A lining knocking sound spectrum intelligent detection system and method

Abstract

The application relates to the technical field of tunnel engineering detection, and discloses a lining knocking sound spectrum intelligent detection system, which comprises a vehicle-mounted mobile platform, a multi-degree-of-freedom mechanical arm knocking system, an acoustic signal acquisition system, a visual auxiliary positioning system, a laser radar and a structured light three-dimensional reconstruction module and a central processing unit. The central processing unit is connected with the other modules in bidirectional communication and control connection through a CAN bus, a data acquisition card and a control signal line. The laser radar and the structured light three-dimensional reconstruction module transmit tunnel three-dimensional data to the central processing unit. The vehicle-mounted mobile platform uniformly drives at a planned path speed. The multi-degree-of-freedom mechanical arm is unfolded to perform grid knocking. The angle adjusting mechanism dynamically maintains the optimal knocking angle, realizes full-section non-repeated and non-dead-angle coverage of the lining, does not need manual intervention, the detection speed is far higher than manual knocking, the detection period is greatly shortened, and the labor intensity and safety risk are reduced.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering inspection technology, specifically to an intelligent detection system and method for the acoustic spectrum of lining impact. Background Technology

[0002] The quality of tunnel concrete lining is directly related to the long-term operational safety of the tunnel. Common lining defects include cracks, honeycombing, pitting, voids, construction joint cracking (crescent-shaped cracks), insufficient concrete strength, and loose surface.

[0003] Currently, the quality inspection of lining mainly relies on the manual tapping method (also known as the "echo method"), where inspectors use a hammer to tap the lining surface and judge the presence of internal defects by listening to the crisp or dull sound. However, the traditional manual tapping method has significant drawbacks:

[0004] 1. Highly subjective: The test results are highly dependent on the personal experience of the testers, resulting in poor reliability;

[0005] 2. Inefficient: It requires full-coverage hammering of the tunnel wall, which is time-consuming and labor-intensive;

[0006] 3. Difficulty in recording: After a defect is discovered, its location and description must be recorded manually, which is prone to errors and not intuitive;

[0007] 4. High safety risks: Inspection personnel need to work at heights inside the tunnel, facing traffic safety and health hazards (such as dust and noise), and there is an urgent need for a detection system and method with better overall performance. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent detection system and method for the acoustic spectrum of lining impact, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an intelligent detection system for the acoustic spectrum of lining impact, comprising a vehicle-mounted mobile platform, a multi-degree-of-freedom robotic arm impact system, an acoustic signal acquisition system, a vision-assisted positioning system, a lidar and structured light 3D reconstruction module, and a central processing unit. The central processing unit establishes bidirectional communication and control connections with the other modules through a CAN bus, a data acquisition card, and control signal lines. The lidar and structured light 3D reconstruction module transmits tunnel 3D data to the central processing unit. The central processing unit outputs control commands to the multi-degree-of-freedom robotic arm impact system, the acoustic signal acquisition system, and the vision-assisted positioning system. The modules work together to achieve full-coverage intelligent detection, ensuring precise linkage between data acquisition, analysis, and storage.

[0010] Preferably, the motion control card of the central processing unit establishes an electrical connection with the servo motors of the robotic arm joints, the hammer drive motor, and the track movement servo motor of the multi-degree-of-freedom robotic arm striking system through a servo driver. The angle sensor and grating ruler closed-loop feedback module built into the robotic arm transmit angle and position data to the central processing unit in real time through signal cables, forming a closed-loop control loop. The robotic arms exchange position and attitude data through a CAN bus, and the central processing unit coordinates the movement using a centralized scheduling algorithm to avoid collision risks. The formula for the closed-loop control error of the robotic arm joint position is: ,in Position error (unit: mm) Indicates the command position (unit: mm). The position is provided by the grating ruler (unit: mm); the formula for calculating the joint speed of the robotic arm is as follows: ,in The velocity of joint movement (unit: mm / s). , For adjacent sampling times (unit: s); the formula for determining the collision risk of the robotic arm is: ,in For robotic arms With robotic arm The minimum spatial distance (unit: mm), when < When the obstacle avoidance action is triggered, The safe distance threshold (unit: mm). , These are the three-dimensional coordinates (unit: mm) of key parts of the two robotic arms.

[0011] Preferably, the directional microphone array of the acoustic signal acquisition system is connected to the input terminal of the data acquisition card through a low-noise preamplifier, and the bandpass filter is integrated into the acquisition circuit to form a signal processing link with the data acquisition card; the synchronous pulse output terminal of the hammer driving mechanism is connected to the trigger terminal of the data acquisition card through a signal line to achieve precise synchronization between the striking action and the sound signal acquisition, and the acquired sound signal is transmitted to the central processing unit through the data acquisition card for preprocessing and feature extraction;

[0012] The acoustic signal acquisition system's four-microphone array is connected to a preamplifier via a signal aggregator. The acoustic signals acquired by each microphone are integrated by the aggregator and then input to the preamplifier for gain adjustment. After being filtered by a bandpass filter, the signals are finally converted into digital signals by the data acquisition card and transmitted to the central processing unit. The microphone array layout is fixed via cable connections to ensure directional acoustic signal acquisition. The preamplifier output signal formula is as follows: ,in This refers to the amplifier output voltage (unit: V). Microphone input voltage (unit: V). The amplifier gain is given by (in dB); the frequency response formula for the bandpass filter is... ,in For frequency The filter gain at that point, Low cutoff frequency (unit: Hz) High cutoff frequency (unit: Hz). The quality factor; the formula for calculating the directional gain of the microphone array is: ,in Directional gain (unit: dB). The axial signal sound pressure level is measured in Pa. The sound pressure level is for non-axial signals (unit: Pa).

[0013] Preferably, the high-definition industrial camera of the vision-assisted positioning system is connected to the output of the AI ​​recognition module of the central processing unit via a GPIO signal line to receive abnormal result trigger signals; the camera's image transmission interface is connected to the image acquisition card of the central processing unit to transmit defect images in real time; the lidar and structured light 3D reconstruction module communicate with the central processing unit via an Ethernet interface to transmit the tunnel's global 3D coordinates and local point cloud data, providing basic data support for defect pixel coordinate transformation; wherein, the pixel coordinate to camera coordinate transformation formula is: ,in For the defect pixel coordinates, These are coordinates in the camera coordinate system (unit: mm). , Camera focal length (unit: mm). , Here are the coordinates of the camera's principal point; the formula for transforming camera coordinates to world coordinates is: ,in The three-dimensional coordinates of the defect in the world coordinate system (unit: mm). It is a 3×3 rotation matrix. The translation vector is 3×1; the formula for calculating the point cloud registration error is:

[0014] ,in The average registration error is expressed in mm. To register the number of points, For visual positioning coordinates, These are the coordinates for the lidar.

[0015] Preferably, the central processing unit establishes a time synchronization connection with the local high-precision clock module through the GPS module, and distributes the unified timestamp to the acoustic signal acquisition system, visual auxiliary positioning system and lidar module through the control signal line to ensure accurate alignment of the spatiotemporal stamps of the acoustic signal, image data and three-dimensional coordinate data; the data storage and visualization module is connected to the industrial computer of the central processing unit through the SATA interface to store the raw data, analysis results and defect association information in real time, and the visualization module receives the defect distribution data of the central processing unit and generates a two-dimensional unfolded distribution map;

[0016] The AI ​​model calculation module of the central processing unit is connected to the acoustic signal processing module and the image processing module through an internal data bus. The 18 acoustic feature data extracted by the acoustic signal processing module and the defect image feature data output by the image processing module are synchronously transmitted to the AI ​​model calculation module. The model calculation results are fed back to the scheduling module of the central processing unit through the control bus to trigger subsequent actions such as camera shooting, data storage and report generation.

[0017] Preferably, the angle adjustment mechanism of the multi-degree-of-freedom robotic arm striking system establishes a signal connection with the central processing unit through an angle sensor. The angle sensor collects the current striking angle data in real time. The central processing unit calculates the adjustment amount based on the lining surface data transmitted by the lidar using a PID algorithm, and then controls the servo motor of the angle adjustment mechanism to dynamically maintain the optimal striking angle through a servo driver. The striking angle error formula is as follows: ,in For angular error (unit: °). The optimal striking angle is 15°-20°. The angle is fed back by the angle sensor (unit: °); the formula for calculating the normal vector of the lining surface is: ,in The equation for the lining surface is... , respectively the curved surface in , The partial derivative of the direction, and the striking angle is the angle between the striking direction and the normal vector.

[0018] Preferably, the driving control module of the vehicle-mounted mobile platform is connected to the central processing unit via a CAN bus. The central processing unit controls the driving speed and direction of the vehicle-mounted mobile platform according to the driving commands output by the path planning algorithm. Simultaneously, the mileage counting module of the vehicle-mounted mobile platform transmits real-time mileage data to the central processing unit, which correlates it with the mileage information in the three-dimensional coordinates of the defect to form a complete defect location data chain. The formula for the uniform speed driving control error of the vehicle-mounted platform is: ,in Speed ​​error (unit: km / h) Command speed (unit: km / h) The odometer feedback speed (unit: km / h); the real-time mileage calculation formula is: ,in This represents the current mileage (in meters). Initial mileage (unit: m). The driving time (in hours) is given; the defect mileage correlation formula is as follows: ,in The tunnel mileage (in meters) corresponding to the defect. The longitudinal distance between the current position of the vehicle platform and the defect detection point (unit: m).

[0019] Preferably, the lidar and the structured light 3D reconstruction module are connected via a synchronization signal line. After the lidar scans and acquires global 3D data of the tunnel, the synchronization signal triggers the structured light module to perform a high-precision local scan of the suspected defect area. The point cloud data from the structured light module and the global data from the lidar are transmitted to the central processing unit via an Ethernet interface. After being fused by a point cloud registration algorithm, a high-precision 3D model is provided to support the defect coordinate transformation. The lidar point cloud density calculation formula is as follows: ,in Point cloud density (unit: points / mm³). To determine the number of point clouds within the scanned area, The volume of the scanning area is (unit: mm³); the formula for calculating the accuracy of structured light local scanning is: ,in Scanning precision (unit: mm). For the first The deviation between the actual distance and the measured distance at each point The average deviation is given; the iterative formula for the point cloud registration rotation matrix is: ,in For the first The rotation matrix after the nth iteration For structured light point cloud points, These are the points corresponding to the lidar point cloud.

[0020] Another technical problem to be solved by the present invention is to provide an intelligent detection method for the sound spectrum of lining impact, characterized in that it is applied to the detection system and includes the following steps:

[0021] Step 1: Device initialization and path planning. The lidar scans the tunnel cross-section, and the central processing unit dynamically plans the layout of the robotic arm and the gridded detection path, completing microphone calibration, camera calibration, angle sensor calibration, and system self-test.

[0022] Step 2: The vehicle-mounted mobile platform travels at a constant speed along the planned path. The multi-degree-of-freedom robotic arm unfolds and maintains a 15°-20° striking angle through the angle adjustment mechanism. The striking hammer moves along the track with a positioning accuracy of ±1mm to perform full-coverage, non-repetitive grid-like striking.

[0023] Step 3: Synchronously acquire sound signals and three-dimensional coordinate information. The sound signals are stored after bandpass filtering and noise reduction processing. The synchronous trigger signal ensures that the sound signals are precisely synchronized with the striking action.

[0024] Step 4: Extract 18 acoustic features from the sound signal, input them into the CNN and Transformer hybrid model for intelligent classification, and output the probability of defect type, severity score and anomaly judgment result.

[0025] Step 5: If a defect is identified, the camera is triggered in real time to capture an image and convert it into absolute three-dimensional coordinates, storing the "sound signal-image-three-dimensional coordinates" associated data;

[0026] Step 6: Integrate all detection data to generate a structured detection report containing detailed defect information, acoustic spectrograms, image evidence, and a two-dimensional unfolded distribution map.

[0027] Preferably, the dataset used in the model training process includes eight common defects in tunnel lining, such as voids, peeling, cracks, honeycombing, and construction joint cracking, with ≥10,000 samples for each defect type, covering different lining thicknesses, concrete strengths, and environmental noise scenarios. Data augmentation employs various noise types, including Gaussian white noise, pink noise, and mechanical noise, with the signal-to-noise ratio controlled within the range of 10-30 dB. Simultaneously, it expands sample diversity through methods such as time-domain stretching, spectrum shifting, and cepstral enhancement. The model training uses the Adam optimizer with dynamically adjusted learning rate and iterative training for ≥100 rounds. An early stopping strategy is used to prevent overfitting. Finally, accurate identification and quantitative scoring of various defects are achieved on an independent test set, providing accurate basis for tunnel lining maintenance and management.

[0028] This invention provides an intelligent detection system and method for the acoustic spectrum of lining impact. It has the following beneficial effects:

[0029] 1. This invention uses a vehicle-mounted mobile platform to travel at a constant speed along a planned path. After the multi-degree-of-freedom robotic arm is deployed, it performs grid-like tapping. The angle adjustment mechanism dynamically maintains the optimal tapping angle, achieving full-section coverage of the lining without repetition or blind spots. No manual intervention is required, and the detection speed is far greater than manual tapping, significantly shortening the detection cycle and reducing labor intensity and safety risks.

[0030] 2. The acoustic signal acquisition system of this invention synchronously acquires the impact sound signal, and after filtering and noise reduction, extracts 18 acoustic features. These features are then input into a hybrid CNN and Transformer model for intelligent classification, accurately identifying 8 types of defects such as voids and peeling, and outputting the probability and severity score of the defect type. This avoids the subjectivity of manual judgment, significantly improving the detection accuracy and consistency, and providing an objective quantitative basis for maintenance decisions.

[0031] 3. This invention provides high-precision three-dimensional data of the tunnel globally and locally through a lidar and structured light three-dimensional reconstruction module. The visual-assisted positioning system triggers the capture of defect images and converts them into absolute three-dimensional coordinates. This forms a "sound signal-image-three-dimensional coordinate" associated data, which fully records the acoustic characteristics, visual morphology and spatial location of the defects, providing comprehensive information for defect assessment.

[0032] 4. The detection process of this invention is fully automated from path planning, intelligent tapping, data collection to AI analysis and report generation, and strictly follows unified standards. Finally, it generates a structured report containing detailed defect information, acoustic spectrograms, image evidence and two-dimensional unfolded distribution map. The data is traceable and the results are quantifiable, promoting the standardization and normalization of tunnel inspection. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating the overall system architecture logic of the present invention.

[0034] Figure 2 This is a flowchart of the closed-loop control and obstacle avoidance logic of the robotic arm in this invention;

[0035] Figure 3 This is a flowchart of the acoustic signal acquisition logic of the present invention;

[0036] Figure 4 This is a flowchart of the visual-assisted positioning logic of the present invention;

[0037] Figure 5 This is a flowchart of the core logic of the central processing unit of the present invention;

[0038] Figure 6 This is a flowchart illustrating the logic for adjusting the striking angle of the robotic arm in this invention.

[0039] Figure 7 This is a flowchart of the control logic of the vehicle-mounted mobile platform of the present invention;

[0040] Figure 8 This is a flowchart illustrating the collaborative logic of lidar and structured light in this invention.

[0041] Figure 9 This is a flowchart illustrating the steps of the detection method of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention.

[0044] Example 1

[0045] A preferred embodiment of the intelligent detection system and method for the sound spectrum of lining impact provided by the present invention is as follows: Figure 1-9 As shown: A lining impact acoustic spectrum intelligent detection system includes a vehicle-mounted mobile platform, a multi-degree-of-freedom robotic arm impact system, an acoustic signal acquisition system, a vision-assisted positioning system, and a central processing unit. The central processing unit establishes bidirectional communication and control connections with the other modules through a CAN bus, a data acquisition card, and control signal lines. The lidar and structured light 3D reconstruction module transmit tunnel 3D data to the central processing unit. The central processing unit outputs control commands to the multi-degree-of-freedom robotic arm impact system, the acoustic signal acquisition system, and the vision-assisted positioning system. The modules work together to achieve full-coverage intelligent detection, ensuring precise linkage between data acquisition, analysis, and storage.

[0046] The vehicle-mounted mobile platform carries various systems to move within the tunnel. The multi-degree-of-freedom robotic arm tapping system dynamically maintains the optimal tapping angle of 15°-20° through an angle adjustment mechanism, achieving non-repetitive full-coverage tapping of the tunnel lining. The acoustic signal acquisition system integrates a high-performance directional microphone with noise reduction and synchronous acquisition structure. The visual-assisted positioning system links with the AI ​​recognition module to trigger shooting and complete coordinate transformation. The central processing unit intelligently analyzes the data through a specific AI model, solving the problems of strong subjectivity, low efficiency, difficulty in recording, and high safety risks associated with traditional manual tapping methods.

[0047] The directional microphone array of the acoustic signal acquisition system is connected to the input terminal of the data acquisition card through a low-noise preamplifier. The bandpass filter is integrated into the acquisition circuit, forming a signal processing link with the data acquisition card. The synchronous pulse output terminal of the hammer drive mechanism is connected to the trigger terminal of the data acquisition card through a signal line to achieve precise synchronization between the hammering action and the sound signal acquisition. The acquired sound signal is transmitted to the central processing unit through the data acquisition card for preprocessing and feature extraction.

[0048] The acoustic signal acquisition system's 4-microphone array is connected to the preamplifier via a signal aggregator. The acoustic signals acquired by each microphone are integrated by the aggregator and then input into the preamplifier for gain adjustment. After being filtered by a bandpass filter, they are finally converted into digital signals by the data acquisition card and transmitted to the central processing unit. The layout of the microphone array is fixed by cable connections to ensure the directional acquisition effect of the acoustic signals.

[0049] Furthermore, in this embodiment, the directional microphone of the acoustic signal acquisition system has a frequency response range of 20Hz-20kHz, a cardioid directivity, and a sensitivity of ≤-38dBV / Pa. The microphone is equipped with a physical soundproof cover and adopts a 4-microphone array technology combined with an adaptive active noise reduction algorithm to effectively suppress ambient noise. The acquisition circuit integrates a low-noise preamplifier and an 80Hz-8kHz bandpass filter. The preamplifier gain is adjustable from 20-60dB, and the filter stopband attenuation is ≥40dB / decade. The sound signal acquisition and the striking action are precisely synchronized through the synchronous trigger signal of the central processing unit. When the hammer driving mechanism moves, it outputs a synchronization pulse to trigger the data acquisition card to acquire the sound signal at a sampling rate of 48kHz, with a time synchronization error of ≤10μs.

[0050] The high-definition industrial camera of the vision-assisted positioning system is connected to the output of the AI ​​recognition module of the central processing unit via GPIO signal lines to receive abnormal result trigger signals; the camera's image transmission interface is connected to the image acquisition card of the central processing unit to transmit defect images in real time; the lidar and structured light 3D reconstruction module communicate with the central processing unit via Ethernet interface to transmit the tunnel's global 3D coordinates and local point cloud data, providing basic data support for defect pixel coordinate transformation.

[0051] Furthermore, in this embodiment, the high-definition industrial camera of the vision-assisted positioning system is triggered to capture images in real time by the abnormal results output by the AI ​​recognition module. It also supports backup shooting at preset intervals. The image and acoustic data are precisely aligned with the local high-precision clock through GPS timestamps, with a time deviation of ≤50μs. The system integrates a lidar and a structured light 3D reconstruction module. The lidar acquires the global 3D coordinates of the tunnel lining, and the structured light scanning constructs a local high-precision point cloud. The coordinates of defect pixels in the image are converted into absolute 3D coordinates of mileage + circumferential angle + radial depth in the tunnel through camera intrinsic parameter calibration, extrinsic parameter calibration, and point cloud registration algorithms. The camera resolution is ≥5 million pixels and the frame rate is ≥30fps to ensure clear capture of defect images.

[0052] The angle adjustment mechanism (universal joint) of the multi-degree-of-freedom robotic arm striking system establishes a signal connection with the central processing unit through an angle sensor. The angle sensor collects the current striking angle data in real time. The central processing unit calculates the adjustment amount through a PID algorithm based on the lining surface data transmitted by the lidar, and then controls the servo motor of the angle adjustment mechanism through a servo driver to dynamically maintain the optimal striking angle.

[0053] Furthermore, in this embodiment, the multi-degree-of-freedom robotic arm striking system includes multiple independently controllable robotic arms. The central processing unit has a built-in full-coverage path planning algorithm. It obtains the tunnel cross-sectional dimensions (radius, central angle) through LiDAR scanning, dynamically plans the number and layout of the robotic arms, and divides the tunnel into non-overlapping detection areas according to the circumferential and longitudinal directions. Each robotic arm operates independently in its zone and avoids motion coupling through an anti-collision algorithm. The longitudinal track at the end of the robotic arm moves in coordination with the striking hammer to form a gridded striking path, ensuring that there are no blind spots and no repetitions in the detection. Each robotic arm is equipped with a 2-meter-long track at its end. The striking hammer is driven by a servo motor to move along the track. It adopts closed-loop feedback control with a grating ruler, and the movement positioning accuracy reaches ±1mm. The track and the robotic arm body are connected by a hinge joint to achieve multi-dimensional posture adjustment.

[0054] Furthermore, in this embodiment, the optimal striking angle of 15°-20° is determined based on the theory of sound wave reflection and a large amount of experimental data. Within this angle range, the reflection path of sound waves inside the lining is most conducive to the propagation of defect signals. Compared with vertical striking, the sound signal-to-noise ratio is improved by more than 30%, and the defect detection rate is improved by 25%. The angle adjustment mechanism at the end of the robotic arm is a universal joint combined with a rotary joint driven by a servo motor. The built-in angle sensor collects the current angle in real time. The central processing unit dynamically adjusts the servo motor through a PID algorithm based on the lining surface data obtained by the lidar, ensuring that the angle is always kept within the set range during the striking process, with an angle control accuracy of ≤0.5°.

[0055] The central processing unit's artificial intelligence analysis algorithm adopts a hybrid CNN and Transformer model architecture. The input layer is a spectrum (256×256 dimensions) obtained by FFT transformation of a 48kHz sampling rate acoustic signal. The hidden layers include 8 convolutional layers (3×3 kernel size), 4 Transformer encoders (8 attention heads), and 2 fully connected layers. The output layer is a multi-class probability distribution. The extracted acoustic features include 18 key features such as Mel frequency cepstral coefficients (24 dimensions), spectral centroid, spectral bandwidth, peak factor, kurtosis, and impulse factor.

[0056] The model is trained on a database of over 100,000 labeled knocking sounds of different defect types (cavities, peeling, cracks, honeycombs, etc.). Data augmentation techniques such as noise addition, temporal stretching, and spectral enhancement are used to address on-site noise. On the test set, the defect classification accuracy is ≥95%, recall is ≥93%, and F1 score is ≥94%. The output results include the probability distribution of defect types and a quantitative score of severity (0-10 points).

[0057] The collaborative working logic among multiple robotic arms adopts a combination of distributed control and centralized scheduling. The central processing unit acts as the scheduling center, allocating independent detection areas to each robotic arm. The robotic arms have built-in local controllers to perform motion control, and at the same time, they transmit position and attitude data in real time via the CAN bus. When there is a risk of overlapping of the robotic arm movement trajectories, the centralized scheduling algorithm avoids collisions by replanning the obstacle avoidance path. The number of robotic arms can be dynamically increased or decreased according to the tunnel cross-section size to adapt to the detection needs of tunnels with different diameters. The swing angle range of the robotic arms is -90° to 90°, and the vertical movement covers the tunnel arch to the invert arch, ensuring full cross-section coverage.

[0058] Furthermore, in this embodiment, the hammer is driven by a servo motor, and the striking force can be adjusted through a closed-loop current control, with an adjustment range of 5-50N. The striking frequency is adjustable from 1-5Hz to ensure effective excitation of defect signals from linings of different strengths. The robotic arm body is made of lightweight carbon fiber material and equipped with a multi-joint torque sensor to monitor motion resistance in real time and avoid rigid collisions with the lining surface. The system also includes a data storage and visualization module. The data storage module uses an industrial-grade SSD with a storage capacity of ≥2TB, supporting long-term retention of original acoustic signals, defect images, three-dimensional coordinates, and analysis results. The visualization module generates a two-dimensional unfolded distribution map of tunnel lining defects, intuitively displaying the distribution, density, type, and severity score of the defects.

[0059] Example 2

[0060] Please see Figures 1-9 Furthermore, based on Example 1, another technical problem to be solved by the present invention is to provide an intelligent detection method for the sound spectrum of lining impact, applied to a detection system, comprising the following steps:

[0061] Step 1: Device initialization and path planning. The lidar scans the tunnel cross-section, and the central processing unit dynamically plans the layout of the robotic arm and the gridded detection path, completing microphone calibration, camera calibration, angle sensor calibration, and system self-test.

[0062] Step 2: The vehicle-mounted mobile platform travels at a constant speed along the planned path. The multi-degree-of-freedom robotic arm unfolds and maintains a 15°-20° striking angle through the angle adjustment mechanism. The striking hammer moves along the track with a positioning accuracy of ±1mm to perform full-coverage, non-repetitive grid-like striking.

[0063] Step 3: Synchronously acquire sound signals and three-dimensional coordinate information. The sound signals are stored after bandpass filtering and noise reduction processing. The synchronous trigger signal ensures that the sound signals are precisely synchronized with the striking action.

[0064] Step 4: Extract 18 acoustic features from the sound signal, input them into the CNN and Transformer hybrid model for intelligent classification, and output the probability of defect type, severity score and anomaly judgment result.

[0065] Step 5: If a defect is identified, the camera is triggered in real time to capture an image and convert it into absolute three-dimensional coordinates, storing the "sound signal-image-three-dimensional coordinates" associated data;

[0066] Step 6: Integrate all detection data to generate a structured detection report containing detailed defect information, acoustic spectrograms, image evidence, and a two-dimensional unfolded distribution map.

[0067] In this embodiment, the dataset used in the model training process includes eight common defects in tunnel lining, such as voids, peeling, cracks, honeycombing, and construction joint cracking. Each defect category has ≥10,000 samples, covering different lining thicknesses, concrete strengths, and environmental noise scenarios. Data augmentation employs various noise types, including Gaussian white noise, pink noise, and mechanical noise, with the signal-to-noise ratio controlled within the range of 10-30 dB. Simultaneously, it expands sample diversity through temporal stretching (0.8-1.2 times), spectral shifting, and cepstral enhancement. The model training uses the Adam optimizer with dynamically adjusted learning rate and iterative training for ≥100 rounds. An early stopping strategy prevents overfitting. Finally, it achieves accurate identification and quantitative scoring of various defects on an independent test set, providing accurate basis for tunnel lining maintenance and management.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart detection system for the sound spectrum of lining impact, characterized in that, The system includes a vehicle-mounted mobile platform, a multi-degree-of-freedom robotic arm striking system, an acoustic signal acquisition system, a vision-assisted positioning system, a lidar and structured light 3D reconstruction module, and a central processing unit. The central processing unit establishes bidirectional communication and control connections with the other modules through a CAN bus, a data acquisition card, and control signal lines. The lidar and structured light 3D reconstruction module transmits tunnel 3D data to the central processing unit, and the central processing unit outputs control commands to the multi-degree-of-freedom robotic arm striking system, the acoustic signal acquisition system, and the vision-assisted positioning system.

2. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The motion control card of the central processing unit establishes an electrical connection with the servo motors of the robotic arm joints, the hammer drive motor, and the track movement servo motor of the multi-degree-of-freedom robotic arm striking system through a servo driver. The angle sensor and grating ruler closed-loop feedback module built into the robotic arm transmit angle and position data to the central processing unit in real time through signal cables, forming a closed-loop control loop. The robotic arms interact with each other through a CAN bus to realize position and attitude data exchange. The central processing unit coordinates the movement through a centralized scheduling algorithm to avoid collision risks.

3. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The directional microphone array of the acoustic signal acquisition system is connected to the input terminal of the data acquisition card through a low-noise preamplifier. The bandpass filter is integrated into the acquisition circuit, forming a signal processing link with the data acquisition card. The synchronous pulse output terminal of the hammer drive mechanism is connected to the trigger terminal of the data acquisition card through a signal line to achieve precise synchronization between the hammering action and the sound signal acquisition. The acquired sound signal is transmitted to the central processing unit through the data acquisition card for preprocessing and feature extraction. The acoustic signal acquisition system's 4-microphone array is connected to a preamplifier via a signal aggregator. The acoustic signals acquired by each microphone are integrated by the aggregator and then input into the preamplifier for gain adjustment. After being filtered by a bandpass filter, they are finally converted into digital signals by the data acquisition card and transmitted to the central processing unit. The layout of the microphone array is fixed by cable connections.

4. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The high-definition industrial camera of the vision-assisted positioning system is connected to the output of the AI ​​recognition module of the central processing unit via a GPIO signal line to receive abnormal result trigger signals; the camera's image transmission interface is connected to the image acquisition card of the central processing unit to transmit defect images in real time; the lidar and structured light 3D reconstruction module communicate with the central processing unit via an Ethernet interface to transmit the tunnel's global 3D coordinates and local point cloud data, providing basic data support for defect pixel coordinate transformation.

5. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The central processing unit establishes a time synchronization connection with the local high-precision clock module through the GPS module, and distributes the unified timestamp to the acoustic signal acquisition system, visual auxiliary positioning system and lidar module through the control signal line to ensure accurate alignment of the spatiotemporal stamps of the acoustic signal, image data and three-dimensional coordinate data; the data storage and visualization module is connected to the industrial computer of the central processing unit through the SATA interface to store raw data, analysis results and defect correlation information in real time, and the visualization module receives the defect distribution data of the central processing unit and generates a two-dimensional unfolded distribution map; The AI ​​model calculation module of the central processing unit is connected to the acoustic signal processing module and the image processing module through an internal data bus. The 18 acoustic feature data extracted by the acoustic signal processing module and the defect image feature data output by the image processing module are synchronously transmitted to the AI ​​model calculation module. The model calculation results are fed back to the scheduling module of the central processing unit through the control bus to trigger subsequent actions such as camera shooting, data storage and report generation.

6. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The angle adjustment mechanism of the multi-degree-of-freedom robotic arm striking system establishes a signal connection with the central processing unit through an angle sensor. The angle sensor collects the current striking angle data in real time. The central processing unit calculates the adjustment amount through a PID algorithm based on the lining surface data transmitted by the lidar, and then controls the servo motor of the angle adjustment mechanism through a servo driver to dynamically maintain the optimal striking angle.

7. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The driving control module of the vehicle-mounted mobile platform is connected to the central processing unit via a CAN bus. The central processing unit controls the driving speed and direction of the vehicle-mounted mobile platform according to the driving instructions output by the path planning algorithm. At the same time, the mileage counting module of the vehicle-mounted mobile platform transmits real-time mileage data to the central processing unit and associates it with the mileage information in the three-dimensional coordinates of the defect to form a complete defect location data chain.

8. The intelligent detection system for the sound spectrum of lining impact according to claim 1, characterized in that, The lidar and the structured light 3D reconstruction module are connected via a synchronization signal line. After the lidar scans and acquires the global 3D data of the tunnel, the structured light module is triggered by the synchronization signal to perform a local high-precision scan of the suspected defect area. The point cloud data of the structured light module and the global data of the lidar are transmitted to the central processing unit through the Ethernet interface. After being fused by the point cloud registration algorithm, a high-precision 3D model is provided to support the defect coordinate transformation.

9. A method for intelligent detection of the sound spectrum of lining impact, characterized in that, The detection system according to any one of claims 1-8 includes the following steps: Step 1: Device initialization and path planning. The lidar scans the tunnel cross-section, and the central processing unit dynamically plans the layout of the robotic arm and the gridded detection path, completing microphone calibration, camera calibration, angle sensor calibration, and system self-test. Step 2: The vehicle-mounted mobile platform travels at a constant speed along the planned path. The multi-degree-of-freedom robotic arm unfolds and maintains a 15°-20° striking angle through the angle adjustment mechanism. The striking hammer moves along the track with a positioning accuracy of ±1mm to perform full-coverage, non-repetitive grid-like striking. Step 3: Synchronously acquire sound signals and three-dimensional coordinate information. The sound signals are stored after bandpass filtering and noise reduction processing. The synchronous trigger signal ensures that the sound signals and the striking action are precisely synchronized. Step 4: Extract 18 acoustic features from the sound signal, input them into the CNN and Transformer hybrid model for intelligent classification, and output the probability of defect type, severity score and anomaly judgment result. Step 5: If a defect is identified, the camera is triggered in real time to capture an image and convert it into absolute three-dimensional coordinates, storing the "sound signal-image-three-dimensional coordinates" associated data; Step 6: Integrate all detection data to generate a structured detection report containing detailed defect information, acoustic spectrograms, image evidence, and a two-dimensional unfolded distribution map.

10. The intelligent detection method for the sound spectrum of lining impact according to claim 9, characterized in that, The dataset used in the model training process includes eight common defects in tunnel lining: voids, peeling, cracks, honeycombing, and construction joint cracking. Data augmentation is achieved by adding various noise types, such as Gaussian white noise, pink noise, and mechanical noise, with the signal-to-noise ratio controlled within the range of 10-30 dB. At the same time, the diversity of samples is expanded through time-domain stretching, spectral shifting, and cepstral enhancement.

Images