Concrete crack comprehensive detection method and system based on wall-climbing robot

Abstract

The invention discloses a concrete crack comprehensive detection method and system based on a wall-climbing robot. The system comprises a double-cylinder negative pressure adsorption mobile platform with a self-adaptive pressure control function, a visual tight coupling real-time positioning module, a crack image acquisition and intelligent identification module, an ultrasonic depth detection module and a high-pressure mooring power supply system. The method comprises the steps that the robot is stably attached to a concrete facade through the adsorption platform; accurate positioning in a GPS-free environment is realized by using a visual tight coupling system; automatically identifying cracks through a deep learning network based on camouflage target identification; an ultrasonic sensor integrated with an elastic pressing mechanism clings to the surface to collect data; the crack depth is solved by adopting an algorithm based on multi-mode noise separation and variational mode decomposition; and finally fusing the multi-source data to generate a detection report. According to the invention, integrated accurate detection of concrete structure cracks from appearance to depth is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of robot detection systems, non-destructive testing equipment, concrete structure detection technology applications, and specifically relates to a concrete crack comprehensive detection method and system based on a wall-climbing robot. BACKGROUND

[0002] Modern buildings, bridges, dams and other large infrastructure generally use concrete as the main structural material. Such structures have the advantages of strong bearing capacity, economical cost, and strong plasticity, but cracks will inevitably occur in concrete structures during long-term service due to multiple factors such as load, temperature change, shrinkage and creep, and environmental erosion. Fine cracks are not only the first channel for internal steel corrosion, but also the key indicator of structural damage and load capacity decline. The apparent characteristics such as width and length of the cracks, together with the depth information, constitute the core basis for evaluating the health status of the structure and predicting its remaining life. Therefore, for concrete structures, especially for fine cracks in high-altitude, water-side or personnel-difficult-to-reach positions, comprehensive apparent identification and accurate depth detection are crucial and extremely challenging tasks in the safe operation of infrastructure.

[0003] However, efficient and accurate crack detection of the above structures faces severe technical challenges and practical problems. First, the detection environment is dangerous and difficult to access. Bridge piers, dam bodies, and building outer walls often lack reliable working surfaces, and traditional methods rely heavily on baskets, scaffolding or aerial work platforms, which not only take a long time to build and are costly, but also pose significant safety risks. Second, the detection object has weak features. Fine cracks are easily obscured by complex stains, water marks, color differences and other textures on the concrete surface, forming a "camouflage effect", which can lead to missed detection and misjudgment by manual visual inspection or traditional image processing algorithms. Third, the detection dimension needs to be extended from two-dimensional to three-dimensional. Complete crack evaluation not only requires identifying its apparent form, but also quantifying its depth, a key three-dimensional information. Existing depth detection methods (such as ultrasonic method) usually require the probe to be stably and fully coupled with the concrete surface, which is difficult to achieve in mobile and non-contact automated inspection, and the detection signal is easily affected by system noise and environmental interference, making it difficult to analyze.

[0004] To address these challenges, automated detection technologies, especially unmanned aerial vehicles (UAVs) and robots, are expected to play a crucial role. In recent years, there have been numerous studies on using UAVs for structural detection. However, when it comes to detecting structural cracks, especially small cracks and crack depths, UAV detection has limitations. Although UAVs have good maneuverability, they have inherent shortcomings such as poor wind resistance, short endurance, and positioning accuracy limited by GPS signals, making it difficult to achieve the sustained and stable contact and compression necessary for ultrasonic flaw detection. In terms of robots, wall-climbing robots provide a new platform for disease detection. However, existing wall-climbing robots are designed for smooth walls (such as glass curtain walls and smooth steel plates), and their adsorption methods and movement mechanisms are not suitable for rough and dusty concrete surfaces, which may result in adsorption failure or movement difficulties. More importantly, existing systems have single functions, either equipped with only visual sensors for surface inspection, lacking depth detection capability, or integrated with depth detection sensors but lack precise mechanisms to maintain constant coupling with rough surfaces during movement, resulting in poor data quality.

[0005] At the data processing level, existing technologies also have shortcomings. Traditional image processing-based crack identification methods have poor robustness under complex background interference, and early deep learning models have limited feature extraction capabilities for subtle and disguised targets. For ultrasonic signal analysis, conventional filtering and spectral analysis methods often struggle to separate complex signals mixed with multiple noise patterns, making it difficult to accurately extract effective modal components representing crack depth.

[0006] Therefore, there is an urgent need in the art for a comprehensive solution that can simultaneously address the four major challenges of reliable adsorption and movement in harsh environments, accurate identification of subtle surface diseases, stable acquisition of depth data in a moving state, and accurate calculation of depth information in a complex noise background. Developing a wall-climbing robot system that integrates adaptive adsorption platforms, high-precision signal-free autonomous positioning, robust contact-type sensing mechanisms, and intelligent multi-modal data analysis algorithms is of urgent practical need and important technical value for improving the automation level, accuracy, and efficiency of concrete structure detection and ensuring the safe operation of major infrastructure. SUMMARY

[0007] The present invention aims to overcome the shortcomings of existing technologies and provide a concrete crack comprehensive detection method and system based on a wall-climbing robot. The system can automatically identify and accurately detect the internal depth of small cracks on the surface of concrete structures, effectively addressing the low efficiency and high risk of manual detection in dangerous environments such as high-rise and near-water environments, as well as the poor adaptability and single detection dimension of existing automated equipment in environments without GPS signals and rough surfaces, thereby serving the efficient, automated, and safe inspection of large concrete infrastructure.

[0008] To solve the above technical problems, the present application adopts the following technical solutions:

[0009] A concrete crack comprehensive detection method based on a wall-climbing robot, comprising the following steps:

[0010] Step 1: Place the wall-climbing robot platform on the surface of the concrete structure, and stably adsorb it on the complex facade such as the concrete wall surface, pier column or bridge bottom through its double-cylinder negative pressure adsorption system.

[0011] Step 2: Use the binocular camera and inertial navigation sensor carried by the robot to real-time locate the position of the robot in the satellite signal-free environment through the visual tight coupling system.

[0012] Step 3: In the process of robot inspection, control the binocular camera to collect crack images, and automatically identify the fine cracks under complex stain interference through the deep learning network based on camouflage target recognition.

[0013] Step 4: Synchronously control the ultrasonic array sensor to tightly adhere to the concrete surface through the elastic compression module at the tail of the robot, and emit and receive ultrasonic data.

[0014] Step 5: Analyze the ultrasonic data through the algorithm based on multi-modal noise separation and variational modal decomposition to accurately solve the crack depth.

[0015] Step 6: Integrate the apparent information and depth data of the crack, combine the robot positioning information, and generate a comprehensive detection report.

[0016] Further, the specific implementation of step 1 is that the wall-climbing robot platform comprises a double-cylinder negative pressure adsorption unit, a four-wheel drive mobile chassis, a binocular camera, an ultrasonic array sensor and an on-board computer. The mobile chassis adopts horizontal gear transmission, so that the driving wheel has self-locking function to prevent slipping when power is interrupted. The driving tire adopts dense tooth tire skin made of silicone material to enhance the grip and obstacle crossing ability on rough concrete surface. The negative pressure adsorption system has self-adaptive pressure control function, and a circle of flexible foam is distributed around the negative pressure suction area at the bottom, which forms a closed cavity with the concrete surface during work, so that the system can generate stable adsorption force with lower power consumption, ensuring reliable adhesion of the robot in vertical, inclined and other attitudes.

[0017] Further, the robot platform adopts a high-voltage tethered power supply system, which includes a ground mobile power supply, a voltage boosting module, a light power supply cable, and a voltage reducing module located on the robot body; the ground mobile power supply provides 220V alternating current, which is converted into 500V direct current high voltage via the voltage boosting module; the 500V direct current high voltage is transmitted to the robot body through the light power supply cable; the voltage reducing module converts the input 500V high voltage into a safe direct current working voltage of 24V or 48V to power the drive, sensing and computing units of the robot. By increasing the transmission voltage to 500V, the transmission current is significantly reduced at the same power, so that the power supply cable can use a thinner wire diameter, thereby effectively reducing the weight of the cable, reducing the heating of the line, and avoiding the drag on the movement of the robot caused by the weight and stiffness of the cable, realizing the long-time continuous operation of the robot.

[0018] Further, the adaptive pressure control of the double-cylinder negative pressure adsorption system adopts an algorithm based on fuzzy PID control, using the pressure values collected by the pressure sensors on the four drive wheels in real time as feedback signals, the deviation of the collected real-time pressure values from the target pressure and its rate of change as the inputs of the fuzzy controller, and the inputs being fuzzified and fuzzy reasoning, and the correction amounts of the PID controller proportion, integral and differential parameters being dynamically output according to the preset fuzzy rules; the PID controller calculates the control signal for adjusting the power of the suction motor using the parameters that have been dynamically corrected, and stabilizes the negative pressure of the adsorption system at the target set value by changing the driving power of the motor, so as to cope with the adsorption force fluctuation caused by the roughness change of the concrete surface and the change of the robot posture.

[0019] Further, the visual tight coupling positioning method used in step 2 specifically includes that the system fuses binocular vision images and inertial navigation data to perform real-time pose estimation using an error state iterative Kalman filter. This algorithm is based on the high-frequency motion prediction provided by the inertial navigation data, uses the images collected by the binocular camera for low-frequency but absolutely accurate measurement updates, aligns the images by minimizing the photometric error, effectively suppresses the cumulative drift of inertial navigation, and thus realizes high-precision, continuous real-time positioning and trajectory tracking of the robot in environments without GPS such as bridge bottoms and indoor environments.

[0020] Further, the deep learning network based on camouflage target recognition in step 3 is further defined as follows: the network adopts an encoding-decoding structure. The network is designed to overcome the limitations of traditional crack segmentation models in terms of uneven gray scale, texture confusion of stains, and missed detection of micro-cracks. A local contrast enhancement module and a multi-scale gradient perception module are embedded in the encoder. The former dynamically enhances the contrast difference between cracks and background by adaptive histogram equalization and regional gray standard deviation analysis, and suppresses the homogeneous interference of stain areas. The latter extracts crack directional gradient features using the Sobel operator, and captures multi-scale context information using a hollow convolution pyramid, thereby strengthening the model's ability to recognize crack topological structures. The decoder introduces a dual-path feature fusion mechanism to promote cross-layer interaction between shallow detail features and deep semantic features, thereby effectively resisting the interference of complex backgrounds such as surface stains and color differences while preserving the complete morphology of fine cracks, and achieving high-precision pixel-level crack segmentation and recognition.

[0021] Further, the elastic compression module structure and working mode of the ultrasonic array sensor in step 4 include: the module adopts a parallel four-bar mechanism as shown in the accompanying Figure 2 The upper cross bar and the lower cross bar are hinged at one end to the rear of the robot and at the other end to the vertical rod, allowing the vertical rod to move freely up and down. The ultrasonic array sensor is fixed to the bottom of the vertical rod and can always remain parallel to the bottom surface of the robot. When the robot moves, the module, under the weight and controllable pressure provided by the damper, allows the sensor panel to adapt to the slight undulations of the concrete surface and closely adhere to it. When the robot is in a relaxed state, the ultrasonic array sensor naturally droops; when the robot is in a working state, the ultrasonic array sensor moves upward under the pressure of the concrete surface, compressing the damper, and the reaction force generated by the damper causes the sensor panel to be stably pressed against the concrete surface, ensuring effective transmission and reception of ultrasonic waves and avoiding measurement errors caused by insufficient coupling.

[0022] Further, the algorithm based on multi-modal noise separation and variational modal decomposition in step 5 specifically includes: to solve the problem of sound signal aliasing caused by rotor vibration, motor noise and airflow disturbance of the suction motor during wall climbing robot detection, a multi-dimensional noise fingerprint database is established, covering time domain waveform, frequency domain energy distribution and statistical characteristics; a variational modal decomposition-independent component analysis joint algorithm is used to adaptively determine the number of decomposition layers by introducing a kurtosis-envelope entropy joint optimization criterion, and the best parameter combination is dynamically searched by a particle swarm optimization algorithm; an independent component analysis algorithm based on negative entropy maximization is used to extract independent components, and a noise reference channel verification mechanism is designed to construct a noise feature matching model combined with motor speed and other working condition parameters, realizing efficient separation of ultrasonic echo signals and interference noise. Subsequently, the effective modal components representing crack depth information are analyzed, and finally the accurate calculation of crack depth in a high noise background is realized.

[0023] Further, the detection report generation method in step 6 comprises: the system fuses and maps the apparent information such as the crack position, length, width and the like identified in step 3, the crack depth data calculated in step 5, and the accurate positioning information of the robot obtained in step 2. Finally, a visual detection report containing the spatial distribution, three-dimensional morphology and quantitative parameters of the crack is generated, data query, statistical analysis and historical comparison are supported, and comprehensive and intuitive data support is provided for the health condition evaluation and maintenance decision of the concrete structure.

[0024] Another object of the present application is to provide a wall climbing robot system for implementing the above method, comprising:

[0025] 1. A double-cylinder negative pressure adsorption mobile platform with adaptive pressure control, suitable for stable climbing on rough concrete vertical surfaces;

[0026] 2. A visual tight-coupling real-time positioning module, comprising a binocular camera, an inertial navigation system and an embedded algorithm, and realizing autonomous navigation in a GPS-free environment;

[0027] 3. A crack image acquisition and intelligent identification module, carrying a high-performance onboard computer and a deep learning network, and realizing real-time detection of fine cracks;

[0028] 4. An ultrasonic depth detection module, integrating an elastic compression mechanism and an array sensor, and ensuring stable data acquisition during movement;

[0029] 5. A high-voltage tethered power supply system, realizing long-time continuous operation of the robot through ground voltage boosting and onboard voltage dropping modes;

[0030] 6. A data fusion and report generation module, integrating multi-source information, and outputting visual and quantifiable comprehensive detection results.

[0031] Through the above system and method, the present application realizes all-around crack detection of concrete structures from the surface to the interior and from qualitative to quantitative, and is particularly suitable for automatic fine inspection of large concrete infrastructure in complex, high-risk and satellite signal-free environments, and has significant engineering application value and promotion prospect.

[0032] Compared with the prior art, the present application has the following advantages:

[0033] 1. The existing wall climbing robot or unmanned aerial vehicle platform is mostly single function, or can only collect surface images, or is difficult to stabilize the implementation of contactless damage detection during movement. The present application innovatively integrates the appearance visual detection system based on deep learning and the ultrasonic depth detection module integrated with the elastic compression mechanism in the same robot platform, synchronously obtains the two-dimensional surface information and three-dimensional depth data of the crack in one inspection operation, and realizes comprehensive evaluation of the health status of the concrete structure crack in all directions and multiple dimensions.

[0034] 2. In the satellite signal denial environment of the bottom of the bridge, indoor venues and the like, conventional automated equipment cannot effectively position. The present application adopts a visual tight coupling positioning system based on an error state iterative Kalman filter, fuses binocular vision and inertial navigation data, realizes real-time and high-precision autonomous positioning and navigation independent of external signals, and provides a solid foundation for spatial registration of detection data and autonomous inspection of the robot.

[0035] 3. The traditional image processing algorithm or ordinary deep learning model has limited recognition ability for fine cracks similar to background texture and high stains, and has a high missed detection rate. The deep learning network based on camouflage target recognition adopted in the present application significantly enhances the extraction and recognition ability of the model for fine crack features in complex backgrounds through the introduction of local contrast enhancement, multi-scale gradient perception and double-path feature fusion mechanisms, and effectively reduces missed detection and misjudgment.

[0036] 4. In the ultrasonic detection on the mobile platform, the coupling stability of the probe and the surface is the key difficulty to ensure data quality. The present application designs an elastic compression module based on parallel mechanism and damper, which can adapt to the slight unevenness of the concrete surface and always provide constant and close adhesion pressure for the ultrasonic sensor during the movement of the robot, effectively avoiding data distortion caused by poor coupling, and providing high-quality signal source for subsequent depth calculation.

[0037] 5. Strong noise generated by robot body vibration, motor operation and the like seriously interferes with the effective analysis of ultrasonic echo signals. The present application adopts an advanced signal processing algorithm based on multi-modal noise separation and variational modal decomposition, which can accurately separate and extract key modal components related to crack depth from mixed signals with extremely low signal-to-noise ratio, so as to realize accurate and reliable calculation of crack depth in complex working conditions.

[0038] 6. The application will automatically identify the apparent information of the cracks, accurately calculate the crack depth data, and deeply fuse with the real-time positioning information of the robot, and map it into the three-dimensional model to generate a visual comprehensive detection report. This method completely changes the disadvantages of data isolation and lack of spatial correlation in traditional detection, realizes the visualization query, statistical analysis and trend prediction of all detection results in a unified digital model, greatly improves the intuitiveness and decision support value of the detection results.

[0039] 7. Compared with the traditional mode relying on multiple equipment step-by-step operation, the application can complete the whole process from discovery, identification to quantitative analysis by a single robot in a single operation, greatly improving the detection efficiency, reducing the operation risk and labor cost in high-altitude, water and other dangerous environments, and providing an innovative technical means for efficient and safe operation and maintenance of large concrete infrastructure. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The figure is a composition diagram of the wall-climbing robot system for concrete structure crack appearance and depth detection involved in the application, wherein 11 is a double-cylinder negative pressure adsorption unit, 12 is a dense tooth tire, 13 is a binocular camera, 14 is an on-board computer inside the robot body, 15 is a horizontal gear driving module, 16 is an elastic extension mechanism, 17 is an ultrasonic array sensor, 18 is a power cable, and 19 is a high-voltage power supply module with a built-in battery;

[0041] Figure 2 The figure is a schematic diagram of the elastic extension mechanism of the robot ultrasonic array module in the embodiment of the application, wherein 21 is an upper horizontal rod, 22 is a damper, 23 is a vertical rod, 24 is a lower horizontal rod, and 25 is an ultrasonic array signal transmitting and receiving module;

[0042] Figure 3 The figure is a schematic diagram of the crack segmentation network based on camouflage target recognition in the embodiment of the application;

[0043] Figure 4 The figure is a general flowchart of the concrete structure crack appearance and depth detection method involved in the application. DETAILED DESCRIPTION

[0044] The technical solutions in the application will be described below in conjunction with the embodiments and the drawings. Obviously, the described embodiments are only a part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0045] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0046] The application will be further described in connection with specific embodiments, but not as a limitation of the application.

[0047] The wall-climbing robot system disclosed in the embodiment comprises a hardware component and an attached Figure 1 As shown in the figure, the core comprises: a robot moving platform with a double-cylinder negative pressure adsorption unit 11 and a four-wheel drive moving chassis with a dense toothed tire 12; a binocular camera 13 arranged at the front of the platform; an on-board computer 14 located in the body; a horizontal gear drive module 15 used for driving wheels; an elastic telescopic mechanism 16 integrated at the tail of the robot and an ultrasonic array sensor 17 thereon; and a high-voltage power supply module 19 with a built-in battery connected through a power supply cable 18.

[0048] The embodiment of the application discloses a concrete crack comprehensive detection method based on a wall-climbing robot. Figure 4 As shown in the figure, the total flow of the method comprises the following steps:

[0049] S1: Equipment inspection and program preparation, start the inspection.

[0050] Before the inspection operation, the robot platform, the sensor and the power supply system are inspected, it is confirmed that the binocular camera 13, the ultrasonic array sensor 17, the on-board computer 14 and other units are working normally, and the system program is started, and the inspection task is prepared to start.

[0051] S2: The robot is placed on the surface of the structure, and the double-cylinder negative pressure adsorption system is self-adaptively adsorbed on the surface of the concrete.

[0052] The wall-climbing robot platform is placed on the surface of the concrete structure, the double-cylinder negative pressure adsorption system 11 is started, and it is self-adaptively and stably adsorbed on the wall surface, the pier column or the bottom of the bridge.

[0053] In this embodiment, the wall-climbing robot platform adopts a four-wheel drive wheeled mobile chassis. Each drive wheel uses a horizontal gear drive module 15 to transmit power, and this transmission structure allows the drive wheel to be self-locking when there is no servo motor signal input, effectively preventing the robot from sliding down the inclined or vertical wall surface due to power interruption. The drive tire uses a dense toothed tire skin 12 made of silicone material to enhance adhesion and passability on rough concrete surfaces. The double-cylinder negative pressure suction system 11 generates suction force through two negative pressure cavities located at the bottom of the robot, and a ring of flexible foam sealing strips is arranged around the cavities. When the robot approaches the wall, the foam deforms under pressure to form an effective sealing space. This system is equipped with a fuzzy PID-based adaptive pressure control function: real-time monitoring of the suction state is performed through pressure sensors distributed around the four drive wheels, and the deviation of the collected real-time pressure value from the target pressure and its rate of change are used as inputs to the fuzzy controller. The dynamic correction amount of the PID controller parameters is output through fuzzy reasoning. The PID controller calculates the control signal using the corrected parameters to dynamically adjust the power of the suction motor to respond to changes in concrete surface roughness and robot attitude, ensuring stable suction and movement of the robot in vertical, inclined, and even inverted states such as the bottom of a bridge.

[0054] S3: Real-time determination of robot pose in a GPS-free environment using a vision tight coupling positioning system.

[0055] The robot is equipped with a binocular camera 13 and an inertial navigation sensor (IMU). The on-board computer 14 runs a vision tight coupling positioning algorithm based on the error state iterative Kalman filter (ES-IKF). This algorithm uses high-frequency angular velocity and acceleration data provided by the IMU as the basis for system state prediction, and performs measurement updates by fusing consecutive image frames captured by the binocular camera 13. In the update phase, the algorithm achieves accurate image alignment by minimizing the re-projection error or photometric error between adjacent frames, effectively estimating and correcting the cumulative error (drift) of inertial navigation. This system does not rely on external satellites or pre-set markers, and can achieve real-time, continuous, and high-precision pose estimation and path tracking of the robot in environments completely without GPS signals such as bridge tunnels and indoor venues, providing accurate spatial position labels and coordinate system conversion basis for all subsequent detection data.

[0056] S4: This step is the core execution link of detection, as shown in Figure 4 The apparent image acquisition and recognition and the ultrasonic depth data acquisition and solving are performed in parallel, and the specific sub-steps include:

[0057] S41: The binocular camera acquires crack images.

[0058] During the robot inspection process, the binocular camera 13 continuously acquires high-definition images of the concrete surface. The image data is transmitted in real time to the on-board computer 14 for processing.

[0059] S42: Identify and analyze the apparent micro-cracks based on the deep learning network.

[0060] The on-board computer 14 runs a deep learning crack segmentation network based on camouflage target recognition as shown in the attached Figure 3 The network adopts an encoding-decoding architecture. In the encoder: the local contrast enhancement module dynamically enhances the gray difference between the crack area and the background through adaptive histogram equalization technology and regional gray standard deviation analysis, and suppresses the homogeneous interference of stains; the multi-scale gradient perception module extracts the directional gradient features of the cracks using the Sobel operator, and captures the multi-scale context information from micro to macro by combining the hollow convolution pyramid, to strengthen the model's recognition ability of the crack topological structure. In the decoder: the dual-path feature fusion mechanism performs cross-scale interaction and fusion of the high-resolution detail features (including crack edges, textures) in the shallow layer of the encoder and the rich semantic features in the deep layer, finally realizing pixel-level crack segmentation. The network is trained on a large number of concrete images containing complex backgrounds such as stains, water marks, and color differences, and has very high recognition accuracy and robustness for micro-cracks. The recognition output includes information such as the pixel position, estimated pixel width, and connected domain length of the cracks.

[0061] S43: Collect echo data by elastically compressing the surface of the ultrasonic sensor.

[0062] The synchronous control ultrasonic array sensor 17, 25 is tightly attached to the concrete surface by the elastic compression module 16, and collects ultrasonic data.

[0063] As shown in the attached Figure 2 The elastic compression module 16 specifically consists of an upper crossbar 21, a damper 22, a vertical rod 23, and a lower crossbar 24, forming a parallel four-bar system. The back plate of the ultrasonic array sensor 25 is fixed to the bottom of the vertical rod 23. This structure ensures that the detection surface is always parallel to the bottom surface of the robot body. When the robot is in a relaxed state, the entire module naturally droops under the action of gravity. When the robot moves to the detection area, the sensor panel begins to contact the concrete surface (contact state), and as the robot continues to move forward, the reaction force of the concrete surface on the panel pushes the vertical rod 23 upward, compressing the damper 22. The reaction force generated by the damper 22 then forms a stable and continuous compression force, causing the system to enter a tightly attached working state. This force overcomes the small undulations of the surface, ensuring that the sensor panel and the uneven concrete surface always maintain close and uniform acoustic coupling, thereby ensuring stable transmission of ultrasonic signals and high signal-to-noise ratio of received echo signals.

[0064] S44: Calculate the crack depth based on noise separation and modal decomposition.

[0065] The crack depth is calculated by analyzing the ultrasonic data using advanced signal processing algorithms.

[0066] The airborne computer 14 receives the original ultrasonic echo signal collected in step S43. Since the signal is mixed with strong noise from multiple sources such as robot body vibration, air suction motor operation, etc., direct analysis is difficult. The present application uses an algorithm based on multi-modal noise separation and variational modal decomposition (VMD-ICA) for processing. The algorithm first establishes a multi-dimensional noise fingerprint database covering time domain waveform, frequency domain energy distribution and statistical characteristics as a reference benchmark. Then, the variational modal decomposition (VMD) is used to adaptively decompose the complex mixed signal into several intrinsic modal functions (IMF), in which the kurtosis-envelope entropy joint optimization criterion is introduced, and the particle swarm optimization (PSO) algorithm is used to dynamically search for the optimal combination of the decomposition layer number (K) and the penalty factor (α) parameters of VMD, so as to accurately separate the noise and effective echo of different modes. Further, the independent component analysis (ICA) algorithm based on negative entropy maximization is used to further extract statistically independent components from the IMF components. By designing a noise reference channel verification mechanism and combining with the motor speed and other working condition parameters to build a noise feature matching model, the effective modal components representing crack information in the ultrasonic echo signal and various interference noise modes are accurately separated and removed. Finally, the effective signal components with significantly improved signal-to-noise ratio are analyzed, the diffraction echo or energy attenuation characteristics generated by the ultrasonic wave at the crack tip are identified, and the depth of the crack is accurately calculated according to the known propagation speed of the ultrasonic wave in the concrete.

[0067] S5: Integrate apparent information and depth data.

[0068] The system integrates the apparent information such as the crack identified in step S42 (including its pixel coordinates, geometric dimensions) and the crack depth data calculated in step S44, according to the accurate pose information of the robot provided in step S3, through coordinate transformation for spatial alignment and fusion, and unifies all crack data to a global coordinate system.

[0069] S6: Generate a visual comprehensive detection report.

[0070] All fused data are mapped to a unified data management platform, which can generate a visual comprehensive detection report. The report can clearly show the spatial distribution, geometric dimensions and depth information of the cracks in the form of a two-dimensional map or a three-dimensional model, and can support statistical analysis and historical data comparison by region and by crack size, providing intuitive and quantitative decision-making basis for evaluating the health status of the concrete structure and developing maintenance plans.

[0071] In summary, the embodiments of the present application use the above-mentioned system and method, combined with the attached Figure 1 to the attached Figure 4The specific structure and flow shown realize integrated and automatic detection of concrete structure from apparent identification to internal deep quantification, and are particularly suitable for efficient and safe inspection of large concrete infrastructures such as building outer walls, bridge piers, dam bodies and the like.

[0072] The above merely describes preferred embodiments of the present application, but does not limit the embodiments and protection scope of the present application. It should be understood by those skilled in the art that any equivalent replacement and obvious change made according to the content of the present application should be included in the protection scope of the present application.

Claims

1. A concrete crack comprehensive detection method based on a wall-climbing robot, characterized in that, The method comprises the following steps: Step 1: Place the wall-climbing robot platform on the surface of the concrete structure, and stably adsorb on the concrete facade through the self-adaptive double-cylinder negative pressure adsorption system thereof; Step 2: Use the binocular camera and inertial navigation sensor carried by the robot to realize real-time positioning of the robot through a visual tight coupling system; Step 3: In the process of robot inspection, control the binocular camera to collect crack images, and automatically identify fine cracks through a deep learning network based on camouflage target recognition; Step 4: Synchronously control the ultrasonic array sensor to tightly adhere to the concrete surface through the elastic compression module at the tail of the robot, and emit and receive ultrasonic data; Step 5: Analyze the ultrasonic data through an algorithm based on multi-modal noise separation and variational modal decomposition, and solve the crack depth; Step 6: Integrate the crack apparent information identified in step 3, the crack depth data solved in step 5, and the robot positioning information obtained in step 2 to generate a comprehensive detection report.

2. The method of claim 1, wherein, The double-cylinder negative pressure adsorption system in step 1 has a self-adaptive pressure control function, and the pressure control thereof adopts an algorithm based on fuzzy PID control, uses the pressure value collected by the pressure sensor in real time as feedback, dynamically corrects the PID parameters through the fuzzy controller, and outputs a control signal to adjust the power of the suction motor, so as to stabilize the adsorption pressure.

3. The method of claim 1, wherein, The visual tight coupling system in step 2 fuses binocular vision images and inertial navigation data, and uses an error state iterative Kalman filter to realize real-time pose estimation.

4. The method of claim 1, wherein, The deep learning network based on camouflage target recognition in step 3 adopts an encoding-decoding structure; The encoder thereof embeds a local contrast enhancement module and a multi-scale gradient perception module; The decoder thereof introduces a double-path feature fusion mechanism.

5. The method of claim 1, wherein, The elastic compression module in step 4 comprises an upper cross rod, a lower cross rod, a vertical rod and a damper, and constitutes a parallel four-bar mechanism; the ultrasonic array sensor is fixed to the bottom of the vertical rod, so that the panel thereof can always be parallel to the bottom surface of the robot; in the working state, the damper is compressed to generate a reaction force, so that the sensor panel is stably compressed on the concrete surface.

6. The method of claim 1, wherein, The algorithm based on multi-modal noise separation and variational modal decomposition in step 5 firstly establishes a multi-dimensional noise fingerprint database; a variational modal decomposition-independent component analysis joint algorithm is adopted, and a kurtosis-envelope entropy joint optimization criterion is introduced to adaptively determine the number of decomposition layers; an independent component analysis algorithm based on negative entropy maximization is adopted to extract independent components, and a noise feature matching model constructed by combining a noise reference channel verification mechanism and working condition parameters is used to separate the ultrasonic echo signal from the interference noise.

7. A wall-climbing robot system for implementing the method of any one of claims 1-6, characterized by It comprises: a double-cylinder negative pressure adsorption mobile platform with self-adaptive pressure control, which is used to stably adsorb the robot on the concrete facade; a visual tight coupling real-time positioning module, which comprises a binocular camera, an inertial navigation sensor and a processing algorithm, and is used to realize real-time positioning of the robot in a GPS-free environment; a crack image acquisition and intelligent identification module, which comprises the binocular camera, an on-board computer and the deep learning network based on camouflage target recognition, and is used to acquire images and automatically identify fine cracks; The ultrasonic depth detection module comprises an ultrasonic array sensor and an elastic compression module integrated in the tail of the robot, which is used to emit and receive ultrasonic data closely to the concrete surface. The data processing and report generation module is used to run the algorithm based on multi-modal noise separation and variational modal decomposition to calculate the crack depth, and integrate the crack apparent information, depth data and robot positioning information to generate a comprehensive detection report.

8. The system of claim 7, wherein, The mobile platform comprises a double-cylinder negative pressure adsorption unit and a four-wheel drive mobile chassis; each drive wheel adopts a horizontal gear to transmit power to realize self-locking without signal input; the drive tire adopts a dense tooth tire skin made of silica gel material; a circle of flexible foam is distributed around the negative pressure suction area at the bottom of the double-cylinder negative pressure adsorption unit.

9. The system of claim 7, wherein, The system further comprises a high-voltage tethered power supply system, which comprises a ground mobile power supply, a voltage boosting module, a light power supply cable and a voltage reducing module located on the robot body, which is used to supply power to each unit of the system through high-voltage transmission and on-board voltage reduction.

10. The system of claim 7, wherein, The elastic compression module specifically comprises an upper cross bar, a lower cross bar, a vertical rod and a damper, forming a parallel four-bar mechanism; the ultrasonic array sensor is fixed to the bottom of the vertical rod, so that the panel can always be parallel to the bottom surface of the robot; the damper is used to be compressed in the working state to generate a counterforce to stabilize the compression of the sensor panel to the concrete surface.

Images