Diagnosis system for internal void of ballastless track based on bionic leg-foot air coupling excitation

Abstract

The application discloses a ballastless track internal void diagnosis system based on bionic leg-foot air coupling excitation, relates to the void diagnosis technical field, and comprises a quadruped robot as a carrier platform, an impact excitation device embedded in the foot end of the quadruped robot, which is used for knocking the track plate surface to generate elastic waves when walking, and a echo signal acquisition device, which is a MEMS microphone fixed on the mechanical leg calf position of the quadruped robot through a rigid support, and the outside of the MEMS microphone is wrapped with a sound insulation cover. The application directly uses the metal ball embedded in the foot end of the quadruped robot as the excitation device, uses the foot falling action of the robot when walking as the impact source, does not need to be equipped with an independent electromagnetic exciter or an artificial knocking device, and remarkably reduces the system energy consumption and equipment complexity. Meanwhile, the double-layer sound insulation cover is arranged outside the MEMS microphone, and the robot itself vibration noise, motor running noise and external environmental wind noise are effectively isolated.

Description

Technical Field

[0001] This invention belongs to the field of void diagnosis technology, specifically a void diagnosis system for ballastless tracks based on biomimetic leg and foot air coupling excitation. Background Technology

[0002] With the continuous expansion and optimization of my country's rail transit network, ballastless track, with its core advantages of high smoothness and low maintenance costs, has become the mainstream track structure for high-speed railways and urban rail transit. By the end of 2025, the operating mileage of my country's high-speed railways had exceeded 50,000 kilometers, with a very high application rate of ballastless track. However, during long-term operation, ballastless track is highly susceptible to interlayer voids between the track slab and the filling layer due to repeated train loads, temperature gradient changes, and subgrade settlement.

[0003] The presence of voids disrupts the interlayer contact state of the track structure, altering the stress pattern of the track slab. Under dynamic train loads, this can easily lead to secondary disasters such as track slab cracking and fracture, severely impacting train safety, stability, and passenger comfort. Therefore, accurately detecting and quantifying void areas within the track slab is crucial for implementing preventative track maintenance and ensuring the long-term performance of the line.

[0004] Currently, the impact echo method has become one of the main methods for detecting hidden defects such as internal voids in ballastless tracks due to its advantages of minimal impact from reinforcing bars and high detection accuracy. Traditional impact echo methods mainly employ two excitation modes: manual tapping and electromagnetic excitation by track inspection vehicles. Manual tapping has low testing efficiency and is difficult to meet the needs of rapid screening over large areas. Track inspection vehicle mode has poor operational flexibility under complex conditions and requires an additional independent electromagnetic exciter module, increasing hardware complexity. In recent years, quadruped robot technology has developed rapidly. Its strong adaptability to complex terrain and compact size make it a promising and efficient carrier for track inspection operations.

[0005] Therefore, there is an urgent need to develop a ballastless track derailment detection system that can move autonomously, be efficiently vibrated, and has strong anti-interference capabilities, so as to provide technical support for the high-quality maintenance and intelligent development of my country's rail transit infrastructure. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a diagnostic system for internal voiding of ballastless track based on biomimetic leg and foot air coupling excitation.

[0007] A diagnostic system for internal voiding of ballastless tracks based on biomimetic leg and foot decoupling excitation includes: Quadruped robots serving as carrier platforms; An impact excitation device embedded in the foot of the quadruped robot is used to strike the surface of the track plate to generate elastic waves when walking; An echo signal acquisition device, wherein the echo signal acquisition device is a MEMS microphone fixed to the lower leg of a quadruped robot by a rigid bracket, and the MEMS microphone is covered with a soundproof cover; An oscilloscope and data analysis device, comprising a data processing unit, a data transmission unit, and a data storage unit, and communicatively connected to the echo signal acquisition device; The data processing unit adopts a void-free diagnosis algorithm based on the fusion of simulation and measured data. The algorithm adopts a two-branch wavelet feature alignment architecture, extracts time-frequency features and sends them to the void-free classifier to obtain the void-free existence probability, and outputs the void-free plane size and depth through regression branch.

[0008] Compared with the prior art, the beneficial effects of the present invention are: This application directly embeds the excitation device into a small metal ball at the foot of a quadruped robot, using the robot's foot-stepping motion as the impact source. This eliminates the need for a separate electromagnetic vibrator or manual striking device, significantly reducing system energy consumption and equipment complexity.

[0009] Meanwhile, a double-layer soundproof cover (outer layer of rigid steel plate reflective layer + inner layer of soft sound-absorbing and vibration-damping cotton layer) is set outside the MEMS microphone to effectively isolate the robot's own vibration noise, motor operation noise and external environmental wind noise, creating a pure acoustic environment for impact echo signal acquisition.

[0010] Furthermore, it adopts a dual-branch architecture that combines fixed wavelet packet transform and learnable wavelet packet transform. By using consistency loss to force theoretical simulation data and engineering measured data to align in the feature space, the deep learning model can learn the essential features that are independent of the data source, thereby improving the model's generalization ability on measured data.

[0011] Furthermore, this application simultaneously outputs the void-free probability of the deep learning classifier and the void-free index calculated based on the weighted sum of three explicit physical features, and performs a fusion judgment by combining the void-free rate, thus avoiding misjudgment or missed judgment by a single method in extreme cases. Attached Figure Description

[0012] Figure 1 This is a diagram showing the components and details of the detection device for a quadruped robot. Figure 2 This is a typical schematic diagram for diagnosing internal voids in ballastless tracks. Figure 3 This is a diagram illustrating the diagnosis of internal damage. Figure 4 This is a diagram illustrating real-time operation. Figure 5 This is a detailed diagram of the echo signal acquisition device; Figure 6A schematic diagram and principle of ballastless track clearance detection based on biomimetic leg and foot excitation; Figure 7 Finite element model diagram for detecting ballastless track slippage; Figure 8 The features for biomimetic leg and foot excitation and the detection features for echo signals with and without disengagement; Figure 9 This is a flowchart of the de-void diagnosis process based on the fusion of simulation and actual measurement. Detailed Implementation

[0013] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0014] Please see Figures 1-9 This application provides a diagnostic system for internal voiding of ballastless track based on biomimetic leg and foot air coupling excitation; As an embodiment of this application, it specifically includes: Quadruped robots serving as carrier platforms; An impact excitation device embedded in the foot of the quadruped robot is used to strike the surface of the track plate to generate elastic waves when walking; An echo signal acquisition device, wherein the echo signal acquisition device is a MEMS microphone fixed to the lower leg of a quadruped robot by a rigid bracket, and the MEMS microphone is covered with a soundproof cover; An oscilloscope and data analysis device, comprising a data processing unit, a data transmission unit, and a data storage unit, and communicatively connected to the echo signal acquisition device; The data processing unit adopts a void-free diagnosis algorithm based on the fusion of simulation and measured data. The algorithm adopts a two-branch wavelet feature alignment architecture, extracts time-frequency features and sends them to the void-free classifier to obtain the void-free existence probability, and outputs the void-free plane size and depth through regression branch.

[0015] As a second embodiment of this application, it specifically includes: Impact excitation device, echo signal acquisition device, oscilloscope and data analysis device, and quadruped robot as carrier platform.

[0016] like Figures 2-4 As shown, the quadruped robot has excellent terrain adaptability and can cross obstacles such as track fasteners and bosses, and walk autonomously along the longitudinal direction of the track. like Figure 5 As shown, the system of the present invention includes: An impact excitation device 1, preferably made of high-carbon steel or steel, is installed on the carrier platform. This device consists of a small metal ball embedded in the foot of the platform. The design utilizes the robot's foot-landing motion as the excitation source. When the robot walks in a preset gait, its feet periodically strike the surface of the track plate, generating elastic waves (including P-waves, S-waves, and R-waves) that propagate into the plate. Compared to traditional electromagnetic vibrators, this method requires no additional energy; the excitation energy is determined by the robot's weight and speed. Controllable impact excitation can be achieved by controlling the robot's step frequency and stride. In this embodiment, the carrier platform is a quadruped robot.

[0017] The preset gait is a diagonal trotting gait with a step frequency of 2-3 steps / second and a stride of 0.2-0.3 meters, to ensure that the feet excite the track surface with appropriate energy, avoiding signal saturation due to excessive energy or insufficient echo signal-to-noise ratio due to insufficient energy.

[0018] The echo signal acquisition device 2, mounted on the carrier platform, is a MEMS microphone fixed to any position on the lower leg of the robotic arm on the platform via a rigid bracket. The microphone is positioned near the ground on the rigid bracket, 5-10 cm above the impact point, to receive airborne sound waves radiated from the track surface. A short-time energy threshold triggering method is used, setting a threshold based on the ambient noise amplitude to eliminate certain noise interference. Waveforms exceeding the threshold are automatically saved. The rigid bracket prevents vibration from causing loose connections that could affect echo signal acquisition and transmission. The MEMS microphone is encased in a soundproof enclosure to block external interference sound waves such as robot joint motor noise, its own vibration noise, and outdoor wind noise, creating a clean acoustic environment for signal acquisition.

[0019] like Figure 6 As shown, the outer layer of the soundproof enclosure is a 0.03mm thick Q235 steel plate, forming a rigid reflective layer to reflect and isolate clutter from the external environment. The inner layer is sound-absorbing cotton, forming a flexible wave-absorbing and vibration-damping layer. Utilizing its porous fiber structure, it inherently dissipates clutter energy, absorbing residual clutter transmitted through the outer layer or generated internally. This prevents clutter from being reflected multiple times between rigid interfaces and its amplitude from being amplified due to coherent superposition, ensuring the purity of the effective peak data. Simultaneously, the sound-absorbing cotton effectively isolates mechanical vibration and provides flexible support for the components inside the enclosure.

[0020] The oscilloscope and data analysis device 3, mounted on the carrier platform, includes a data processing unit, a data transmission unit, and a data storage unit, all integrated within the carrier device. This device is communicatively connected to the echo signal acquisition device, receiving echo signal data output by the acquisition device in real time. In online mode, the data is processed by a trained neural network and then transmitted back to the cloud server; in offline mode, the signal data and processing results are stored in the local storage unit.

[0021] like Figure 7 As shown, the detection principle in this embodiment is as follows: based on the principle of impact echo, when an elastic wave propagates within a concrete slab, it will be reflected when it encounters an interface with a difference in wave impedance. If the track slab structure is intact, i.e., well bonded, most of the energy is transmitted into the underlying structure, resulting in a weak reflected echo with complex frequency components. If there is a void, i.e., the bottom of the slab is an air interface, the wave impedance difference is extremely large, and the elastic wave will be reflected back and forth in the thickness direction of the slab, forming a significant thickness resonance.

[0022] The wave impedance difference interface includes, but is not limited to, the interface between the bottom of the plate and the substrate, the interface of cracks within the plate, and the interface between the reinforcing steel and the concrete, as well as other structural interfaces that may affect the propagation of elastic waves.

[0023] The acquired signals are marked as acquired data and transmitted to the data processing unit; the data processing unit processes the acquired data in the following ways: First, low-frequency mechanical noise and high-frequency electromagnetic interference are removed by using a bandpass filter, and then the time-domain signal is converted into a frequency-domain signal using a fast Fourier transform. For a track slab of thickness T, the relationship between its thickness frequency f and the P-wave velocity Cp is as follows: f = β × Cp / 2T; Where β is the structural correction factor, which is generally taken as 0.96 for concrete structures; if the track slab material is not concrete, such as steel fiber concrete or resin-based material, the value of β should be adjusted according to the actual measurement and calibration results of the material, and the value range is 0.92-0.98. In this embodiment, unless otherwise specified, it is still taken as 0.96. like Figure 8 As shown, a void-free diagnostic algorithm based on the fusion of simulation and measured data is adopted, and a cross-domain deep learning detection and quantitative evaluation framework for void-free defects is used to automatically identify and determine the void-free status. The specific process is as follows: First, a dual-source database is constructed, containing theoretical simulation data with complete empty labels as the source domain, and engineering field measured data with scarce labels as the target domain; the empty labels include the empty location, planar dimensions, and depth; The gap-out diagnosis algorithm adopts a dual-branch wavelet feature alignment architecture: The learnable wavelet packet branch contains 3 to 5 one-dimensional convolutional layers, preferably 3. The kernel size of the first convolutional layer ranges from 32 to 128, preferably 64. The wavelet packet decomposition has 4 layers, which can effectively remove the sensitive frequency bands related to de-emptiness. The gap-out diagnostic algorithm adopts a dual-branch network architecture that integrates theoretical simulation data and engineering measured data:

[0024] In the formula, and These are the deep features extracted from simulation and measured data, respectively. It is a penalty factor, with an optimal range of 0.1-1.0; It is the classification cross-entropy loss; Maximum mean difference loss.

[0025] The upper branch performs a fixed wavelet packet transform on the theoretical simulation signal and extracts time-frequency domain wavelet coefficient features based on physical priors through a preset low-pass and high-pass filter bank. The lower branch employs a learnable wavelet packet transform on the measured signal and utilizes a trainable convolutional encoder-decoder to adaptively extract learnable wavelet coefficient features that fit the distribution of the measured data.

[0026] The two-branch features are constrained across domains through the consistency loss function, which forces theoretical data and measured data to align in the feature space, thereby reducing the inter-domain differences between simulation and measurement, and enabling the model to learn the essential feature expression that is not affected by the data source.

[0027] The aligned and fused features are fed in parallel into the declassifier and the domain discriminator. The classifier calculates the classification loss based on the supervised information of the labeled theoretical data to achieve declassification qualitative discrimination; the domain discriminator confuses the feature distributions of the source and target domains through adversarial training and calculates the adversarial loss to enhance the model's generalization ability on the measured data.

[0028] Simultaneously, a regression branch is extended at the classifier output, using the size and thickness labels of the overlying structure in the theoretical data for supervision, and then outputting the result. The entire framework is jointly optimized end-to-end by consistency loss, classification loss, adversarial loss, and regression loss.

[0029] The training process of the above neural network model uses the Adam optimizer, with an initial learning rate of 0.0001, 100 training rounds, and a batch size of 64. The ratio of simulated data to actual test data in the training data is 3:1. The actual test data needs to be manually pre-screened to remove invalid samples with a signal-to-noise ratio of less than 10dB.

[0030] During the inference phase, the system automatically extracts cross-domain aligned time-frequency features from the measured impact echo signal and inputs them into a classifier to obtain the probability of void existence. If the probability exceeds a preset threshold (set to 0.75 in this embodiment), the system combines the regression branch to output the planar range and depth value of the void below the measuring point. It also automatically generates a void defect cloud map of the track slab, using color depth to represent the severity of the void, forming an internal quality health record for the ballastless track and providing precise coordinates for grouting repair by the engineering department.

[0031] The preset threshold of 0.75 is an empirical value and can be adjusted according to different track slab structure types and environmental noise levels. The adjustment range is 0.6-0.9. The automatic identification of the main spectral peak is implicit in the frequency band energy distribution of wavelet packet decomposition. The model automatically focuses on the theoretical thickness frequency shift characteristics and high-amplitude resonance peak characteristics that are sensitive to gaps, such as... Figure 9 As shown, no manual setting of the frequency offset threshold is required, thus enabling intelligent, robust, automatic diagnosis and quantitative assessment of void defects.

[0032] Example 3, as another embodiment of this application, is implemented based on Example 1, except that the method by which the data processing unit processes the collected data is different. The specific method of this embodiment is as follows: The impact echo signal at each measuring point is processed by short-time Fourier transform and wavelet packet decomposition to extract three key features from the spectrum: The degree of deviation of the measured main peak frequency relative to the theoretical thickness frequency at the measurement point; The attenuation ratio of the main peak amplitude to the normal reference amplitude in the same region; the reference amplitude is determined by the average main peak amplitude of 10 measuring points automatically collected by the quadruped robot at the start of the detection in a known track plate area without any gaps, through artificial air coupling impact echo verification area. The relative significance of the amplitude of the secondary peak to that of the primary peak.

[0033] By comprehensively weighting the above three characteristic quantities, the void separation index is obtained, which effectively overcomes the problem of inconsistent benchmarks caused by the difference in structural thickness between the rail support platform and the flat plate area, and reduces misjudgments and omissions caused by noise or local thickness fluctuations.

[0034] Simultaneously, the voiding index of each measuring point is linked to the real-time global positioning coordinates and odometer data recorded by the robot. Spatial interpolation methods are used to expand the discrete measuring point data into a continuous distribution field, and each location is classified into severe voiding, suspected voiding, or normal states based on a preset index threshold range. Replacing binary judgment results with a continuously quantified voiding index more accurately depicts the voiding boundary and reflects the graded changes in defect severity, providing clear quantitative guidance for the subsequent grouting repair workload and work priorities.

[0035] The formula for calculating the de-vacancy index is as follows: DI=ω1·Δf+ω2·α+ω3·γ; Where Δf is the main peak frequency offset, α is the amplitude attenuation ratio, and γ is the relative amplitude of the secondary peak. The weights ω1, ω2, and ω3 are set to 0.6, 0.3, and 0.1, respectively. The threshold range for the vacancy index is as follows: greater than 0.7 indicates severe vacancy, 0.4-0.7 indicates suspected vacancy, and less than 0.4 indicates a normal state.

[0036] Example 4 is implemented based on Examples 2 and 3, except that this example includes both methods of determining whether the object is emptied, as used in Examples 2 and 3. After obtaining the probability of exiting the void and the void-exit index, and removing the dimensions from both, the comprehensive void-exit rate is calculated using the following formula: Overall emptying rate = h × probability of emptying + (1-h) × emptying index; In this embodiment, h is set to 0.5. When the overall vacuolation rate is greater than 0.7, it is considered severe vacuolation; 0.35-0.7 is considered suspected vacuolation; and less than 0.35 is considered normal. Example 5 is an implementation based on Example 4, except that if the overall void-free rate is in a suspected void-free situation, the data processing unit needs to perform detection path planning. The specific planning method is as follows: The stride is reduced to 0.1m to ensure that the landing point interval is less than the common small gap size, which is less than 0.15m in diameter; At the same time, the walking path was changed to a serpentine full-coverage path, planned in an S-shaped path along the track slab to ensure that the foot landing point covers the entire width of the track slab, including the edge area. Adjust the cadence to 3-4 steps / second to adapt to dynamic stability under small strides; The signals collected during the re-inspection were processed again as described in Part Four of Example 4 to obtain the overall void removal rate of the high-density measurement points.

[0037] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A diagnostic system for internal voiding of ballastless track based on biomimetic leg-foot decoupling excitation, characterized in that, include: Quadruped robots serving as carrier platforms; An impact excitation device embedded in the foot of the quadruped robot is used to strike the surface of the track plate to generate elastic waves when walking; An echo signal acquisition device, wherein the echo signal acquisition device is a MEMS microphone fixed to the lower leg of a quadruped robot by a rigid bracket, and the MEMS microphone is covered with a soundproof cover; An oscilloscope and data analysis device, comprising a data processing unit, a data transmission unit, and a data storage unit, and communicatively connected to the echo signal acquisition device; The data processing unit adopts a void-free diagnosis algorithm based on the fusion of simulation and measured data. The algorithm adopts a two-branch wavelet feature alignment architecture, extracts time-frequency features and sends them to the void-free classifier to obtain the void-free existence probability, and outputs the void-free plane size and depth through regression branch.

2. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The impact excitation device is a small metal ball embedded in the foot of a quadruped robot; The quadruped robot's default gait is a diagonal trotting gait with a step frequency of 2-3 steps per second and a stride of 0.2-0.3 meters.

3. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The outer layer of the soundproof enclosure is a rigid reflective layer made of 0.03mm thick Q235 steel plate, while the inner layer is a flexible wave-absorbing and vibration-damping layer made of sound-insulating cotton.

4. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The data processing unit first filters the acquired signal using a bandpass filter, then converts the time-domain signal into a frequency-domain signal using a fast Fourier transform, and calculates the theoretical thickness frequency of the track slab according to the formula f=β×Cp / 2T, where β is the structural correction coefficient, which is taken as 0.96 for concrete structures; T is the thickness of the track slab, and Cp is the P-wave velocity.

5. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The neural network model with dual-branch wavelet feature alignment architecture uses the Adam optimizer with an initial learning rate of 0.0001, 100 training rounds, a batch size of 64, and a simulation data to experimental data ratio of 3:

1. Invalid samples with a signal-to-noise ratio below 10dB are removed from the experimental data.

6. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The preset threshold for the probability of void presence output by the void classifier is 0.75, and the threshold is adjusted within the range of 0.6-0.

9.

7. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The data processing unit is also used to: perform short-time Fourier transform and wavelet packet decomposition on the impact echo signal of each measuring point, extract three key feature quantities: the degree of deviation of the measured main peak frequency relative to the theoretical thickness frequency, the attenuation ratio of the main peak amplitude to the reference amplitude, and the relative significance of the secondary peak amplitude to the main peak amplitude, and calculate the voiding index according to the formula DI=0.6×Δf+0.3×α+0.1×γ; where DI>0.7 indicates severe voiding, 0.4≤DI≤0.7 indicates suspected voiding, and DI<0.4 indicates normal state; Δf is the main peak frequency deviation rate, α is the amplitude attenuation ratio, and γ is the relative amplitude of the secondary peak.

8. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 7, characterized in that, The data processing unit is also used to: remove the dimensions from the probability of void existence and the void index, and calculate the comprehensive void rate according to the formula: comprehensive void rate = h x probability of void existence + (1-h) x void index; when the comprehensive void rate is greater than 0.7, it is a serious void, 0.35-0.7 is a suspected void, and less than 0.35 is a normal state.

9. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 8, characterized in that, When the overall clearance rate is in a suspected clearance state, the data processing unit controls the quadruped robot to perform a re-inspection path planning: the stride is reduced to 0.1m, the walking path is switched to serpentine full coverage, the step frequency is adjusted to 3-4 steps / second, and the overall clearance rate is recalculated on the signals collected during the re-inspection; the value of h is 0.

5.

10. The ballastless track internal void diagnosis system based on biomimetic leg and foot air coupling excitation according to claim 1, characterized in that, The quadruped robot has the ability to adapt to terrain by crossing track fasteners and bosses, and can walk autonomously along the longitudinal direction of the track.

Images