A digital twin-based urban rail vehicle driving analysis system

Abstract

This invention relates to the field of rail transit monitoring technology and discloses a digital twin-based urban rail vehicle operation analysis system, comprising: real-time acquisition of vehicle vibration and traction motor current signals through an onboard sensing unit; calculation of the energy ratio characteristics of a predefined fault frequency band by an edge feature extraction unit; and transmission of the feature data and operating condition information to a digital twin platform by a data transmission unit. The platform has a built-in digital fault tree model and identifies cross-physical domain fault propagation paths through multi-source data coupling analysis. This invention achieves collaborative diagnosis of mechanical vibration and electrical anomalies. By combining frequency band energy ratio characteristics with fault tree logic reasoning, it effectively captures electromechanical coupling fault risks that cannot be detected by traditional isolated monitoring. At the same time, it dynamically adjusts the diagnostic threshold according to the wheel-rail adhesion state, improving the diagnostic reliability under complex operating conditions.

Description

Technical Field

[0001] This invention relates to a digital twin-based urban rail vehicle driving analysis system, belonging to the field of rail transit monitoring technology. Background Technology

[0002] Currently, urban rail vehicle condition monitoring mainly adopts a subsystem independent alarm mechanism: physical quantities such as mechanical vibration and traction current are set with fixed thresholds in their respective monitoring domains. When a single signal exceeds the limit, an isolated alarm is triggered. This method exposes a fundamental limitation in typical scenarios such as electrical phase separation sections—instantaneous current fluctuations in the traction system are easily misjudged as electrical faults, and the mechanical resonance of the transmission chain excited by this is missed because it does not reach the vibration threshold. The root cause is that existing technology cannot capture the energy transfer path across physical domains.

[0003] As urban rail networks evolve towards higher frequency and more complex operating conditions, the system faces deeper contradictions: 1. Sudden changes in wheel-rail adhesion, such as rain or snow, cause distortion in the transmission efficiency of vibration signals, rendering diagnostic benchmarks based on fixed thresholds ineffective; 2. Resonance frequency shifts caused by track foundation settlement lead to the gradual inaccuracy of preset frequency band monitoring over time; 3. Industry attempts to integrate multi-source data to improve accuracy have been hampered by insufficient original signal transmission bandwidth and limitations in edge computing power.

[0004] Specifically, existing technologies suffer from the following shortcomings: fragmented processing of mechanical, electrical, and track data, making it impossible to construct a logical chain for the transmission of fault risks across physical domains; statically fixed diagnostic thresholds, leading to severe data distortion under dynamic conditions such as sudden changes in wheel-rail adhesion; and characteristic frequency shifts caused by slow deformation of infrastructure, causing preset frequency band analysis to gradually deviate from the actual fault-sensitive area. Therefore, how to establish cross-physical domain energy transfer path perception capabilities and achieve autonomous adaptation of diagnostic benchmarks to complex environments and infrastructure evolution becomes the technical problem this invention aims to solve. Summary of the Invention

[0005] This invention provides a digital twin-based urban rail vehicle driving analysis system, the main purpose of which is to solve the problems that existing domain-based monitoring mechanisms cannot capture fault propagation paths across physical domains, and that diagnostic benchmarks lack dynamic adaptability to complex environments and infrastructure evolution.

[0006] To achieve the above objectives, the present invention provides a digital twin-based urban rail vehicle driving analysis system, comprising:

[0007] The on-board sensing unit is configured to collect multi-source data in real time, including at least vehicle vibration signals and traction motor current signals.

[0008] The edge feature extraction unit is configured to process the raw data collected by the on-board sensing unit in real time, and calculate the frequency band energy ratio of the vehicle vibration signal and the frequency band energy ratio of the traction motor current signal based on multiple target frequency bands that are predefined by engineering practice and associated with the fault mode of urban rail vehicles.

[0009] The data transmission unit is configured to transmit the frequency band energy ratio of the vehicle vibration signal, the frequency band energy ratio of the traction motor current signal, and real-time vehicle operating condition data to the digital twin platform. The digital twin platform is configured to: have a built-in digital fault tree model containing multiple logic gates to define the cross-physical domain coupling logic relationship between the frequency band energy ratio and the vehicle operating condition data; receive the frequency band energy ratio and real-time vehicle operating condition data, and use the digital fault tree model to perform logical reasoning to identify potential fault risk transmission paths caused by the coupling of multiple physical domain factors; and generate diagnostic conclusions and maintenance recommendations that include cross-physical domain causal chain analysis.

[0010] Preferably, in the edge feature extraction unit, the lightweight algorithm used to calculate the frequency band energy ratio is configured to be selected from the Goertzel algorithm or a combination of digital bandpass filtering and root mean square operation. This lightweight algorithm achieves extreme simplification and efficient representation of the complexity of the original signal by calculating the energy aggregation features for a specific frequency band, and can run on a microcontroller.

[0011] Preferably, in the digital twin platform, the digital fault tree model is configured to include a logic branch for determining the risk of coupled resonance between traction motor current fluctuations and mechanical vibrations. This branch identifies the risk of coupled resonance by comparing the bandwidth energy ratio of the traction motor current signal with a first preset threshold and comparing the bandwidth energy ratio of the vehicle vibration signal with a second preset threshold.

[0012] Preferably, the digital twin platform is also configured to have an adaptive adjustment function. When the real-time vehicle operating data indicates that the vehicle is in rainy or snowy weather or passing through an electrical phase-splitting section, the digital twin platform automatically calls the corresponding sub-logic tree in the digital fault tree model or dynamically adjusts the judgment threshold used for comparison with the frequency band energy ratio.

[0013] Preferably, the edge feature extraction unit is further configured to calculate the high-frequency ripple density of the traction motor current signal and use the high-frequency ripple density as a proxy variable for the wheel-rail adhesion state; the digital twin platform is further configured to adjust the judgment threshold for comparing the frequency band energy ratio of the vehicle vibration signal in real time according to the current ripple density.

[0014] Preferably, in the digital twin platform, the adjustment module for determining the frequency band energy ratio of vehicle vibration signals is configured such that when the current ripple density... Exceeding the adhesion state threshold determined by engineering practice At the same time, the threshold for determining the frequency band energy ratio of the vehicle vibration signal is dynamically lowered. Their adjustment relationship satisfies: ,in, The adjusted judgment threshold. This is the adjustment coefficient between the sticky state proxy variable and the decision threshold. For current ripple density, The threshold for the sticky state. The threshold value before adjustment.

[0015] Preferably, the vehicle-mounted sensing unit is further configured to control the vehicle vibration sensor to operate in a preset low-frequency sampling mode to collect track residual vibration signals when the real-time vehicle operating data indicates that the vehicle is in a parked state; the edge feature extraction unit is further configured to calculate the temporal arrangement entropy that characterizes the signal complexity based on the track residual vibration signal; the digital twin platform is further configured to dynamically adjust the frequency range of the target frequency band on which the frequency band energy ratio calculation of the vehicle vibration signal is based according to the long-term change trend of the temporal arrangement entropy.

[0016] Preferably, the data transmission unit is configured to transmit the frequency band energy ratio and real-time vehicle operating condition data via a low-power wide area network (LPWAN), wherein the LPWAN is selected from LTE-M or NB-IoT.

[0017] Compared with the prior art, the beneficial effects of the present invention are:

[0018] 1. By synchronously extracting the frequency band energy ratio of vibration and current signals at the edge side, and combining it with the multi-logic gate coupling analysis of the digital fault tree on the platform side, the system for the first time dynamically correlates the energy accumulation characteristics of mechanical vibration anomalies and electrical fluctuations in specific frequency bands. When the traction motor current BER exceeds the threshold and forms a logical AND gate relationship with the vibration BER anomaly in a specific frequency band, the system automatically identifies the electromechanical resonance conduction path that cannot be captured in traditional isolated monitoring. Thus, without increasing hardware costs, it avoids the problems of missed and false alarms caused by cross-domain coupling effects. Furthermore, by using the traction motor current ripple density as a proxy variable for wheel-rail adhesion state, the platform side dynamically corrects the vibration BER judgment threshold based on this variable. When rain and snow cause a decrease in wheel-rail transmission efficiency, the system actively compensates for signal attenuation by lowering the vibration alarm threshold, keeping the diagnostic logic stable under complex working conditions. This mechanism enables the current signal to simultaneously generate implicit environmental parameters for calibrating mechanical diagnostic benchmarks in addition to undertaking the main function of monitoring, achieving a synergistic multiplication of multi-source data efficiency.

[0019] 2. By reusing vibration sensors collected during station stops to collect residual track vibrations and quantifying track foundation state changes based on temporal permutation entropy, the system transforms ultra-low frequency vibration signals into long-term trend indicators characterizing track stiffness. When the PE value in a specific section continues to rise, the platform automatically expands the target frequency band range for vibration BER analysis, enabling core diagnostic parameters to follow the resonant frequency movement state adjustment caused by track settlement, thereby avoiding the timeliness bottleneck of traditional fixed frequency band analysis decaying with operating time. Under the constraint of zero increase in onboard hardware, the system automatically switches sensors to low-frequency sampling mode based on vehicle parking status, turning idle periods into track status monitoring windows. Utilizing the extremely low computing power of the permutation entropy algorithm, real-time calculation of track foundation health is achieved on the microcontroller, enabling a single onboard terminal to simultaneously perform the dual functions of dynamic fault diagnosis and infrastructure status perception, forming a closed loop of health management for mobile nodes and fixed facilities.

[0020] 3. The digital fault tree directly outputs solutions for the source of resonant energy based on the identified risk transmission path. For example, when it is determined that the wheel-rail resonance in a specific section is induced by rail corrugation, the system automatically generates rail grinding suggestions accurate to the section number, upgrading the operation and maintenance decision from component repair to blocking the energy transmission path, effectively curbing the occurrence of multi-physical domain chain failures. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall architecture of a digital twin-based urban rail vehicle driving analysis system according to the present invention.

[0022] Figure 2 This is a graph showing the changes in the ratio of traction current to vibration energy and the fault logic output of the present invention.

[0023] Figure 3 This is a sequence diagram of the workflow of the digital twin-based urban rail vehicle driving analysis system of the present invention.

[0024] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0026] This application provides a digital twin-based urban rail vehicle operation analysis system. Its overall architecture is based on the collaborative operation of both the vehicle and cloud ends. On the vehicle terminal side, an on-board sensing unit, an edge feature extraction unit, and a data transmission unit are deployed, while a core digital twin platform is constructed in the cloud. The system's workflow begins with the on-board sensing unit synchronously collecting key physical signals of the vehicle, followed by lightweight feature extraction at the edge side. Then, the data transmission unit uploads highly condensed feature information and operating condition data. Finally, the digital twin platform's built-in digital fault tree model performs deep fusion analysis and logical reasoning, outputting diagnostic conclusions and maintenance suggestions. In actual operating scenarios, massive amounts of raw vibration and current waveform data pose a severe challenge to the computing power and transmission bandwidth of the on-board terminal. To address this challenge, the present invention is configured to perform efficient feature characterization at the source of data generation. Specifically, the vehicle-mounted sensing unit continuously collects multi-source data in real time, including at least vehicle vibration signals and traction motor current signals. Then, the edge feature extraction unit does not directly transmit these raw signals. Instead, it invokes a lightweight algorithm configured to either use the Goertzel algorithm or a combination of digital bandpass filtering and root mean square (RMS) arithmetic, which can stably run on resource-constrained vehicle microcontrollers. This algorithm calculates the band energy ratio of the vehicle vibration signal and the traction motor current signal for multiple target frequency bands predefined through extensive engineering practice and closely associated with specific fault modes. Thus, by performing energy aggregation feature calculations for specific frequency bands at the edge, effective dimensionality reduction and information representation of high-dimensional time-series data are achieved, thereby solving the dual constraints of computing power and data transmission bandwidth of the vehicle terminal without sacrificing key fault information.

[0027] Furthermore, the fundamental limitation of traditional domain-based monitoring lies in its inability to discern potential fault propagation relationships between different physical domains, such as mechanical and electrical domains. Therefore, this invention constructs a core mechanism for cross-domain coupling analysis. The data transmission unit is responsible for efficiently and reliably transmitting the bandwidth energy ratio of the vehicle vibration signal and traction motor current signal generated by the aforementioned edge computing, along with real-time vehicle operating condition data obtained from the vehicle bus network, to a remote digital twin platform via a low-power wide-area network (e.g., LTE-M or NB-IoT). The core of this platform incorporates a digital fault tree model containing multiple logic gates to rigorously define the cross-physical-domain coupling between the bandwidth energy ratio and the vehicle operating condition data. Logical relationships, for example, include a key logical branch specifically configured to determine the risk of coupled resonance between traction motor current fluctuations and mechanical vibrations. The determination rule is that when the frequency band energy ratio of the traction motor current signal received by the platform exceeds a first preset threshold, and at the same time or within a very short time window, the frequency band energy ratio of the vehicle vibration signal corresponding to the target frequency band also exceeds a second preset threshold, a logical AND gate is triggered. Through this logical reasoning based on multi-source data coupling and fault tree models, the system can accurately identify the fault risk transmission path that cannot be detected by a single monitoring method due to the coupling of multiple physical domain factors, and generate diagnostic conclusions that include cross-physical domain causal chain analysis.

[0028] Furthermore, the operating environment of urban rail vehicles is complex and variable, especially in rainy or snowy weather, where the adhesion state between the wheel and rail changes significantly, leading to distortion in the transmission efficiency of vibration signals. If a fixed static threshold is still used for diagnosis, the reliability of the monitoring results will be greatly reduced. To establish an environmentally adaptive diagnostic benchmark, this invention introduces a proxy variable for the wheel-rail adhesion state. When processing the traction motor current signal, the edge feature extraction unit is also configured to calculate its high-frequency ripple density. This high-frequency ripple density, as an effective proxy variable for the wheel-rail adhesion state, is transmitted to the digital twin platform. After receiving this ripple density data, the platform initiates an adaptive adjustment function, adjusting the judgment threshold used for comparison with the frequency band energy ratio of the vehicle vibration signal in real time based on the current ripple density. The specific adjustment procedure is determined as a precise mathematical relationship: when the current ripple density... Exceeding an adhesion state threshold determined through extensive circuit testing and engineering practice At that time, the platform dynamically lowers the threshold for determining the frequency band energy ratio of the vehicle vibration signal. Adjusted threshold Satisfy the formula ,in, This is a pre-calibrated adjustment coefficient representing the sensitivity between the adhesive state proxy variable and the judgment threshold; with this mechanism, when external operating conditions cause vibration signal attenuation, the system can actively compensate by reducing the alarm threshold, ensuring the stability and reliability of the diagnostic logic under complex operating conditions.

[0029] Furthermore, to avoid monitoring attenuation issues caused by long-term, slow deformation of infrastructure, such as resonance frequency shift caused by track foundation settlement, this invention designs a resource-reusable track state perception and model update mechanism. When the onboard perception unit detects that the vehicle's real-time operating data indicates that the vehicle is in a parked state, it automatically controls the vehicle's vibration sensor to switch to a preset low-frequency sampling mode to specifically collect residual track vibration signals. The edge feature extraction unit then calculates the temporal permutation entropy of this low-frequency signal. As a time-series complexity quantification index with extremely low computational complexity, temporal permutation entropy can run effortlessly on the onboard microcontroller. The digital twin platform can effectively characterize the changes in the stiffness and constraints of the track foundation. After long-term collection of temporal arrangement entropy data from various vehicles corresponding to specific track sections, the platform can infer the evolution of the track foundation state based on its long-term trend. When the temporal arrangement entropy of a specific section shows a continuous unidirectional deviation or abnormal fluctuation, the platform will automatically and dynamically adjust the frequency range of the target frequency band used to calculate the frequency band energy ratio of the vehicle vibration signal in that section. By transforming the idle time of vehicle parking into a track state monitoring window, the system constructs a dynamic feedback closed loop, ensuring the long-term effectiveness of the core diagnostic model.

[0030] Ultimately, the value of this invention lies not only in risk identification but also in providing closed-loop decision-making that directly guides operational practices. When the digital fault tree model identifies a clear fault risk transmission path based on the above analysis—for example, determining that the persistent wheel-rail resonance in a certain section is induced by rail corrugation of a specific wavelength—the platform is also configured to generate and output maintenance recommendations for rail grinding in that specific section based on this analysis result. These recommendations can be accurate to the track section number. In this way, operational decisions can evolve from the traditional component maintenance mode to an active intervention mode that blocks the transmission path of specific resonant energy, thereby effectively curbing the occurrence and deterioration of multi-physical domain cascading faults and achieving integrated management of the health status of vehicles and tracks.

[0031] In this scheme, the target frequency band associated with specific fault modes, such as the cardiac frequency, is... With bandwidth This is a statistical method that distributes signal power in the frequency domain by performing power spectral density analysis on bench test data or historical maintenance diagnostic data containing known faults. It identifies characteristic frequency points and their fluctuation ranges of energy concentration; the adjustment coefficient used for adaptively adjusting the diagnostic threshold is set accordingly. Its value is the current ripple density exceeding the threshold value simultaneously measured on a wheel-rail simulation test bench during the process of systematically reducing the viscosity coefficient by changing the lubricating medium. Linear regression analysis of the relative attenuation rate of vibration signal energy is a statistical method used to establish a linear relationship model between variables, resulting in the slope of the fitted straight line; while the adjustment coefficient is used to dynamically adjust the target frequency band. In the finite element simulation environment, by gradually reducing the stiffness parameters of the track foundation support, the resulting resonant frequency shift of the wheel-rail system and the slope of the rate of change of the temporal arrangement entropy of the track residual vibration are analyzed. The embedding dimension used in the calculation of temporal permutation entropy is determined by regression analysis. With time delay These two core calculation parameters are fixed at... and This setting is based on the analysis of the dynamic characteristics of vehicle vibration signals. This dimension is sufficient to unfold the main dynamic modes of the track system while the computational load is moderate. This delay can preserve the details of the signal in the high-frequency band. At the same time, considering the manufacturing tolerances and wear differences of different vehicles, the system performs a baseline calibration procedure during the initial deployment phase. This procedure requires the specified vehicle to run completely on a reference line containing typical working conditions such as traction, braking and coasting. The system automatically collects full-time domain data during this process and builds a background noise statistical database specific to that vehicle number. Subsequently, all initial judgment thresholds in the digital fault tree model will be finally individualized based on the statistical distribution of this personalized database, such as the mean plus three times the standard deviation, thereby completing the deployment of the diagnostic model from general to specific.

[0032] Example 1: In a continuously operating urban rail line, a long-standing challenge lies in a special section with both steep gradients and small-radius curves. Especially after sudden heavy rain, the rail surface in this section becomes drastically affected by the mixture of rainwater and accumulated oil, causing a sharp deterioration in wheel-rail adhesion. When a train passes through this section under traction conditions, the wheels are prone to severe dynamic slippage. This condition presents a dilemma for traditional monitoring systems: the traction motor's rapidly increasing current to resist slippage is easily misinterpreted as an electrical fault by isolated electrical monitoring systems; while the intermittent high-frequency mechanical impacts generated by slippage excitation may not have enough energy to trigger the statically set high vibration alarm threshold, leading to missed detection of potential damage to the transmission system. When a train equipped with the solution of this invention enters the aforementioned slippery curve section, its onboard sensing unit simultaneously collects the traction motor current signal and the vehicle vibration signal. At this time, the vehicle's edge feature extraction unit, while processing the traction motor current signal, simultaneously calculates its high-frequency ripple density. Due to the violent slippage of the wheels, The value spiked instantaneously and significantly exceeded the adhesion state threshold calibrated by engineering practice. After receiving this data, the digital twin platform uses its built-in digital fault tree model based on... far beyond This fact determines severe wheel-rail adhesion as the primary boundary condition of the current system. This determination then triggers the system's adaptive adjustment function, and the platform automatically invokes... The deterministic procedure, based on the real-time input The value was dynamically adjusted to lower the threshold for judging vehicle vibration signals. Thus, in this scenario, the high-frequency ripple density of the traction motor current signal is no longer regarded as an isolated electrical disturbance, but is used as an environmental state parameter for calibrating a mechanical vibration diagnostic benchmark, thereby avoiding the lack of adaptability of the static threshold under complex working conditions.

[0033] Furthermore, at the vibration detection threshold After being adaptively reduced, the mechanical vibration of the transmission chain induced by wheel-rail slippage, which was previously negligible due to its small energy amplitude, now becomes prominent and exceeds the new adjusted threshold. At the same time, the traction motor current fluctuates in response to slippage, and its frequency band energy ratio also exceeds the first preset threshold. The digital fault tree model in the digital twin platform captures the synchronous anomaly of these two in the time and frequency domains. Its logical branch for determining the risk of coupled resonance is satisfied, and the system thus generates a diagnostic conclusion that includes cross-physical domain causal chain analysis, clearly indicating that the risk originates from electromechanical coupled resonance caused by poor wheel-rail adhesion, rather than an isolated electrical or mechanical fault.

[0034] Example 2: This example aims to reproduce the risks of vibration signal distortion and missed fault detection faced by traditional fixed threshold diagnostic methods when the wheel-rail adhesion of urban rail vehicles deteriorates through a controlled bench test. Based on this, it verifies the ability of the present invention to maintain diagnostic reliability by dynamically adjusting the diagnostic threshold. The test utilizes a full-size wheel-rail simulation test platform, which can reproduce the vehicle's wheelsets, traction drive, and suspension system at a 1:1 scale. It is equipped with a high-precision torque and speed controller and can accurately simulate wheel-rail contact conditions from dry to varying degrees of wetness on the rail wheel surface through a controllable liquid spraying system. The platform is equipped with vibration sensors of the same type as the onboard sensing unit at the axle box and a current sensor at the traction motor output to collect vehicle vibration signals and traction motor current signals in real time during the test. First, to simulate a constant early mechanical fault, an external vibrator applies a weak, continuous vibration signal of a specific frequency to the axle box. The vibration energy is precisely set at the static diagnostic threshold under normal dry conditions. About 80% of that, The value itself was determined using the standard deviation method after statistical analysis of background noise data from long-term operation of the system under dry conditions. The technical trade-off in its setting lies in ensuring sensitivity to minor faults while avoiding false alarms caused by fluctuations during normal operation. Secondly, the current ripple density serves as a proxy variable for wheel-rail adhesion. The functional relationship between it and the actual adhesion coefficient and key parameters and The viscosity coefficient was determined through a series of calibration tests before the formal start of the experiment. These calibration tests involved gradually decreasing the viscosity coefficient and recording the corresponding values. Stable values ​​were obtained, thus establishing a mapping model between the two.

[0035] After the test process was started, the wheel-rail simulation test platform first operated stably under the reference conditions of dryness and high adhesion coefficient. At this time, the energy of the weak fault vibration signal at the axle box was stable, but below the static threshold. The system diagnostic status was normal. Subsequently, the liquid spraying system began injecting a water-oil mixture onto the surface of the rail wheel to simulate a continuous and gradual decrease in the viscosity coefficient. During this process, the system recorded the wheel-rail viscosity coefficient and current ripple density in real time. The experiment observed a key phenomenon: as the adhesion coefficient decreases, due to the deterioration of wheel-rail transmission efficiency, the vibration signal energy generated by the same fault source collected by the sensor shows a significant attenuation trend. Simultaneously, the current ripple density, characterizing the increased wheel-rail slippage, increases. (The data includes vibration signal energy, diagnostic results based on static thresholds and the adaptive threshold of this invention, and the results of the experiment.) The values ​​then rise significantly, and Table 1 shows a typical set of data records during this process.

[0036] Table 1: Comparison of diagnostic results under different adhesion states.

[0037]

[0038] As shown in Table 1, when the wheel-rail adhesion coefficient drops to 0.2 or below, the energy of the actual collected vibration signal has attenuated to below 6.8, which is lower than the static threshold of 10 units. This leads to missed detections in traditional diagnostic methods. However, the digital twin platform in this invention, based on the synchronously received current ripple density that has climbed to over 8.5, can detect these errors. It calculates the dynamic threshold in real time through its built-in adaptive adjustment mechanism. When the viscosity coefficient is 0.1, The value has been automatically lowered to 5.2. At this moment, the attenuated vibration signal energy of 5.9 can still effectively break through the dynamic threshold, thereby triggering an accurate alarm. The underlying mechanism of this result is that the present invention does not treat the vibration signal as an isolated observation, but uses the characteristics of the current signal to correct the diagnostic model of the vibration signal in real time, thereby compensating for the information attenuation caused by changes in the physical environment.

[0039] Example 3: This example combines Figures 1 to 3 This paper describes the implementation of a digital twin-based urban rail vehicle driving analysis system, such as... Figure 1 As shown, the system is divided into two main parts: the vehicle-mounted end and the cloud end. The vehicle-mounted end includes a vehicle-mounted sensing unit, an edge feature extraction unit, and a data transmission unit. The vehicle-mounted sensing unit is configured to collect raw signals in real time, including vehicle vibration signals collected by vibration sensors and traction motor current signals collected by current sensors. The edge feature extraction unit processes the received raw signals in real time and calculates the frequency band energy ratio, current ripple density, and temporal arrangement entropy. The data transmission unit is responsible for transmitting the feature data generated by the edge feature extraction unit, including the frequency band energy ratio, current ripple density, and temporal arrangement entropy, as well as the real-time vehicle operating data, to the cloud via a low-power wide area network (LTE-M / NB-IoT). The core component of the cloud end is a digital twin platform. This platform has a built-in digital fault tree model and uses this model to perform cross-physical domain coupled logic analysis. The digital twin platform also has an adaptive threshold adjustment function and optimizes the system through a feedback adjustment mechanism. Finally, the system can output diagnostic conclusions and maintenance suggestions.

[0040] like Figure 2 As shown in the figure, the horizontal axis represents time (seconds), and the vertical axis represents the frequency band energy ratio / logic output. The figure contains three curves: the solid black box represents the traction motor current frequency band energy ratio, the dashed box represents the vehicle vibration frequency band energy ratio, and the dashed dotted box represents the fault detection logic output. The figure shows the changing trends of the traction motor current frequency band energy ratio and the vehicle vibration frequency band energy ratio within a specific time period, as well as the output results of the fault detection logic after coupling analysis based on these energy ratios.

[0041] like Figure 3As shown, the process begins with the onboard sensing unit acquiring vibration and current signals in real time, transmitting the raw signal data to the edge feature extraction unit. The edge feature extraction unit then calculates the vibration frequency band energy ratio and the current frequency band energy ratio, and sends the feature data to the data transmission unit. The data transmission unit acquires vehicle operating condition data and transmits the data, including feature data and vehicle operating condition data, to the digital twin platform via a low-power network. The digital twin platform invokes fault diagnosis logic, performs cross-domain coupling analysis using the fault model, identifies risk transmission paths, and then returns the diagnostic results, generates a diagnostic report, and finally outputs maintenance recommendations.

[0042] Example 4: During long-term operation of a railway line, slow, non-uniform settlement will occur in the track foundation of specific sections. This settlement will lead to changes in track stiffness, which in turn will cause a drift in the resonant frequency of the wheel-rail system. For a monitoring system with a preset fixed fault characteristic frequency band, this drift will cause the preset monitoring frequency band to gradually deviate from the actual fault energy accumulation area, thereby reducing the system's detection sensitivity. To address this problem of reduced monitoring timeliness, the present invention incorporates a model adaptive adjustment closed loop based on infrastructure state perception. The starting point of this closed loop is the analysis of track residual vibration signals performed by the onboard terminal during each vehicle stop. Specifically, when the vehicle is stopped, the onboard sensing unit switches to a preset low-frequency sampling mode to collect track residual vibration signals. The edge feature extraction unit then performs time-domain permutation entropy calculation on the onboard microcontroller. The calculation procedure is as follows: First, the collected length is... Phase space reconstruction is performed on the residual vibration time series to generate a series of First, a delay vector is generated; second, the elements in each delay vector are sorted by size to determine their corresponding dimensions. One of the possible permutation patterns is determined; next, the frequency of each permutation pattern is statistically analyzed to obtain its probability distribution; finally, the entropy value of this probability distribution is calculated according to the definition of information entropy, which is the quantified value of the permutation entropy in this time domain, and this value characterizes the constraint state of the orbital system.

[0043] On the digital twin platform side, the core digital fault tree model undergoes an offline construction and parameter calibration phase before deployment. Its logical structure is based on the analysis of systemic fault modes and effects of urban rail vehicles. Specific component-level faults are treated as underlying basic events, and system-level hazards as top-level events. Fault propagation topology is constructed through AND and OR gate logic relationships. The calibration procedure for the judgment thresholds of each logic gate in the model is as follows: A large amount of current data from healthy vehicles under various typical operating conditions is collected, and the frequency band energy ratio of each target frequency band is calculated to form a benchmark database. Statistical analysis of this database is performed to determine the background noise distribution of each frequency band energy ratio under each operating condition. Finally, the judgment threshold for that frequency band is set to the mean of the background noise under its corresponding operating condition plus... Double the standard deviation, The integer value is greater than or equal to 3 to balance the detection rate and false alarm rate. Regarding the resonance frequency shift problem caused by track foundation settlement, the digital twin platform executes a long-cycle adaptive adjustment program. The platform continuously collects and stores the temporal arrangement entropy data with precise location tags uploaded by each vehicle. For each independent track segment, linear regression analysis is performed on its temporal arrangement entropy sequence within a rolling long-term window to calculate the slope of its changing trend. When the platform determines that the slope of the temporal arrangement entropy of a certain segment continuously exceeds a preset foundation condition deterioration rate threshold within multiple consecutive calculation windows, it triggers an adjustment to the monitoring parameters for that segment. The adjustment procedure is as follows: the system will adjust the center frequency of the target frequency band originally set for vehicle vibration signal analysis for that segment according to the degree of slope exceeding the limit. Fine-tuning was performed, resulting in the new center frequency. satisfy ,in, This is the numerical value of the difference between the measured slope and the degradation rate threshold. It is an adjustment coefficient that characterizes the relationship between track stiffness changes and resonance frequency shift, calibrated in advance through simulation or bench testing. Through this data-driven slow variable feedback adjustment mechanism, the present invention ensures that the diagnostic capability can autonomously adjust with the evolution of the infrastructure, thereby maintaining its effectiveness throughout its life cycle.

[0044] Example 5: To further ensure the universality and high reliability of the present invention solution under different vehicle conditions and special track environments, a baseline characteristic calibration must be performed on each vehicle to which the system will be deployed. This procedure requires the vehicle to run completely on a test track with representative operating conditions, according to a preset test program, covering traction, braking, and coasting states at different speed levels. During this period, the on-board sensing unit and edge feature extraction unit operate in learning mode, continuously collecting and calculating the full-band energy ratio of the vehicle's vibration and traction motor current on the baseline track. The collected data is transmitted to the digital twin platform, which generates a personalized background noise database for the specific vehicle. Subsequently, the platform will use this database to fine-tune the individual judgment thresholds in the general digital fault tree model, thereby avoiding individual differences caused by different vehicle manufacturing tolerances or component wear conditions, and ensuring the accuracy of the diagnostic logic.

[0045] Furthermore, to address special operating conditions such as passing through electrical phase-splitting sections, the digital fault tree model pre-sets several operating condition logic suppression branches. The real-time vehicle operating condition data received by the system through the data transmission unit includes a status flag indicating whether the vehicle is currently in an electrical phase-splitting section, obtained from the train network control system. When the digital twin platform detects that the flag is true, it automatically activates the corresponding suppression logic branch. The function of this branch is to adjust the trigger threshold of specific fault judgment branches related to sudden changes in traction motor current according to preset rules during the period when the flag is valid. When the vehicle leaves the electrical phase-splitting section and the status flag returns to false, all judgment thresholds immediately and automatically return to their normal or adaptively adjusted levels. Through this deterministic operating condition identification and logic suppression mechanism, false diagnostic reports caused by large fluctuations in electrical signals due to normal operation can be effectively avoided. These are all extended implementation methods known to those skilled in the art.

[0046] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A digital twin-based urban rail vehicle driving analysis system, characterized in that, include: The on-board sensing unit is configured to collect multi-source data in real time, including at least vehicle vibration signals and traction motor current signals. The edge feature extraction unit is configured to process the raw data collected by the on-board sensing unit in real time, and calculate the frequency band energy ratio of the vehicle vibration signal and the frequency band energy ratio of the traction motor current signal based on multiple target frequency bands that are predefined by engineering practice and associated with the fault mode of urban rail vehicles. The data transmission unit is configured to transmit the frequency band energy ratio of the vehicle vibration signal, the frequency band energy ratio of the traction motor current signal, and the real-time operating data of the vehicle to the digital twin platform. The digital twin platform is configured to: have a built-in digital fault tree model, which contains multiple logic gates to define the cross-physical domain coupling logic relationship between the frequency band energy ratio and vehicle operating condition data; receive the frequency band energy ratio and real-time vehicle operating condition data, and use the digital fault tree model to perform logical reasoning to identify potential fault risk transmission paths caused by the coupling of multiple physical domain factors; Generate diagnostic conclusions and maintenance recommendations that include cross-physical domain causal chain analysis; The edge feature extraction unit is also configured to calculate the high-frequency ripple density of the traction motor current signal and use the high-frequency ripple density as a proxy variable for the wheel-rail adhesion state; the digital twin platform is also configured to adjust the judgment threshold for comparing the frequency band energy ratio of the vehicle vibration signal in real time based on the current ripple density.

2. The urban rail vehicle driving analysis system based on digital twins according to claim 1, characterized in that, In the edge feature extraction unit, the lightweight algorithm used to calculate the frequency band energy ratio is configured to be selected from the Goertzel algorithm or a combination of digital bandpass filtering and root mean square operation. This lightweight algorithm calculates the energy aggregation features for a preset frequency band.

3. The urban rail vehicle driving analysis system based on digital twins according to claim 1, characterized in that, In the digital twin platform, the digital fault tree model is configured to include a logic branch for determining the risk of coupled resonance between traction motor current fluctuations and mechanical vibrations. It identifies the risk of coupled resonance by comparing the bandwidth energy ratio of the traction motor current signal with a first preset threshold and comparing the bandwidth energy ratio of the vehicle vibration signal with a second preset threshold.

4. The urban rail vehicle driving analysis system based on digital twin according to claim 1, characterized in that, The digital twin platform is also configured with adaptive adjustment capabilities. When real-time vehicle operating data indicates that the vehicle is in rainy or snowy weather or passing through an electrical phase-splitting section, the digital twin platform automatically calls the corresponding sub-logic tree in the digital fault tree model or dynamically adjusts the judgment threshold used for comparison with the frequency band energy ratio.

5. The urban rail vehicle driving analysis system based on digital twins according to claim 1, characterized in that, In the digital twin platform, the adjustment module for determining the frequency band energy ratio of vehicle vibration signals is configured to adjust the threshold when the current ripple density... Exceeding the adhesion state threshold determined by engineering practice At the same time, the threshold for determining the frequency band energy ratio of the vehicle vibration signal is dynamically lowered. Their adjustment relationship satisfies: ,in, The adjusted judgment threshold. This is the adjustment coefficient between the sticky state proxy variable and the decision threshold. For current ripple density, The threshold for the sticky state. The threshold value before adjustment.

6. The urban rail vehicle driving analysis system based on digital twins according to claim 1, characterized in that, The vehicle-mounted sensing unit is also configured to control the vehicle vibration sensor to operate in a preset low-frequency sampling mode to collect residual track vibration signals when the real-time vehicle operating data indicates that the vehicle is in a parked state; the edge feature extraction unit is also configured to calculate the temporal arrangement entropy that characterizes the signal complexity based on the residual track vibration signal; the digital twin platform is also configured to dynamically adjust the frequency range of the target frequency band on which the frequency band energy ratio calculation of the vehicle vibration signal is based, according to the long-term change trend of the temporal arrangement entropy.

7. The urban rail vehicle driving analysis system based on digital twins according to claim 1, characterized in that, The data transmission unit is configured to transmit the frequency band energy ratio and real-time vehicle operating condition data via a low-power wide area network (LPWAN), which is selected from LTE-M or NB-IoT.

Images