Multi-scene six-degree-of-freedom training simulation method based on decoupling of high-speed train measured data

Abstract

The present application relates to the field of rail transit simulation technology, in particular to a multi-scene six-degree-of-freedom training simulation method based on decoupling of measured data of a motor train, comprising: accessing multi-channel measured data of real route operation of a motor train set, and outputting standardized multi-channel time sequence motion data; based on timestamp matching, train operation control logs, line geographic information and meteorological data, the multi-channel time sequence motion data is divided into multiple typical working condition scene segments; for each working condition scene segment, independent excitation sources of the corresponding scene are obtained through adaptive decoupling separation; motion mapping is performed on each independent excitation source after decoupling to generate platform motion control instructions conforming to hardware safety constraints; through collection of actual motion trajectories of the platform, dynamic time warping similarity and Pearson correlation coefficient are used as double indexes to quantitatively evaluate the simulation reproduction accuracy; and based on the unscented Kalman filter algorithm, real-time closed-loop compensation is performed on the platform attitude error.

Description

Technical Field

[0001] This invention relates to the field of rail transit simulation technology, specifically a multi-scenario six-degree-of-freedom training simulation method based on decoupling from actual train measurement data. Background Technology

[0002] Train driver training is a crucial aspect of ensuring railway operational safety. Current driver training primarily focuses on standardized operating procedures and signal responses. While multi-degree-of-freedom motion simulation technology is maturely applied in aviation and automotive driver training, in railway driving simulation, it largely remains at the level of basic vibration alerts. A systematic motion simulation training system addressing the swaying characteristics of trains in various scenarios has not yet been developed, resulting in limited engineering applications.

[0003] Existing high-speed train driver training simulators mainly have the following problems in terms of motion sensing: (1) Single data source: Most simulators rely on dynamic simulation software to generate synthetic data, which has a single feature and deviates significantly from the complex coupled swaying characteristics of real trains during operation, and cannot meet the requirements of accurate physical sensation reproduction; (2) Insufficient scene coverage: It cannot systematically cover the swaying characteristics of typical multi-scenes such as cross wind, tunnel micro-pressure waves, wheel-rail mismatch, and passing vehicles, and the training effect is significantly different from the actual operation. (3) Single dimension of body sensation: It mainly consists of single-axis or simple three-axis vibration, lacking comprehensive reproduction of pitch, roll and yaw three-axis attitude sway; (4) Lack of accuracy guarantee mechanism: After long-term operation, the motion deviation caused by mechanical wear cannot be automatically detected and compensated, and the motion accuracy gradually degrades with the running time, resulting in inappropriate compensation strategy. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a multi-scenario six-degree-of-freedom training simulation method based on decoupling from actual EMU test data, applied to a six-degree-of-freedom training simulation system for EMUs. The method includes: S1. Access multi-channel measured data from the actual operation of the EMU on the real route, preprocess the multi-channel measured data, and output standardized multi-channel time-series motion data; S2. Based on timestamp matching of train operation control logs, line geographic information and meteorological data, the multi-channel time-series motion data is divided into multiple typical working condition scenario segments; for each working condition scenario segment, the independent excitation source of the corresponding scenario is obtained through adaptive decoupling and separation. S3. Obtain the hardware limit constraint parameters of the six-degree-of-freedom motion platform, perform motion mapping on each decoupled independent excitation source, and generate platform motion control commands that conform to hardware safety constraints. S4. The actual motion trajectory of the platform is collected by MEMS sensors installed on the six-degree-of-freedom motion platform. The simulation reproduction accuracy is quantitatively evaluated by using dynamic time warping similarity and Pearson correlation coefficient as dual indicators. S5. Based on the unscented Kalman filter algorithm, the platform attitude error is compensated in real time using the measured data of the six-degree-of-freedom motion platform and the sensor.

[0005] The typical operating scenarios described in S2 include crosswind scenario, tunnel micro-pressure wave scenario, turnout / wheel-rail mismatch scenario, and train meeting scenario. The rules for dividing each operating scenario segment are as follows: In the crosswind scenario, meteorological data and abrupt changes in lateral acceleration are matched for extraction, and the excitation features are mainly coupled with lateral acceleration and vehicle body tilt. In the tunnel micro-pressure wave scenario, the tunnel entry and exit coordinates in the line geographic information are matched for extraction, and the excitation characteristics are mainly vertical and longitudinal transient aerodynamic impacts. Turnout / wheel-rail mismatch scenario: The turnout position of the matched line is selected, and the excitation characteristics are mainly high-frequency multi-axis vibration and vertical impact; Meeting scenario: Matching and capturing the meeting signals of trains on adjacent tracks, the excitation features are mainly transient lateral and yaw disturbances.

[0006] S2 employs a dual-track adaptive decoupling algorithm combining ICA and PCA to separate the independent excitation sources for the corresponding scenario. The specific operation is as follows: S201. Input multi-channel temporal motion data of the current scene segment; S202. Calculate the kurtosis of the measured data for each channel within the current time window; S203. Calculate the proportion of channels with kurtosis values ​​greater than a preset Gaussian threshold to obtain the ICA channel rate; if the ICA channel rate exceeds a preset proportion threshold, it is determined that the ICA mode is used for data decoupling; otherwise, the PCA mode is used to complete the data decoupling. S204. Calculate the covariance matrix of the multi-channel data, solve for the eigenvalues ​​and calculate the contribution rate of each principal component, and determine the number of independent excitation sources output after decoupling based on the cumulative contribution rate of the principal components. S205. The FastICA algorithm is used for iterative calculation to separate the independent components corresponding to the number of independent excitation sources.

[0007] In a specific implementation, the preset Gaussian threshold is 0.5, and the preset ratio threshold is 60%.

[0008] The hardware limit constraint parameters of the six-degree-of-freedom motion platform include: Translational parameters: , , ; Attitude parameters: , , ; System natural frequency: .

[0009] The motion mapping of each independent excitation source after decoupling, as described in S3, is specifically performed as follows: S301, Amplitude Normalization: Extract the original signal of the decoupled excitation source, determine the maximum amplitude of the signal within the current time window, and perform linear mapping according to the hardware displacement / angle limit of the corresponding degree of freedom to obtain the normalized excitation signal. S302, Frequency Domain Constrained Filtering: A resonant band-stop filter is used to filter out interference frequency bands in the normalized excitation signal that are close to or exceed the system's natural frequency; S303, Translational Degree of Freedom Constraint: The high-frequency components with a frequency greater than 0.5Hz in the normalized excitation signal are converted into velocity commands through a first integration and into displacement commands through a second integration; if the displacement exceeds the corresponding translational parameter limit threshold, the hyperbolic tangent function is used for nonlinear compression limiting; S304, Attitude Degrees of Freedom Constraint: Convert low-frequency components in the signal with frequencies less than 0.3Hz into attitude tilt coordination commands, and obtain angle commands through integration; if the angle exceeds the corresponding attitude parameter limit threshold, perform smoothing and amplitude limiting processing. S305, Composite Scenario Overlay: When multiple operating scenarios exist simultaneously in the same operating section, they are weighted and overlaid according to the energy proportion of each excitation source to generate composite motion control commands.

[0010] As described in S4, the simulation reproduction accuracy is quantitatively evaluated using a dual index of dynamic time warping similarity and Pearson correlation coefficient. Specifically: S401. Use the platform motion control command generated by S3 as a reference trajectory and align it precisely with the measured trajectory collected by the MEMS sensor on the time axis. S402. Based on the reference trajectory and the measured trajectory, calculate the Dynamic Time Warping (DTW) distance respectively, and obtain the trajectory similarity S after normalization. DTW And calculate the Pearson correlation coefficient r for the corresponding channels of the two trajectories; S403, When the trajectory similarity S DTW If the similarity is greater than the threshold and the Pearson correlation coefficient r is greater than the reference value, then both indicators are considered to meet the standard.

[0011] The formula for calculating the Pearson correlation coefficient r is as follows: Where n is the total number of sampling points; X i This represents the value of the i-th sampling point in the X channel of the reference trajectory; This is the arithmetic mean of all sampled points in the X channel of the reference trajectory; Y i This represents the value of the i-th sampling point in the channel corresponding to the measured trajectory Y; This represents the arithmetic mean of all sampling points in the channel corresponding to the measured trajectory Y.

[0012] The specific operation of S5 is as follows: S501, Set the initial state vector of the system The state vector includes the platform's three-dimensional translational position, three-dimensional attitude angles, and three-dimensional linear velocity; and the initial covariance matrix P(0) is set. S502, Based on the current state estimate With covariance matrix Multiple Sigma sampling points are generated according to the unscented transformation rules; S503. Substitute all Sigma sampling points into the system process model to complete state propagation and calculate the predicted state vector. With the predicted covariance matrix ; S504. Obtain measured data from the platform's end effector and construct a platform observation model; calculate the residual between the predicted observation value and the measured data based on the observation model; calculate the Kalman gain according to the prediction covariance matrix, map the residual to the state space for iterative updating, and obtain the optimal attitude estimate of the platform. ; S505: Collect static attitude data of the platform in real time through a sliding average window, and calculate and deduct the static zero drift error of the system. S506, The optimal attitude estimate value With the motion commands issued by the system The difference is calculated to obtain the real-time attitude error. The real-time attitude error is fed back to S3 for inverse kinematics parameter correction, thereby achieving closed-loop error compensation.

[0013] The present invention also provides a multi-scenario six-degree-of-freedom training simulation system for implementing the multi-scenario six-degree-of-freedom training simulation method based on decoupling from actual train test data as described above, comprising: The data preprocessing module is used to input multi-channel measured data from the actual operation of the EMU on the real route, preprocess the measured data, and output standardized multi-channel time-series motion data. The scenario decoupling module is used to match train operation control logs, line geographic information and meteorological data based on timestamps, and divide multi-channel time-series motion data into multiple typical working condition scenario segments; for each working condition scenario segment, the independent excitation source of the corresponding scenario is obtained through adaptive decoupling and separation. The motion mapping module is used to obtain the hardware limit constraint parameters of the six-degree-of-freedom motion platform, perform motion mapping on each decoupled independent excitation source, and generate platform motion control commands that conform to hardware safety constraints. The trajectory verification module is used to collect the actual motion trajectory of the platform through MEMS sensors installed on the six-degree-of-freedom motion platform. It uses dynamic time warping similarity and Pearson correlation coefficient as dual indicators to quantitatively evaluate the simulation reproduction accuracy. The compensation module is used to perform real-time closed-loop compensation for the platform attitude error based on the unscented Kalman filter algorithm and the measured data of the six-degree-of-freedom motion platform and sensors.

[0014] Beneficial effects: This invention is a multi-scenario six-degree-of-freedom training simulation method based on decoupling of actual train test data. This method directly uses actual road network test data as input, abandons the traditional dynamic simulation synthetic data, and accurately removes complex coupled excitation sources through dual-track adaptive decoupling, so as to restore the swaying characteristics of the actual train operation to the greatest extent. The body perception reproduction fit is much higher than that of traditional simulation data schemes.

[0015] Meanwhile, four typical special working conditions were identified: lateral wind, tunnel micro-pressure waves, turnout wheel-rail mismatch, and train encounters. Specific motion characteristics for each scenario were extracted and reproduced, achieving comprehensive coverage of training scenarios. Through full six-degree-of-freedom motion reproduction of X / Y / Z translation and roll / pitch / yaw attitude, combined with hardware constraint parameters, the simulation maximized the reproduction of real vehicle motion within safe boundaries, improving the immersion and realism of the training simulation. MEMS sensors were used as independent verification nodes, and the simulation reproduction accuracy was evaluated using bifinite standard quantities. Unscented Kalman filtering was used to achieve real-time closed-loop compensation for attitude errors, solving the problem of accuracy decay after long-term platform operation and ensuring long-term stability of simulation reproduction accuracy. Detailed Implementation

[0016] Exemplary embodiments of this disclosure will now be described in more detail.

[0017] Example This embodiment provides a multi-scenario six-degree-of-freedom training simulation method based on decoupling from actual EMU test data. It is applied to a six-degree-of-freedom training simulation system for EMUs, which includes a six-degree-of-freedom motion platform, an industrial control computer, MEMS sensors, a data interface unit, and a motion control card. The specific implementation steps of the method are as follows: S1. Access multi-channel measured data from the actual operation of the EMU on the real route, preprocess the multi-channel measured data, and output standardized multi-channel time-series motion data; In this embodiment, the measured data comes from the test EMU of the China Academy of Railway Sciences. The data supports binary continuous recording format and CSV table format. The multi-channel measured data includes six channels of raw data: lateral acceleration, vertical acceleration, longitudinal acceleration, roll rate, pitch rate, and yaw rate.

[0018] The system accesses experimental data from the China Academy of Railway Sciences' test EMU and automatically parses the data header information to preprocess the measured data. Its operations include: Byte order conversion: Convert the measured data from its original byte order to little-endian byte order; Physical dimension calibration: Multiply the original ADC value by the calibration coefficient to convert it into the corresponding physical quantity; Median debiasing: Calculate the median of the data from each channel and subtract it from the original data to eliminate the sensor's static zero bias; Outlier handling: For outlier data points that exceed the reasonable range of physical quantities, linear interpolation of the preceding and following valid data is used for replacement.

[0019] S2. Based on timestamp matching of train operation control logs, line geographic information and meteorological data, the multi-channel time-series motion data is divided into multiple typical working condition scenario segments; for each working condition scenario segment, the independent excitation source of the corresponding scenario is obtained through adaptive decoupling and separation. After preprocessing, based on timestamp matching of train operation control logs, line GIS information and meteorological data, the continuous measured data is divided into multiple scenarios, into four typical working condition scenario segments, including crosswind scenario, tunnel micro-pressure wave scenario, turnout / wheel-rail mismatch scenario and passing scenario.

[0020] The criteria for dividing each work scenario segment are as follows: In the crosswind scenario, when the instantaneous wind speed in the meteorological data is greater than or equal to the preset wind speed threshold and there is a sudden change in the amplitude of the crosswind acceleration, the data segments before and after the sudden change are extracted as the crosswind scenario; the core excitation of this scenario is the coupled motion of crosswind acceleration and vehicle body tilt caused by lateral aerodynamic force.

[0021] The tunnel micro-pressure wave scenario is based on the tunnel start and end coordinates in the line GIS information, matched with the train running mileage, and the data segments of 3 seconds before and after the train enters and exits the tunnel are extracted as the tunnel micro-pressure wave scenario; the excitation of this scenario is mainly vertical and longitudinal transient aerodynamic impact.

[0022] Turnout / wheel-rail mismatch scenario: Based on the turnout position coordinates, match the train running mileage and extract data segments 2 seconds before and after the turnout passes; the core excitation in this scenario is high-frequency multi-axis vibration and vertical impact caused by wheel-rail impact.

[0023] Meeting scenario: Based on the meeting signal of trains on adjacent tracks, data segments of 2 seconds before and after the two trains meet are captured; the core excitation of this scenario is the transient lateral sway and yaw disturbance caused by the air pressure wave of the meeting.

[0024] This embodiment takes a crosswind scenario as an example and uses a dual-track adaptive decoupling algorithm of ICA and PCA to separate the independent excitation sources for the corresponding scenario. The specific operation is as follows: S201. Input multi-channel time-series motion data of the crosswind scene, including lateral acceleration, vertical acceleration, and yaw rate.

[0025] S202. Calculate the kurtosis of the measured data for each channel within the current time window. The calculation formula is as follows: In the formula, x i Here, μ represents the sampled data, σ represents the standard deviation of the data, and N represents the total number of sampled points within the window.

[0026] Calculations show that in this embodiment, the lateral acceleration kurtosis Kurt=0.82 (>0.5, non-Gaussian distribution), the vertical acceleration kurtosis Kurt=0.65 (>0.5, non-Gaussian distribution), and the yaw rate kurtosis Kurt=0.31 (≤0.5, approximately Gaussian distribution).

[0027] S203. Calculate the ICA channel rate by counting the proportion of channels with a kurtosis greater than 0.5. If the ICA channel rate exceeds the preset proportion threshold, it is determined that the ICA mode is used for data decoupling; otherwise, the PCA mode is used to complete the data decoupling.

[0028] Based on the calculated peak values ​​of the measured data for each channel, the ICA channel rate = 2 / 3 > 60%, therefore it is determined that the ICA mode is used to complete the data decoupling.

[0029] S204. Calculate the multi-channel data covariance matrix, solve for the eigenvalues ​​and calculate the contribution rate of each principal component, and determine the number of independent excitation sources output after decoupling based on the cumulative contribution rate of the principal components.

[0030] The formula for calculating the contribution rate is: In the formula, l i Let i represent the i-th eigenvalue.

[0031] S205. The FastICA algorithm is used for iterative calculation to separate the independent components corresponding to the number of independent excitation sources, namely crosswind excitation source S1, vertical vibration source S2, and yaw excitation source S3. Based on the physical characteristics of the scene, it can be confirmed that S1 is the main crosswind excitation source, S2 is the vertical coupling source related to crosswind tilt, and S3 is the residual interference component.

[0032] In addition, after separating the independent excitation sources for the corresponding scenario, S2 also includes verifying the coupling interference by calculating the mutual information value between each independent excitation source.

[0033] In this embodiment, the mutual information values ​​between each independent excitation source are calculated, and I(S1;S2)=0.02bit, I(S1;S3)=0.01bit, and I(S2;S3)=0.03bit are obtained. All mutual information values ​​are less than 0.05bit, indicating that there is no obvious coupling interference between each excitation source and the decoupling quality is qualified.

[0034] Similarly, for other working conditions such as tunnel micro-pressure waves, the decoupling process is the same as above, ultimately separating the independent excitation source for the corresponding scenario.

[0035] S3. Obtain the hardware limit constraint parameters of the six-degree-of-freedom motion platform, perform motion mapping on each decoupled independent excitation source, and generate platform motion control commands that comply with hardware safety constraints.

[0036] The hardware limit constraint parameters of the six-degree-of-freedom motion platform include: Translational parameters: , , ; Attitude parameters: , , ; System natural frequency: .

[0037] The motion mapping of each independent excitation source after decoupling, as described in S3, is specifically performed as follows: S301, Amplitude Normalization: Extract the original signal of the decoupled excitation source, determine the maximum amplitude of the signal within the current time window, and perform linear mapping according to the hardware displacement / angle limit of the corresponding degree of freedom to obtain the normalized excitation signal. S302, Frequency Domain Constrained Filtering: A resonant band-stop filter is used to filter out interference frequency bands in the normalized excitation signal that are close to or exceed the system's natural frequency; S303, Translational Degree of Freedom Constraint: The high-frequency components with a frequency greater than 0.5Hz in the normalized excitation signal are converted into velocity commands through a first integration and into displacement commands through a second integration; if the displacement exceeds the corresponding translational parameter limit threshold, the hyperbolic tangent function is used for nonlinear compression limiting; S304, Attitude Degrees of Freedom Constraint: Convert low-frequency components in the signal with frequencies less than 0.3Hz into attitude tilt coordination commands, and obtain angle commands through integration; if the angle exceeds the corresponding attitude parameter limit threshold, perform smoothing and amplitude limiting processing. S305, Composite Scenario Overlay: When multiple operating scenarios exist simultaneously in the same operating section, they are weighted and overlaid according to the energy proportion of each excitation source to generate composite motion control commands.

[0038] S4. The actual motion trajectory of the platform is collected by MEMS sensors installed on the six-degree-of-freedom motion platform. The simulation reproduction accuracy is quantitatively evaluated by using dynamic time warping similarity and Pearson correlation coefficient as dual indicators.

[0039] At least two sets of MEMS sensors are fixedly installed at diagonal positions at the end of a six-degree-of-freedom motion platform to synchronously collect actual motion data of the platform in six degrees of freedom.

[0040] The quantitative evaluation of the simulation reproduction accuracy is specifically as follows: S401. Use the platform motion control command generated by S3 as a reference trajectory and align it precisely with the measured trajectory collected by the MEMS sensor on the time axis.

[0041] The standard platform motion control command, after decoupling and mapping the crosswind scene, is used as the reference trajectory S. ref The platform's measured trajectory S, collected by the sensor. mems Perform precise timeline alignment to eliminate time deviations caused by data transmission and execution delays.

[0042] S402. Based on the reference trajectory and the measured trajectory, calculate the Dynamic Time Warping (DTW) distance respectively, and obtain the trajectory similarity S after normalization. DTW And to calculate the Pearson correlation coefficient r for the corresponding channels of the two trajectories, the formula is: Where n is the total number of sampling points; X i This represents the value of the i-th sampling point in the X channel of the reference trajectory; This is the arithmetic mean of all sampled points in the X channel of the reference trajectory; Y i This represents the value of the i-th sampling point in the channel corresponding to the measured trajectory Y; This represents the arithmetic mean of all sampling points in the channel corresponding to the measured trajectory.

[0043] S403, When the trajectory similarity S DTW If the correlation coefficient is greater than 0.85 and the Pearson correlation coefficient r is greater than 0.80, then both indicators are considered to meet the standards and the simulation reproduction accuracy meets the training requirements.

[0044] S5. Based on the unscented Kalman filter algorithm, real-time closed-loop compensation is performed on the platform attitude error according to the measured data of the six-degree-of-freedom motion platform and sensors. The specific operation is as follows: S501, Set the initial state vector of the system The state vector includes the platform's three-dimensional translational position, three-dimensional attitude angles, and three-dimensional linear velocity, wherein, The platform's three-dimensional translational position. These are the roll, pitch, and yaw attitude angles, respectively. It is a triaxial linear velocity; Simultaneously, the initial covariance matrix P(0) is set to be a diagonal matrix, and the initial variance of the position terms is set to 10. -3 m 2 The initial variance of the attitude angle term is set to 10. -4 rad 2 The initial variance of the velocity term is set to 10. -2 m 2 / s 2 .

[0045] S502, Based on the current state estimate With covariance matrix Multiple Sigma sampling points are generated according to the unscented transformation rules; S503. Substitute all Sigma sampling points into the system process model to complete state propagation and calculate the predicted state vector. With the predicted covariance matrix ; The system process model is constructed based on platform spatial kinematics, specifically: updating the three-dimensional position by integrating the three-dimensional linear velocities from the previous moment over time, and updating the three-dimensional attitude angles and three-dimensional linear velocities in conjunction with the issued motion commands. Its discretized nonlinear state transition equation is: in, Let be the platform's kinematic state transition function. Input for platform motion control commands. This refers to system process noise.

[0046] The system process model is used to perform a nonlinear mapping on the Sigma sampling points, and the weighted mean and covariance of the mapped point set are calculated to obtain the predicted state vector. With the predicted covariance matrix .

[0047] S504, Obtain actual test data from the platform terminal. The platform observation model Y(k) is constructed; the residual between the predicted observation value and the measured data is calculated based on the observation model; the Kalman gain is calculated according to the prediction covariance matrix, and the residual is mapped to the state space for iterative updating to obtain the optimal attitude estimate of the platform. ; The measured data This includes the actual three-dimensional position output from the forward calculation of the platform's underlying servo encoder, and the three-dimensional attitude angles output after being acquired by the MEMS sensor and processed by the built-in algorithm; the actual three-dimensional position and the three-dimensional attitude angles together constitute a 6-dimensional observation vector. .

[0048] Constructing a platform observation model ,in To extract the linear observation matrix of position and attitude from the state vector, Measure the noise of the system.

[0049] Based on this observation model, the residual between the observed predicted values ​​and the measured data is calculated. The Kalman gain is then calculated using the predicted covariance matrix. The residual is mapped to the state space for iterative updates, yielding the optimal platform attitude estimate after filtering out high-frequency mechanical vibrations and sensor noise. .

[0050] Since the residual is the deviation of the observation space dimension, the matrix transformation is performed by using the Kalman gain matrix to convert the residual in the observation domain into the correction offset of each state component in the state vector, thereby realizing the projection of the observation error onto the space where the system state vector is located, and then iteratively correcting the predicted state vector.

[0051] S505. Collect the platform's static attitude data in real time through a sliding average window, and calculate and deduct the system's static zero drift error. S506, The optimal attitude estimate value With the motion commands issued by the system The difference is calculated to obtain the real-time attitude error. The real-time attitude error is fed back to S3 for inverse kinematics parameter correction, thereby achieving closed-loop error compensation.

[0052] Meanwhile, S5 also includes the ability to separate error sources through spectrum analysis, distinguish between thermal deformation errors and mechanical wear errors, and monitor equipment health status. By statistically analyzing the daily average platform attitude error E, a 30-day error time series is formed; this series is then subjected to Fast Fourier Transform (FFT) spectrum analysis. If the measured error spectrum shows a significant peak with an amplitude ratio >60% in the 24-hour period, and the cumulative increase rate of the average error over 30 days is <5%, the current error is determined to be mainly caused by thermal deformation due to changes in ambient temperature, with no mechanical wear on the equipment, and no hardware replacement is required. If the error spectrum does not show a significant daily periodic peak, and the cumulative increase rate of the average error over 30 days is ≥15%, the error source is determined to be mechanical wear and aging of transmission components, triggering a hardware maintenance warning and enabling predictive maintenance.

[0053] Furthermore, this embodiment also provides a multi-scenario six-degree-of-freedom training simulation system based on decoupling from actual train test data, including: The data preprocessing module is used to input multi-channel measured data from the actual operation of the EMU on the real route, and to preprocess the multi-channel measured data to output standardized multi-channel time-series motion data. The scenario decoupling module is used to match train operation control logs, line geographic information and meteorological data based on timestamps, and divide multi-channel time-series motion data into multiple typical working condition scenario segments; for each working condition scenario segment, the independent excitation source of the corresponding scenario is obtained through adaptive decoupling and separation. The motion mapping module is used to obtain the hardware limit constraint parameters of the six-degree-of-freedom motion platform, perform motion mapping on each decoupled independent excitation source, and generate platform motion control commands that conform to hardware safety constraints. The trajectory verification module is used to collect the actual motion trajectory of the platform through MEMS sensors installed on the six-degree-of-freedom motion platform. It uses dynamic time warping similarity and Pearson correlation coefficient as dual indicators to quantitatively evaluate the simulation reproduction accuracy. The compensation module is used to perform real-time closed-loop compensation for the platform attitude error based on the unscented Kalman filter algorithm and the measured data of the six-degree-of-freedom motion platform and sensors.

Claims

1. A multi-scenario six-degree-of-freedom training simulation method based on decoupling from actual EMU test data, applied to a six-degree-of-freedom training simulation system for EMUs, characterized in that, The method includes: S1. Access multi-channel measured data from the actual operation of the EMU on the real route, preprocess the multi-channel measured data, and output standardized multi-channel time-series motion data; S2. Based on timestamp matching of train operation control logs, line geographic information and meteorological data, multi-channel time-series motion data is divided into multiple typical working condition scenario segments; for each working condition scenario segment, independent excitation sources for the corresponding scenario are obtained through adaptive decoupling and separation. S3. Obtain the hardware limit constraint parameters of the six-degree-of-freedom motion platform, perform motion mapping on each decoupled independent excitation source, and generate platform motion control commands that conform to hardware safety constraints. S4. The actual motion trajectory of the platform is collected by MEMS sensors installed on the six-degree-of-freedom motion platform. The simulation reproduction accuracy is quantitatively evaluated by using dynamic time warping similarity and Pearson correlation coefficient as dual indicators. S5. Based on the unscented Kalman filter algorithm, the platform attitude error is compensated in real time using the measured data of the six-degree-of-freedom motion platform and the sensor.

2. The multi-scenario six-degree-of-freedom training simulation method according to claim 1, characterized in that, The typical operating scenarios described in S2 include crosswind scenario, tunnel micro-pressure wave scenario, turnout / wheel-rail mismatch scenario, and train meeting scenario. The rules for dividing each operating scenario segment are as follows: In the crosswind scenario, meteorological data and abrupt changes in lateral acceleration are matched for extraction, and the excitation features are mainly coupled with lateral acceleration and vehicle body tilt. In the tunnel micro-pressure wave scenario, the tunnel entry and exit coordinates in the line geographic information are matched for extraction, and the excitation characteristics are mainly vertical and longitudinal transient aerodynamic impacts. Turnout / wheel-rail mismatch scenario: The turnout position of the matched line is selected, and the excitation characteristics are mainly high-frequency multi-axis vibration and vertical impact; Meeting scenario: Matching and capturing the meeting signals of trains on adjacent tracks, the excitation features are mainly transient lateral and yaw disturbances.

3. The multi-scenario six-degree-of-freedom training simulation method according to claim 1, characterized in that, S2 employs a dual-track adaptive decoupling algorithm combining ICA and PCA to separate the independent excitation sources for the corresponding scenario. The specific operation is as follows: S201. Input multi-channel temporal motion data of the current scene segment; S202. Calculate the kurtosis of the measured data for each channel within the current time window; S203. Calculate the proportion of channels with kurtosis values ​​greater than the preset Gaussian threshold to obtain the ICA channel rate; if the ICA channel rate exceeds the preset proportion threshold, it is determined that the ICA mode is used for data decoupling. Otherwise, the PCA pattern is used to decouple the data; S204. Calculate the covariance matrix of the multi-channel data, solve for the eigenvalues ​​and calculate the contribution rate of each principal component, and determine the number of independent excitation sources output after decoupling based on the cumulative contribution rate of the principal components. S205. The FastICA algorithm is used for iterative calculation to separate the independent components corresponding to the number of independent excitation sources.

4. The multi-scenario six-degree-of-freedom training simulation method according to claim 3, characterized in that, The preset Gaussian threshold is 0.5, and the preset ratio threshold is 60%.

5. The multi-scenario six-degree-of-freedom training simulation method according to claim 1, characterized in that, The hardware limit constraint parameters of the six-degree-of-freedom motion platform include: Translational parameters: , , ; Attitude parameters: , , ; System natural frequency: .

6. The multi-scenario six-degree-of-freedom training simulation method according to claim 1, characterized in that, The motion mapping of each independent excitation source after decoupling, as described in S3, is specifically performed as follows: S301, Amplitude Normalization: Extract the original signal of the decoupled excitation source, determine the maximum amplitude of the signal within the current time window, and perform linear mapping according to the hardware displacement / angle limit of the corresponding degree of freedom to obtain the normalized excitation signal. S302, Frequency Domain Constrained Filtering: A resonant band-stop filter is used to filter out interference frequency bands in the normalized excitation signal that are close to or exceed the system's natural frequency; S303, Translational Degree of Freedom Constraint: The high-frequency components with a frequency greater than 0.5Hz in the normalized excitation signal are converted into velocity commands through a first integration and into displacement commands through a second integration; if the displacement exceeds the corresponding translational parameter limit threshold, the hyperbolic tangent function is used for nonlinear compression limiting; S304, Attitude Degrees of Freedom Constraint: Convert low-frequency components in the signal with frequencies less than 0.3Hz into attitude tilt coordination commands, and obtain angle commands through integration; if the angle exceeds the corresponding attitude parameter limit threshold, perform smoothing and amplitude limiting processing. S305, Composite Scenario Overlay: When multiple operating scenarios exist simultaneously in the same operating section, they are weighted and overlaid according to the energy proportion of each excitation source to generate composite motion control commands.

7. The multi-scenario six-degree-of-freedom training simulation method according to claim 1, characterized in that, As described in S4, the simulation reproduction accuracy is quantitatively evaluated using a dual index of dynamic time warping similarity and Pearson correlation coefficient. Specifically: S401. Use the platform motion control command generated by S3 as a reference trajectory and align it precisely with the measured trajectory collected by the MEMS sensor on the time axis. S402. Based on the reference trajectory and the measured trajectory, calculate the Dynamic Time Warping (DTW) distance respectively, and obtain the trajectory similarity S after normalization. DTW And calculate the Pearson correlation coefficient r for the corresponding channels of the two trajectories; S403, When the trajectory similarity S DTW If the similarity is greater than the threshold and the Pearson correlation coefficient r is greater than the reference value, then both indicators are considered to meet the standard.

8. The multi-scenario six-degree-of-freedom training simulation method according to claim 7, characterized in that, The formula for calculating the Pearson correlation coefficient r is as follows: Where n is the total number of sampling points; X i This represents the value of the i-th sampling point in the X channel of the reference trajectory; This is the arithmetic mean of all sampled points in the X channel of the reference trajectory; Y i This represents the value of the i-th sampling point in the channel corresponding to the measured trajectory Y; This represents the arithmetic mean of all sampling points in the channel corresponding to the measured trajectory Y.

9. The multi-scenario six-degree-of-freedom training simulation method according to claim 1, characterized in that, The specific operation of S5 is as follows: S501, Set the initial state vector of the system The state vector includes the platform's three-dimensional translational position, three-dimensional attitude angles, and three-dimensional linear velocity; and the initial covariance matrix P(0) is set. S502, Based on the current state estimate With covariance matrix Multiple Sigma sampling points are generated according to the unscented transformation rules; S503. Substitute all Sigma sampling points into the system process model to complete state propagation and calculate the predicted state vector. With the predicted covariance matrix ; S504. Obtain measured data from the platform's end effector and construct a platform observation model; calculate the residual between the predicted observation value and the measured data based on the observation model; calculate the Kalman gain according to the prediction covariance matrix, map the residual to the state space for iterative updating, and obtain the optimal attitude estimate of the platform. ; S505: Collect static attitude data of the platform in real time through a sliding average window, and calculate and deduct the static zero drift error of the system. S506, The optimal attitude estimate value With the motion commands issued by the system The difference is calculated to obtain the real-time attitude error. The real-time attitude error is fed back to S3 for inverse kinematics parameter correction, thereby achieving closed-loop error compensation.

10. A multi-scenario six-degree-of-freedom training simulation system that implements the multi-scenario six-degree-of-freedom training simulation method based on decoupling from actual train test data as described in claim 1, characterized in that, include: The data preprocessing module is used to input multi-channel measured data from the actual operation of the EMU on the real route, and to preprocess the multi-channel measured data to output standardized multi-channel time-series motion data. The scenario decoupling module is used to match train operation control logs, line geographic information and meteorological data based on timestamps, and divide multi-channel time-series motion data into multiple typical working condition scenario segments. For each work scenario segment, an independent excitation source for the corresponding scenario is obtained through adaptive decoupling and separation; The motion mapping module is used to obtain the hardware limit constraint parameters of the six-degree-of-freedom motion platform, perform motion mapping on each decoupled independent excitation source, and generate platform motion control commands that conform to hardware safety constraints. The trajectory verification module is used to collect the actual motion trajectory of the platform through MEMS sensors installed on the six-degree-of-freedom motion platform. It uses dynamic time warping similarity and Pearson correlation coefficient as dual indicators to quantitatively evaluate the simulation reproduction accuracy. The compensation module is used to perform real-time closed-loop compensation for the platform attitude error based on the unscented Kalman filter algorithm and the measured data of the six-degree-of-freedom motion platform and sensors.

Images