the arch-climbing robot trolley (1), and rotates every certain range; after stabilization, the knocking device (3) starts to work, and the knocking coverage is the upper half of the outer surface of the arch; the arch-climbing robot trolley (1) moves every certain distance; after one knocking is completed, the sound and the measured point image are transmitted back to the remote computer in real time; the depth camera (4) synchronously transmits the video, and through the provided RGB data, it assists the detector to identify the color, texture and other characteristics of the apparent damage; through the depth data provided by the depth camera (4), it assists the detector to detect the concave and convex morphological changes in the damaged area;

[0017] S3. The remote computer is built-in with a signal processing algorithm, which can extract the knocking sound, filter out the noise, and at the same time identify whether the signal is abnormal; after judging the abnormality, the detector manually controls the robot to knock in place to determine the debonding edge and range; the wireless signal transmission system can transmit the four-wheel coordinates at the same time, and the detector can determine the location of the damage based on this;

[0018] S4. The arch-climbing robot (1) walks along the top of the outer surface of the arch once, and can complete the detection of the upper half of the outer surface of the arch; then it sucks back and walks along the bottom of the outer surface of the arch once, and repeats operations S1-S3 on the lower half of the outer surface of the arch. After walking these two times, one arch ring can be detected;

[0019] S5. After the first detection of the bridge, use the data set collected for the first time as the source data set for transfer learning, effectively utilize the existing data and model knowledge, and improve the efficiency and accuracy of subsequent detections. The steps are as follows:

[0020] S5.1. Use the data set collected from the first concrete-filled steel tube arch bridge as the source data set D 1 for pre-training a deep learning model; these data include the signal data of the bridge and its defect labels, expressed as:

[0021]

[0022] Among them, represents the signal data of the i-th sample in the source data set, represents the label of the i-th sample, 0 represents no defect, 1 represents a defect, N 1 is the number of samples in the source data set;

[0023] S5.2. The data set collected subsequently is used as the target data set D2 , to meet the inspection requirements of different bridges, expressed as:

[0024]

[0025] Among them, represents the signal data of the i-th sample in the subsequent dataset, represents the label of the i-th sample, 0 represents no defect, 1 represents a defect, N 2 is the number of samples in the subsequent dataset;

[0026] S5.3. Based on the source dataset D 1 Train to obtain the pre-trained model M 1 , and the subsequent new model M 2 is expressed as:

[0027] M 2 = M 1 + ΔM (3)

[0028] Among them, ΔM is the newly added or adjusted part during the migration process;

[0029] S5.4. Use the model M 2 to detect the real-time data, and the rapid and accurate positioning of the debonding defect with few samples can be realized.

[0030] Advantages of the present invention:

[0031] (1) For the complexity of the inspection of the concrete-filled steel tube arch structure, devices such as permanent magnet adsorption wheels and extended robotic arms are equipped. There is no need for manual marking of measurement points for path planning, and the entire bridge can be inspected by walking once at the top and bottom of the arch respectively.

[0032] (2) A depth camera is equipped to simultaneously detect the apparent damage of the concrete-filled steel tube arch structure.

[0033] (3) The knocking signal and the corresponding position image can be transmitted back in real time. After identifying an anomaly, control the device to knock near the anomaly in situ to determine the edge and range of debonding and delamination, improving the inspection accuracy.

[0034] (4) Based on the first inspection dataset for transfer learning, the inspection efficiency and accuracy are improved, and the data requirements and training costs are significantly reduced. Description of the Drawings

[0035] Figure 1 is a schematic diagram of the present invention for inspecting a concrete-filled steel tube arch bridge;

[0036] Figure 2 is a schematic diagram of the robot for detecting and positioning the apparent damage and internal debonding and delamination of the concrete-filled steel tube arch structure of the present invention;

[0037] Figure 3 Schematic diagram of the walking chassis of the detection and positioning robot of the present invention;

[0038] Figure 4 Schematic diagram of the knocking device of the detection and positioning robot of the present invention;

[0039] Figure 5 Schematic flow diagram of the transfer learning method of the present invention;

[0040] Figure 6 Typical sound signal waveform diagram collected on a certain bridge;

[0041] Figure 7 Results diagram of anomaly detection on a certain bridge: (a) Anomaly point recognition result in area 1; (b) Anomaly point recognition result in area 2; (c) Anomaly point recognition result in area 3; (d) Anomaly point recognition result in area 4;

[0042] Figure 8 Results diagram (top view) of comparing the knocking sound method with ultrasonic phased array on a certain bridge.

[0043] Markings in the figure: 1. Climbing arch robot trolley; 2. Extended robotic arm; 3. Knocking device; 4. Depth camera; 5. Signal acquisition device; 6. Control module;

[0044] 11. Carbon fiber chassis; 12. Permanent magnet adsorption wheel; 13. Electromagnet; 14. Battery; 15. Wireless signal transmitter;

[0045] 21. Horizontal connecting arm; 22. Vertical connecting arm;

[0046] 31. Knocking hammer; 32. Spring drive device; 33. Connecting rod;

[0047] 51. Wireless audio transmission device; 52. Microphone;

[0048] 61. Printed circuit board; 62. Chip; 63. Ultrasonic ranging sensor; 64. Motor driver. Specific embodiments

[0049] The present invention will be further described in detail below in combination with test examples and specific embodiments.

[0050] Refer to Figures 1 to 8 , taking a certain real bridge as an example, the invented device is used to detect the debonding and delamination defects and apparent damages of the bridge.

[0051] According to the method and steps proposed by the present invention, some areas of the bridge are detected, and some of the sound data collected during the detection is as Figure 6As shown, the anomaly detection function of the convolutional autoencoder model is used to detect this area, and the anomaly detection results are as Figure 7 shown. It can be seen that in the corresponding Figure 7 (a), Figure 7 (b), Figure 7 (c) of Region 1, Region 2, and Region 3, only a very small number of detection points exceed the anomaly threshold, and the distribution is discrete, which does not conform to the distribution law of void points, so it can be determined as a false positive detection result. While in the corresponding Figure 7 (d) of Region 4, multiple detection points continuously exceed the anomaly threshold, indicating that there is a relatively large area of void in Region 4. The shape of this void area is as Figure 8 shown in "the area identified by this method" in Figure 8 . At the same time, comparing this method with the ultrasonic phased array imaging results, as Figure 8 shown, the detection results are highly consistent, further verifying the accuracy of the method proposed in the present invention.

[0052] After obtaining the sound data, taking the convolutional autoencoder model as an example, the transfer learning steps are described in detail as follows:

[0053] (1) Manually label the normal data of the first bridge as the training set, convert the audio signal into a Mel spectrogram, and use it to train the convolutional autoencoder model to obtain the pre-trained model M 1 . Subsequently, use all the data of the first bridge as the test set D 1 , and use the model M 1 to identify abnormal sounds and locate defects, and complete the defect detection of the first bridge.

[0054] (2) The robot collects the audio dataset D 2 of the second bridge.

[0055] (3) Retain and freeze the encoder part of the pre-trained model M 1 , and unfreeze the decoder part to enable it to adapt to the data characteristics of the target task.

[0056] (4) Add a small classification network (such as a fully connected layer) after the encoder of the autoencoder to perform defect classification based on the features extracted by the encoder. Freeze the encoder and only train the classification head and the decoder part to adapt to the reconstruction task of the target bridge.

[0057] (5) Partially unfreeze the upper layers of the encoder and perform fine-tuning to better capture the features of the target bridge. According to the number of labeled data and the training situation, gradually unfreeze more encoder layers to complete the full fine-tuning and obtain the new model M 2 . Finally, use the new model M 2 to perform defect detection on the second bridge.

Claims

1. A robot for detecting and positioning surface damage and internal debonding and degassing of steel tube concrete arch structures, characterized in that: It comprises an arch climbing robot vehicle (1), an extended mechanical arm (2), a knocking device (3), a depth of field camera (4), a signal acquisition device (5) and a control module (6); The arch climbing robot car (1) comprises a carbon fiber chassis (11), a permanent magnetic adsorption wheel (12), an electromagnet (13), a battery (14) and a wireless communication system (15); the permanent magnetic adsorption wheel (12) is connected to the four corners of the carbon fiber chassis (11), and a motor is provided on each side, and each wheel is controlled by a motor alone; the electromagnet (13) is placed at the bottom center of the carbon fiber chassis (11); the battery (14) is placed at the top of the carbon fiber chassis (11); the permanent magnetic adsorption wheel (12), the electromagnet (13) and the motor work together to control the movement of the car on the arch; the adsorption force of the electromagnet (13) is controlled by controlling the magnitude of the current to ensure the safety and stability of the car on the arch; The extended mechanical arm (2) is fixed at the top center of the carbon fiber chassis (11), and comprises a horizontal connecting arm (21) and a vertical connecting arm (22); when the arch climbing robot car (1) is fixed on the arch surface, the knocking range of the extended mechanical arm (2) can cover 1 / 2 of the outer circle of the arch cross section; The knocking device (3) is fixed to the end of the extended mechanical arm (2), and comprises a knocking hammer (31), a spring transmission device (32) and a connecting rod (33), all of which are controlled by a motor; The depth-of-field camera (4) is fixed on the top of the carbon fiber chassis (11) and is located on the side of the arch-climbing robot vehicle (1) that moves forward; the depth-of-field camera (4) and the pan / tilt platform equipped with a camera are connected to a wireless signal transmitter (15) via wires to transmit image information; The signal collection device (5) is fixed to the bottom of the carbon fiber chassis (11), located on the side where the arch climbing robot car (1) moves forward, and is composed of a wireless audio transmission device (51) and a microphone (52), and is fixed to the end of the mechanical arm, close to the side of the robot car body; The control module (6) is fixed on the top of the carbon fiber chassis (11); the control module (6) comprises a printed circuit board (61), a chip (62), an ultrasonic distance sensor (63) and a motor driver (64); the motor driver (64) is connected to the chip (62) and the motor via two sides of a wire respectively, so as to control the movement of the arch climbing robot trolley (1).

2. An operating method of a robot for detecting and positioning surface damage and internal debonding and degassing of a steel tube concrete arch structure, characterized in that: The following steps are involved: S1, placing the detection and positioning robot on the outer surface of the arch, and the detection personnel send instructions through the wireless remote control; S2, the extended mechanical arm (2) starts to rotate from one side of the arch climbing robot car (1), and rotates once every certain range; after stabilization, the knocking device (3) starts to work, and the knocking coverage range is the upper outer surface of the arch; the arch climbing robot car (1) moves once every certain distance; after the knocking is completed, the sound and the image of the measuring point are transmitted back to the remote computer in real time; the depth of field camera (4) synchronously transmits the video, and through the provided RGB data, assists the inspection personnel to identify the color, texture and other characteristics of the apparent damage; The depth data provided by the depth camera (4) assists the inspector in detecting the changes in the concave and convex shapes of the damaged area; S3. The remote computer has a built-in signal processing algorithm, which can extract the knocking sound, filter out the noise, and identify whether the signal is abnormal. After determining the abnormality, the inspection personnel manually control the robot to knock in place to determine the edge and range of the air gap. The wireless signal transmission system can simultaneously send back the coordinates of the four wheels, and the inspection personnel can determine the location of the damage based on this. S4, the arch climbing robot car (1) walks along the top of the outer surface of the arch once, and can complete the detection of the upper half of the outer surface of the arch; then it walks along the bottom of the outer surface of the arch once, and repeats the operations S1-S3 on the lower half of the outer surface of the arch. After walking twice, one arch circle can be detected; S5. After the first inspection of the bridge, the first collected data set is used as the source data set for transfer learning, effectively utilizing the existing data and model knowledge to improve the efficiency and accuracy of subsequent inspections. The steps are as follows: S5.

1. The data set collected from the first steel tube concrete arch bridge is used as the source data set D1 for pre-training the deep learning model; these data contain the signal data of the bridge and its defect labels, expressed as: in, represents the signal data of the i-th sample in the source data set, represents the label of the i-th sample, 0 represents no defect, 1 represents defect, and N1 is the number of samples in the source data set; S5.

2. The subsequently collected data set is used as the target data set D2 to meet the detection requirements of different bridges, which is expressed as: in, represents the signal data of the i-th sample in the subsequent data set, represents the label of the i-th sample, 0 represents no defect, 1 represents defect, and N2 is the number of samples in the subsequent data set; S5.3, based on the source data set D1, the pre-trained model M1 is trained, and the new model M2 obtained later is expressed as: M2=M1+ΔM (3) Among them, ΔM is the part added or adjusted during the migration process; S5.

4. Use model M2 to detect real-time data to achieve fast and accurate positioning of void defects with a small number of samples.

Images