Automatic detection device for tunnel lining cavity

Abstract

The invention relates to an automatic detection device for tunnel lining cavities, and belongs to the technical field of tunnel detection. A sectional type moving unit (with the curvature radius of 3-15 m) is adopted to cooperate with a multi-knocking unit (tungsten steel hammer head impact of 3 m / s), and sound wave detection (96 kHz sampling + 48-order FIR filtering) and a deep learning model (ResNet-1D + random forest) are integrated. Through a dynamic positioning system (IMU / encoder + / -1cm), an environment compensation module (temperature drift compensation + / -0.1 mV / DEG C) and an online learning mechanism (200 groups of data increment update), hollowing detection precision + / -2mm is realized, and error lt is marked; and self-adaptive detection (-20 DEG C to 60 DEG C) in a complex tunnel environment is supported. The BIM model marks defects in real time, the detection efficiency reaches 50 m < 2 > / min, and the detection rate of small cavities (phi 5 cm) is gt; and the time consumed by model reasoning is 22 ms.

Description

Technical Field

[0001] The invention belongs to the technical field of tunnel detection, and relates to an automatic detection device for tunnel lining cavities. Background Art

[0002] As a core component of modern transportation infrastructure, the structural safety of tunnels is directly related to the reliability of transportation systems. In the long-term operation process, tunnel linings are affected by geological subsidence, water seepage erosion, construction defects and other factors, and are prone to hidden defects such as internal voids. Such voids will significantly weaken the bearing capacity of the lining, causing concrete spalling and even landslides. Traditional detection methods mainly rely on manual hammering, where technicians knock on the lining surface and judge the sound difference based on experience, but this method has three major defects:

[0003] (1) The detection efficiency is low, and manual tapping is time-consuming and labor-intensive;

[0004] (2) It is highly subjective, and the test results are significantly affected by the hearing sensitivity and fatigue status of the personnel;

[0005] (3) Dangerous area coverage is insufficient, and manual inspection is difficult to carry out in high-risk locations such as tunnel vaults.

[0006] In recent years, although non-destructive testing technologies (such as geological radar and ultrasonic testing) have been used in tunnel engineering, they still have obvious limitations. Geological radar is susceptible to electromagnetic interference from steel mesh and has insufficient resolution for shallow tiny cavities (<5cm); ultrasonic testing requires coupling agents and has strict requirements on surface flatness, making it difficult to adapt to the complex curved surface environment of tunnel linings. In addition, existing equipment mostly relies on manual operation, and the detection speed and degree of automation cannot meet the needs of rapid inspection of long-distance tunnels.

[0007] With the development of artificial intelligence technology, image recognition methods based on deep learning have begun to be applied to tunnel surface disease detection, but there are still technical bottlenecks in the analysis of acoustic signals of internal cavities. Specifically,

[0008] (1) It is difficult to extract the characteristics of acoustic wave signals. The time-frequency characteristics of the knock echo are affected by multiple factors such as material density, ambient temperature and humidity;

[0009] (2) Annotated data is scarce, and the cost of building a hollow voiceprint database with positioning annotations in actual projects is high;

[0010] (3) The model has weak generalization ability, and the differences in acoustic responses of different tunnel structures lead to a decrease in the adaptability of the algorithm.

[0011] Therefore, it is urgent to develop a detection system that integrates automated mechanical devices and intelligent algorithms to achieve: (1) high-precision positioning of knocks and synchronous signal acquisition; (2) enhanced extraction of voiceprint features in a multi-physical field environment; (3) deep learning model optimization under small sample conditions. Through mechatronic design and algorithm innovation, we can break through the technical barriers of traditional methods in detection efficiency, accuracy and adaptability, and provide a reliable solution for tunnel structure health monitoring. Summary of the invention

[0012] In view of this, an object of the present invention is to provide an automatic detection device for tunnel lining voids, which can timely and accurately detect tunnel lining voids on the basis of controlling costs.

[0013] In order to achieve the above object, the present invention provides the following technical solutions:

[0014] An automatic detection device for tunnel lining cavities, comprising two sets of section slide rails arranged in parallel on both sides of the tunnel floor;

[0015] A mobile unit is slidably mounted on the road section slide rail, wherein the mobile unit is an annular structure and maintains a preset distance from the tunnel lining surface;

[0016] A striking unit, comprising an elastic hammer, an energy storage spring and a release device, wherein the striking unit is distributed along the circumference of the moving unit and is used to vertically strike the lining surface;

[0017] A collection unit, including a signal amplifier and a sound sensor, the collection unit is arranged beside the knocking unit and in contact with the lining surface, and is used to collect the knocking echo and send a signal through a wireless transmission module;

[0018] An analysis unit, configured to receive the signal from the acquisition unit and compare and analyze the echo features based on a deep learning algorithm to identify the location of the cavity;

[0019] The marking unit comprises a spray gun device, and the marking unit is arranged corresponding to the knocking unit and is used to mark the cavity position according to the instruction of the analysis unit.

[0020] Furthermore, the mobile unit is composed of a multi-section segmented structure connected by loops, each section is 0.3 to 1 meter long, and the overall shape is adapted to the cross-sectional profile of the tunnel.

[0021] Furthermore, the mobile unit is equipped with an automatic sliding positioning device, including a distance sensor and a timing module, which is used to record the position coordinates of the mobile unit in real time and generate cross-section positioning data.

[0022] Furthermore, the striking unit also includes a force rope, the energy storage spring is pre-tightened by the force rope and then triggered by an electromagnetic release device, and the striking end of the elastic hammer is provided with a pressure feedback sensor.

[0023] Furthermore, the signal amplifier is a multi-band filter amplifier, and the sound sensor adopts a contact piezoelectric ceramic sensor array with an array spacing of 5 to 10 cm.

[0024] Further, the deep learning algorithm includes a convolutional neural network model, and the analysis unit is configured with:

[0025] A feature extraction module, used to extract the time-frequency domain feature vector of the echo;

[0026] The comparison and analysis module matches the feature vector with the preset lining health status voiceprint database for similarity;

[0027] The decision module determines the probability of existence and spatial distribution of voids based on the matching results.

[0028] Furthermore, the spray gun device is a pneumatic adjustable nozzle structure, the direction of the nozzle is controlled by a servo motor, and the marking color is switched by a multi-color ink tank.

[0029] Furthermore, each movable unit is provided with 3 to 8 knocking units, the interval between adjacent knocking units is 20 to 50 cm, and the knocking force of each knocking unit can be adjusted independently.

[0030] Furthermore, it also includes an environmental compensation module, which includes a temperature and humidity sensor and a noise detector, and is used to correct the environmental interference of the collected signal.

[0031] Furthermore, the voiceprint database is dynamically updated through transfer learning, and the updated data includes fused annotation data of on-site detection data and geological radar detection results.

[0032] The beneficial effects of the present invention are:

[0033] (1) Through the automatic sliding of the mobile unit along the preset slide rail (speed 0.2-1m / s) and the parallel operation of multiple knocking units (5 units per section), the full-section inspection can be completed in a single trip, which is more than 10 times more efficient than traditional manual inspection. It is especially suitable for continuous inspection of long tunnels with a length of more than 5km. The ring-shaped mobile unit structure can adapt to the shape of the tunnel section (curvature radius 3-15m), and cooperate with the hydraulic adjustment mechanism to ensure that the knocking unit is at a constant distance of 5cm in complex locations such as the arch and side wall, and the detection blind area is reduced to <2%.

[0034] (2) The voiceprint feature fusion analysis technology is used: the standardized percussion with an elastic hammer (energy 50±5J) is combined with bandpass filtering (100Hz-5kHz) and signal amplification (gain 60dB) to effectively suppress environmental noise interference and increase the signal-to-noise ratio to more than 35dB. The deep learning model (1D-CNN+GRU hybrid network) jointly analyzes the time-frequency features, making the cavity recognition accuracy reach 98.7% (experimental data), which is 13.5 percentage points higher than the traditional spectrum analysis method (85.2%), and the minimum detectable cavity diameter is reduced from 10cm to 3cm.

[0035] (3) The environmental compensation module (4G / 5G transmission delay <50ms) realizes unmanned detection, preventing personnel from entering high-risk areas (such as landslide risk sections), and reducing the risk of accidents by more than 90%. Through LiDAR real-time modeling (accuracy ±1cm), the concave and convex areas of the lining surface are automatically identified and the knocking force is adjusted (adjustable from 20 to 100N). A lightweight knocking strategy is adopted in the concrete spalling area to prevent secondary structural damage.

[0036] (4) The time-space synchronous positioning system (GNSS + encoder) can achieve centimeter-level positioning of the defect position (error < 2cm), and automatically generate a cavity distribution heat map in combination with the BIM model, which improves the positioning efficiency by 20 times compared with manual recording. Fluorescent paint (visible at night) and RFID tags (ID-related detection data) are collaboratively marked to provide accurate coordinates and damage degree data for subsequent repairs (cavity depth prediction error < 5%).

[0037] (5) The modular design reduces the assembly and disassembly time of the device to less than 4 hours (traditional equipment requires 2 days), and reduces the transportation volume by 60%, making it suitable for confined spaces such as mountain tunnels. The transfer learning mechanism (pre-trained model + field data fine-tuning) reduces the number of training samples for the new tunnel model from 1,000 to 200, reducing the data annotation cost by 80%.

[0038] Effect realization mechanism related explanation

[0039] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0041] Figure 1 This is a logic block diagram of the first embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0043] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0044] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0045] See also Figure 1, is an automatic detection device for tunnel lining voids, including a section slide rail, a mobile unit, a knocking unit, a collection unit and an analysis unit; there are two groups of section slide rails, which are fixed on the left and right sides of the tunnel floor and are mainly used for the movement of the mobile unit; the mobile unit is fixedly installed on the section slide rail, and the mobile unit is 0.5 sections, which are knotted into the shape of the tunnel section with ring buckles, so that it can be close to the tunnel lining at 5cm; the mobile unit is equipped with an automatic sliding distance time recording device; the automatic sliding distance time recording device mainly records the time position of the mobile unit moving, and is used to autonomously locate the knocking section position It is convenient for comparison and recording; 5 knocking units are installed on each mobile unit. The collection unit is mainly an elastic hammer, which knocks the lining vertically after being released; a signal amplifier collection unit is installed next to each knocking unit. The collection unit is close to the lining before knocking, used to collect the echo after knocking, and transmit it to the collection device through the 4G transmission module; the processing unit is used to compare and analyze the collected echoes and select the sound of the hollow lining; the marking unit is mainly used to mark the fixed position of the hollow lining found by the processing unit. The marking unit is a paint spray gun, and each knocking unit is equipped with a marking unit. This device can detect the hollow lining of the tunnel in a timely, accurate and fast manner on the basis of controlling costs.

[0046] Example 1: Basic automated testing process

[0047] High-precision aluminum alloy slide rails (parallelism error <1mm) are installed on both sides of the tunnel floor, and 0.5m long mobile unit modules are connected by quick-release buckles to form a ring frame that matches the tunnel section (the curvature radius has an adaptive adjustment range of 3 to 15m);

[0048] Calibrate the position of the striking unit so that the elastic hammer head is 5 cm away from the lining surface (real-time feedback from the laser rangefinder, adjustment error ±0.2 cm).

[0049] Workflow:

[0050] The mobile unit slides at a constant speed of 0.5m / s, and every 10cm of movement triggers the positioning system (encoder + IMU combined navigation) to record the three-dimensional coordinates (X, Y, Z accuracy ±1cm);

[0051] When reaching the detection point, the solenoid valve releases the energy storage spring (energy storage 50J), driving the tungsten steel hammer (diameter 30mm) to vertically strike the lining at a speed of 3m / s, with a contact time of 5ms;

[0052] The piezoelectric sensor array (8 sensors distributed in a ring, sampling rate 96kHz) is started synchronously to collect acoustic signals, which are then transmitted to the analysis terminal through the LoRa wireless module after noise reduction by a 48th-order FIR filter (passband 100Hz-8kHz).

[0053] The analysis unit calls the pre-trained ResNet-1D model to extract the Mel spectrum (frame length 25ms, frame shift 10ms) and MFCC features (13-dimensional coefficients) of the sound wave signal;

[0054] Compare the healthy sample database (including 5000 sets of voiceprint data), trigger the cavity alarm when the cosine similarity is <0.8, and calculate the cavity equivalent diameter (formula: D = 3.2 × (energy decay rate) 0.45 );

[0055] After locating the cavity, the adjacent rotary spray gun (controlled by a stepper motor, with an angle resolution of 0.1°) sprays the UV fluorescent paint (spraying pressure 0.3 MPa, marking diameter 15 cm);

[0056] The defect location is simultaneously marked in the BIM model (coordinate error <2cm) and a test report is generated (including acoustic wave waveform and confidence score).

[0057] Embodiment 2: Environmental Adaptive Detection Mode

[0058] Workflow:

[0059] The 3D line laser scanner (scanning frequency 100 Hz) at the front end of the mobile unit acquires the lining surface morphology in real time and constructs a triangular mesh model (accuracy 0.5 mm);

[0060] The temperature and humidity sensor (measurement range -20℃~60℃, accuracy ±0.5℃) and microphone array (4 channels, A-weighted noise monitoring) synchronously collect environmental parameters;

[0061] When the surface undulation is detected to be >2cm, the six-degree-of-freedom robotic arm automatically adjusts the posture of the striking unit to ensure that the deviation between the hammer head axis and the lining normal direction is less than 3°;

[0062] In high temperature areas (>40°C), reduce the tapping frequency to 1 time / second and start the thermoelectric cooler to compensate for the temperature drift of the sensor (compensation accuracy ±0.1mV / °C);

[0063] According to the intensity of environmental noise (Leq>75dB), switch to the impact echo method: the preload of the energy storage spring is increased to 80N, and the first three echo peaks are collected after the knock (time window 0-20ms);

[0064] Establish the transfer function H(f)=S out(f) / S in(f) , the structural resonance frequency offset Δf is extracted by cepstrum analysis, and when Δf>15Hz, it is judged as a void;

[0065] The acoustic data and laser point cloud data are aligned in the space-time coordinate system (time synchronization error <1ms), and a fused feature vector is generated and input into the random forest model, which improves the detection rate of small-sized cavities (diameter <5cm) to 95%.

[0066] Example 3: Deep learning model online optimization system

[0067] Workflow:

[0068] Deploy a pre-trained WaveNet model (input layer: 500ms sound wave segments, output layer: void probability values ​​0-1), which has been trained on 30,000 sets of data from 10 tunnel projects;

[0069] During real-time detection, the review mechanism is automatically triggered for suspected samples with a confidence level of 0.7 to 0.9:

[0070] Start tapping the same point three times repeatedly to collect the consistency of the data envelope (correlation coefficient>0.9 is considered valid);

[0071] Valid samples are added to the training set and associated with the ground penetrating radar verification data of the point (dielectric constant abnormal area annotation);

[0072] After accumulating 200 sets of new data, start model fine-tuning: freeze the first 5 convolution kernels and only update the parameters of the fully connected layer (learning rate 1×e -5 , Adam optimizer);

[0073] Using knowledge distillation technology, the old model output is used as a soft label (temperature coefficient T = 2) and jointly trained with the newly labeled data to prevent catastrophic forgetting;

[0074] The optimized model was pruned (neurons with contribution < 0.1% were removed) and converted to the TensorRT engine, reducing the inference time from 85ms to 22ms.

[0075] After each kilometer of tunnel inspection is completed, a model performance report (including precision-recall curve and feature importance ranking) is automatically generated.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. An automatic detection device for tunnel lining voids, characterized in that: Two sets of section slide rails are arranged in parallel on both sides of the tunnel floor; A mobile unit is slidably mounted on the road section slide rail, wherein the mobile unit is an annular structure and maintains a preset distance from the tunnel lining surface; A striking unit, comprising an elastic hammer, an energy storage spring and a release device, wherein the striking unit is distributed along the circumference of the moving unit and is used to vertically strike the lining surface; A collection unit, including a signal amplifier and a sound sensor, the collection unit is arranged beside the knocking unit and in contact with the lining surface, and is used to collect the knocking echo and send a signal through a wireless transmission module; An analysis unit, configured to receive the signal from the acquisition unit and compare and analyze the echo features based on a deep learning algorithm to identify the location of the cavity; The marking unit comprises a spray gun device, and the marking unit is arranged corresponding to the knocking unit and is used to mark the cavity position according to the instruction of the analysis unit.

2. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: The mobile unit is composed of multiple segmented structures connected by ring buckles, each segment is 0.3 to 1 meter long, and the overall shape is adapted to the cross-sectional profile of the tunnel.

3. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: The mobile unit is equipped with an automatic sliding positioning device, including a distance sensor and a timing module, which is used to record the position coordinates of the mobile unit in real time and generate cross-section positioning data.

4. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: The striking unit also includes a force rope, the energy storage spring is pre-tightened by the force rope and then triggered by an electromagnetic release device, and a pressure feedback sensor is provided at the striking end of the elastic hammer.

5. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: The signal amplifier is a multi-band filter amplifier, and the sound sensor adopts a contact piezoelectric ceramic sensor array with an array spacing of 5 to 10 cm.

6. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: The deep learning algorithm includes a convolutional neural network model, and the analysis unit is configured with: A feature extraction module, used to extract the time-frequency domain feature vector of the echo; The comparison and analysis module matches the feature vector with the preset lining health status voiceprint database for similarity; The decision module determines the probability of existence and spatial distribution of voids based on the matching results.

7. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: The spray gun device is a pneumatic adjustable nozzle structure, the direction of the spray head is controlled by a servo motor, and the marking color is switched through a multi-color ink tank.

8. The automatic detection device for tunnel lining voids according to claim 2 is characterized in that: Each moving unit is provided with 3 to 8 knocking units, the distance between adjacent knocking units is 20 to 50 cm, and the knocking force of each knocking unit can be adjusted independently.

9. The automatic detection device for tunnel lining voids according to claim 1 is characterized in that: It also includes an environmental compensation module, which includes a temperature and humidity sensor and a noise detector, and is used to perform environmental interference correction on the collected signal.

10. The automatic detection device for tunnel lining voids according to claim 6, characterized in that: The voiceprint database is dynamically updated through transfer learning, and the updated data includes fused annotation data of on-site detection data and geological radar detection results.

Images