Rail transit vehicle multi-channel signal acquisition and processing system

Abstract

This invention belongs to the field of signal acquisition and data processing technology, specifically relating to a multi-channel signal acquisition and processing system for rail transit vehicles. The system includes: an analog feature sensing device that uses an event-triggered mechanism to achieve data noise reduction and redundancy removal at the physical front end; a phase calibration processing device that constructs a virtual time axis based on a dynamic model and uses a cross-correlation algorithm to achieve logical phase synchronization of the signal; a tensor edge computing device that directly extracts deep operational features in a multi-dimensional tensor space using a lightweight neural network; and a dynamic resource scheduling device that dynamically optimizes computing power output according to the importance level of the features. Through the above solutions, this invention achieves a significant reduction in data dimensionality, eliminates clock drift caused by electromagnetic interference, improves synchronization accuracy while enabling on-demand allocation of computing resources, and enhances the system's ability to extract complex spatial correlation features of rail transit vehicles, providing a solid foundation for highly reliable predictive maintenance.

Description

Technical Field

[0001] This invention belongs to the field of signal acquisition and data processing technology, specifically relating to a multi-channel signal acquisition and processing system based on rail transit vehicles. Background Technology

[0002] With the intelligent transformation of the rail transit industry, onboard signal acquisition and processing systems play a crucial role in ensuring safe train operation and preventative maintenance. Modern rail transit vehicles typically integrate complex monitoring networks to ensure high reliability during long-term operation by monitoring the traction system, bogies, and key components in real time. In high-speed or heavy-load transportation scenarios, the system needs to simultaneously acquire and deeply analyze multi-dimensional heterogeneous signals related to operational status. This is not only crucial for early warning of vehicle malfunctions but also affects the overall operation and maintenance efficiency and operational safety of the rail transit system.

[0003] Multi-channel signal acquisition and processing technology is a key path to realizing vehicle condition monitoring. It aims to construct a feature map reflecting the vehicle's operational health by uniformly acquiring multi-source data such as vibration, temperature, and current. This technology requires the system to coordinate the sampling frequency and timing of multiple independent channels, ensuring logical alignment of multi-source data streams in both time and space dimensions under complex electromagnetic interference and variable workloads. Through the digital transformation and feature encapsulation of massive physical quantity signals, the system provides a high-fidelity data foundation for subsequent fault diagnosis and edge reasoning.

[0004] Existing technologies generally employ a full-data acquisition and linear transmission architecture, resulting in a large amount of redundant data across the entire frequency band flooding the vehicle bus. This easily leads to severe data lag and computational delays under limited onboard computing resources. Traditional acquisition schemes lack effective identification of the sparsity of key characteristic signals, causing significant resource waste when processing stable signals, while transient impact signals cannot receive millisecond-level responses due to bus congestion. Simultaneously, physical clock synchronization mechanisms are prone to non-eliminable clock drift in long-distance distribution and strong electromagnetic environments, leading to decoupling of phase consistency between multi-channel signals. This makes it difficult to accurately reconstruct the physical coupling relationships between different components, resulting in insufficient accuracy in fault location. Furthermore, traditional solutions focus on scalar data processing, failing to capture the spatial correlation between multi-dimensional signals and making it difficult to achieve robust signal feature fusion in complex operating environments. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-channel signal acquisition and processing system based on rail transit vehicles, which can solve the contradiction between full-band acquisition redundancy and key feature sparsity in the above-mentioned background technology, as well as the cross-channel phase consistency decoupling problem caused by multi-channel asynchronous sampling, and eliminate the computational delay and clock drift bias caused by data swamp.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A multi-channel signal acquisition and processing system for rail transit vehicles includes an analog feature sensing device, a phase calibration processing device, a tensor quantization edge computing device, a dynamic resource scheduling device, and a communication interaction device, wherein:

[0008] The simulated feature sensing device is used to perform real-time sensing and preliminary feature extraction of multi-source heterogeneous signals generated by rail transit vehicles during operation through a built-in cross-channel shared simulated feature extraction matrix. The multi-source heterogeneous signals include vibration signals reflecting mechanical operating conditions, temperature signals reflecting thermodynamic conditions, and current signals reflecting electrical characteristics.

[0009] The analog feature sensing device is also used to dynamically allocate sampling weights according to the preset physical properties of the signal using reconfigurable logic circuits, and monitor the rate of change of the signal through a built-in event triggering mechanism. When the rate of change of the signal exceeds the preset dynamic fluctuation threshold, it automatically switches from low-frequency monitoring mode to high-frequency capture mode, converting uniform streaming data into sparse event data, and realizing preliminary noise reduction and redundancy removal of data at the physical front end.

[0010] The phase calibration processing device is used to establish physical coupling constraints between signals from different monitoring channels based on the dynamic model of rail transit vehicles, and to construct a signal correlation phase calibration model. By analyzing the causal timing logic between sudden changes in motor current and vehicle vibration response, a unified virtual time axis is constructed at the data processing level.

[0011] The phase calibration processing device is also used to reconstruct the time domain of each channel signal with physical sampling asynchronous by using a cross-correlation algorithm. By compensating for clock drift caused by electromagnetic interference or transmission line delay on the virtual time axis, it achieves absolute phase synchronization of multi-channel signals at the logic level, ensuring the accuracy of subsequent fault source location.

[0012] The tensor edge computing device is used to encapsulate the synchronization signal stream output by the phase calibration processing device in multiple granularities according to the spatial and temporal dimensions, construct a multidimensional data tensor, and use the built-in lightweight neural network model to directly perform convolution operations in the tensor space to extract depth running features, avoiding the processing step of unpacking the tensor into scalar data and preserving the spatial correlation between signals.

[0013] The dynamic resource scheduling device is used to monitor the operating status of rail transit vehicles and the occupancy of onboard computing resources. Based on the feature importance level extracted by the tensor edge computing device, it dynamically adjusts the execution priority of computing tasks, reduces the overall computing power output of the system when the vehicle is running smoothly, and instantly improves computing performance to meet real-time processing needs when abnormal features are detected.

[0014] The communication interaction device is used to establish a logical connection between the vehicle system and the remote monitoring platform. Based on the decision instructions of the dynamic resource scheduling device, it selectively uploads key fault feature tensors to the cloud for archiving and analysis, or triggers immediate security protection actions locally.

[0015] Preferably, the analog feature sensing device includes a multiplexer matrix and a programmable gain control unit. The multiplexer matrix is ​​used to dynamically adjust the channel order of access to the analog-to-digital conversion unit within a preset time slice according to the urgency of each channel signal. The programmable gain control unit is used to automatically adjust the amplification factor according to the amplitude range of the signal to ensure that weak early fault signals can be completely captured.

[0016] Furthermore, the event triggering mechanism in the simulated feature sensing device adopts an adaptive threshold algorithm. The adaptive threshold is dynamically updated based on the vehicle's preset operating speed, environmental noise level, and historical sampling average. When the detected signal slope or energy integral undergoes a step change within a predetermined time and exceeds the adaptive threshold, the system immediately activates a high-precision continuous sampling process.

[0017] Furthermore, the signal correlation phase calibration model in the phase calibration processing device compensates for the spatial propagation delay of sensors at different physical locations by introducing a nonlinear dynamic constraint term. The nonlinear dynamic constraint term is determined based on the train's wheelbase, bogie geometric parameters, and the preset propagation speed of the signal in the metal components.

[0018] Furthermore, when constructing the virtual time axis, the phase calibration processing device identifies the characteristic anchor points in the signal stream, calculates the time offset of each channel's characteristic anchor point relative to the reference channel, and uses a multi-order interpolation algorithm to resample the data points of non-reference channels to eliminate the accumulated phase error caused by the inconsistency of the physical crystal oscillator frequency.

[0019] Preferably, the lightweight neural network in the tensor edge computing device adopts a depthwise separable convolutional structure to maintain sensitivity to multidimensional tensor features while reducing the number of parameters. The input layer dimension of the neural network matches the number of effective channels of the simulated feature perception device, and its output layer corresponds to different vehicle operating health levels.

[0020] Furthermore, the tensor quantization edge computing device also includes a tensor reduction module, which is used to reduce the dimensionality of multidimensional data tensors by means of singular value decomposition or principal component analysis when the signal features are highly sparse, and extract the core components that reflect the system's operating state, so as to further reduce the computational load of subsequent convolution operations.

[0021] Furthermore, the dynamic resource scheduling device is equipped with an energy efficiency ratio optimization model. This model balances the real-time requirements of the system with energy consumption through a preset cost function. When the on-board battery voltage is lower than a predetermined safety threshold, the dynamic resource scheduling device will automatically compress the sampling depth of non-critical monitoring channels to prioritize signal processing related to traction safety and braking systems.

[0022] Furthermore, the communication interaction device supports a multi-path redundancy transmission protocol, which can automatically and seamlessly switch between different frequency bands or communication standards according to the signal strength of the vehicle wireless network, ensuring that critical alarm information can be sent out first through a preset low-bandwidth narrowband channel when passing through weak signal areas such as tunnels or complex terrain.

[0023] Furthermore, the system also includes a self-diagnosis and self-repair unit, which is used to monitor the electrical connection status of each sensor channel in real time. When a signal interruption or abnormal data fluctuation is detected in a specific channel and does not conform to physical laws, the self-diagnosis and self-repair unit will automatically switch to a preset redundant backup channel, or use the signal correlation phase calibration model to reverse deduce the missing physical quantity data through the signal characteristics of the associated channel.

[0024] Preferably, the reconfigurable logic circuit in the analog feature sensing device adopts a field-programmable gate array architecture, and the interconnection relationship of its internal logic units can be dynamically reorganized according to the configuration stream issued by the dynamic resource scheduling device to achieve consistent adaptation to sensor networks with different topologies.

[0025] Furthermore, when executing the cross-correlation algorithm, the phase calibration processing device sets a sliding observation window, performs a fast Fourier transform on the multi-channel signals within the window, analyzes the phase gradient of the signals in the frequency domain, and performs precise time alignment correction by inverse transforming back to the time domain.

[0026] Furthermore, the tensor-based edge computing device is equipped with a local knowledge base, which stores standard feature mode tensors of rail transit vehicles under different operating conditions. By comparing the similarity between the real-time generated feature tensors and the standard feature mode tensors, the device can achieve second-level identification of predefined faults such as common mechanical wear, electrical overheating, and sensor failure.

[0027] Furthermore, the system is also equipped with an environmental sensing unit, which is used to acquire information on the ambient temperature, humidity and track gradient outside the train, and input these environmental parameters as bias terms into the event triggering mechanism of the analog feature sensing device to eliminate the interference of environmental factors on the recognition of signal change rate.

[0028] Furthermore, during the execution of braking or emergency acceleration commands by the vehicle, the dynamic resource scheduling device will automatically lower the event trigger threshold of all channels to a preset sensitivity range, so as to achieve lossless recording of the entire transient process under drastic operating conditions.

[0029] Preferably, the communication interaction device uses a data encryption and compression algorithm when interacting with the cloud platform. Different compression ratios are selected according to the category attributes of the signal. For waveform raw data, a high proportion of lossy compression is used, while for fault judgment logic and key status parameters, lossless compression is used to ensure that information transmission efficiency is maximized under limited bandwidth conditions.

[0030] Furthermore, during the initialization phase, the system automatically performs a multi-channel synchronous performance evaluation by sending a set of preset wideband excitation simulation signals to detect the response time difference and phase offset of each channel, and then solidifies the detection results into the initial compensation parameters of the phase calibration processing device.

[0031] Furthermore, the system introduces data verification bits at each processing stage to ensure that no logical jumps or data corruption occur throughout the entire lifecycle of the signal, from front-end sensing and mid-end calibration to back-end calculation. Once verification fails, the system will automatically trigger resampling or recalculation logic to ensure the determinism of the output result.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. Achieved a significant reduction in data dimensionality. Through front-end analog feature sensing devices and a built-in event triggering mechanism, the system can effectively identify key features of signals at the physical source, transforming the originally massive full-band uniform sampling data into sparse event data containing key information. While ensuring no fault features are lost, the amount of redundant data transmitted to the bus is reduced by a predetermined percentage, relieving bandwidth pressure on the vehicle communication bus and resolving the computational latency problem caused by data swamps.

[0034] 2. Achieved synchronization accuracy exceeding physical hardware limitations. This invention abandons the traditional approach of relying solely on physical clock synchronization, constructing a virtual timeline based on a vehicle dynamics model through a phase calibration processing device. Utilizing the physical coupling relationships between signals for logical alignment, the phase synchronization error of multi-channel signals is improved from the microsecond level to the nanosecond level at the logic level for correction. Even in complex electromagnetic interference and long-distance cable transmission environments, it can still ensure phase consistency during the fusion of heterogeneous signal features, reducing the false alarm and false negative rates for fault location.

[0035] 3. Dynamic adaptive optimization of onboard computing resources was achieved. Through a dynamic resource scheduling device, the computing power output was finely controlled, improving the system's energy efficiency. During the stable phase of normal vehicle operation, the system maintains a low-power monitoring state; while during the transient process of fault evolution, the system can automatically unleash high computing power. This dynamically fluctuating computing power allocation mode, which depends on the operating state, maximizes the efficiency of the onboard embedded computing system under limited energy supply.

[0036] 4. Enhanced system capability for extracting complex spatial correlation features. By directly performing feature extraction in the multidimensional tensor space through a tensor-based edge computing device, the system can capture spatial coupling relationships between sensors that are unrecognizable under traditional scalar processing modes. Combined with the parallel processing advantages of lightweight neural networks, it achieves in-depth analysis of complex operating conditions of rail transit vehicles, providing a solid data processing foundation for achieving highly reliable predictive maintenance. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0038] Figure 2 This is a schematic diagram of the core principle framework of cross-channel phase calibration and virtual time axis construction based on physical coupling constraints in this invention;

[0039] Figure 3 This is a logical flowchart of the simulated feature perception and data sparsity processing based on the event triggering mechanism in this invention.

[0040] Figure 4 This is a flowchart illustrating the feature extraction logic based on multidimensional tensor encapsulation and lightweight neural networks in this invention.

[0041] Figure 5 This is a flowchart illustrating the logical process of dynamic scheduling and energy efficiency optimization of computing resources based on operational status awareness in this invention.

[0042] Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the vehicle system and the cloud monitoring platform in this invention. Detailed Implementation

[0043] Example 1: Please refer to the appendix Figure 1 To be continued Figure 6 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0044] The system is based on a multi-channel signal acquisition and processing system for rail transit vehicles, including an analog feature sensing device, a phase calibration processing device, a tensor quantization edge computing device, a dynamic resource scheduling device, a communication interaction device, a self-diagnosis and self-repair unit, and an environmental sensing unit.

[0045] The analog feature sensing device is used to perform real-time sensing and preliminary feature extraction of multi-source heterogeneous signals generated by rail transit vehicles during operation through a built-in cross-channel shared analog feature extraction matrix. These multi-source heterogeneous signals include vibration signals reflecting mechanical operating states, temperature signals reflecting thermodynamic states, and current signals reflecting electrical characteristics. The analog feature sensing device is connected via a physical interface layer to various sensors arranged in the vehicle bogie, traction motor, gearbox, and on-board power distribution cabinet. Vibration signals are typically acquired using piezoelectric sensors or microelectromechanical systems (MEMS) sensors. The acquired raw analog voltage signals first enter the front-end conditioning circuit of the analog feature sensing device. This front-end conditioning circuit is configured to perform impedance matching and common-mode rejection on signals with different amplitude ranges.

[0046] The analog feature sensing device is also used to dynamically allocate sampling weights based on preset physical properties of the signal using reconfigurable logic circuits. The reconfigurable logic circuits are physically implemented using a field-programmable gate array (FPGA) architecture. By constructing multiple parallel processing pipelines within the FPGA, the system can allocate higher clock frequencies and bit width weights to high-frequency vibration signals, while allocating lower polling priorities to slowly changing temperature signals. This allocation mechanism is not fixed but is reconfigured in real-time by the configuration stream issued by the dynamic resource scheduling device.

[0047] The analog feature sensing device integrates an event triggering mechanism to monitor the rate of change of the signal. When the rate of change of the signal exceeds a preset dynamic fluctuation threshold, the system automatically switches from a low-frequency monitoring mode to a high-frequency acquisition mode. This switching is manifested at the hardware level as a step increase in the sampling trigger pulse frequency of the analog-to-digital converter. In this way, the system can transform streaming data that is originally uniformly distributed along the time axis into high-density sparse event data generated only at critical moments. At the physical front end, this mechanism effectively eliminates redundant data. For example, when the train is running at a constant speed in a straight line without abnormal vibration, the system records the baseline state at a low frequency. However, once a wheel-rail impact or current surge occurs, the system can activate full-band acquisition within nanoseconds, ensuring that transient features are not missed.

[0048] The analog feature sensing device specifically includes a multiplexer matrix and a programmable gain control unit. The multiplexer matrix dynamically adjusts the channel order accessing the analog-to-digital converter (ADC) within a preset time slice based on the urgency of each channel signal. For example, for a pressure signal channel related to braking safety, its access frequency in the multiplexer matrix is ​​configured to be higher than that of ordinary auxiliary system monitoring channels. The programmable gain control unit automatically adjusts the amplification factor based on the real-time amplitude range of the signal. When the input vibration signal exhibits weak ripples due to minor wear, the programmable gain control unit automatically increases the gain to ensure that the weak early fault signal can fill the quantization range of the ADC, improving the signal-to-noise ratio after digitization.

[0049] In the simulated feature sensing device, the event triggering mechanism adopts an adaptive threshold algorithm.

[0050] In the simulated feature sensing device, the adaptive threshold algorithm is specifically implemented as follows:

[0051] In one specific embodiment, the system maintenance length is The sliding time window. For any signal channel. Real-time calculation of the root mean square value of the signal within the window. This serves as a short-term energy representation. Simultaneously, it is considered in conjunction with the vehicle's current operating speed. External ambient temperature and historical average Dynamically update the event trigger threshold for this channel. Its update logic is defined by the following formula:

[0052] ;

[0053] in, The standard deviation of the signal within the sliding window is used to characterize the current background noise benchmark. This is the first adjustment factor, used to set the trigger sensitivity relative to background noise; This is a threshold compensation function built based on a vehicle dynamics model. This function outputs a bias term based on speed, temperature, and historical averages. This is the second adjustment factor, used to balance the impact of environmental factors on the trigger threshold.

[0054] Data flow: and It is calculated in real time by the logic unit inside the simulated feature sensing device. Obtained from the vehicle communication bus (such as MVB or TRDP). Provided by the environmental sensing unit, Retrieved from system memory. Calculated. It is written in real time into the comparator for comparison with the rate of change or energy integral of the signal. When the slope of the detected signal... or energy integral At the scheduled time More than When the event occurs, it is considered a valid triggering event.

[0055] The adaptive threshold is not a fixed value, but is dynamically updated by internal logic based on the vehicle's preset operating speed, ambient noise level, and historical sampling average. The system calculates the signal standard deviation within a past sliding time window in real time and uses it as the background noise benchmark. When the detected signal slope changes rapidly within a predetermined time, and the change is proportional to the background noise benchmark by more than a preset multiple, the system determines that a valid event has occurred. Furthermore, the ambient temperature, humidity, and track gradient information acquired by the environmental sensing unit are also input as bias terms into the event triggering mechanism. For example, when the train is traveling on a steep gradient, the traction current benchmark value will naturally increase. At this time, the system will automatically raise the trigger threshold of the current channel to avoid frequent false triggers caused by normal load changes.

[0056] The phase calibration processing device is used to establish physical coupling constraints between signals from different monitoring channels based on the dynamics model of rail transit vehicles, and to construct a signal correlation phase calibration model.

[0057] When performing phase synchronization, the phase calibration processing device achieves absolute synchronization at the logical level through the following steps:

[0058] First, select the channel with the highest signal-to-noise ratio as the reference channel. For any non-reference channel The system calculates its time offset relative to the reference channel using a sliding cross-correlation algorithm. The cross-correlation function is defined as follows:

[0059] ;

[0060] in, and The reference channel and the channel to be calibrated are respectively in time. The signal value. The corresponding value when the maximum value is obtained The value is the optimal time offset between the two channels. .this It consists of two parts: the theoretical spatial delay determined by physical distance and propagation speed. And inherent errors caused by hardware clock drift. ,Right now .

[0061] Data flow: The results, calculated using a cross-correlation algorithm, are provided as input to the resampling module. Subsequently, the system uses a cubic spline interpolation algorithm to perform resampling on the non-reference channels. The data points are resampled. Assume the original sampling time is... The signal value is After calibration Signal value at time Determined by the following formula:

[0062] ;

[0063] Among them, coefficient The interpolation curve is determined by solving a system of linear equations using four adjacent raw data points, ensuring the continuity of the first and second derivatives. In this way, the time series from the non-reference channel is mapped onto a virtual time axis that references the reference channel.

[0064] For frequency domain correction, the phase calibration processing device performs a fast Fourier transform on the multi-channel signals within the sliding observation window to obtain the frequency domain representation. The phase gradient is calculated by analyzing the linear relationship of the phase spectra between different channels. Ideally, a fixed delay. This will cause the phase to change linearly with frequency, that is... The system fits the slope of this linear relationship using the least squares method. ,this This refers to the fixed time delay deviation that needs to be compensated. Finally, the time-domain compensation and frequency-domain compensation results are fused to obtain the final calibration signal.

[0065] In rail transit vehicles, the excitation sources experienced by sensors at different physical locations are often correlated. The phase calibration processing device analyzes the causal timing logic between sudden changes in motor current and the vehicle vibration response, constructing a virtual time axis at the data processing level that is unaffected by physical sampling offsets.

[0066] The signal correlation phase calibration model in the phase calibration processing device compensates for the spatial propagation delay of sensors at different physical locations by introducing nonlinear dynamic constraints. These nonlinear dynamic constraints are determined based on the train's wheelbase, bogie geometry, and a preset propagation speed of the signal within the metal components. For example, when an impact occurs on the front axle of the bogie, the impact signal propagates rearward along the car body structure. The phase calibration processing device calculates the physical distance between the front and rear axle sensors and combines this with the material's sound velocity to calculate the theoretical delay. By comparing the actually captured signal phase with this theoretical delay, the system can correct deviations caused by asynchronous sampling clocks.

[0067] When constructing the virtual time axis, the phase calibration processing device identifies characteristic anchor points in the signal stream and calculates the time offset of each channel's characteristic anchor point relative to the reference channel. These characteristic anchor points can be zero-crossing points, peak points, or specific spectral feature points of the signal.

[0068] The system utilizes a multi-order interpolation algorithm to resample data points from non-reference channels, eliminating accumulated phase errors caused by inconsistent physical crystal oscillator frequencies. During the cross-correlation algorithm execution, the phase calibration processing device performs a Fast Fourier Transform on the multi-channel signals within a sliding observation window, analyzing the signal's phase gradient in the frequency domain. By observing the linear phase shifts of different frequency components, the system can accurately identify fixed time delay deviations and perform precise time alignment correction by inverse transforming back to the time domain. This correction capability improves the phase synchronization error of multi-channel signals from the traditional microsecond level to the logic-level nanosecond level, enhancing the accuracy of fault location during subsequent multi-source information fusion.

[0069] The tensor edge computing device is used to encapsulate the synchronization signal stream output by the phase calibration processing device in multiple granularities according to the spatial and temporal dimensions to construct a multidimensional data tensor.

[0070] The tensor edge computing device constructs a multidimensional data tensor. .in, and

[0071] These represent the height and width of the spatial grid mapped from vehicle spatial coordinates (such as carriage number and bogie position), respectively. This represents the sensor type dimension (such as vibration, temperature, current). This represents the length of the time series. For example, for a vehicle with... Each carriage has [number] sections. A train with one sensor, the spatial dimension can be set to... .

[0072] The lightweight neural network employs a depthwise separable convolutional structure, and its core computation process is as follows:

[0073] First, a channel-wise convolution layer is applied to the input tensor. Each channel is convolved independently. Let the kernel size be... Then for the first Each channel outputs a feature map. for:

[0074] ;

[0075] in, For the first The convolution kernel has channels. The total number of parameters in this step is . .

[0076] Subsequently, a pointwise convolution layer uses The convolutional kernel linearly combines the feature maps output by each channel convolution along the channel dimension to generate new features. The calculation process is as follows:

[0077] ;

[0078] in, for convolution kernel, Indicates the output channel index. This is a bias term. The total number of parameters in this step is... Through this decomposition, the total number of parameters is reduced from that of traditional convolution. Reduce to This achieves lightweight model design.

[0079] Traditional signal processing often treats the data from each channel as independent one-dimensional arrays. However, this system integrates spatial coordinate information from different sensor locations, sensor type dimensional information, and continuous sampling time dimension information to form a third-order or higher-order tensor structure. This tensor representation can fully preserve the spatial correlation between signals.

[0080] The tensor-based edge computing device utilizes a built-in lightweight neural network model to directly perform convolution operations within the tensor space to extract depth-based operational features. This lightweight neural network employs a depthwise separable convolution structure, which decomposes traditional convolution operations into channel-wise and pointwise convolutions. This reduces the number of computational parameters and computational complexity while maintaining high sensitivity to multi-dimensional tensor features. The input layer dimension of the neural network is strictly matched to the number of effective channels in the simulated feature perception device, while its output layer corresponds to different vehicle operational health levels, such as normal operation, minor wear, warning status, or serious malfunction.

[0081] The tensor-based edge computing device also includes a tensor reduction module. In cases where signal features are extremely sparse, such as during prolonged stable cruising, tensor data contains a large number of zero or repetitive values. The tensor reduction module is configured to perform dimensionality reduction on the multidimensional data tensors using singular value decomposition or principal component analysis to extract the core principal components reflecting the system's operating state. This further reduces the computational load of subsequent convolution operations, enabling the edge computing unit to maintain high-frequency feature monitoring with extremely low power consumption.

[0082] The tensor-based edge computing device is also equipped with a local knowledge base. This local knowledge base stores standard feature mode tensors of rail transit vehicles under different operating conditions (such as starting, braking, high-speed travel, and curve passage). The system calculates the Euclidean distance or cosine similarity between the real-time generated feature tensor and the standard feature mode tensor to achieve second-level identification of predefined faults such as common mechanical wear, electrical overheating, and sensor failure. When the similarity is lower than a preset safety threshold, the system immediately identifies it as an unknown anomaly and triggers a deep feature extraction process.

[0083] The dynamic resource scheduling device is used to monitor the operating status of rail transit vehicles and the occupancy of onboard computing resources.

[0084] The energy efficiency ratio optimization model built into the dynamic resource scheduling device minimizes a cost function.

[0085] This balances the system's real-time requirements with energy consumption. The cost function is defined as follows:

[0086] ;

[0087] in, It's about CPU operating frequency. and task queue The function represents the system processing delay; It is about and vehicle battery voltage The function represents the system power consumption; It is about the feature set extracted by the tensor quantization edge computing device. The function is set to negative when the feature importance level is higher than a preset threshold, in order to incentivize the system to reduce latency. These are weighting coefficients used to balance the influence of the three factors.

[0088] When the vehicle is running smoothly, the system solves... To obtain the optimal processor frequency, typically the lowest operating frequency. When anomalies are detected, The item dropped rapidly, leading to The value surged, and the system response was to Upgrade to the highest setting.

[0089] The dynamic resource scheduling device incorporates an energy efficiency ratio optimization model, which balances the system's real-time requirements with energy consumption through a preset cost function. When the vehicle is running smoothly, the system detects a low rate of signal change. At this time, the scheduling device reduces the processor's operating frequency and shuts down unnecessary computing units, maintaining the overall computing power output at an extremely low level. However, once the feature importance level extracted by the tensor edge computing device reaches the alarm threshold, or if the vehicle is detected to be executing a braking command or emergency acceleration command, the scheduling device automatically lowers the event trigger threshold of all channels to a preset sensitivity range and instantly increases the operating frequency of the computing core to ensure lossless recording of transient processes under drastic operating conditions.

[0090] The dynamic resource scheduling device also monitors the voltage status of the vehicle's battery in real time. When the battery voltage falls below a predetermined safety threshold, the scheduling device automatically executes a task compression strategy to ensure driving safety. Specifically, this involves reducing the sampling depth and processing frequency of non-critical monitoring channels (such as air conditioning and lighting systems) while prioritizing signal processing tasks related to traction safety, communication safety, and braking systems.

[0091] The communication interaction device is used to establish a logical connection between the vehicle-mounted system and the remote monitoring platform. This device supports a multi-path redundancy transmission protocol and can automatically and seamlessly switch between different frequency bands or communication standards (such as rail communication bands, public mobile communication networks, or satellite communication links) based on the signal strength of the vehicle-mounted wireless network. When traversing weak signal environments such as tunnels, deep ravines, or areas with complex electromagnetic interference, the communication interaction device, according to the decision instructions of the dynamic resource scheduling device, selectively prioritizes transmitting the most critical fault feature tensors through a preset low-bandwidth narrowband channel, while temporarily storing the non-real-time raw waveform data in local memory.

[0092] When interacting with the cloud platform, the communication device employs a data encryption and compression algorithm. The system selects different compression strategies based on the signal's category. For waveform-type raw data requiring detailed retention, a high-proportion lossy compression algorithm is used to significantly reduce data volume while ensuring no loss of envelope features. For core logic instructions, key status parameters, and alarm codes related to fault diagnosis, a strict lossless compression algorithm is employed, along with multiple layers of encryption to ensure the security and accuracy of information transmission.

[0093] The self-diagnosis and self-repair unit is used to monitor the electrical connection status of each sensor channel in real time.

[0094] When a signal interruption is detected in a specific channel, the self-diagnosis and self-repair unit uses the nonlinear dynamic constraint term in the signal correlation phase calibration model to inversely deduce the missing data. Specifically, assuming the channel...

[0095] The data is missing, but it is associated with multiple channels. There exists a physical coupling relationship determined by the dynamic model. The system reconstructs the missing signal by solving the following optimization problem using calibrated and normal correlation channel signals. :

[0096] ;

[0097] in, It is derived from the aforementioned dynamic model, which involves the channel. Signal mapping to channel The transfer function, These are model parameters (such as wheelbase, mass, stiffness, etc.). This is a regularization term used to ensure the smoothness of the reconstructed signal. is the regularization coefficient. Through iterative solving, the system can estimate the missing data and use this reconstructed data as a backup until the physical channel is restored.

[0098] For example, when the vibration sensor of a certain axle box fails, the system can infer the vibration characteristics of the failed axle box based on the vibration signals of adjacent axle boxes and the transmission relationship determined by the vehicle dynamics model (such as the two-stage suspension model), thereby maintaining the continuity of monitoring.

[0099] The unit determines whether a sensor has experienced a disconnection, short circuit, or physical damage by detecting the loop resistance, capacitance distribution, and statistical distribution characteristics of the signal in the detection channel. When a signal interruption or abnormal data fluctuation is detected in a specific channel, and this fluctuation does not physically and logically conform to the coupling constraints in the phase calibration processing device, the self-diagnosis and self-repair unit automatically switches to a preset redundant backup channel. If a redundant channel is lacking, the unit uses a signal correlation phase calibration model, combined with the signal characteristics of other normally functioning associated channels, to reverse-engineer the missing physical quantity data, maintaining the continuity of the monitoring system.

[0100] The system automatically performs a multi-channel synchronization performance evaluation during the initialization phase. Specifically, the system's internal signal generation circuit sends a set of preset wideband excitation analog signals to each acquisition front-end. By detecting the response time difference and phase shift of each channel to the reference signal, the inherent latency of the entire system's hardware and software link is obtained. These detection results are then embedded into the initial compensation parameters of the phase calibration processing device to offset consistency issues caused by differences in hardware manufacturing processes.

[0101] Furthermore, the system incorporates multi-dimensional data check bits at each processing stage to ensure that no logical jumps or data corruption occur throughout the entire lifecycle of the signal, from front-end sensing and mid-end calibration to back-end computation. Each processing module verifies the integrity of the check bits when receiving data from the previous stage. If verification fails, the system will automatically trigger resampling or recalculation logic according to a strategy, or extract historical backup data from the circular buffer for overwriting, ensuring the determinism and reliability of the final output result.

[0102] Example 2: This example provides a multi-channel signal acquisition and processing system for rail transit vehicles based on distributed edge node collaboration. Unlike the centralized processing architecture of Example 1, this example focuses more on the distributed monitoring scenario of large, long-formation trains. By deploying independent edge sensing nodes in each carriage, it achieves synchronous processing of a larger number of channels.

[0103] The system is based on a multi-channel signal acquisition and processing system for rail transit vehicles, including analog feature sensing modules distributed in each carriage, a distributed phase alignment and synchronization bridge, a tensor feature aggregation center, a distributed resource coordinator, and redundant communication links.

[0104] The analog feature sensing module is deployed in the bogie control box of each carriage, and it contains multiple independent cross-channel shared analog feature extraction matrices. Due to the long wiring distances of long-formation trains, analog signals are highly susceptible to high-frequency electromagnetic interference generated by the traction inverter during transmission. Therefore, the analog feature sensing module integrates a high-performance digital isolation unit at the front end, using optocoupler or magnetic isolation technology to electrically isolate sensitive analog circuits from digital computing circuits.

[0105] Within each sensing module, reconfigurable logic circuits analyze the signal characteristics of local sensors and autonomously trigger events. The event triggering mechanism is configured with local dynamic thresholds. For example, when the axle box vibration of the fifth carriage suddenly exceeds a locally preset threshold of three times the standard deviation, the module immediately upgrades its state to high-frequency sampling mode. Unlike Embodiment 1, this module also sends a warning broadcast command to adjacent carriage modules via a high-speed industrial Ethernet bus, enabling the acquisition systems of adjacent carriages to collaboratively enter a sensitivity-enhanced state, achieving cross-space capture of vibration propagation between carriages.

[0106] The distributed phase-aligned synchronization bridge is used to achieve nanosecond-level time alignment across train carriages in the absence of a unified hardwired synchronization trigger signal. The bridge utilizes an improved precision time protocol to calculate the physical clock deviation between edge sensing nodes by periodically exchanging synchronization messages with high-precision hardware timestamps. Simultaneously, combining the dynamics model of rail transit vehicles, the bridge establishes a mechanical coupling model of signals between carriages using the physical stiffness constraints of the connecting passageways. By identifying continuous impact feature points generated when the train passes through track joints, the bridge can calculate the time-domain lag caused by the train formation length and perform dynamic alignment on a virtual time axis using a cross-correlation algorithm.

[0107] The tensor feature aggregation center is located in the central control unit of the train's head car or middle section. Preprocessed sparse event data from the edge nodes of each carriage are aggregated to the aggregation center via a backbone network. The aggregation center concatenates the local feature tensors transmitted from each node according to the vehicle formation order, constructing a global multidimensional data tensor covering the entire train's physical space.

[0108] The tensor feature aggregation center stores a global vehicle health assessment model built based on deep learning. This model is configured to perform convolution calculations directly within the global tensor space. Because the tensor structure contains the sequential numbers and relative positions of the carriages, the model can identify beat frequency phenomena or cooperative vibration modes unique to the carriages, which is impossible in a single-carriage processing architecture. To improve computational efficiency, the aggregation center adopts a heterogeneous computing architecture, with a high-performance general-purpose processor handling task flow control, while the tensor processing unit performs high-dimensional convolution operations.

[0109] The distributed resource coordinator is used to balance the computing load and communication bandwidth requirements of the entire train monitoring system. When the train is running at high speed, the signal activity of all nodes increases due to significant aerodynamic vibrations. At this time, the resource coordinator dynamically allocates the backbone network bandwidth quota based on the salience of the characteristics reported by each carriage. For carriages exhibiting obvious abnormal characteristics, the resource coordinator grants them the highest level of communication priority, allowing them to upload complete tensor data; while for carriages running smoothly, only core state parameters are required to be reported. This content-based dynamic scheduling mechanism solves the bandwidth congestion problem faced by ultra-long train formations during multi-channel full-data acquisition.

[0110] The distributed resource scheduling device also features collaborative protection logic. When the processor temperature of an edge sensing node is too high or the load exceeds the limit, the resource coordinator will transfer the node's non-core computing tasks (such as deep neural network inference) to a neighboring node with a lower load via the network for execution. The processing results are then sent back to the original node for local processing. This spatial shifting of computing tasks enhances the system's robustness in extreme environments.

[0111] The redundant communication links include an onboard fiber optic ring network and a wireless backup link. The fiber optic ring network serves as the backbone channel, responsible for high-bandwidth tensor data transmission. The wireless backup link utilizes short-range wireless connection technology between carriages to automatically establish point-to-point relays between adjacent carriages when the fiber optic ring network breaks due to mechanical force, ensuring that basic safety monitoring data can be transmitted back to the aggregation center.

[0112] In the self-diagnosis and self-repair unit of the system, in addition to the sensor health monitoring function in Embodiment 1, a monitoring function for the network topology is added. When a node is lost from the synchronization bridge, the self-diagnosis unit triggers a heartbeat reconnection mechanism and automatically performs blind alignment processing based on historical synchronization parameters. At the same time, by utilizing the global correlation model stored in the tensor feature aggregation center and through interpolation calculation of the signals of adjacent carriages, the system can complete the missing key operating features of the faulty carriage within a certain error range.

[0113] The environmental sensing unit features multi-point sampling in a distributed architecture. Environmental sensors distributed at the front and rear of the train can monitor wind speed, rainfall, and ambient temperature in real time. For example, when the train enters a tunnel, the lead train senses a violent fluctuation in pressure waves, and this information is quickly synchronized to all nodes in the train via the resource coordinator. After receiving the environmental bias command, each edge node automatically adjusts the reconfigurable logic within its analog feature sensing layer to increase the cutoff frequency of the anti-interference filter, thereby filtering out non-fault-related low-frequency vibration components caused by sudden changes in tunnel air pressure.

[0114] The initial synchronization performance evaluation of the system employs an echo positioning strategy. The system initiates a global synchronization pulse at the aggregation center, and each edge node immediately responds upon receiving the pulse. The aggregation center automatically calculates the logical topology diagram of signal flow throughout the entire train by measuring the round-trip time delay of each node and combining this with the dynamic coupling coefficients between the nodes. This process eliminates the need for manual pre-setting of the train formation sequence; the system can automatically identify the mounting positions and directions of the carriages, greatly facilitating the flexible formation and maintenance of rail transit vehicles.

[0115] Furthermore, the tensor edge computing device in this embodiment also integrates a fault evolution trend prediction module. This module, by analyzing the evolution path of characteristic tensors over a continuous time period, can predict the remaining lifespan of components in advance using a trend extrapolation algorithm before a fault reaches its trigger threshold. This predictive capability based on tensor evolution trajectories provides scientific data support for condition-based maintenance of rail transit, avoiding the problems of over-maintenance or under-maintenance.

[0116] This system also ensures data consistency through a distributed verification mechanism. Each edge node generates a comprehensive checksum before sending data, including its local physical state, timestamp, and processing logic version. Upon receiving the data, the aggregation center uses this checksum to verify whether bit flips due to electromagnetic interference occurred during distributed network transmission. If a checksum anomaly is detected, the system determines, based on the data's real-time requirements, whether to request the edge node to recalculate and resend, or to utilize the spatial redundancy of the global tensor model for error correction.

[0117] In summary, regardless of whether it's the centralized architecture of Embodiment 1 or the distributed architecture of Embodiment 2, this invention constructs a highly efficient, high-precision, and robust signal processing system from the physical layer to the logical layer by introducing simulated feature perception, phase alignment calibration, and tensor edge processing. The system achieves a leapfrog dimensionality reduction of data through an event-triggered mechanism, solving the data swamp problem; it achieves synchronization accuracy exceeding the physical limits of hardware by utilizing a virtual time axis with physical coupling constraints; and it captures complex multi-source heterogeneous features through deep learning inference in tensor space. The synergistic combination of these technologies enhances the early identification and precise location capabilities of rail transit vehicles for minor faults in complex operating environments, providing solid technical support for the safe operation of trains.

[0118] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0119] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "multiple" means two or more, unless otherwise explicitly specified.

[0120] In this specification, all references to "for" can be understood as hardware structures, firmware configurations, logic gate interconnections, or software code segments capable of performing specific functions. Their implementation can be flexibly selected and combined among field-programmable gate arrays, digital signal processors, application-specific integrated circuits, or general-purpose processors according to specific engineering requirements.

[0121] Those skilled in the art will understand that although terms such as device, module, and unit are used herein to describe the components of the system, these components are not limited to a specific physical package, but can be logically combined or physically separated as needed. For example, the analog feature sensing device and the phase calibration processing device can be implemented on the same high-performance field-programmable gate array board, or they can be distributed in different embedded processing nodes and interact with each other via a bus.

[0122] Finally, the textual description of the algorithmic logic and formulaic relationships described in this specification aims to clearly express the physical essence and engineering logic of the processing procedure. In specific implementation, all mathematical operations involved should be converted into a set of computational instructions and executed on the appropriate computing device. For example, the mentioned convolution operation should be understood as a multiplication-accumulation operation performed within a multidimensional array space and its corresponding sliding window control logic. The mentioned virtual timeline should be understood as a constant-rate digitized counting sequence used internally by the system to mark the order of data. The conventional implementation methods of these technical details are self-evident to those skilled in the art.

[0123] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-channel signal acquisition and processing system for rail transit vehicles, including: The device includes a simulated feature sensing device, a phase calibration processing device, a tensor quantization edge computing device, a dynamic resource scheduling device, and a communication interaction device. The device is characterized in that the analog feature sensing device is used to perform real-time sensing and feature extraction of multi-source heterogeneous signals of rail transit vehicles during operation through a built-in cross-channel shared analog feature extraction matrix, and to dynamically allocate sampling weights according to the physical properties of the signals using reconfigurable logic circuits, and to monitor the rate of change of the signals through a built-in event triggering mechanism to perform sampling mode switching. The phase calibration processing device is used to establish physical coupling constraints between signals from different monitoring channels based on the dynamics model of rail transit vehicles, construct a signal correlation phase calibration model, build a unified virtual time axis at the data processing level, and realize phase synchronization of multi-channel signals. The tensor edge computing device is used to encapsulate the synchronization signal stream output by the phase calibration processing device into a multidimensional data tensor, and to use a lightweight neural network model to perform convolution operations in the tensor space to extract depth running features. The dynamic resource scheduling device is used to monitor the operating status of rail transit vehicles and the occupancy of onboard computing resources, and dynamically adjust the execution priority of computing tasks according to the importance level of the deep operation characteristics. The communication interaction device is used to establish a logical connection between the vehicle system and the remote monitoring platform, and selectively upload key fault feature tensors to the cloud or trigger security protection actions locally.

2. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 1, characterized in that, The analog feature sensing device includes a multiplexer switch matrix and a programmable gain control unit; The multiplexer switch matrix is ​​used to receive urgency instructions from each channel signal and dynamically adjust the order and frequency of each channel accessing the analog-to-digital conversion unit within a preset time slice. The programmable gain control unit is used to monitor the amplitude range of the signal in real time and automatically adjust the analog amplification factor according to the amplitude range so that the amplitude of the early fault signal fills the quantization range of the analog-to-digital converter. The reconfigurable logic circuit adopts a field-programmable gate array architecture, which contains multiple parallel processing pipelines to receive the configuration stream issued by the dynamic resource scheduling device, so as to achieve consistent adaptation to sensor networks with different topologies, and to assign differentiated clock frequency weights and data bit width weights for high-frequency vibration signals and low-frequency temperature signals.

3. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 2, characterized in that, The event triggering mechanism in the simulated feature sensing device adopts an adaptive threshold algorithm; The logic unit inside the simulated feature sensing device is used to calculate the signal standard deviation within the sliding time window in real time, and uses the signal standard deviation as a background noise benchmark. Combined with the vehicle's preset running speed, historical sampling average, and environmental parameters, the adaptive threshold is dynamically updated. The analog feature sensing device is used to control the sampling trigger pulse frequency of the analog-to-digital converter to increase stepwise when the detected signal slope or energy integral changes abruptly within a predetermined time and exceeds the adaptive threshold, thereby converting uniform streaming data into sparse event data.

4. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 3, characterized in that, The signal correlation phase calibration model in the phase calibration processing device is used to compensate for the spatial propagation delay of sensors at different physical locations by introducing nonlinear dynamic constraints. The nonlinear dynamic constraint term is determined based on the train's wheelbase parameters, bogie geometry parameters, and the preset propagation speed of signals in the metal components; The phase calibration processing device is specifically used to: calculate the time offset of each channel's feature anchor point relative to the reference channel by identifying feature anchor points in the signal stream; The feature anchor points include zero-crossing points, peak points, or spectral feature points of the signal; Multi-order interpolation algorithms are used to resample data points in non-reference channels to eliminate accumulated phase errors caused by inconsistent physical crystal oscillator frequencies, thereby achieving absolute phase synchronization at the logic level.

5. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 4, characterized in that, When performing phase synchronization, the phase calibration processing device is also used to: set a sliding observation window and perform a fast Fourier transform on the multi-channel signals within the sliding observation window to analyze the phase gradient of the signals in the frequency domain. By observing the linear shift relationship of different frequency components in the phase, the fixed time delay deviation caused by electromagnetic interference or transmission line delay is identified, and time alignment correction is performed by backtracking to the time domain through inverse transformation.

6. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 5, characterized in that, The tensor-based edge computing device includes a tensor order reduction module and a local knowledge base; The tensor edge computing device is used to encapsulate spatial coordinate information from different sensor locations, sensor type dimensional information, and continuous sampling time dimensional information at multiple granularities to construct a multidimensional data tensor with an order greater than or equal to the third order. The lightweight neural network employs a depthwise separable convolutional structure, which includes channel-wise convolutional layers and pointwise convolutional layers, for directly performing multiply-accumulate operations within the tensor space. The tensor reduction module is used to reduce the dimensionality of the multidimensional data tensor by means of singular value decomposition or principal component analysis when the sparsity of signal features is higher than a preset ratio, and extract the core principal components that reflect the system's operating state. The local knowledge base is used to store standard feature mode tensors of rail transit vehicles under different operating conditions. By calculating the cosine similarity or Euclidean distance between the real-time generated feature tensor and the standard feature mode tensor, predefined faults are identified.

7. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 6, characterized in that, The dynamic resource scheduling device is equipped with an energy efficiency ratio optimization model, which balances the system's real-time requirements and energy consumption through a preset cost function. The dynamic resource scheduling device is used to: reduce the processor's operating frequency and shut down unnecessary computing units when the vehicle is running smoothly; When an abnormal feature is detected and its importance level reaches the alarm threshold, or when a vehicle executes a braking command or emergency acceleration command, the event trigger threshold is lowered to the preset sensitivity range, and the operating frequency of the computing core is increased. The dynamic resource scheduling device is also used to monitor the on-board battery voltage in real time. When the on-board battery voltage is lower than a predetermined safety threshold, it automatically reduces the sampling depth and processing frequency of non-critical monitoring channels to prioritize the signal processing tasks of the traction system, communication system, and braking system.

8. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 7, characterized in that, The communication interaction device supports a multi-path redundancy transmission protocol, which is used to automatically and seamlessly switch between dedicated track communication frequency bands, public mobile communication networks and satellite communication links based on the signal strength of the vehicle-mounted wireless network. The communication interaction device uses a data encryption and compression module when interacting with the cloud platform. The data encryption and compression module is used to: use a lossy compression algorithm for waveform raw data to reduce the data volume while preserving the envelope characteristics. The fault determination logic instructions, key status parameters and alarm codes are subjected to lossless compression algorithm and encryption processing. The communication interaction device is also used to, in a weak signal environment, according to the decision instructions of the dynamic resource scheduling device, prioritize the transmission of key fault feature tensors through a low-bandwidth narrowband channel, and transfer the non-real-time raw waveform data to the local memory.

9. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 8, characterized in that, The system also includes a self-diagnosis and self-repair unit and an environmental sensing unit; The environmental sensing unit is used to acquire information on the ambient temperature, humidity and track gradient outside the train, and inputs the environmental parameters as bias terms into the analog feature sensing device to eliminate the interference of environmental factors on the recognition of signal change rate. The self-diagnosis and self-repair unit is used to monitor the electrical connection status of each sensor channel in real time, and to determine the sensor status by detecting the loop resistance and signal statistical distribution characteristics of the channel. When a channel signal interruption or abnormal data fluctuation is detected and does not meet the coupling constraints of the signal correlation phase calibration model, the self-diagnosis and self-repair unit is used to automatically switch to the redundant backup channel, or use the signal correlation phase calibration model in combination with the signal characteristics of other related channels to reverse deduce the missing physical quantity data.

10. The multi-channel signal acquisition and processing system for rail transit vehicles according to claim 9, characterized in that, The system is equipped with an initialization synchronization performance evaluation module and a data verification module; The initialization synchronization performance evaluation module is used to control the signal generation circuit to send a set of wideband excitation analog signals to each acquisition front end during the system initialization phase. By detecting the response time difference and phase shift of each channel to the wideband excitation analog signals, the inherent delay of the entire link is obtained, and the inherent delay of the entire link is solidified as the initial compensation parameter of the phase calibration processing device. The data verification module is used to introduce multi-dimensional data verification bits in the front-end sensing, mid-end calibration and back-end calculation stages of the signal. When each processing module receives data, it verifies the integrity of the verification bits and triggers resampling logic, recalculation logic or extracts historical backup data from the circular buffer to overwrite when the verification fails.

Images