A track state detection system based on a non-contact power supply system

Abstract

The application discloses a track state detection system based on a non-contact power supply system, belongs to the field of track transportation operation and maintenance, and comprises an energy supply isolation module, a signal monitoring module, a temperature sensing monitoring module, an optical fiber monitoring module, a temperature rise prediction module, a detection classification module, a multi-mode fusion module, a digital twin module, a network communication module, a synchronous integration module and a closed-loop correction module. The application avoids the wear and electric shock risk of contact energy supply, improves the system operation stability and safety, realizes multi-angle perception of the track operation environment, can detect foreign matters, temperature rise and structural degradation at the same time, improves the preventive maintenance capability, significantly improves the foreign matter identification precision and robustness, provides quantitative basis for the operation and maintenance department, automatically generates a maintenance level, improves the decision scientificity, and realizes the track whole life cycle management combining virtuality and reality.

Description

Technical Field

[0001] This invention relates to the field of rail transit operation and maintenance, and in particular to a track condition detection system based on a non-contact power supply system. Background Technology

[0002] With the widespread application of rail transit in urban clusters and long-distance high-speed transportation, the safety and reliability of rail systems have become core concerns for operating departments and research institutions. Traditional track condition monitoring relies on manual inspections or contact-based detection methods, which suffer from low detection efficiency, insufficient real-time performance, and high maintenance costs. Especially in high-speed railways, subways, and unmanned lines, track faults such as abnormal temperature rise, foreign object intrusion, localized electromagnetic interference, and structural fatigue are often difficult to detect in their early stages. Once they develop into serious accidents, they will cause severe economic losses and social impacts.

[0003] A search revealed that Chinese patent CN105667540B discloses a track condition detection system. While this invention solves the technical problem of the inability to monitor track changes in real time due to the large workload and long inspection cycle of manual track inspection, it still suffers from wear and electric shock risks associated with contact power supply, reducing system stability and safety. It also fails to provide multi-angle perception of the track operating environment, reducing preventive maintenance capabilities, and cannot provide quantitative data for operation and maintenance departments, thus reducing the scientific basis of decision-making. Therefore, a track condition detection system based on a non-contact power supply system is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a track condition detection system based on a non-contact power supply system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A track condition detection system based on a non-contact power supply system includes a power supply isolation module, a signal monitoring module, a temperature sensing monitoring module, an optical fiber monitoring module, a temperature rise prediction module, a detection classification module, a multi-mode fusion module, a digital twin module, a network communication module, a synchronization integration module, and a closed-loop correction module.

[0007] The power supply isolation module is used to prevent the induced current in the temperature sensing line from causing an overload impact on the subsequent circuit.

[0008] The signal monitoring module is used to collect track voltage and current signals in real time and establish basic operating status judgment data;

[0009] The temperature sensing monitoring module is used to collect track temperature in real time through a temperature sensing wire buried on the side of the track.

[0010] The fiber optic monitoring module is used to deploy a distributed fiber optic temperature sensing network along the track and collect continuous spatial temperature distribution data.

[0011] The temperature rise prediction module identifies abnormal temperature rise trends and establishes an early fire warning mechanism based on the collected track temperature and temperature distribution data.

[0012] The detection and classification module is used to detect metallic or non-metallic foreign objects around the track and to automatically classify them.

[0013] The multi-mode fusion module is used to receive data from different sensors and perform fusion modeling;

[0014] The digital twin module is used to establish a digital twin model of the track, receive multi-source sensor data in real time, reconstruct the electromagnetic field and temperature field distribution, and generate a track health score.

[0015] The network communication module performs adaptive networking communication and distributed acquisition of various signals through low-power sensor nodes deployed along the track.

[0016] The synchronization integration module is used to uniformly store and format the multi-source heterogeneous data generated by each module.

[0017] The closed-loop correction module is based on joint modeling of historical and real-time data, and dynamically corrects the parameters of the track digital twin model and the anomaly threshold.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] This invention drives a Litz wire to form a primary-side loop using an AC constant current source, and achieves power supply and signal isolation through an isolation transformer and rectifier bridge. A resistor network is used to suppress high-frequency transient voltages, and the processed signal is finally sent to a DSP circuit for track state analysis. Subsequently, based on temperature sensing and fiber optic data, data preprocessing, sliding window segmentation, and standardization are completed. An LSTM model is used for training and prediction, and abnormal temperature rise detection and early warning are performed using residual threshold judgment. Simultaneously, in terms of foreign object detection, mutual inductance and impedance changes are monitored in real time by excitation and receiving coils, and the mutual inductance difference and impedance phase difference are extracted to distinguish between metallic and non-metallic foreign objects. An impedance phase spectrum curve is constructed, and support vector machines and convolutional neural networks are used to classify the material of the foreign object. Then, a comprehensive state vector is established, and a Kalman filter and neural network fusion unit are combined to output the location, size, and material classification probability of the foreign object. Furthermore… A digital twin model of the track is established, mapping sensor data to a geometric model. Electromagnetic and temperature fields are reconstructed through finite element analysis. Residual calibration parameters are used to generate a risk health score, and maintenance levels and operation and maintenance recommendations are generated based on the score. This avoids wear and electric shock risks associated with contact power supply, improves system stability and safety, and enables multi-angle perception of the track operating environment. It can simultaneously detect foreign objects, temperature rise, and structural degradation, enhance preventive maintenance capabilities, significantly improve the accuracy and robustness of foreign object identification, provide quantitative basis for operation and maintenance departments, and automatically generate maintenance levels to improve the scientific nature of decision-making, achieving a virtual-physical integrated full life cycle management of the track. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0021] Figure 1 This is a system block diagram of a track condition detection system based on a non-contact power supply system proposed in this invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0023] Reference Figure 1 A track condition detection system based on a non-contact power supply system includes a power supply isolation module, a signal monitoring module, a temperature sensing monitoring module, an optical fiber monitoring module, a temperature rise prediction module, a detection classification module, a multi-mode fusion module, a digital twin module, a network communication module, a synchronization integration module, and a closed-loop correction module.

[0024] The power supply isolation module is used to prevent the induced current in the temperature sensing line from causing an overload impact on the subsequent circuit.

[0025] The signal monitoring module is used to collect track voltage and current signals in real time and establish basic operating status judgment data;

[0026] The temperature sensing monitoring module is used to collect track temperature in real time through a temperature sensing wire buried on the side of the track.

[0027] The fiber optic monitoring module is used to deploy a distributed fiber optic temperature sensing network along the track and collect continuous spatial temperature distribution data.

[0028] The temperature rise prediction module identifies abnormal temperature rise trends and establishes an early fire warning mechanism based on the collected track temperature and temperature distribution data.

[0029] The detection and classification module is used to detect metallic or non-metallic foreign objects around the track and to automatically classify them.

[0030] The multi-mode fusion module is used to receive data from different sensors and perform fusion modeling;

[0031] The digital twin module is used to establish a digital twin model of the track, receive multi-source sensor data in real time, reconstruct the electromagnetic field and temperature field distribution, and generate a track health score.

[0032] The network communication module performs adaptive networking communication and distributed acquisition of various signals through low-power sensor nodes deployed along the track.

[0033] The synchronization integration module is used to uniformly store and format the multi-source heterogeneous data generated by each module.

[0034] The closed-loop correction module is based on joint modeling of historical and real-time data, and dynamically corrects the parameters of the track digital twin model and the anomaly threshold.

[0035] The power supply isolation module is used to prevent the induced current from the temperature sensing line from causing an overload impact on the subsequent circuit.

[0036] Specifically, a Litz wire is driven by an AC constant current source to form the primary track circuit. Based on the current amplitude and driving frequency set by the AC constant current source, the magnitude of the AC input current on the primary track is obtained in real time. An isolation transformer is added between the Litz wire circuit and the detection circuit. By adjusting the turns ratio of the isolation transformer, the secondary track signal is stepped down or stepped up. The isolated AC signal is then fed into a full-wave rectifier bridge composed of four diodes for voltage rectification. A resistor network is used to suppress high-frequency transient voltages in the rectified voltage. At the same time, the resistance value of the absorption resistor network is adjusted according to actual needs to dynamically adjust the energy absorption power of the resistor network. Finally, the output signal, after isolation and rectification by the isolation transformer, is sent to the DSP circuit for track state analysis.

[0037] The power supply isolation module of this invention avoids overload impact on subsequent circuits caused by induced current from the temperature sensing line. The specific steps are as follows:

[0038] S1.1: The Litz wire is driven by an AC constant current source to form a primary side loop of the track, and the magnitude of the AC input current on the primary side of the track is obtained in real time according to the current amplitude and driving frequency set by the AC constant current source.

[0039] The specific formula for calculating the AC input current on the primary side of the track is as follows: In the formula, Representing time The primary side of the track is supplied with AC current, in amperes (A). This represents the current amplitude set by the constant current source, in amperes (A). Represents the operating angular frequency, with units of rad / s, where, , This represents the driving frequency, measured in Hz.

[0040] S1.2: An isolation transformer is added between the Litz line loop and the detection circuit. By adjusting the turns ratio of the isolation transformer, the secondary side signal of the track is stepped down or stepped up. The isolated AC signal is then fed into a full-wave rectifier bridge composed of four diodes for voltage rectification.

[0041] The specific calculation formula for the isolation transformer to step down or step up the secondary signal of the track is as follows: In the formula, Representing time The secondary voltage, in volts (V); Representing time The primary voltage; Represents the number of turns in the primary winding; This represents the number of turns in the secondary winding.

[0042] The specific calculation formula for voltage rectification is as follows: In the formula, Representing time The instantaneous rectified output voltage, in volts (V); This represents the peak value of the secondary side voltage, measured in volts (V).

[0043] S1.3: The high-frequency transient voltage in the rectified voltage is suppressed by the resistor network. At the same time, the resistance value of the absorption resistor network is adjusted according to actual needs to dynamically adjust the energy absorption power of the resistor network. The output signal after isolation and rectification absorption by the isolation transformer is then sent to the DSP circuit for track state analysis.

[0044] The specific formula for calculating the energy absorption power of a resistive network is as follows: In the formula, This represents the average power consumed by the absorption resistor, measured in watts (W). This represents the resistance value of the absorption resistor, in Ω; This represents the period of the rectified voltage, measured in seconds (s).

[0045] The signal monitoring module of this invention is used to collect track voltage and current signals in real time and establish basic operating status judgment data; the temperature sensing monitoring module is used to collect track temperature in real time through temperature sensing wires buried on the side of the track; the optical fiber monitoring module is used to deploy a distributed optical fiber temperature sensing network along the track and collect continuous spatial temperature distribution data.

[0046] The temperature rise prediction module of this invention identifies abnormal temperature rise trends based on collected track temperature and temperature distribution data, and establishes an early fire warning mechanism. The specific steps of the temperature rise prediction module in identifying abnormal temperature rise trends are as follows:

[0047] S2.1: The temperature rise prediction module receives the raw temperature data collected by the temperature sensing monitoring module and the fiber optic monitoring module, and constructs a continuous time-series temperature data according to a unified sampling interval. Then, it performs missing value imputation, short-term outlier smoothing and standardization processing on the established time-series temperature data.

[0048] S2.2: Based on the preset sliding window and prediction step size, the continuous standardized temperature data is divided into multiple sample-label pairs, and each sample-label pair is divided into a training set and a test set. A temperature prediction model is built based on the LSTM model architecture, and then the training set is divided into multiple training subsets.

[0049] S2.3: Input each training subset into the temperature prediction model in sequence. The temperature prediction model processes the input training subset step by step through the forward propagation algorithm and calculates the output results of the input gate, forget gate and output gate in the temperature prediction model respectively. Then, based on the output results, calculate the candidate cell state, cell state and hidden state at the current time. Finally, at the last time step, map the final hidden state to the corresponding predicted temperature.

[0050] The specific calculation formula for the input gate is as follows: In the formula, Represents time step The activation vector of the input gate at that time; Represents the input sequence at time step The input vector at that time; The temperature prediction model represents the time step The hidden state at that time; Represents the input gate bias vector; as well as These represent the input weight matrix and the hidden weight matrix of the input gate, respectively; This represents the Sigmoid activation function.

[0051] The specific calculation formula for the forgetting gate is as follows: In the formula, Represents time step The activation vector of the forget gate at that time; Represents the forget gate bias vector; as well as These represent the input weight matrix and the hidden weight matrix of the forget gate, respectively.

[0052] The specific calculation formula for the output gate is as follows: In the formula, Represents time step The activation vector of the output gate at that time; Represents the output gate bias vector; as well as These represent the input weight matrix and the hidden weight matrix of the output gate, respectively.

[0053] The specific formula for calculating the candidate cell state is as follows: In the formula, Represents time step The state of candidate cells at that time; Represents the hyperbolic tangent activation function; The bias vector representing the candidate cell; as well as The input weight matrix and the hidden weight matrix represent candidate cells.

[0054] The specific formula for calculating cell state is as follows: In the formula, Represents time step The state of the cells at that time; Represents time step The state of the cells at that time; Represents element-wise product.

[0055] The specific formula for calculating the hidden state is as follows: In the formula, Represents time step The hidden state at that time.

[0056] S2.4: Calculate the error between the predicted temperature and the corresponding real label using the root mean square error function. Then, input the calculated error values ​​into the temperature prediction model sequentially according to the backpropagation algorithm, calculate the gradient values ​​of the error values ​​with respect to each model parameter, and optimize each model parameter by gradient descent using the Adam optimizer.

[0057] S2.5: After each round of training, the validation set is input into the temperature prediction model, and the error value of the temperature prediction model on the validation set is calculated to evaluate the performance index of the temperature prediction model after the current training is completed. If the performance index of the temperature prediction model does not reach the preset passing threshold, the model is retrained until the performance index of the temperature prediction model reaches the passing threshold.

[0058] The specific formula for calculating the residual value is as follows: In the formula, Representing the Time residuals; Represents the first on the orbit Each sensor point at time point Standardized temperature values; Representing the Each sensor point at time point Standardized temperature prediction values;

[0059] The specific calculation formula for the mean-standard deviation method is as follows: In the formula, This represents the threshold for anomaly detection; This represents the mean of the training residuals; This represents a preset multiple used to control the false positive rate; This represents the standard deviation of the training residuals.

[0060] S2.6: The newly collected temperature data are preprocessed and input into the temperature prediction model. The temperature data are processed step by step based on the forward propagation algorithm, and the temperature prediction value is output. The temperature prediction value is then standardized, the residual value at the current time is calculated, and the current anomaly judgment threshold is calculated by the mean-standard deviation method.

[0061] S2.7: Based on the continuous window strategy, check the residual values ​​of multiple sliding windows. If the number of times each residual value is higher than the current anomaly judgment threshold is greater than the preset threshold, and there are at least a preset number of adjacent sensing points that meet the continuous anomaly within the current time interval, it indicates that there is an abnormal temperature rise in the current track and triggers an anomaly alarm.

[0062] The detection and classification module of this invention is used to detect metallic or non-metallic foreign objects around the track and automatically classify them. The specific steps are as follows:

[0063] S3.1: An excitation coil is arranged around the track Litz line and an alternating magnetic field is generated by driving it with a constant current source. At the same time, a receiving coil is arranged on the other side of the track. When there are no foreign objects, the electromotive force induced in the receiving coil is determined only by the coupling magnetic flux. At this time, the electromotive force induced in the receiving coil is the reference state. At the same time, a set of test signals of different frequencies are applied to the excitation coil and the equivalent impedance values ​​at each frequency are recorded.

[0064] S3.2: Real-time monitoring of the equivalent mutual inductance coefficient of the excitation coil and the receiving coil, as well as the change in the impedance of the excitation coil itself. When a metallic or non-metallic foreign object enters the track magnetic field region, the equivalent mutual inductance coefficient of the excitation coil and the receiving coil shifts, and the equivalent impedance of the excitation coil changes.

[0065] S3.3: When a metallic or non-metallic foreign object is detected, extract the equivalent mutual inductance coefficient between the excitation coil and the receiving coil, as well as the mutual inductance difference and impedance phase difference generated by the change in the impedance of the excitation coil itself.

[0066] S3.4: When the mutual inductance difference is higher than the preset offset threshold and the impedance phase difference is greater than the preset change limit, the foreign object is determined to be a high conductivity metallic foreign object. When the mutual inductance difference is less than the preset offset threshold but the impedance phase difference is less than the preset change range, the foreign object is determined to be a low conductivity or non-metallic foreign object.

[0067] S3.5: Based on the equivalent impedance values ​​of the excitation coil impedance at different frequencies, generate the corresponding phase spectrum curves, then combine the impedance phase values ​​at different frequency points into a phase feature vector, then normalize the phase feature vector, use the normalized phase feature vector to train the support vector machine, and obtain the preliminary classification results through the support vector machine.

[0068] S3.6: The phase spectrum curve is subjected to multi-layer convolution and non-linear activation function through CNN network, and the probability distribution of each material category is output through fully connected layer. If the classification result of support vector machine is consistent with the material category with the highest probability of CNN network, it is taken as the preliminary material judgment result of the corresponding foreign object. Otherwise, the material category with the highest probability of CNN network is taken as the preliminary material judgment result of the corresponding foreign object.

[0069] The multi-mode fusion module of this invention is used to receive data from different sensors and perform fusion modeling. The specific steps are as follows:

[0070] S4.1: Collect the response of the receiving coil measured at multiple frequencies or a single frequency, and initially extract a set of electromagnetic features through the EM feature extraction operator. Then, use the thermal image processing operator to perform target detection on the thermal image frame and extract thermal image features such as thermal core, maximum temperature and hot spot area. After that, use the vibration extraction operator to extract vibration features such as vibration signal arrival time difference, amplitude spectrum and energy center for positioning and size estimation.

[0071] S4.2: Establish a comprehensive state vector for the foreign object, which includes latent variables corresponding to position information, size and material. Based on simple linear dynamic evolution, obtain the state change of the non-contact power system at the next moment. Then, based on electromagnetic features, thermal imaging features and vibration features, establish overall measurement features.

[0072] S4.3: Calculate the prior state and prior covariance matrix of the foreign object at the current time. Then, based on the prior state and prior covariance matrix at the current time, establish the innovation covariance matrix through Kalman gain and posterior estimation. At the same time, calculate the posterior state and posterior covariance matrix at the current time.

[0073] S4.4: The posterior state and posterior covariance matrix of the foreign object are concatenated with various feature data to generate comprehensive input features. Then, the comprehensive input features are imported into the neural network fusion device based on the NN structure. At the same time, the neural network fusion device processes the current comprehensive input features layer by layer and outputs the location, size and material classification probability of the foreign object.

[0074] S4.5: If the material classification probability is greater than the preset recognition threshold and is consistent with the preliminary material judgment result, the location, size, material category and material probability distribution map of the current foreign object will be directly output. If the material classification probability is greater than the preset recognition threshold but is inconsistent with the preliminary material judgment result, or if the material classification probability is less than the preset recognition threshold, manual verification will be performed.

[0075] The digital twin module of this invention is used to establish a digital twin model of an orbit, receive multi-source sensor data in real time, reconstruct the electromagnetic field and temperature field distribution, and generate an orbital health score. The specific steps are as follows:

[0076] S5.1: Establish a three-dimensional geometric model of the track and surrounding structure using simulation software, set corresponding material partitions, and set initial physical properties based on the preset spatial magnetic permeability distribution, spatial electrical conductivity distribution, and spatial thermal conductivity distribution in the simulation software;

[0077] S5.2: Map the data from various sensors to the geometric coordinate system of the digital twin model of the track, and use various data as observation vectors. Then, through finite element analysis, reconstruct the electromagnetic field distribution under a given excitation, and use the electromagnetic dissipation obtained from the electromagnetic field simulation as a heat source term. The evolution of the temperature field over time is obtained through the heat conduction equation.

[0078] S5.3: Compare the simulation results of electromagnetic field and temperature field with the actual observation results, construct a cost function, and use the minimization of the cost function value as the objective to calibrate the parameters in the orbital digital twin model, while updating them in real time.

[0079] S5.4: Using the calibrated digital twin model of the orbit, establish the temperature rise risk component, electromagnetic field strength risk component, and structural risk component respectively. Then, through weighted fusion, synthesize the risk health score of the current orbit by combining the risk components. Based on the risk health score of the current orbit, generate the corresponding maintenance level and recommendations.

[0080] The network communication module of this invention performs adaptive networking communication and distributed acquisition of various signals by deploying low-power sensor nodes along the track.

[0081] The synchronous integration module of this invention is used to uniformly store and format multi-source heterogeneous data generated by each module.

[0082] The closed-loop correction module of this invention is based on joint modeling of historical data and real-time data, and dynamically corrects the parameters of the track digital twin model and the anomaly threshold.

[0083] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A track condition detection system based on a non-contact power supply system, characterized in that, It includes a power supply isolation module, a signal monitoring module, a temperature sensing monitoring module, an optical fiber monitoring module, a temperature rise prediction module, a detection and classification module, a multi-mode fusion module, a digital twin module, a network communication module, a synchronization integration module, and a closed-loop correction module; The power supply isolation module is used to prevent the induced current from the temperature sensing line from causing an overload impact on the subsequent circuit; the specific steps are as follows. S1.1: The Litz wire is driven by an AC constant current source to form a primary side loop of the track, and the magnitude of the AC input current of the primary side of the track is obtained in real time according to the current amplitude and driving frequency set by the AC constant current source. S1.2: An isolation transformer is added between the Litz line loop and the detection circuit. By adjusting the turns ratio of the isolation transformer, the secondary side signal of the track is stepped down or stepped up. The isolated AC signal is then fed into a full-wave rectifier bridge composed of four diodes for voltage rectification. S1.3: The high-frequency transient voltage in the rectified voltage is suppressed by the resistor network. At the same time, the resistance value of the absorption resistor network is adjusted according to the actual needs to dynamically adjust the energy absorption power of the resistor network. The output signal after isolation and rectification absorption by the isolation transformer is then sent to the DSP circuit for track state analysis. The signal monitoring module is used to collect track voltage and current signals in real time and establish basic operating status judgment data; The temperature sensing monitoring module is used to collect track temperature in real time through a temperature sensing wire buried on the side of the track. The fiber optic monitoring module is used to deploy a distributed fiber optic temperature sensing network along the track and collect continuous spatial temperature distribution data. The temperature rise prediction module identifies abnormal temperature rise trends and establishes an early fire warning mechanism based on the collected track temperature and temperature distribution data; the specific steps are as follows: S2.1: The temperature rise prediction module receives the raw temperature data collected by the temperature sensing monitoring module and the fiber optic monitoring module, and constructs a continuous time-series temperature data according to a uniform sampling interval. Then, the established time-series temperature data is processed by missing value imputation, short-term outlier smoothing and standardization. S2.2: Based on the preset sliding window and prediction step size, the continuous standardized temperature data is divided into multiple sample-label pairs, and each sample-label pair is divided into a training set and a test set. A temperature prediction model is built based on the LSTM model architecture, and then the training set is divided into multiple training subsets. S2.3: Input each training subset into the temperature prediction model in sequence. The temperature prediction model processes the input training subset step by step through the forward propagation algorithm and calculates the output results of the input gate, forget gate and output gate in the temperature prediction model respectively. Then, based on each output result, calculate the candidate cell state, cell state and hidden state at the current time. In the last time step, map the final hidden state to the corresponding predicted temperature. S2.4: Calculate the error between the predicted temperature and the corresponding real label using the root mean square error function. Then, input the calculated error values ​​into the temperature prediction model sequentially according to the backpropagation algorithm, calculate the gradient values ​​of the error values ​​with respect to each model parameter, and optimize each model parameter by gradient descent using the Adam optimizer. S2.5: After each round of training, the validation set is input into the temperature prediction model, and the error value of the temperature prediction model on the validation set is calculated to evaluate the performance index of the temperature prediction model after the current training is completed. If the performance index of the temperature prediction model does not reach the preset passing threshold, the model is retrained until the performance index of the temperature prediction model reaches the passing threshold. S2.6: The newly collected temperature data are preprocessed and input into the temperature prediction model. The temperature data are processed step by step based on the forward propagation algorithm, and the temperature prediction value is output. The temperature prediction value is then standardized, the residual value at the current time is calculated, and the current anomaly judgment threshold is calculated by the mean-standard deviation method. S2.7: Based on the continuous window strategy, check the residual values ​​of multiple sliding windows. If the number of times each residual value is higher than the current anomaly judgment threshold is greater than the preset threshold, and there are at least a preset number of adjacent sensing points that meet the continuous anomaly within the current time interval, it indicates that there is an abnormal temperature rise in the current track and triggers an anomaly alarm. The detection and classification module is used to detect metallic or non-metallic foreign objects around the track and to automatically classify them. The multi-mode fusion module is used to receive data from different sensors and perform fusion modeling; The digital twin module is used to establish a digital twin model of the track, receive multi-source sensor data in real time, reconstruct the electromagnetic field and temperature field distribution, and generate a track health score. The network communication module performs adaptive networking communication and distributed acquisition of various signals through low-power sensor nodes deployed along the track. The synchronization integration module is used to uniformly store and format the multi-source heterogeneous data generated by each module. The closed-loop correction module is based on joint modeling of historical and real-time data, and dynamically corrects the parameters of the track digital twin model and the anomaly threshold.

2. The track condition detection system based on a non-contact power supply system according to claim 1, characterized in that, The specific steps of the detection and classification module in detecting and automatically classifying metallic or non-metallic foreign objects around the track are as follows: S3.1: An excitation coil is arranged around the track Litz line and an alternating magnetic field is generated by driving it with a constant current source. At the same time, a receiving coil is arranged on the other side of the track. When there are no foreign objects, the electromotive force induced in the receiving coil is determined only by the coupling magnetic flux. At this time, the electromotive force induced in the receiving coil is the reference state. At the same time, a set of test signals of different frequencies are applied to the excitation coil and the equivalent impedance values ​​at each frequency are recorded. S3.2: Real-time monitoring of the equivalent mutual inductance coefficient of the excitation coil and the receiving coil, as well as the change in the impedance of the excitation coil itself. When a metallic or non-metallic foreign object enters the track magnetic field region, the equivalent mutual inductance coefficient of the excitation coil and the receiving coil shifts, and the equivalent impedance of the excitation coil changes. S3.3: When a metallic or non-metallic foreign object is detected, extract the equivalent mutual inductance coefficient between the excitation coil and the receiving coil, as well as the mutual inductance difference and impedance phase difference generated by the change in the impedance of the excitation coil itself. S3.4: When the mutual inductance difference is higher than the preset offset threshold and the impedance phase difference is greater than the preset change limit, the foreign object is determined to be a high conductivity metallic foreign object. When the mutual inductance difference is less than the preset offset threshold but the impedance phase difference is less than the preset change range, the foreign object is determined to be a low conductivity or non-metallic foreign object. S3.5: Based on the equivalent impedance values ​​of the excitation coil impedance at different frequencies, generate the corresponding phase spectrum curves, then combine the impedance phase values ​​at different frequency points into a phase feature vector, then normalize the phase feature vector, use the normalized phase feature vector to train the support vector machine, and obtain the preliminary classification results through the support vector machine. S3.6: The phase spectrum curve is subjected to multi-layer convolution and non-linear activation function through CNN network, and the probability distribution of each material category is output through fully connected layer. If the classification result of support vector machine is consistent with the material category with the highest probability of CNN network, the material category of the corresponding foreign object is marked. Otherwise, the material category with the highest probability of CNN network is used as the material category of the corresponding foreign object.

3. The track condition detection system based on a non-contact power supply system according to claim 2, characterized in that, The specific steps of the multi-mode fusion module for fusion modeling are as follows: S4.1: Collect the response of the receiving coil measured at multiple frequencies or a single frequency, and initially extract a set of electromagnetic features through the EM feature extraction operator. Then, use the thermal image processing operator to perform target detection on the thermal image frame and extract thermal image features such as thermal core, maximum temperature and hot spot area. After that, use the vibration extraction operator to extract vibration features such as vibration signal arrival time difference, amplitude spectrum and energy center for positioning and size estimation. S4.2: Establish a comprehensive state vector for the foreign object, which includes latent variables corresponding to position information, size and material. Based on simple linear dynamic evolution, obtain the state change of the non-contact power system at the next moment. Then, based on electromagnetic features, thermal imaging features and vibration features, establish overall measurement features. S4.3: Calculate the prior state and prior covariance matrix of the foreign object at the current time. Then, based on the prior state and prior covariance matrix at the current time, establish the innovation covariance matrix through Kalman gain and posterior estimation. At the same time, calculate the posterior state and posterior covariance matrix at the current time. S4.4: The posterior state and posterior covariance matrix of the foreign object are concatenated with various feature data to generate comprehensive input features. Then, the comprehensive input features are imported into the neural network fusion device based on the NN structure. At the same time, the neural network fusion device processes the current comprehensive input features layer by layer and outputs the location, size and material classification probability of the foreign object. S4.5: If the material classification probability is greater than the preset recognition threshold and is consistent with the preliminary material judgment result, the location, size, material category and material probability distribution map of the current foreign object will be directly output. If the material classification probability is greater than the preset recognition threshold but is inconsistent with the preliminary material judgment result, or if the material classification probability is less than the preset recognition threshold, manual verification will be performed.

4. The track condition detection system based on a non-contact power supply system according to claim 1, characterized in that, The specific steps by which the digital twin module reconstructs the electromagnetic field and temperature field distribution to generate an orbital health score are as follows: S5.1: Establish a three-dimensional geometric model of the track and surrounding structure using simulation software, set corresponding material partitions, and set initial physical properties based on the preset spatial magnetic permeability distribution, spatial electrical conductivity distribution, and spatial thermal conductivity distribution in the simulation software; S5.2: Map the data from various sensors to the geometric coordinate system of the digital twin model of the track, and use various data as observation vectors. Then, through finite element analysis, reconstruct the electromagnetic field distribution under a given excitation, and use the electromagnetic dissipation obtained from the electromagnetic field simulation as a heat source term. The evolution of the temperature field over time is obtained through the heat conduction equation. S5.3: Compare the simulation results of electromagnetic field and temperature field with the actual observation results, construct a cost function, and use the minimization of the cost function value as the objective to calibrate the parameters in the orbital digital twin model, while updating them in real time. S5.4: Using the calibrated digital twin model of the orbit, establish the temperature rise risk component, electromagnetic field strength risk component, and structural risk component respectively. Then, through weighted fusion, synthesize the risk health score of the current orbit by combining the risk components. Based on the risk health score of the current orbit, generate the corresponding maintenance level and recommendations.

Images