Rail crack identification method and system based on ultrasonic guided waves and support vector machine

Abstract

The invention relates to a rail crack identification method and system based on ultrasonic guided waves and a support vector machine. The method comprises the steps that the ultrasonic guided waves are emitted at the rail waist position of a to-be-detected rail through an excitation transducer; receiving transducers arranged at preset intervals are used for collecting the propagated guided wave signals and digitalizing the propagated guided wave signals; preprocessing the digital signal, and extracting a multi-dimensional feature vector containing a time domain feature and a frequency domain feature; the feature vector is input into a pre-training model, a rail health state classification result is output, an alarm is triggered, the pre-training model is based on a support vector machine classifier, and a multi-dimensional feature vector composed of a time domain feature and a frequency domain feature extracted from an ultrasonic guided wave signal of a standard rail sample is used as input data; and taking a preset damage label corresponding to the crack damage position and depth as output data, and training and optimizing a kernel function and a hyper-parameter to obtain the data. According to the invention, a detection blind area can be broken through, real-time on-line monitoring is realized, and the problem of complexity of guided wave signal analysis is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail transit safety monitoring, in particular to a rail crack identification method and system based on ultrasonic guided waves and support vector machines. BACKGROUND

[0002] With the acceleration of urbanization, underground transportation engineering has become a key infrastructure for relieving ground traffic pressure and protecting the ecological environment. As the core component of the transportation system, the rail is prone to cracks due to stress concentration in complex construction environments such as shield tunnels, and there is a risk of structural failure. The traditional rail flaw detection technology has significant defects:

[0003] Manual inspection relies on visual inspection and knocking sound to distinguish, which is low in efficiency and strong in subjectivity, and cannot detect internal early cracks.

[0004] Although large flaw detection vehicles integrate ultrasonic and eddy current detection technologies, the probe is difficult to cover the fixed area of the splint, the turnout and the blocked area of the fastener, forming a detection blind area. In addition, periodic detection cannot provide real-time early warning for sudden damage.

[0005] Although the structural health monitoring (SHM) technology supports online monitoring, the ultrasonic guided wave (UGW) technology faces core problems in rail applications: the irregular cross-section of the rail causes severe frequency dispersion and multi-modal conversion during guided wave propagation, and the multi-mode wave packets are superimposed on each other in the received signal. The damage characteristics are submerged in background noise, and traditional signal processing methods cannot achieve high-precision identification.

[0006] Existing guided wave detection schemes generally rely on professionals to manually interpret complex time / frequency domain signals, which is low in automation and insufficient in sensitivity to early micro-cracks. Therefore, there is an urgent need for an intelligent rail crack identification technology that can break through the detection blind area, achieve real-time online monitoring, and solve the complexity of guided wave signal analysis. SUMMARY

[0007] Therefore, it is necessary to provide a rail crack identification method and system based on ultrasonic guided waves and support vector machines, which can break through the detection blind area, achieve real-time online monitoring, and solve the complexity of guided wave signal analysis.

[0008] The rail crack identification method based on ultrasonic guided waves and support vector machines provided by the present application comprises:

[0009] An excitation transducer is used to emit ultrasonic guided waves at the rail waist position of the rail to be tested;

[0010] A receiving transducer arranged at a predetermined interval collects the guided wave signals after propagation and digitizes them;

[0011] The digital signals are preprocessed to extract a multi-dimensional feature vector containing time domain features and frequency domain features.

[0012] inputting the feature vector into a pre-trained SVM model to output a rail health state classification result and trigger an alarm, the pre-trained SVM model being based on a support vector machine classifier, a multi-dimensional feature vector composed of time domain features and frequency domain features extracted from ultrasonic guided wave signals of standard rail samples being input data, and a damage label corresponding to a crack damage position and depth being output data, the pre-trained SVM model being obtained through training and optimization of a kernel function and hyperparameters.

[0013] In one of the embodiments, the excitation of the ultrasonic guided wave comprises:

[0014] generating an electrical pulse signal with a center frequency of 160 kHz;

[0015] amplifying the electrical pulse signal by a high-fidelity power amplifier;

[0016] converting the amplified electrical pulse signal into mechanical vibration by an excitation transducer and coupling the mechanical vibration to the rail web centerline.

[0017] In one of the embodiments, the arrangement of the receiving transducers satisfies:

[0018] pasting the receiving transducers on the rail web centerline of the rail to be tested at intervals of 1 meter;

[0019] performing band-pass filtering and gain-adjustable amplification processing on the received guided wave signal by a sensing signal filtering-amplifying module.

[0020] In one of the embodiments, the preprocessing of the digital signal comprises:

[0021] adopting an adaptive filtering algorithm to remove environmental noise and system noise;

[0022] performing baseline correction and signal smoothing operations to improve the signal-to-noise ratio.

[0023] In one of the embodiments, the construction of the pre-trained SVM model comprises:

[0024] setting known damage states at different positions for standard rail samples;

[0025] collecting guided wave signals under the damage states and extracting multi-dimensional feature vectors;

[0026] adopting a Sigmoid kernel function or a Gaussian radial basis kernel function to optimize hyperparameters for training with the damage label as the output target.

[0027] In one of the embodiments, the triggering of the alarm comprises:

[0028] when the output result is a crack damage, starting a sound-light alarm device in real time;

[0029] transmit the damage position classification result to the upper computer through the communication interface.

[0030] The application also provides a rail crack identification device based on ultrasonic guided waves and a support vector machine, which comprises:

[0031] The sending module is configured to emit ultrasonic guided waves at the rail web position of the rail to be measured by exciting the transducer.

[0032] The receiving module is configured to collect and digitize the guided wave signals after propagation by the receiving transducers arranged at intervals of a preset interval.

[0033] The extraction module is configured to pre-process the digital signals and extract a multi-dimensional feature vector containing time domain features and frequency domain features.

[0034] The output module is configured to input the feature vector into a pre-trained SVM model, output a rail health state classification result, and trigger an alarm.

[0035] The application also provides a rail crack identification system, which comprises:

[0036] The excitation amplification module is configured to receive and amplify input electrical signals and is connected to the excitation transducer at the output end to act on the rail web.

[0037] The hardware integration unit is electrically connected to the excitation amplification module and is configured to generate a signal excitation instruction and control the excitation timing.

[0038] The sensing filtering-amplification module receives the guided wave signals after propagation of the rail through the sensing transducer at the input end and sequentially performs filtering and noise reduction and signal amplification processing.

[0039] The algorithm integration unit receives the output signals of the sensing filtering-amplification module and integrates the following functions:

[0040] The real-time monitoring process is triggered by the "start test" button.

[0041] The support vector machine classification model training is performed by the "model training" button.

[0042] The input signals are subjected to feature extraction and damage classification identification.

[0043] The identification state is output on the "result judgment" interface.

[0044] In one of the embodiments, the excitation amplification module, the hardware integration unit and the sensing filter-amplification module are physically connected through a cable.

[0045] The algorithm integration unit is an embedded device with a display screen, and its interface contains a "parameter setting" button for configuring the detection period.

[0046] The sensing filter-amplification module connects at least two sensors to receive guided wave signals of different sections of the rail in a multi-channel mode.

[0047] The algorithm integration unit is connected to the remote monitoring system through a multi-protocol communication interface to realize real-time transmission of alarm information.

[0048] The above rail crack identification method and system based on ultrasonic guided waves and support vector machines, by fixing the transducer on the rail waist, combined with cyclic monitoring, breaks through the space coverage limit of the flaw detection vehicle and the periodic detection of defects, realizes full-time domain coverage of key areas and real-time early warning of sudden damage, eliminates the detection blind area and realizes real-time monitoring; a multi-dimensional feature vector extraction mechanism is used to extract the crack-sensitive characteristic quantity from the complex frequency dispersion signal, replacing the traditional artificial waveform interpretation, solving the problem of damage feature submersion caused by multi-modal wave packet superposition, overcoming the complexity of guided wave signal analysis; through the pre-trained SVM model for nonlinear pattern recognition of the feature vector, high-precision classification of the crack position is realized, subjective misjudgment is avoided, the dependence on professional personnel is reduced, the identification reliability is significantly improved, the identification accuracy and intelligent level are improved; the time domain energy attenuation and the frequency domain nonlinear harmonic feature are fused, so that the system is sensitive to early micro-cracks that cause weak signal distortion, breaks through the limitation of traditional methods to macro-damage, realizes early warning ability, and enhances the sensitivity of early damage. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0050] Figure 1 A flowchart of a rail crack identification method based on ultrasonic guided waves and support vector machines for one embodiment;

[0051] Figure 2 A schematic diagram of a rail crack identification device for one embodiment based on ultrasonic guided waves and support vector machines;

[0052] Figure 3 A schematic diagram of a rail crack identification system for one embodiment;

[0053] Figure 4 An internal structure diagram of a computer device of an embodiment. DETAILED DESCRIPTION

[0054] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely explain the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0055] With the continuous advancement of urbanization in China, the scale and speed of urban construction are accelerating, but the problems brought by urbanization are also increasingly prominent. Traffic congestion, building space shortage, environmental pollution and other problems, especially in large cities, are particularly serious. In order to solve these contradictions, the development and utilization of urban underground space has become an important choice. Through the construction of urban underground traffic engineering, not only can the pressure of ground traffic be relieved, but also the ecological environment of the city can be effectively protected, and the overall sustainable development ability of the city can be improved. However, in the construction of underground traffic engineering, the rail is an important part of the transportation system, especially in shield tunnel engineering, because of the complex construction process and variable environmental conditions, if the rail is damaged, structural failure will easily occur, and there is a great safety risk. In order to ensure the smooth progress of the project and minimize accidents, it is particularly important to research and develop an effective intelligent monitoring and control system. Through real-time monitoring of the track structure and the running state of the locomotive, potential safety hazards can be found and solved in time, so as to ensure the safe operation of the urban traffic system.

[0056] Therefore, regular detection and long-term health monitoring of the rail are necessary means to ensure the safe operation of rail transit. At present, the main rail flaw detection technologies include:

[0057] Manual inspection: mainly relies on visual inspection and knocking sound judgment of inspection personnel, which has high labor intensity, low efficiency, is greatly affected by subjective factors, and cannot find early internal cracks in the rail.

[0058] Large flaw detection vehicle: integrated with ultrasonic flaw detection (such as B-scan), eddy current flaw detection and other non-destructive testing technologies, is the current mainstream detection method. It has high automation degree and fast detection speed.

[0059] Structural health monitoring (SHM) technology: In recent years, online monitoring methods based on fiber Bragg grating, piezoelectric sensor, acoustic emission and other technologies have been proposed, aiming to realize real-time and continuous monitoring of the structure state. Among them, the ultrasonic guided wave (UGW) technology is considered as an ideal technical means for long-distance online monitoring of long strip structures such as rails, because of its long propagation distance, small attenuation in waveguide structures such as plates and rods, and sensitivity to defects in the whole cross-section of the structure.

[0060] Although some progress has been made, there are still the following defects and deficiencies that need to be solved urgently:

[0061] Detection blind area exists, and full coverage monitoring cannot be achieved: When detecting, the probe of the large-scale flaw detection vehicle cannot cover the joint area fixed by the plate, the turnout area and the area blocked by the fastener, forming a detection blind area. These areas are exactly the weak links where stress is concentrated and fatigue damage is prone to occur.

[0062] Non-real-time monitoring, unable to warn sudden damage: The detection of the flaw detection vehicle is periodic, usually once every few months, and cannot warn the damage occurring between two detection periods or the rapid expansion of existing damage, lacking real-time performance.

[0063] Complex guided wave signal analysis, low recognition accuracy: The cross-sectional shape of the rail is extremely irregular, resulting in severe frequency dispersion and multi-modal conversion phenomena when ultrasonic guided waves propagate therein. The received signal waveform is complex, and multiple mode wave packets are superimposed on each other, making it difficult to effectively identify the damage characteristic signal in the complex background signal through simple manual interpretation or traditional signal processing methods, resulting in low recognition accuracy and reliability.

[0064] Low degree of intelligence, relying on professional analysis: The existing guided wave-based detection methods mostly stay at the signal analysis stage, requiring experienced professional technicians to interpret and judge complex time-domain and frequency-domain signals, with low automation and intelligence levels, which is not conducive to large-scale popularization and application.

[0065] In order to solve the problems of detection blind area, non-real-time monitoring, complex signal analysis and low degree of intelligence, a rail crack recognition method and system based on ultrasonic guided wave and support vector machine are provided.

[0066] The rail crack recognition method and system based on ultrasonic guided wave and support vector machine will be described below. Figures 1-4 The rail crack recognition method and system based on ultrasonic guided wave and support vector machine will be described below.

[0067] As shown in Figure 1 In one embodiment, a rail crack recognition method based on ultrasonic guided wave and support vector machine includes the following steps:

[0068] Step S110, excite the transducer to emit ultrasonic guided waves at the rail web position to be measured.

[0069] The excitation process of the ultrasonic guided wave is to emit guided waves at the rail web center line through a piezoelectric ceramic transducer, which is implemented as follows:

[0070] The hardware integration unit generates an electrical pulse signal with a center frequency of 160 kHz. The signal is a 5-cycle Hanning window modulated sinusoidal pulse signal with a bandwidth of 120 kHz to 200 kHz to ensure energy concentration and reduce dispersion effects. Hanning window modulation optimizes the time-frequency characteristics of the signal and improves the signal-to-noise ratio.

[0071] The high-fidelity power amplifier amplifies the electrical pulse signal with a gain range of 20 dB to 60 dB adjustable. The peak voltage of the amplified signal is set to 100 V to drive the transducer to work efficiently. In actual implementation, the gain can be dynamically adjusted according to the rail material (such as steel rail or alloy), which is configured through the "parameter setting" interface of the algorithm integration unit.

[0072] The excitation transducer (with a resonance frequency matching 160 kHz) converts the amplified electrical signal into mechanical vibration and couples it to the rail web center line. The coupling method uses industrial-grade coupling agent (such as silicone grease) to ensure efficient energy transfer. In implementation, the installation position of the excitation transducer is accurately calibrated to avoid signal attenuation caused by deviation.

[0073] By limiting the excitation mode of the ultrasonic guided wave (generating a 160 kHz electrical pulse signal, high-fidelity power amplification, and piezoelectric ceramic transducer coupling to the rail web center line), a guided wave signal with pure excitation mode and weak dispersion effect is ensured, reducing energy loss and signal distortion, and providing a stable excitation source for subsequent high-precision detection. The use of a 5-cycle Hanning window modulated sinusoidal pulse signal (bandwidth 120-200 kHz) and 100 V peak voltage amplification optimizes the signal spectral concentration, reduces spectral leakage, enhances the capture ability of non-linear effects caused by small cracks in the rail, and improves the sensitivity of early damage detection.

[0074] Step S120, the receiving transducers arranged at intervals of a predetermined distance collect and digitize the propagated guided wave signals.

[0075] The receiving transducer arrangement and signal acquisition process supports multi-channel detection, covering different sections of the rail. The receiving transducers are pasted at intervals of 1 meter on the rail web center line to be measured. Each receiving transducer is a wideband piezoelectric sensor with a response frequency covering 100 kHz to 250 kHz to ensure the capture of fundamental frequency (160 kHz) and double frequency (320 kHz) signals. In actual implementation, the transducer array can be expanded to 4-8 channels to cover blind areas in tunnels (such as switch or fastener blocking areas).

[0076] The sensing signal filtering-amplifying module performs band-pass filtering (center frequency 160 kHz, bandwidth 80 kHz) and gain-adjustable amplification processing on the received guided wave signals. The gain is independently adjustable to adapt to different environmental noise levels. For example, in a high-noise underground tunnel, the gain is set to 40 dB or more; in a low-noise road section, the gain is reduced to 20 dB. The module outputs a digital signal, and the acquisition synchronization is realized through a hardware integrated unit (such as FPGA or MCU), with an error control within microseconds.

[0077] By arranging wide-band piezoelectric sensors at 1-meter intervals and band-pass filtering, blind areas such as rail joints and switches are covered, environmental noise is effectively suppressed, and the complete collection of weak guided wave signals and the improvement of signal-to-noise ratio are ensured.

[0078] In step S130, the digital signal is preprocessed to extract a multi-dimensional feature vector containing time domain features and frequency domain features.

[0079] The preprocessing stage improves signal quality and extracts key feature vectors, laying the foundation for SVM classification. Adaptive filtering algorithms are used to remove environmental noise and system noise. The digital signal is processed using the normalized least mean square (NLMS) algorithm, which has a fast convergence speed and is computationally efficient. The algorithm parameters (such as step size μ) are optimized through experiments and are executed in real time on embedded devices. For example, in benchmark testing, NLMS improves the signal-to-noise ratio by 15 dB or more, effectively suppressing environmental vibration noise. Baseline correction and signal smoothing operations are performed to improve the signal-to-noise ratio. Baseline correction uses polynomial fitting to remove DC offset; signal smoothing uses Savitzky-Golay filtering (window size 5 points) to retain high-frequency features while smoothing random noise. In implementation, the preprocessing time is controlled within 10 ms, meeting the real-time requirements. The combination of adaptive filtering (LMS / NLMS) and signal smoothing preprocessing specifically removes rail vibration noise and system interference, significantly improving signal quality and providing a high-reliability data foundation for subsequent feature extraction.

[0080] A multi-dimensional feature vector is extracted from the preprocessed signal, including time domain features (at least two of maximum amplitude, variance, and effective value) and frequency domain features (160 kHz fundamental frequency amplitude and 320 kHz second harmonic amplitude). For example, when a crack exists, the second harmonic amplitude significantly increases (due to nonlinear effects), and this feature is used as a key input. The feature vector has a dimension of 5-7, which is calculated by an algorithm integration unit. The combination of time domain features (maximum amplitude, variance, and effective value) and frequency domain features (160 kHz fundamental frequency amplitude and 320 kHz second harmonic amplitude) constructs a multi-dimensional feature vector, which simultaneously captures the energy attenuation and nonlinear harmonic response caused by cracks, enhancing the sensitivity and discrimination of early micro-damage and cracks at different locations.

[0081] Step S140, input the feature vector into the pre-trained SVM model, output the rail health state classification result, and trigger the alarm. The pre-trained SVM model is based on a support vector machine classifier, and the multi-dimensional feature vector composed of the time domain features and the frequency domain features extracted from the ultrasonic guided wave signals of the standard rail sample is used as the input data, and the pre-set damage label corresponding to the crack damage position and depth is used as the output data, and the kernel function and hyperparameters are optimized by training to obtain.

[0082] The pre-trained SVM model is based on standard rail samples. First, set the known damage state (such as no damage, near-end damage, middle-end damage, and far-end damage) at different positions of the sample, collect the guided wave signals, and extract the feature vector. During training, the Gaussian radial basis function (RBF) is used, and the kernel parameters (such as C and γ) are optimized by the Bayesian optimization method to minimize the cross-validation error. For example, on a data set (1000 samples), the accuracy of the optimized model is 98.5%. Based on the known damage label and the optimized kernel function (Sigmoid / RBF), the SVM model is trained, combined with grid / random / Bayesian hyperparameter optimization, which significantly improves the classification accuracy and generalization ability of the model for damage patterns in complex guided wave signals.

[0083] The feature vector is input into the SVM model, and the four-classification result is output. If it is identified as a crack damage (such as near-end or middle-end damage), the audible and visual alarm device (such as LED flickering and buzzer) is started in real time. At the same time, the damage position is transmitted to the upper computer (railway maintenance management system) through the communication interface (RS485, CAN bus or 4G / 5G wireless). In implementation, the alarm delay is less than 100ms. Through real-time audible and visual alarm and multi-protocol communication interface transmission of damage position, second-level response and remote monitoring are realized, ensuring timely warning of sudden cracks and rapid intervention of the maintenance system.

[0084] Automatic cycle monitoring is performed to realize long-term continuous monitoring. The interval time can be dynamically adjusted, such as shortening the interval to 30 minutes in high-load sections (such as freight lines), and extending the interval to 2 hours when the environmental conditions are stable. The cycle process records logs for subsequent optimization. Cycle monitoring is performed and dynamic adjustment period is supported, breaking through the cycle limit of traditional flaw detection vehicles, realizing continuous monitoring of the rail in the full time domain, adapting to load and environmental changes, and ensuring long-term operation safety.

[0085] The rail crack identification method based on ultrasonic guided waves and support vector machines in the embodiment breaks through the space coverage limitation and periodic defect detection of the flaw detection vehicle by fixing and installing the transducer on the rail waist (especially for traditional blind areas such as joints and switches) and combining with cyclic monitoring, realizes full-time domain coverage of key areas and real-time early warning of sudden damage, eliminates the detection blind area and realizes real-time monitoring; the multi-dimensional feature vector extraction mechanism is adopted to strip out the crack-sensitive characteristic quantity from the complex dispersion signal, replace the traditional artificial waveform interpretation, solve the damage characteristic submergence problem caused by multi-modal wave packet superposition and overcome the complexity of guided wave signal analysis; the pre-trained SVM model is used for nonlinear pattern recognition of the feature vector, high-precision classification of the crack position (near-end / middle / remote) is realized, subjective misjudgment is avoided, the dependence on professional personnel is reduced, the identification reliability is significantly improved, the identification accuracy and intelligent level are improved; the time domain energy attenuation and the frequency domain nonlinear harmonic features (such as the amplitude of the second harmonic) are fused, the system is sensitive to early micro-cracks causing weak signal distortion, the limitation of traditional methods to macro-damage is broken through, the early warning capability is realized, and the early damage sensitivity is enhanced.

[0086] The rail crack identification device based on ultrasonic guided waves and support vector machines provided by the application is described below, and the rail crack identification device based on ultrasonic guided waves and support vector machines described below can be correspondingly referred to the rail crack identification method based on ultrasonic guided waves and support vector machines described above.

[0087] As shown in Figure 2 In one embodiment, a rail crack identification device based on ultrasonic guided waves and support vector machines includes a sending module 210, a receiving module 220, an extraction module 230 and an output module 240.

[0088] The sending module 210 is used to emit ultrasonic guided waves at the rail waist position to be measured by exciting the transducer;

[0089] The receiving module 220 is used to collect and digitize the guided wave signals propagated by the receiving transducers arranged at intervals of a preset interval;

[0090] The extraction module 230 is used to pre-process the digital signals and extract multi-dimensional feature vectors containing time domain features and frequency domain features;

[0091] The output module 240 is used to input the feature vectors into the pre-trained SVM model, output the rail health state classification result, and trigger the alarm, the pre-trained SVM model is based on the support vector machine classifier, the multi-dimensional feature vectors composed of the time domain features and the frequency domain features extracted from the ultrasonic guided wave signals of the standard rail sample are used as the input data, and the pre-set damage labels corresponding to the crack damage position and depth are used as the output data, and the kernel function and the hyperparameter are obtained after training and optimization.

[0092] In addition, the application also provides a rail crack identification system.

[0093] As shown in the drawings, Figure 3 In one embodiment, a rail crack identification system includes an excitation amplification module 310, a hardware integrated unit 320, a sensing filter-amplification module 330, and an algorithm integrated unit 340. The excitation amplification module 310, the hardware integrated unit 320, and the sensing filter-amplification module 330 are physically connected through cables.

[0094] The excitation amplification module 310 is used to receive input electrical signals and amplify them. The output end is connected to an excitation transducer to act on the rail web. The input end of the excitation amplification module 310 receives a signal excitation instruction from the hardware integrated unit 320, and the output end is connected to the excitation transducer. The excitation amplification module 310 contains a high-fidelity power amplifier with a gain of 20-60 dB adjustable and supports a peak voltage of 100 V output. In implementation, it is packaged in a waterproof housing and is suitable for use in a humid tunnel environment.

[0095] The hardware integrated unit 320 is electrically connected to the excitation amplification module 310 and is used to generate a signal excitation instruction and control the excitation timing. The hardware integrated unit 320 is based on a programmable logic device (FPGA) or a microcontroller (MCU) and accurately controls the excitation timing and signal acquisition synchronization as well as multi-channel switching. For example, the FPGA realizes multi-channel switching (supports up to 8 receiving transducers) with a timing error of <1 μs. The unit connects the excitation amplification module and the sensing filter-amplification module through a closed-loop control link to ensure signal integrity.

[0096] The sensing filter-amplification module 330 receives the guided wave signal propagated by the rail through a sensing transducer, and sequentially performs filtering and noise reduction and signal amplification processing. The input end is connected to a receiving transducer to perform band-pass filtering (center 160 kHz, bandwidth 80 kHz) and independent gain amplification. The gain of each channel is adjustable to adapt to different sensors. The module outputs a digital signal to the algorithm integrated unit, and the filter response time is <5 ms.

[0097] The algorithm integrated unit 340 receives the output signal of the sensing filter-amplification module, integrates the following functions: triggers a real-time monitoring process through a "start test" button; performs support vector machine classification model training through a "model training" button; extracts features from the input signal and identifies damage classification; and outputs the identification state (no damage, near-end damage, middle-end damage, and far-end damage) on a "result judgment" interface.

[0098] The algorithm integrated unit 340 is an embedded device (such as an ARM Cortex-A series) with a display screen and integrates the following functions:

[0099] User interface: including "start test" button (trigger real-time monitoring), "model training" button (execute SVM training), "parameter setting" button (configure detection period and gain), "result judgment" interface (display recognition state: no damage, near-end damage, etc.).

[0100] Core algorithm: real-time execution of signal preprocessing, feature extraction and SVM classification. In implementation, the algorithm uses C++ optimization, and the processing delay is <50ms.

[0101] Communication interface: connect remote monitoring system (such as cloud platform) through RS485, Ethernet, etc. Upload alarm information and damage data.

[0102] The rail crack recognition system of the application realizes full-process closed-loop control from excitation to acquisition to processing to output through excitation amplification, sensor filtering-amplification, and modular integration of hardware and algorithm units, improves system reliability, and greatly reduces the use threshold through interface operation (test / training / result display). Multi-channel independent filtering and amplification design supports parallel processing of sensors, ensuring independent optimization of signals in different sections; embedded device parameter configuration function enhances system adaptability to meet the needs of diversified monitoring scenarios. Multi-sensor multi-channel layout covers different sections of the rail, eliminating detection blind spots; remote monitoring (maintenance system / cloud platform) expands global management capabilities, realizing networked monitoring. FPGA / MCU precisely controls excitation timing and acquisition synchronization to suppress signal jitter; multi-protocol communication interface ensures real-time transmission of alarm information, improving system response speed and collaboration efficiency. The hardware integrated unit connects the excitation amplification module and the sensor filtering-amplification module, forming a "excitation to acquisition" closed-loop control link, which strengthens the hardware collaboration of the excitation to acquisition closed-loop link design, ensures strict synchronization of signal excitation and reception, reduces the impact of timing errors on feature extraction, and improves damage recognition consistency.

[0103] The application has the following advantages:

[0104] Real-time online and full-coverage monitoring: The system of the application can be fixedly installed on the rail to realize 7x24 hour uninterrupted monitoring, making up for the shortcomings of periodic detection by the flaw detection vehicle, and enabling timely detection of sudden damage. Sensors can be installed in joints, switches and other traditional detection blind areas to achieve full-time coverage of key areas.

[0105] High recognition accuracy and strong reliability: The application does not rely on direct interpretation of complex waveforms, but extracts multi-dimensional feature vectors sensitive to damage, and then uses the powerful nonlinear classification ability of SVM for pattern recognition. This method has high recognition accuracy for cracks in different positions, effectively overcoming the problems of frequency dispersion and multi-modal interference, and the results are highly reliable.

[0106] High automation and intelligence: the whole process from signal acquisition, feature extraction to state judgment is completed automatically by the system without human intervention, which reduces the requirement for professional skills of the operator, eliminates subjective judgment errors and improves the intelligent level of detection.

[0107] Sensitive to early micro-damage: combined with guided wave energy attenuation (time domain feature) and nonlinear effect (frequency domain feature), the application is not only sensitive to macroscopic cracks that cause significant echo, but also sensitive to early micro-cracks that cause signal nonlinearity, and has the ability of early warning.

[0108] Figure 4 An example of a schematic diagram of the physical structure of an electronic device, which can be a smart terminal, is shown in Figure 4 The electronic device includes a processor, a memory and a network interface connected by a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the electronic device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the rail crack identification method based on ultrasonic guided wave and support vector machine, which includes:

[0109] An ultrasonic guided wave is emitted at the rail waist position to be measured by exciting the transducer;

[0110] The received transducer arranged at intervals of a predetermined distance collects the propagated guided wave signal and digitizes it;

[0111] The digital signal is preprocessed to extract a multi-dimensional feature vector containing time domain features and frequency domain features;

[0112] The feature vector is input into a pre-trained SVM model to output a rail health state classification result and trigger an alarm. The pre-trained SVM model is based on a support vector machine classifier, and the multi-dimensional feature vector composed of time domain features and frequency domain features extracted from the ultrasonic guided wave signal of the standard rail sample is used as the input data, and the pre-set damage label corresponding to the crack damage position and depth is used as the output data, and the kernel function and hyperparameters are obtained after training and optimization.

[0113] Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the application, and does not constitute a limitation on the electronic device to which the scheme of the application is applied. The specific electronic device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0114] In another aspect, the present application also provides a computer storage medium storing a computer program, the computer program being executed by a processor to implement an ultrasonic guided wave and support vector machine based rail crack identification method, the method comprising:

[0115] emitting an ultrasonic guided wave at a rail web position to be measured by exciting a transducer;

[0116] collecting and digitizing the propagated guided wave signals by receiving transducers arranged at intervals with a preset interval;

[0117] preprocessing the digital signals to extract a multi-dimensional feature vector containing time domain features and frequency domain features;

[0118] inputting the feature vector into a pre-trained SVM model, outputting a rail health state classification result, and triggering an alarm, the pre-trained SVM model being based on a support vector machine classifier, a multi-dimensional feature vector composed of time domain features and frequency domain features extracted from ultrasonic guided wave signals of standard rail samples being input data, and a pre-set damage label corresponding to a crack damage position and depth being output data, and the pre-trained SVM model being obtained by training and optimizing a kernel function and hyperparameters.

[0119] In yet another aspect, a computer program product or computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor implements an ultrasonic guided wave and support vector machine based rail crack identification method when executing the computer instructions, the method comprising:

[0120] emitting an ultrasonic guided wave at a rail web position to be measured by exciting a transducer;

[0121] collecting and digitizing the propagated guided wave signals by receiving transducers arranged at intervals with a preset interval;

[0122] preprocessing the digital signals to extract a multi-dimensional feature vector containing time domain features and frequency domain features;

[0123] inputting the feature vector into a pre-trained SVM model, outputting a rail health state classification result, and triggering an alarm, the pre-trained SVM model being based on a support vector machine classifier, a multi-dimensional feature vector composed of time domain features and frequency domain features extracted from ultrasonic guided wave signals of standard rail samples being input data, and a pre-set damage label corresponding to a crack damage position and depth being output data, and the pre-trained SVM model being obtained by training and optimizing a kernel function and hyperparameters.

[0124] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory.

[0125] By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0126] The technical features of the above embodiments can be combined in any way. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0127] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the present application. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method for identifying rail cracks based on ultrasonic guided waves and support vector machines, characterized in that, The method includes: Ultrasonic guided waves are emitted at the web position of the rail under test by an excitation transducer. The propagated guided wave signal is collected and digitized by receiving transducers arranged at preset intervals; Digital signals are preprocessed to extract multidimensional feature vectors containing time-domain and frequency-domain features; The feature vector is input into the pre-trained SVM model, which outputs the classification result of the rail health status and triggers an alarm. The pre-trained SVM model is based on the support vector machine classifier. It uses the multi-dimensional feature vector composed of time-domain and frequency-domain features extracted from the ultrasonic guided wave signal of the standard rail sample as input data and the damage label corresponding to the crack damage location and depth as pre-set output data. It is obtained by training and optimizing the kernel function and hyperparameters.

2. The rail crack identification method based on ultrasonic guided waves and support vector machines according to claim 1, characterized in that, The excitation of the ultrasonic guided wave includes: Generate an electrical pulse signal with a center frequency of 160kHz; The electrical pulse signal is amplified by a high-fidelity power amplifier; The amplified electrical pulse signal is converted into mechanical vibration by the excitation transducer and coupled to the center line of the rail web.

3. The rail crack identification method based on ultrasonic guided waves and support vector machines according to claim 1, characterized in that, The arrangement of the receiving transducers satisfies: Attach receiving transducers at 1-meter intervals along the center line of the web of the rail to be tested. The received guided wave signal is bandpass filtered and amplified with adjustable gain using a sensor signal filtering-amplification module.

4. The rail crack identification method based on ultrasonic guided waves and support vector machines according to claim 1, characterized in that, The preprocessing of the digital signal includes: An adaptive filtering algorithm is used to remove environmental and system noise. Perform baseline correction and signal smoothing operations to improve the signal-to-noise ratio.

5. The rail crack identification method based on ultrasonic guided waves and support vector machines according to claim 1, characterized in that, The construction of the pre-trained SVM model includes: Known damage states at different locations were set for standard rail samples; Acquire guided wave signals under damaged conditions and extract multidimensional feature vectors; Using damage labels as the output target, the hyperparameters are optimized using the Sigmoid kernel function or the Gaussian radial basis kernel function for training.

6. The rail crack identification method based on ultrasonic guided waves and support vector machines according to claim 1, characterized in that, The alarm triggering includes: When the output result indicates crack damage, the audible and visual alarm device will be activated in real time. The damage location classification results are transmitted to the host computer via the communication interface.

7. A rail crack identification device based on ultrasonic guided waves and support vector machines, characterized in that, The device includes: The transmitting module is used to transmit ultrasonic guided waves at the web position of the rail under test via an excitation transducer. The receiving module is used to collect and digitize the propagated guided wave signal by receiving transducers arranged at preset intervals; The extraction module is used to preprocess digital signals and extract multidimensional feature vectors containing time-domain and frequency-domain features; The output module is used to input the feature vector into the pre-trained SVM model, output the classification result of the rail health status, and trigger an alarm. The pre-trained SVM model is based on the support vector machine classifier. It uses the multi-dimensional feature vector composed of time-domain and frequency-domain features extracted from the ultrasonic guided wave signal of the standard rail sample as input data, and the damage labels corresponding to the crack damage location and depth are set in advance as output data. It is obtained by training and optimizing the kernel function and hyperparameters.

8. A rail crack identification system, characterized in that, include: The excitation amplification module is used to receive and amplify the input electrical signal, and its output is connected to the excitation transducer to act on the web of the rail. The hardware integration unit, electrically connected to the excitation amplification module, is used to generate signal excitation commands and control the excitation timing. The sensor filtering-amplification module receives the guided wave signal propagating from the rail through a sensor transducer at the input end, and performs filtering and noise reduction and signal amplification processing in sequence. The algorithm integration unit receives the output signal from the sensing filter-amplification module and integrates the following functions: Trigger the real-time monitoring process by clicking the "Start Test" button; Use the "Model Training" button to train the support vector machine classification model; Feature extraction and damage classification identification of input signals; The recognition status is displayed on the "Result Judgment" interface.

9. The rail crack identification system as described in claim 8, characterized in that: The excitation amplification module, hardware integration unit, and sensing filter-amplification module are physically connected by cables. The algorithm integration unit is an embedded device with a display screen, and its interface includes a "parameter setting" button for configuring the detection cycle. The sensing filter-amplification module is connected to at least two sensors to receive guided wave signals from different sections of the railway track in a multi-channel mode. The algorithm integration unit connects to the remote monitoring system through a multi-protocol communication interface to achieve real-time transmission of alarm information.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the rail crack identification method based on ultrasonic guided waves and support vector machines as described in any one of claims 1 to 6.

Images