Roller coaster working state monitoring method based on multi-source data fusion and application thereof

Abstract

The application discloses a roller coaster working state monitoring method based on multi-source data fusion and application thereof, in which an IMU, a Beidou / GNSS positioning module and an odometer are non-invasively installed on a vehicle body, and running data is collected after unified time synchronization; an inertial error S domain transfer model is established, sensor precision weights are determined, and fusion calculation is carried out in combination with forward robust adaptive filtering, reverse smoothing and graph optimization; then, the vehicle pose is mapped among a global coordinate system, a Frenet coordinate system, a local coordinate system and an arc length coordinate system, track linearity, curvature, displacement and flatness are inversed, working state indexes are generated, and monitoring results are graded and output; and the application can continuously acquire vehicle motion parameters and track geometric states in the normal running process, reduces the risk of manual high-altitude detection, and improves the online detection efficiency of the roller coaster and the state evaluation accuracy.

Description

Technical Field

[0001] This invention relates to the field of roller coaster detection technology, and in particular to a method for monitoring the working status of roller coasters based on multi-source data fusion and its application. Background Technology

[0002] Roller coasters are typical large-scale track-based amusement rides, involving various motion scenarios such as lifting, gliding, diving, turning, looping, braking, and entering stations. Due to their high speeds, dramatic changes in attitude, and significant load variations, any abnormalities in the vehicle's motion, track geometry, or critical structural conditions can significantly impact passenger and operational safety. Therefore, efficient, continuous, and precise monitoring of the operational status of roller coasters is of paramount importance.

[0003] Current roller coaster inspection methods mainly include manual periodic inspection, fixed track measuring point inspection, vehicle vibration inspection, combined posture and acceleration inspection, and threshold warning based on multiple operating parameters. Manual periodic inspection usually requires inspectors to set up measuring points or conduct local inspections near the track, pillars, or high-altitude structures, which has the problems of high-altitude operation risks, long preparation time, and low inspection efficiency. Moreover, this type of inspection mostly occurs when the roller coaster is stopped or at low speed, making it difficult to reflect the track deformation and vehicle response under real high-speed operation, impact loads, and dynamic loads.

[0004] While fixed track monitoring points can collect vibration, stress, or displacement information at specific locations, their detection range is limited by the location and number of monitoring points, making it difficult to cover the entire roller coaster route. For sections with complex spatial structures such as loops, sharp curves, dives, and climb transitions, the deployment of fixed monitoring points is difficult and costly to maintain. Furthermore, the data from fixed monitoring points is often discrete, making it difficult to generate continuous track arc length status curves. Relying solely on a few monitoring points to determine the overall track condition can easily lead to overlooking local alignment changes, localized smoothness deterioration, and dynamic deformation anomalies.

[0005] Vibration and acceleration detection can reflect the dynamic response of a vehicle during operation. However, relying solely on vibration or acceleration signals for status assessment is easily affected by factors such as operating speed, load, passenger distribution, wind load, braking status, and sensor installation stiffness. Without accurate position, attitude, and track arc length references, acceleration, vibration, or attitude data collected simultaneously cannot accurately correspond to the spatial position on the track, leading to insufficient anomaly positioning accuracy. This is especially problematic during the high-speed gliding phase of a roller coaster, where the vehicle traverses long track sections in a short time. Insufficient time synchronization or spatial mapping accuracy of sensor data can easily result in inconsistencies between the "data occurrence location" and the "data analysis location."

[0006] Some existing solutions infer the vehicle's track position using pose information and compare the acceleration at the corresponding position with historical standard values ​​to provide anomaly warnings. However, these solutions typically rely on comparisons of single or a small number of motion parameters, failing to fully consider the propagation of inertial navigation errors over time, and neglecting to provide unified quantitative modeling for IMU bias, GNSS obstruction, odometer slippage, installation extrinsic parameter errors, and time synchronization errors. Under conditions such as GNSS signal obstruction, rapid changes in vertical loop attitude, high-impact vibration, and wheel-rail relative slippage, a single forward calculation can easily generate cumulative drift, leading to a decrease in the accuracy of vehicle pose and track geometry inversion.

[0007] Furthermore, existing multi-parameter early warning schemes typically compare parameters such as operating speed, acceleration, motor current, vibration, axial force, and video data with expert or historical thresholds. While these methods can achieve a certain degree of real-time monitoring, they focus more on whether parameters exceed limits and lack the ability to continuously invert vehicle kinematics and track geometry into a track arc-length coordinate system. They also struggle to directly output spatial state indicators closely related to structural health assessment, such as track three-dimensional alignment, curvature changes, lateral offset, vertical offset, and flatness.

[0008] Therefore, there is an urgent need for a new method for monitoring the working status of roller coasters, which can monitor the vehicle's motion status through the fusion of onboard multi-source data without changing the existing structure of the roller coaster or relying on a large-scale deployment of fixed high-altitude measuring points. Summary of the Invention

[0009] In view of this, the purpose of this invention is to propose a roller coaster operation status monitoring method based on multi-source data fusion and its application. This solution aims to achieve continuous calculation of vehicle motion status through onboard multi-source data fusion, and further invert track geometry status, thereby solving the problems of high risk of high-altitude operation, low measurement efficiency, discrete and discontinuous data, difficulty in synchronously acquiring motion parameters and track status, large drift of single inertial navigation and / or difficulty in reflecting the continuous state of track space in traditional detection.

[0010] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows: A method for monitoring the operational status of a roller coaster based on multi-source data fusion, comprising: S1. Non-intrusive installation of onboard data acquisition components on the roller coaster body. The onboard data acquisition components include at least an inertial measurement unit, a Beidou / GNSS positioning module, an odometer, and a data acquisition controller. At the same time, the installation extrinsic parameter relationship between the onboard data acquisition components and the vehicle body reference point is established. S2. During the operation of the roller coaster, the on-board data acquisition components are synchronized in a unified time, and inertial data, positioning data and mileage data during the operation of the roller coaster are collected to form a multi-source operation data sequence; S3. Based on the error information of the inertial measurement unit and the odometer scale error, establish an inertial sensor error propagation model, and construct an S-domain error propagation model through Laplace transform to determine the noise covariance of the multi-source running data sequence in the fusion solution. S4. Perform data preprocessing and operation scenario segmentation on the multi-source operation data sequence, identify the platform section, lifting section, taxiing section, braking section and entry section, and perform robust weight reduction processing on abnormal observation data based on observation residuals; S5. Based on robust adaptive Kalman filtering, the inertial data, positioning data and odometer data are fused and solved in the forward direction to obtain the forward position, velocity and attitude estimation sequence of the roller coaster body reference point; S6. Based on reverse smoothing and graph optimization, the forward position, velocity and attitude estimation sequence is corrected for consistency throughout the entire operation cycle to obtain a globally consistent vehicle pose trajectory. S7. Map the vehicle pose trajectory from the global coordinate system to the Frenet coordinate system, the vehicle body local coordinate system, and the track arc length coordinate system, and establish the correspondence between the vehicle motion state and the track spatial position based on the track arc length parameter. S8. Based on the vehicle's position trajectory, installation external parameter relationship, and track arc length parameter, the geometric state of the roller coaster track is inverted to obtain track feature information; S9. Using the track arc length parameter as a unified index, the vehicle feature information and track feature information are fused to generate a working state feature vector, and the working state feature vector is compared with the health benchmark model to output the roller coaster working state monitoring results.

[0011] As one possible implementation, further, in step S1 of this solution, the installation external parameter relationship is established in the following way: The position and orientation of the vehicle body reference point in the global coordinate system are obtained, and the installation position vector of the sensor on the vehicle data acquisition component relative to the vehicle body reference point is obtained. The position of the sensor in the global coordinate system is calculated according to the following formula: in, This indicates the sensor's position in the global coordinate system. This indicates the position of the vehicle reference point in the global coordinate system. This represents the rotation matrix from the vehicle coordinate system to the global coordinate system. This represents the sensor's mounting position vector relative to a vehicle body reference point.

[0012] As a possible implementation, further, in step S2 of this scheme, the inertial measurement unit, the Beidou / GNSS positioning module and the odometer are synchronized in a unified time during the operation of the roller coaster.

[0013] As a possible implementation, further, in step S3 of this scheme, an inertial sensor error propagation model is established based on the accelerometer error, gyroscope error, installation error and odometer scale error of the inertial measurement unit, and an S-domain error propagation model is constructed through Laplace transform to determine the noise covariance of the multi-source running data sequence in the fusion solution.

[0014] As a preferred implementation option, preferably, in step S3 of this scheme, the error state vector of the inertial sensor error propagation model includes position error, velocity error, attitude error, accelerometer bias error, gyroscope bias error, and odometer scale error; its error state equation is: After Laplace transform, we get: in, Let be the error state vector. This is the time derivative of the error state vector, i.e., the rate of change of the error over time. Here is the error state transition matrix. For noise driving matrix, For system noise, This is the S-domain representation of the error state vector. This is the initial error state. The S-domain representation of system noise. For the Laplace operator, It is an identity matrix.

[0015] As one possible implementation, further, in step S4 of this solution, the robust weighting process includes: Calculation time Normalized residuals of the observation residuals: The time is determined according to the following weighting function. Observation weights: in, For a moment The observation residuals The residual covariance matrix is... For normalized residuals, For robustness threshold, For a moment The observation weights; when When the weight is increased, the weight of the corresponding observation data in the fusion solution is reduced.

[0016] As one possible implementation, further, in step S5 of this scheme, the robust adaptive Kalman filtering includes the following prediction and update process: in, To predict the state, This is the state as updated in the previous moment. For inertial input, It is a nonlinear state transition function. To predict covariance, The discrete state transition matrix, The covariance updated at the previous time step. For process noise covariance, For Kalman gain, For the observation matrix, The observation noise covariance after robust weighting correction. To update the status, For multi-source observations, For observation functions.

[0017] As a possible implementation, further, in step S6 of this scheme, the constraint factors for graph optimization include at least inertial pre-integration factor, positioning observation factor, odometry factor, orbital motion constraint factor and / or operating scenario constraint factor; In step S6, the graph optimization is achieved through the following objective function: in, The sequence of vehicle states to be optimized. For inertial pre-integration residuals, To locate the observation residual, For odometer residuals, For orbital motion constraint residuals, To constrain residuals for the operating scenario, For robust kernel functions, , and These are the covariance matrices of the inertial pre-integration residual, the positioning observation residual, the odometer residual, the orbital motion constraint residual, and the operational scenario constraint residual, respectively.

[0018] As one possible implementation, further, in step S7 of this solution, mapping the vehicle position to the track arc length coordinate system includes: According to the center line of the track and vehicle location The track arc length parameter corresponding to the vehicle is determined by the following formula: And represented in Frenet coordinate system as: in, For the first The track arc length parameter corresponding to the vehicle at any given time. The orbital normal vector, The orbital binormal vector, For horizontal offset, For vertical offset, The centerline of the track in the arc length parameter The global position vector at that location.

[0019] As one possible implementation, further, in step S8 of this solution, the track feature information includes: track three-dimensional alignment, track curvature, track lateral offset, track vertical offset, and flatness index; In step S8, the orbital geometry is inverted in the following manner: Based on the vehicle body reference point pose and the vehicle body key point installation position vector, calculate the position of the vehicle body key point in the global coordinate system: The key point trajectories obtained from multiple runs are then resampled according to the orbital arc length parameter and weighted and fused to obtain the inverted orbital centerline: in, For the first Key points of the vehicle body in the first Global position at any given moment For the vehicle body reference point at the 1st Global position at any given moment Let be the rotation matrix from the vehicle system to the global coordinate system. For the first The position vectors of each key point relative to the vehicle body reference point To retrieve the orbital centerline, For the first The next run is at the arc length position. Data weights at each location For the first The orbital centerline obtained from the inversion of the second run; The number of runs used for track centerline fusion; In step S9, the vehicle feature information includes vehicle position, velocity, three-dimensional attitude, and multi-axis acceleration; In step S9, using the track arc length parameter as a unified index, the vehicle position, speed, three-dimensional attitude and / or multi-axis acceleration, as well as the track three-dimensional alignment, track curvature, track lateral offset, track vertical offset and / or flatness index are fused to generate a working state feature vector. The working state feature vector is then compared with the health benchmark model to output the roller coaster working state monitoring results.

[0020] Based on the above, this solution also proposes a roller coaster operation status monitoring system based on multi-source data fusion, which includes: The onboard data acquisition module is non-invasively installed on the roller coaster body and includes an inertial measurement unit, a Beidou / GNSS positioning module, an odometer, and a data acquisition controller. The onboard data acquisition module is used to collect multi-source operation data sequences during the operation of the roller coaster. The time synchronization module is used to perform unified time synchronization on the multi-source operation data sequence and identify the roller coaster operation scene segment; The error modeling module is used to establish an error propagation model for inertial sensors and to determine the noise covariance through the S-domain error propagation model. The scene segmentation module is used to preprocess the multi-source operation data sequence and segment the operation scene, identify the platform segment, lifting segment, taxiing segment, braking segment and entry segment, and perform robust weight reduction processing on abnormal observation data based on the observation residual; The forward fusion module is used to perform forward fusion calculation on the multi-source running data sequence based on robust adaptive Kalman filtering; The smoothing optimization module is used to perform full-cycle consistency correction on the forward fusion solution results based on reverse smoothing and graph optimization. The coordinate mapping module is used to map the vehicle's pose trajectory to the Frenet coordinate system, the vehicle body local coordinate system, and the track arc length coordinate system; The track inversion module is used to invert the track geometry based on the vehicle's pose trajectory, installation extrinsic parameter relationships, and track arc length parameters. The status assessment module is used to generate a working status feature vector, compare the working status feature vector with the health benchmark model, and output the roller coaster working status monitoring results.

[0021] Based on the above, this solution also proposes an application of the roller coaster operation status monitoring method based on multi-source data fusion as described above. The method is applied to at least one of the following: online detection of roller coasters, auxiliary detection for periodic inspection of roller coasters, evaluation of roller coaster operation status, inversion of the geometric state of roller coaster tracks, and structural health monitoring of large-scale amusement facilities with track systems.

[0022] By adopting the above technical solution, the present invention has the following advantages compared with the prior art: This solution non-invasively installs onboard data acquisition components (including inertial measurement units, Beidou / GNSS positioning modules, and odometers, etc.) on the roller coaster body, enabling it to collect multi-source operating data during the normal operation of the roller coaster. This eliminates the need to deploy a large number of fixed measuring points along the entire track, reducing the risks of high-altitude operations and the workload of on-site deployment. Regarding the preprocessing of raw data, this solution achieves unified time synchronization and external parameter calibration, enabling the fusion of acceleration, angular velocity, positioning information, and odometer information collected by different sensors under the same time and space reference, avoiding inconsistencies in vehicle state calculations caused by sensor time deviations and installation errors.

[0023] In terms of post-processing data, this scheme establishes an inertial sensor error propagation model and constructs an S-domain error propagation model through Laplace transform. This allows for quantitative analysis of the impact of sensor noise, zero-bias drift, installation errors, and odometer scale errors on the vehicle's position, speed, attitude, and track geometry inversion results. This achieves quantitative matching between sensor level and measurement accuracy, improving the verifiability of the system design. Furthermore, this scheme employs robust adaptive Kalman filtering for forward fusion of multi-source data and reduces the weight of abnormal observations based on observation residuals. This mitigates the impact of GNSS jumps, odometer slippage, high-impact vibrations, and local sensor anomalies on the overall solution. Additionally, this scheme further combines inverse smoothing and graph optimization to perform global consistency correction on the vehicle's pose trajectory within the complete operating cycle. This suppresses the cumulative drift generated by single inertial navigation forward integration, improving trajectory stability in long-distance, high-speed, and multi-attitude change operating scenarios.

[0024] This solution also maps vehicle pose trajectory to the global coordinate system, Frenet coordinate system, vehicle local coordinate system, and track arc length coordinate system. It uses track arc length as a unified index to organize vehicle motion state and track geometry state, enabling spatial alignment and comparison of data under different operating cycles, speed conditions, and load conditions. Simultaneously, this solution inverts the three-dimensional alignment, curvature, lateral offset, vertical offset, and flatness index of the track based on vehicle pose trajectory, installation extrinsic parameters, and track arc length parameters. This not only monitors vehicle operating status but also simultaneously obtains track geometry state, solving the problem that traditional methods struggle to reflect track deformation and track alignment changes during dynamic operation.

[0025] In terms of output, this solution generates a working state feature vector containing vehicle position, speed, three-dimensional attitude, multi-axis acceleration, track alignment, curvature, displacement, and flatness. This vector is then compared with a health baseline model to output graded status results such as normal, watch, maintenance, and shutdown for verification. This facilitates accurate location of abnormal sections and the development of maintenance strategies by inspection personnel. This solution can obtain continuous data throughout the entire process with only a few normal runs. Compared with traditional manual measurement point detection and total station discrete static measurement methods, it has the advantages of high measurement efficiency, complete spatial coverage, realistic dynamic response, and high degree of automation.

[0026] In terms of application prospects, this solution is not only suitable for online inspection of roller coasters, but can also be extended to inspection scenarios of other special equipment running along fixed tracks, and has good engineering adaptability and promotion value. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a simplified implementation flowchart of a roller coaster operation status monitoring method based on multi-source data fusion in this solution; Figure 2 This is a simplified connection diagram of the unit modules of the vehicle-mounted data acquisition component involved in this solution method; Figure 3 This is a schematic diagram of the unit module connection of a roller coaster operation status monitoring system based on multi-source data fusion in this solution. Detailed Implementation

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Combination Figure 1 As shown in the figure, this implementation plan proposes a method for monitoring the working status of roller coasters based on multi-source data fusion, which includes: S1. A non-invasive onboard data acquisition component is installed on the roller coaster body. The onboard data acquisition component includes at least an inertial measurement unit, a BeiDou / GNSS positioning module, an odometer, and a data acquisition controller (see reference). Figure 2 Simultaneously, the installation extrinsic parameter relationship between the vehicle-mounted data acquisition component and the vehicle body reference point is established; S2. During the operation of the roller coaster, the on-board data acquisition components are synchronized in a unified time, and inertial data, positioning data and mileage data during the operation of the roller coaster are collected to form a multi-source operation data sequence; S3. Based on the error information of the inertial measurement unit and the odometer scale error, establish an inertial sensor error propagation model, and construct an S-domain error propagation model through Laplace transform to determine the noise covariance of the multi-source running data sequence in the fusion solution. S4. Perform data preprocessing and operation scenario segmentation on the multi-source operation data sequence, identify the platform section, lifting section, taxiing section, braking section and entry section, and perform robust weight reduction processing on abnormal observation data based on observation residuals; S5. Based on robust adaptive Kalman filtering, the inertial data, positioning data and odometer data are fused and solved in the forward direction to obtain the forward position, velocity and attitude estimation sequence of the roller coaster body reference point; S6. Based on reverse smoothing and graph optimization, the forward position, velocity and attitude estimation sequence is corrected for consistency throughout the entire operation cycle to obtain a globally consistent vehicle pose trajectory. S7. Map the vehicle pose trajectory from the global coordinate system to the Frenet coordinate system, the vehicle body local coordinate system, and the track arc length coordinate system, and establish the correspondence between the vehicle motion state and the track spatial position based on the track arc length parameter. S8. Based on the vehicle's position trajectory, installation external parameter relationship, and track arc length parameter, the geometric state of the roller coaster track is inverted to obtain track feature information; S9. Using the track arc length parameter as a unified index, the vehicle feature information and track feature information are fused to generate a working state feature vector, and the working state feature vector is compared with the health benchmark model to output the roller coaster working state monitoring results.

[0031] Traditional roller coaster testing typically requires the deployment of fixed measuring points at high points on the track or at critical structural locations, which presents challenges such as high-risk high-altitude operations, long installation cycles, and limited coverage. Furthermore, if the onboard sensors are simply fixed to the vehicle body without establishing a spatial extrinsic parameter relationship between them and a reference point on the vehicle body, the subsequently collected inertial data, positioning data, and odometer data cannot be uniformly mapped to the same reference point on the vehicle body, resulting in inconsistencies in the spatial reference for vehicle pose calculation and track geometry inversion.

[0032] As one possible implementation, further, in step S1 of this solution, the installation external parameter relationship is established in the following way: The position and orientation of the vehicle body reference point in the global coordinate system are obtained, and the installation position vector of the sensor on the vehicle data acquisition component relative to the vehicle body reference point is obtained. The position of the sensor in the global coordinate system is calculated according to the following formula: in, This indicates the sensor's position in the global coordinate system. This indicates the position of the vehicle reference point in the global coordinate system. This represents the rotation matrix from the vehicle coordinate system to the global coordinate system. This represents the sensor's mounting position vector relative to a vehicle body reference point.

[0033] As an example of implementation, step S1 of this solution includes the following sub-steps: In this step, the onboard data acquisition component is non-invasively installed on the roller coaster body. The onboard data acquisition component includes at least an inertial measurement unit, a BeiDou / GNSS positioning module, an odometer, and a data acquisition controller.

[0034] In addition, depending on the actual detection accuracy requirements, vibration sensors, temperature sensors, visual positioning units, or wireless communication modules can also be included for assistance. These are all common data acquisition components on the market, and their working principles will not be elaborated here.

[0035] In this scheme, the inertial measurement unit is used to collect the vehicle's three-axis acceleration and three-axis angular velocity; the Beidou / GNSS positioning module is used to provide absolute position observation in open areas; the odometer is used to provide the wheel rolling distance or the relative displacement of the vehicle along the track direction; and the data acquisition controller is used for unified sampling control, timestamp writing, data caching, and data uploading.

[0036] For installation, clamp-type bases, magnetic bases, adhesive bases, or existing bolts can be used to avoid drilling, welding, or permanent modification to the roller coaster's load-bearing structure. After installation, select a rigid point on the car body as the car body reference point, such as the center point of the car chassis, the projection point of the axle center, or the geometric center point of the car body, and establish the following coordinate system: Global coordinate system : Used to represent the three-dimensional location of a roller coaster track in geographic or station space; Vehicle coordinate system : Taking the reference point of the vehicle body as the origin, it moves with the roller coaster vehicle body; IMU coordinate system The coordinate axes are based on the sensitive axes of the inertial measurement unit. GNSS antenna coordinate system The GNSS antenna phase center is used as the reference point; Odometer coordinate system : Use the wheel or axle measuring unit as a reference.

[0037] Since the sensors installed at different locations on the vehicle body all move rigidly with the vehicle body, the position of any sensor in the global coordinate system can be obtained by superimposing the position of the vehicle body reference point with the sensor installation offset after the vehicle body is rotated.

[0038] Assuming a sensor ( ; The mounting position vectors of the IMU, GNSS antenna, and odometer (respectively referring to the vehicle reference point) relative to the vehicle body reference point are: The position of the vehicle body reference point in the global coordinate system is: The rotation matrix of the vehicle coordinate system relative to the global coordinate system is: Then the position of the sensor in the global coordinate system is: This formula represents a rigid body coordinate transformation relationship. That is, the position of the sensor relative to the vehicle reference point is first rotated from the vehicle system to the global coordinate system, and then the global position of the vehicle reference point is superimposed.

[0039] in, Indicates sensor In the global coordinate system The position vector below; This indicates that the vehicle reference point is in the global coordinate system. The position vector below; Represents the vehicle coordinate system To the global coordinate system rotation matrix; Indicates sensor In the vehicle system relative to the vehicle body reference point Based on the installation position vector, the position offset vectors of the IMU, GNSS antenna, and odometer relative to the vehicle reference point are determined. .

[0040] For the IMU, its sensing axis direction may not be completely consistent with the vehicle system. Therefore, it is also necessary to establish a rotational extrinsic parameter from the IMU coordinate system to the vehicle system, which is defined as: in, The three-axis accelerations were collected in the IMU coordinate system. To convert to three-axis acceleration in the vehicle system; The three-axis angular velocities are collected in the IMU coordinate system; To convert to the three-axis angular velocities under the vehicle system; This is the rotation matrix from the IMU coordinate system to the vehicle coordinate system.

[0041] Similarly, by establishing the rotational extrinsic parameters from the IMU coordinate system to the vehicle coordinate system, the rotation matrix from the odometer coordinate system to the vehicle coordinate system can be obtained. .

[0042] Through step S1, this solution can obtain the set of installation extrinsic parameters. Its definition is: This includes the position and attitude offsets of the IMU, GNSS antenna, and odometer relative to the vehicle reference point; These are the position offset vectors of the IMU, GNSS antenna, and odometer relative to the vehicle reference point, respectively. These are the rotation matrices from the coordinate systems of the IMU and odometer to the vehicle coordinate system, respectively. This set of extrinsic parameters will serve as the spatial reference for the unified acquisition of multi-source data in subsequent steps, and as auxiliary data for vehicle pose mapping and track geometry inversion.

[0043] Roller coasters operate at high speeds, and even millisecond-level time deviations can lead to significant spatial positioning errors during sections such as gliding, diving, looping, and sharp turns. For example, if the vehicle is traveling at 20 m / s and there is a time deviation of 10 ms between sensors, the corresponding spatial error is approximately 0.2 m. Therefore, if the IMU, GNSS, and odometer data are not synchronized in time, subsequent fusion calculations will result in inconsistencies between the "acceleration occurrence time," "positioning occurrence time," and "mileage accumulation time."

[0044] To address the issue of inconsistent sampling times among multiple sensor sources, as a possible implementation method, in step S2 of this solution, the inertial measurement unit, BeiDou / GNSS positioning module, and odometer are synchronized in a unified time during the operation of the roller coaster.

[0045] As an example of implementation, step S2 of this solution includes the following sub-steps: In this step, the local high-precision clock or GNSS timing signal of the data acquisition controller is used as the master clock to provide unified timing for the IMU, BeiDou / GNSS positioning module, and odometer. For sensors that support PPS pulses, PPS pulse synchronization is preferred; for sensors that do not support hardware synchronization, software timestamp correction and interpolation resampling are used for time alignment.

[0046] During the roller coaster's operation, according to a unified sampling time sequence : Among them, the definition For the first There are 1 sampling time period, and M is the total number of samples.

[0047] In this scheme, the IMU outputs triaxial acceleration and triaxial angular velocity, the GNSS outputs the antenna phase center position and positioning quality, and the odometer outputs the mileage increment and velocity along the track within the sampling interval; during the roller coaster operation, the following data are collected: in, For the first The IMU observation vector at the sampling time contains all the measurement data output by the IMU at that time, such as the first sampling time. Noisy triaxial acceleration measurements acquired by the IMU at each sampling time point. , No. Noisy triaxial angular velocity measurements acquired by the IMU at each sampling time point. ; For the first The GNSS observation vector at the sampling time contains position and positioning quality related data of the GNSS output at that time, such as the first sampling time. The noisy position measurement of the GNSS antenna phase center in the GNSS global coordinate system (G system, such as WGS84 latitude / longitude / geocentric coordinate system) at each sampling time. ; For the first The positioning quality-related data output by GNSS at each sampling time includes the position solution noise covariance matrix or its equivalent form, which is used to characterize the noise level of the current GNSS position observation; For the first The odometer observation vector at each sampling time includes the odometer increment and along-track velocity data within that sampling interval, such as the first... arrive Within each sampling interval, the odometer measures the noisy mileage increment. , For the first At each sampling time, the odometer measures the noisy velocity along the track direction. For the first The speed / distance measurement quality-related data directly output by the odometer at the sampling time, including the noise covariance or error estimate of the speed / mileage increment, is used to characterize the noise level of the current odometer observation.

[0048] Let the first The original timestamps of each sensor are Since the sensor's local clock has a fixed offset and drift, its relationship with the master clock can be expressed as: in, To correct the timestamp to the master clock; For the first The original timestamp of the i-th sensor; (\alpha_i) is the original timestamp of the i-th sensor. The drift coefficient of the local clock of each sensor; For the first Each sensor has a fixed time offset relative to the master clock; For synchronous residuals.

[0049] When the master clock time series is For IMU data with a high sampling frequency, the closest sample can be directly selected. The sampled values ​​can be obtained by linear interpolation; for GNSS data with low sampling frequencies, interpolation can be used to obtain the observations at the corresponding time, which are defined as: in, To synchronize to The position of the GNSS antenna at any given time; , These are the position observations at two adjacent GNSS sampling times; , The sampling time for adjacent GNSS signals; To unify the target sampling time in the time series.

[0050] Based on the above steps, this scheme obtains multi-source operational data sequences under a unified time base. Its definition is: in, For the first IMU observation vectors at each sampling time; For the first GNSS observation vectors at each sampling time; For the first Odometry observation vectors at each sampling time; To unify the target sampling time in the time series; install the extrinsic parameter set. .

[0051] This data sequence It is used to establish an error propagation model for subsequent steps; and for data preprocessing, anomaly detection and segmentation of the operating scenario.

[0052] When operating in the high-impact, high-angular-velocity, and strong-vibration environment of a roller coaster, the IMU is affected by zero-bias drift, white noise, scale factor error, installation error, and temperature drift. Without establishing an error propagation model, it is impossible to determine whether a certain level of sensor meets the accuracy requirements for track geometry inversion, nor is it possible to reasonably set the noise covariance in subsequent fusion algorithms.

[0053] To address the problem of inertial navigation errors accumulating over time and being difficult to quantify, as a possible implementation method, in step S3 of this scheme, an inertial sensor error propagation model is established based on the accelerometer error, gyroscope error, installation error, and odometer scale error of the inertial measurement unit, and an S-domain error propagation model is constructed through Laplace transform to determine the noise covariance of the multi-source operating data sequence in the fusion solution.

[0054] As a preferred implementation option, preferably, in step S3 of this scheme, the error state vector of the inertial sensor error propagation model includes position error, velocity error, attitude error, accelerometer bias error, gyroscope bias error, and odometer scale error; its error state equation is: After Laplace transform, we get: in, Let be the error state vector. This is the time derivative of the error state vector, i.e., the rate of change of the error over time. Here is the error state transition matrix. For noise driving matrix, For system noise, This is the S-domain representation of the error state vector. This is the initial error state. The S-domain representation of system noise. For the Laplace operator, It is an identity matrix.

[0055] As an example of implementation, step S3 of this solution includes the following sub-steps: In this step, the inertial error state equation is established based on the IMU measurement model. The observation models for the IMU accelerometer and gyroscope are as follows: in, For the first The acceleration measured by the IMU at that moment; For the first Real-time acceleration; To achieve zero bias in the accelerometer; To measure noise for accelerometers; For the first Angular velocity measured by the IMU at time t; For the first Real angular velocity at any given moment; Zero bias for the gyroscope; To measure the noise of the gyroscope.

[0056] Define the error state vector as follows: in, This refers to the positional error; For speed error; This refers to attitude error; This refers to the zero bias error of the accelerometer. This refers to the zero bias error of the gyroscope. This refers to the odometer scale error.

[0057] The basic equation of motion for inertial navigation is defined as follows: in, This refers to the position of the vehicle body reference point, i.e., the position vector. Position vector The first derivative with respect to time, i.e., the linear velocity vector in the global coordinate system; The velocity is the reference point velocity of the vehicle body, i.e., the velocity vector. velocity vector The first derivative with respect to time, i.e., the linear acceleration vector in the global coordinate system. The raw measurement value from the accelerometer. For the bias error of the accelerometer, The measurement noise of the accelerometer; Let be the rotation matrix from the vehicle system to the global coordinate system. Rotation matrix The first derivative with respect to time, i.e., the rate of change of the attitude rotation matrix, These are the raw measurements from the gyroscope; This refers to gravitational acceleration in the global coordinate system. This refers to the bias error of the gyroscope. This refers to the measurement noise of the gyroscope. Let be the antisymmetric matrix corresponding to the vector.

[0058] By linearizing the above nonlinear equations in the vicinity of the current estimated state, the error state equation can be obtained: in, This is the error state transition matrix, used to describe the coupling relationship between position error, velocity error, attitude error, and zero bias error; Let be the error state vector. The time derivative of the error state vector; This is the noise driving matrix; The system noise vector includes acceleration noise, angular velocity noise, zero-bias random walk noise, and odometer scale noise.

[0059] Perform a Laplace transform on the error state equation. According to the properties of the Laplace transform: Substituting, we get: The results were: therefore: in, Let Laplace transform function, This is the S-domain representation of the error state vector; For the Laplace operator; It is the identity matrix; This is the initial error state; This represents the S-domain expression of system noise.

[0060] Therefore, the error output is determined by both the initial error term and the noise propagation term. Further, we define the output quantity to be evaluated: Its S-domain error output is: in, The output error can be position error, attitude error, curvature inversion error, or orbital offset error. Choose a matrix for the error output; This is the S-domain representation of the output error.

[0061] Based on this, the contribution of sensor error to the detection result can be calculated, and the noise covariance matrix can be formed, which is defined as: in, For accelerometer noise variance; This represents the gyroscope noise variance. The variance of the accelerometer's zero-bias random walk; The variance of the gyroscope's zero-bias random walk; This represents the variance of the odometer scale error.

[0062] Due to the presence of noise, abrupt changes, differences in time-phase characteristics, and abnormal observations in the original multi-source data, and considering the significantly different motion patterns of a roller coaster during its lifting, sliding, looping, and braking phases, using the same filtering parameters and thresholds for all data could easily lead to misjudgments. Furthermore, GNSS may exhibit abrupt changes in obstructed areas, odometers may become inaccurate during wheel-rail slippage, and IMUs may experience short-term saturation under impact and vibration. Therefore, step S4 of this scheme performs data segmentation, quality inspection, and robustness processing before fusion.

[0063] As one possible implementation, further, in step S4 of this solution, the robust weighting process includes: Calculation time Normalized residuals of the observation residuals: The time is determined according to the following weighting function. Observation weights: in, For a moment The observation residuals The residual covariance matrix is... For normalized residuals, For robustness threshold, For a moment The observation weights; when When the weight is increased, the weight of the corresponding observation data in the fusion solution is reduced.

[0064] As an example of implementation, step S4 of this solution includes the following sub-steps: First, the IMU data is subjected to anomaly removal, desampling, temperature drift compensation, and zero bias initial value estimation. Second, the GNSS data is checked for satellite number, carrier-to-noise ratio, positioning accuracy factor, and position jump. Third, the odometer data is tested for rolling continuity and slippage. Finally, based on speed, acceleration, mileage increment, and operation control signals, the operation process is divided into platform section, lifting section, taxiing section, braking section, and station entry section.

[0065] For each type of observation data, calculate its observation residual: in, For the first Observe the residuals at all times; For the first Actual observations at any given time; These are theoretical observations calculated based on the predicted state; For the first Predict the state at any time.

[0066] In this step, to determine whether the observed residuals are abnormal, normalized residuals are used: in, No. Normalized residuals at time points; The residual covariance matrix; Residual covariance matrix Defined as follows: in, The observation matrix; To predict the covariance matrix; To observe the noise covariance matrix.

[0067] like A smaller value indicates a high degree of consistency between the actual and predicted observations; if The large value indicates that the current observation may be affected by GNSS jumps, odometer slippage, or vibration shocks.

[0068] Therefore, a robust weighting function is set: in, For a moment The observation weights; For robustness threshold; This is the normalized residual.

[0069] when When (w_k=1), it indicates that the observation is normal; when hour, This indicates a reduction in the impact of the observation. The corresponding corrected observation noise covariance is: in, This is the robustly corrected observation noise covariance matrix; This is the original observation noise covariance matrix.

[0070] IMUs have good short-time dynamic response but suffer from integral drift; GNSS can provide absolute position but is prone to failure in obstructed, vertical loop, and near steel structure sections; odometers can provide relative displacement along the rail direction but may be affected by wheel-rail slippage. Therefore, step S5 of this scheme uses robust adaptive Kalman filtering to complementarily fuse multi-source data to obtain a vehicle pose estimate that is continuously updated over time.

[0071] To address the issue that a single sensor cannot stably provide vehicle position, speed, and attitude over a long period, as a possible implementation method, step S5 of this solution further includes the following prediction and update process: in, To predict the state, This is the state as updated in the previous moment. For inertial input, It is a nonlinear state transition function. To predict covariance, The discrete state transition matrix, The covariance updated at the previous time step. For process noise covariance, For Kalman gain, For the observation matrix, The observation noise covariance after robust weighting correction. To update the status, For multi-source observations, For observation functions.

[0072] As an example of implementation, step S5 of this solution includes the following sub-steps: The system state vector is defined as follows: in, This represents the position of the vehicle body reference point in the global coordinate system. The velocity of the vehicle reference point in the global coordinate system; For the vehicle body attitude quaternion; To achieve zero bias in the accelerometer; Zero bias for the gyroscope; This is the odometer scale factor.

[0073] In this step, the filtering process includes two parts: prediction and update. The prediction process is mainly driven by the IMU, while the update process is mainly corrected by GNSS and odometry observations.

[0074] In this step, at adjacent sampling times and Between, time interval for: State prediction is performed using IMU acceleration and angular velocity. Velocity prediction is as follows: The predicted location is: Attitude prediction is: in, For the first Predict speed in real time. For the first Speed ​​in a given moment; For the first Predicting location at any given time, i.e., the predicted value of the prior location. For the first The location at time, i.e., the posterior location estimate; For the first The attitude quaternion is predicted at each moment, which is the predicted value of the prior attitude quaternion. For the first The time-space attitude quaternion is the posterior attitude quaternion estimate. For the previous moment ( Estimate the rotation matrix corresponding to the attitude at time (time). Acceleration was measured by the IMU; Angular velocity was measured by the IMU; This is the zero bias estimate of the accelerometer at the previous moment; For the previous moment ( (Time) Gyroscope bias estimation; This is quaternion multiplication; This is an exponential mapping from a rotation vector to a quaternion; This refers to gravitational acceleration in the global coordinate system. Covariance prediction is: in, To predict covariance; For the previous moment ( Updated covariance; It is a discrete state transition matrix; The process noise covariance obtained from the aforementioned steps.

[0075] When GNSS observations are available, construct the position observation equations: in, These are GNSS antenna position observations; The installation position of the GNSS antenna phase center relative to the vehicle body reference point; This refers to GNSS observation noise. This represents the position of the vehicle body reference point in the global coordinate system. This is the rotation matrix from the vehicle system to the global coordinate system.

[0076] When odometer observations are available, construct the track-side displacement constraints: in, This represents the actual incremental displacement of the vehicle along the track. The displacement increment measured by the odometer; Odometer scale factor; This refers to odometer noise.

[0077] The filter update uses: in, Kalman gain; The observation matrix; The robust corrected observation noise covariance obtained in step S4; For the predicted state; To update the state, it can be used as the first The state is updated by forward filtering at each time step; For observation functions; The updated state covariance can be used as the first... The filtered covariance at time t.

[0078] After step S5, the forward fusion state sequence is output, which is defined as: in, Indicates the first Update status.

[0079] A roller coaster's operation typically involves a complete closed loop or fixed start and end points, and physical constraints exist between different operating scenarios. Using only forward filtering cannot utilize subsequent observation data to correct earlier errors, nor can it leverage the track and scene constraints of the entire trajectory. Therefore, reverse smoothing and graph optimization are necessary to obtain a globally consistent vehicle pose trajectory.

[0080] To address the problem that forward filtering only utilizes information prior to the current moment, leading to local optima in the trajectory, as a possible implementation method, in step S6 of this scheme, the constraint factors for graph optimization include at least inertial pre-integration factor, positioning observation factor, odometry factor, orbital motion constraint factor, and / or operational scenario constraint factor. In step S6, the graph optimization is achieved through the following objective function: in, The sequence of vehicle states to be optimized. For inertial pre-integration residuals, To locate the observation residual, For odometer residuals, For orbital motion constraint residuals, To constrain residuals for the operating scenario, For robust kernel functions, , and These are the covariance matrices of the inertial pre-integration residual, the positioning observation residual, the odometer residual, the orbital motion constraint residual, and the operational scenario constraint residual, respectively.

[0081] As an example of implementation, step S6 of this solution includes the following sub-steps: First, the forward fusion sequence output in step S5 is subjected to reverse smoothing. Reverse smoothing traces back from the end point to the starting point, using future time-time state estimates to correct the current state. Then, using the vehicle state at each time point as graph nodes and multi-source observations and motion constraints as graph factors, a graph optimization objective function is established, and the globally consistent trajectory is solved using nonlinear least squares.

[0082] In this step, reverse smoothing takes the following form: The smoothing gain is: in, For the first The state after time-smoothing; For the first The state is updated continuously using positive filtering. For the first Smooth state at all times; Predict the state at time (k+1); For inverse smoothing gain; For the first Time-mapping covariance; This is the state transition matrix; For the first Predict covariance at any given time.

[0083] Under the definition logic of this formula, if the first... If the smoothed state at time step 1 has a correction amount compared to the predicted state, this correction amount will be fed back to the state transition correlation at time step 2. time.

[0084] Further construct the graph optimization objective function, which is defined as: in, This is the optimized vehicle state sequence. The sequence of vehicle states to be optimized. For the IMU inertial pre-integration residual, For GNSS positioning observation residuals, For odometer residuals, For orbital motion constraint residuals, To constrain residuals for the operating scenario, This is a robust kernel function used to reduce the impact of abnormal GNSS or abnormal odometry observations on the optimization results. , and These are the covariance matrices of the inertial pre-integration residual, the positioning observation residual, the odometer residual, the orbital motion constraint residual, and the operational scenario constraint residual, respectively.

[0085] The orbital motion constraint residuals can be expressed as: in, For the first Vehicle velocity vector at any given moment; The tangential unit vector of the orbit; This is the projection of the velocity onto the tangential direction of the orbit.

[0086] This constraint states that the main velocity direction of the roller coaster vehicle should be consistent with the tangential direction of the track, and the lateral non-physical velocity component should be suppressed.

[0087] The speed, passenger load, and environmental conditions may vary each time a roller coaster runs. If data is compared only using time as an index, the same point in time may not correspond to the same track position. Therefore, step S7 of this solution transforms the vehicle motion data from the time domain to the track arc length domain, enabling comparisons of vehicle status, track geometry, and anomaly features based on track position.

[0088] To address the issue of difficulty in uniformly aligning operational data according to track spatial location, as a possible implementation method, step S7 of this solution further includes mapping the vehicle position to the track arc length coordinate system, which includes: According to the center line of the track and vehicle location The track arc length parameter corresponding to the vehicle is determined by the following formula: And represented in Frenet coordinate system as: in, For the first The track arc length parameter corresponding to the vehicle at any given time. The orbital normal vector, The orbital binormal vector, For horizontal offset, For vertical offset, The centerline of the track in the arc length parameter The global position vector at that location.

[0089] As an example of implementation, step S7 of this solution includes the following sub-steps: In this step, the track design centerline or the baseline track centerline obtained by fitting historical healthy operation data is defined as follows: in, This is the track arc length parameter, representing the cumulative centerline arc length calculated from the track reference starting point; arc length The three-dimensional position vector of the reference orbit centerline in the global coordinate system. These represent the x, y, and z coordinate components of the centerline in the global coordinate system. The reference track centerline can be obtained directly from track design data or fitted through historical train health operation trajectory data, serving as a reference benchmark for track geometry assessment.

[0090] For the position of the vehicle reference point in the global coordinate system at each moment (i.e., vehicle position), its corresponding track arc length position is determined by the projection of the nearest point. Then, a Frenet coordinate system is established, including the tangent vector, normal vector, and binormal vector.

[0091] In this step, the position of the orbital arc length is determined by the following optimization problem: in, For the first The track arc length parameter corresponding to the vehicle at any given time; For the first The position of the vehicle body reference point at any time (i.e., the vehicle position); The baseline is the centerline of the track. The distance is Euclidean.

[0092] Calculate the orbital tangential unit vector: in, arc length The unit vector tangential to the orbit at that point; It is the first derivative of the orbital centerline with respect to the arc length.

[0093] In three-dimensional space, combined with the orbital normal vector and binormal vector The vehicle's location can be represented as: in, For the first The normal offset of the vehicle relative to the centerline of the track at any given moment; For the first The vehicle's offset relative to the track centerline in the secondary normal direction at any given moment; Position of arc length The normal unit vector at that location; Position of arc length The binormal unit vector at that location.

[0094] Therefore, the time series state The state is converted into an arc-length sequence, defined as: in, For the first The state vector of the arc length sequence corresponding to each time; For the first The track arc length parameter corresponding to the vehicle at any given time; For the first The position of the vehicle body reference point at any time (i.e., the vehicle position); The velocity of the vehicle reference point in the global coordinate system; This is the rotation matrix from the vehicle system to the global coordinate system; , These are the lateral and vertical offsets of the train reference point relative to the centerline of the reference track, also known as the normal offset and sub-normal offset, respectively. If it is necessary to output the state according to equal arc length intervals, it can be handled for non-uniform... Data resampling is defined as follows: in, Equal arc length sampling points obtained through linear interpolation The train state vector at the location; For the first The arc length sequence state at time t. Sampling points of equal arc length; , To and Two adjacent arc length positions.

[0095] Current track inspection methods largely rely on static total stations, manual measuring points, or local sensors, resulting in discrete data that fails to reflect the dynamic track response during actual vehicle operation. This step further maps the vehicle's motion state into three-dimensional track alignment, curvature, lateral offset, vertical offset, and flatness indices by using vehicle pose trajectory, extrinsic parameters installed at key points on the vehicle body, and track arc length parameters.

[0096] To address the problem that traditional detection methods struggle to obtain continuous track geometry under dynamic operating conditions, as a possible implementation method, in step S8 of this solution, the track feature information includes: track three-dimensional alignment, track curvature, track lateral offset, track vertical offset, and flatness index. In step S8, the orbital geometry is inverted in the following manner: Based on the vehicle body reference point pose and the vehicle body key point installation position vector, calculate the position of the vehicle body key point in the global coordinate system: The key point trajectories obtained from multiple runs are then resampled according to the orbital arc length parameter and weighted and fused to obtain the inverted orbital centerline: in, For the first Key points of the vehicle body in the first Global position at any given moment For the vehicle body reference point at the 1st Global position at any given moment Let be the rotation matrix from the vehicle system to the global coordinate system. For the first The position vectors of each key point relative to the vehicle body reference point To retrieve the orbital centerline, For the first The next run is at the arc length position. Data weights at each location For the first The orbital centerline obtained from the inversion of the second run; This refers to the number of runs used for track centerline fusion.

[0097] As an example of implementation, step S8 of this solution includes the following sub-steps: In this step, based on the vehicle body structure and wheel-rail contact relationship, several key points of the vehicle body are selected, such as the center of the front wheel assembly, the center of the rear wheel assembly, the center of the left guide wheel, the center of the right guide wheel, the center of the vehicle body, or the sensor mounting reference point. Based on the installation extrinsic parameters obtained in step S1 and the arc length mapping results obtained in step S7, the position of each key point in the global coordinate system is calculated.

[0098] In this step, for the first The positions of key points on the vehicle body in the global coordinate system are as follows: in, For the first The key point is in the first The global position at any given moment; For the first The position of the vehicle body reference point in the global coordinate system at any given moment; For the first The rotation matrix from the time-tracking system to the global coordinate system; For the first The installation position vector of each key point relative to the vehicle body reference point.

[0099] Because the trajectory of key points on the vehicle is affected by vehicle vibration, wheel-rail clearance, and sensor noise, the results of a single run may exhibit random fluctuations. Therefore, a weighted fusion of data from multiple runs is used to estimate the track centerline. in, To retrieve the orbital centerline, For the first The next run is at the arc length position. Data weights at each location For the first The orbital centerline obtained from the inversion of the second run; This refers to the number of runs used for track centerline fusion.

[0100] The weights can be determined by the trajectory uncertainty and the observation quality, and are defined as follows: in, For the first The next run is at the arc length position. The variance of the trajectory estimation at that location; To prevent extremely small positive numbers with a denominator of zero.

[0101] The orbital geometric offset is: in, Position of arc length The orbital space offset at that location; The centerline of the health baseline track.

[0102] Projecting the offsets onto the Frenet coordinate system yields the horizontal and vertical offsets: in, Lateral offset of the track; This represents the vertical offset of the track. The orbital normal unit vector; Let be the unit vector of the orbital subnormal direction.

[0103] The orbital curvature is calculated as follows: in, Position of arc length The orbital curvature at that point; This is the first derivative of the inverted orbit centerline; The second derivative of the inverted orbit centerline; It is the cross product of vectors.

[0104] The flatness index can be expressed as the rate of change of vertical offset within a unit arc length, which is defined as: in, This refers to the track flatness index; This represents the vertical offset of the track. For the orbital arc length parameter; For arc length The second-order differential operator is used to describe the curvature change of the vertical offset.

[0105] After the aforementioned steps, this solution has obtained various parameters such as vehicle position, speed, attitude, acceleration, and track geometry. However, it is difficult to accurately determine the overall working status of the roller coaster by observing only a single parameter. Therefore, step S9 of this solution uses the track arc length as a unified index to construct a comprehensive working status feature vector, and compares it with the health benchmark model to achieve hierarchical status output and abnormal section location.

[0106] This solution addresses the problem of fragmented multi-source monitoring results making it difficult to form a comprehensive status assessment. In step S9 of this solution, the vehicle characteristic information includes vehicle position, velocity, three-dimensional attitude, and multi-axis acceleration.

[0107] In addition, in step S9, using the track arc length parameter as a unified index, the vehicle position, speed, three-dimensional attitude and / or multi-axis acceleration, as well as the track three-dimensional alignment, track curvature, track lateral offset, track vertical offset and / or flatness index are fused to generate a working state feature vector, and the working state feature vector is compared with the health benchmark model to output the roller coaster working state monitoring results.

[0108] As an example of implementation, step S9 of this solution includes the following sub-steps: In this step, the track arc length is used. and runtime As an index, the vehicle motion state and track geometry state are fused to form a working state feature vector. The working state feature vector includes vehicle speed, multi-axis acceleration, three-dimensional attitude angle, jerk, track curvature, lateral offset, vertical offset, flatness index, and vibration energy index.

[0109] The vehicle speed can be obtained from the velocity vector output in the preceding steps, and it is defined as follows: in, Position of arc length ,time The corresponding vehicle speed; The velocity vector is the reference point of the vehicle body.

[0110] In this scheme, the jerk (also known as the impact force) can be obtained by differentiating the acceleration with respect to time, and is defined as follows: in, To accelerate; This refers to the combined acceleration of the vehicle or the specified axial acceleration.

[0111] This step also constructs a working state feature vector, which is defined as follows: in, This represents the feature vector of the working state. For vehicle speed; The three-axis acceleration in the vehicle system or track coordinate system; This refers to the roll angle; The pitch angle; Yaw angle; To accelerate; For orbital curvature; Lateral offset of the track; This represents the vertical offset of the track. This is an indicator of track flatness.

[0112] A health benchmark model is established based on historical health operation data, and it is defined as follows: in, Position of arc length The orbital health baseline model at the location, Arc length position The mean of the feature vectors in a healthy state; Position of arc length The covariance matrix of eigenvectors under healthy conditions.

[0113] The deviation between the current working state and the health baseline state is calculated using Mahalanobis distance, and it is defined as follows: in, This refers to the degree of state deviation. It is the inverse of the health baseline covariance matrix.

[0114] Under the definition logic of this formula, different features have different dimensions and fluctuation ranges. If Euclidean distance is used directly, features with larger values ​​among velocity, displacement or acceleration may dominate the judgment. After using Mahalanobis distance, the covariance matrix can be used to normalize and correct the correlation of different features, so as to more reasonably measure the comprehensive deviation between the current state and the health baseline state.

[0115] This step also sets grading rules based on the degree of state deviation, which are defined as follows: in, The threshold for attention; This is the maintenance threshold. This is the threshold for shutdown review.

[0116] To pinpoint the source of the anomaly, the contribution of each individual feature can be further calculated, defined as: in, For the first The abnormal contribution of each feature; The first feature in the current feature vector One eigenvalue; The first in the health benchmark The mean of each feature; The first in the health benchmark The standard deviation of each feature; To prevent extremely small positive numbers with a denominator of zero.

[0117] By comparing each The size of the abnormality can be used to determine whether the abnormality is mainly due to velocity abnormality, attitude abnormality, acceleration abnormality, curvature abnormality, lateral offset abnormality, vertical offset abnormality, or flatness abnormality.

[0118] Through step S9 of this scheme, the output monitoring result information Defined as follows: in, For anomalies or to monitor the corresponding track arc length position; This represents the feature vector of the working state. This refers to the degree of state deviation. Status levels include Normal, Attention, Under Maintenance, and Outage Review; Abnormal contribution level; Suggested actions, including continued operation, focused observation, maintenance scheduling, shutdown for verification, or track retesting, can be extracted from a pre-built monitoring and response database.

[0119] Based on the above, this solution also proposes an application of the roller coaster operation status monitoring method based on multi-source data fusion as described above. The method is applied to at least one of the following: online detection of roller coasters, auxiliary detection for periodic inspection of roller coasters, evaluation of roller coaster operation status, inversion of the geometric state of roller coaster tracks, and structural health monitoring of large-scale amusement facilities with track systems.

[0120] Combination Figure 3 As shown, based on the above, this solution also proposes a roller coaster operation status monitoring system based on multi-source data fusion, which includes: The onboard data acquisition module is non-invasively installed on the roller coaster body and includes an inertial measurement unit, a Beidou / GNSS positioning module, an odometer, and a data acquisition controller. The onboard data acquisition module is used to collect multi-source operation data sequences during the operation of the roller coaster. The time synchronization module is used to perform unified time synchronization on the multi-source operation data sequence and identify the roller coaster operation scene segment; The error modeling module is used to establish an error propagation model for inertial sensors and to determine the noise covariance through the S-domain error propagation model. The scene segmentation module is used to preprocess the multi-source operation data sequence and segment the operation scene, identify the platform segment, lifting segment, taxiing segment, braking segment and entry segment, and perform robust weight reduction processing on abnormal observation data based on the observation residual; The forward fusion module is used to perform forward fusion calculation on the multi-source running data sequence based on robust adaptive Kalman filtering; The smoothing optimization module is used to perform full-cycle consistency correction on the forward fusion solution results based on reverse smoothing and graph optimization. The coordinate mapping module is used to map the vehicle's pose trajectory to the Frenet coordinate system, the vehicle body local coordinate system, and the track arc length coordinate system; The track inversion module is used to invert the track geometry based on the vehicle's pose trajectory, installation extrinsic parameter relationships, and track arc length parameters. The status assessment module is used to generate a working status feature vector, compare the working status feature vector with the health benchmark model, and output the roller coaster working status monitoring results.

[0121] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0122] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for monitoring the operational status of a roller coaster based on multi-source data fusion, characterized in that, It includes: S1. Non-intrusive installation of onboard data acquisition components on the roller coaster body. The onboard data acquisition components include at least an inertial measurement unit, a Beidou / GNSS positioning module, an odometer, and a data acquisition controller. At the same time, the installation extrinsic parameter relationship between the onboard data acquisition components and the vehicle body reference point is established. S2. During the operation of the roller coaster, the on-board data acquisition components are synchronized in a unified time, and inertial data, positioning data and mileage data during the operation of the roller coaster are collected to form a multi-source operation data sequence; S3. Based on the error information of the inertial measurement unit and the odometer scale error, establish an inertial sensor error propagation model, and construct an S-domain error propagation model through Laplace transform to determine the noise covariance of the multi-source running data sequence in the fusion solution. S4. Perform data preprocessing and operation scenario segmentation on the multi-source operation data sequence, identify the platform section, lifting section, taxiing section, braking section and entry section, and perform robust weight reduction processing on abnormal observation data based on observation residuals; S5. Based on robust adaptive Kalman filtering, the inertial data, positioning data and odometer data are fused and solved in the forward direction to obtain the forward position, velocity and attitude estimation sequence of the roller coaster body reference point; S6. Based on reverse smoothing and graph optimization, the forward position, velocity and attitude estimation sequence is corrected for consistency throughout the entire operation cycle to obtain a globally consistent vehicle pose trajectory. S7. Map the vehicle pose trajectory from the global coordinate system to the Frenet coordinate system, the vehicle body local coordinate system, and the track arc length coordinate system, and establish the correspondence between the vehicle motion state and the track spatial position based on the track arc length parameter. S8. Based on the vehicle's position trajectory, installation external parameter relationship, and track arc length parameter, the geometric state of the roller coaster track is inverted to obtain track feature information; S9. Using the track arc length parameter as a unified index, the vehicle feature information and track feature information are fused to generate a working state feature vector, and the working state feature vector is compared with the health benchmark model to output the roller coaster working state monitoring results.

2. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 1, characterized in that, In step S1, the installation extrinsic parameter relationship is established in the following way: The position and orientation of the vehicle body reference point in the global coordinate system are obtained, and the installation position vector of the sensor on the vehicle data acquisition component relative to the vehicle body reference point is obtained. The position of the sensor in the global coordinate system is calculated according to the following formula: in, This indicates the sensor's position in the global coordinate system. This indicates the position of the vehicle reference point in the global coordinate system. This represents the rotation matrix from the vehicle coordinate system to the global coordinate system. This represents the sensor's mounting position vector relative to a vehicle body reference point; In step S2, during the operation of the roller coaster, the inertial measurement unit, the Beidou / GNSS positioning module, and the odometer are synchronized in a unified time. In step S3, an inertial sensor error propagation model is established based on the accelerometer error, gyroscope error, installation error, and odometer scale error of the inertial measurement unit, and an S-domain error propagation model is constructed through Laplace transform to determine the noise covariance of the multi-source running data sequence in the fusion solution.

3. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 2, characterized in that, In step S3, the error state vector of the inertial sensor error propagation model includes position error, velocity error, attitude error, accelerometer bias error, gyroscope bias error, and odometer scale error; its error state equation is: After Laplace transform, we get: in, Let be the error state vector. This is the time derivative of the error state vector, i.e., the rate of change of the error over time. Here is the error state transition matrix. For noise driving matrix, For system noise, This is the S-domain representation of the error state vector. This is the initial error state. This represents the S-domain expression of system noise. For the Laplace operator, It is an identity matrix.

4. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 1, characterized in that, In step S4, the robust weighting process includes: Calculation time Normalized residuals of the observation residuals: The time is determined according to the following weighting function. Observation weights: in, For a moment The observation residuals The residual covariance matrix is... For normalized residuals, For robustness threshold, For a moment The observation weights; when When the weight is increased, the weight of the corresponding observation data in the fusion solution is reduced.

5. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 1, characterized in that, In step S5, the robust adaptive Kalman filtering includes the following prediction and update process: in, To predict the state, This is the state as updated in the previous moment. For inertial input, It is a nonlinear state transition function. To predict covariance, The discrete state transition matrix, The covariance updated at the previous time step. For process noise covariance, For Kalman gain, For the observation matrix, The observation noise covariance after robust weighting correction. To update the status, For multi-source observations, For observation functions.

6. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 1, characterized in that, In step S6, the constraint factors for graph optimization include at least inertial pre-integration factor, positioning observation factor, odometry factor, orbital motion constraint factor and / or operating scenario constraint factor; In step S6, the graph optimization is achieved through the following objective function: in, The sequence of vehicle states to be optimized. For inertial pre-integration residuals, To locate the observation residual, For odometer residuals, For orbital motion constraint residuals, To constrain residuals for the operating scenario, For robust kernel functions, , and These are the covariance matrices of the inertial pre-integration residual, the positioning observation residual, the odometer residual, the orbital motion constraint residual, and the operational scenario constraint residual, respectively.

7. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 1, characterized in that, In step S7, mapping the vehicle position to the track arc length coordinate system includes: According to the center line of the track and vehicle location The track arc length parameter corresponding to the vehicle is determined by the following formula: And represented in Frenet coordinate system as: in, For the first The track arc length parameter corresponding to the vehicle at any given time. The orbital normal vector, The orbital binormal vector, For horizontal offset, For vertical offset, The centerline of the track in the arc length parameter The global position vector at that location.

8. The roller coaster operation status monitoring method based on multi-source data fusion as described in claim 1, characterized in that, In step S8, the track feature information includes: track three-dimensional alignment, track curvature, track lateral offset, track vertical offset, and flatness index; In step S8, the orbital geometry is inverted in the following manner: Based on the vehicle body reference point pose and the vehicle body key point installation position vector, calculate the position of the vehicle body key point in the global coordinate system: The key point trajectories obtained from multiple runs are then resampled according to the orbital arc length parameter and weighted and fused to obtain the inverted orbital centerline: in, For the first Key points of the vehicle body in the first Global position at any given moment For the vehicle body reference point at the 1st Global position at any given moment Let be the rotation matrix from the vehicle system to the global coordinate system. For the first The position vectors of each key point relative to the vehicle body reference point To retrieve the orbital centerline, For the first The next run is at the arc length position. Data weights at each location For the first The orbital centerline obtained from the inversion of the second run; The number of runs used for track centerline fusion; In step S9, the vehicle feature information includes vehicle position, velocity, three-dimensional attitude, and multi-axis acceleration; In step S9, using the track arc length parameter as a unified index, the vehicle position, speed, three-dimensional attitude and / or multi-axis acceleration, as well as the track three-dimensional alignment, track curvature, track lateral offset, track vertical offset and / or flatness index are fused to generate a working state feature vector. The working state feature vector is then compared with the health benchmark model to output the roller coaster working state monitoring results.

9. A roller coaster operation status monitoring system based on multi-source data fusion, characterized in that, include: The onboard data acquisition module is non-invasively installed on the roller coaster body and includes an inertial measurement unit, a Beidou / GNSS positioning module, an odometer, and a data acquisition controller. The onboard data acquisition module is used to collect multi-source operation data sequences during the operation of the roller coaster. The time synchronization module is used to perform unified time synchronization on the multi-source operation data sequence and identify the roller coaster operation scene segment; The error modeling module is used to establish an error propagation model for inertial sensors and to determine the noise covariance through the S-domain error propagation model. The scene segmentation module is used to preprocess the multi-source operation data sequence and segment the operation scene, identify the platform segment, lifting segment, taxiing segment, braking segment and entry segment, and perform robust weight reduction processing on abnormal observation data based on the observation residual; The forward fusion module is used to perform forward fusion calculation on the multi-source running data sequence based on robust adaptive Kalman filtering; The smoothing optimization module is used to perform full-cycle consistency correction on the forward fusion solution results based on reverse smoothing and graph optimization. The coordinate mapping module is used to map the vehicle's pose trajectory to the Frenet coordinate system, the vehicle body local coordinate system, and the track arc length coordinate system; The track inversion module is used to invert the track geometry based on the vehicle's pose trajectory, installation extrinsic parameter relationships, and track arc length parameters. The status assessment module is used to generate a working status feature vector, compare the working status feature vector with the health benchmark model, and output the roller coaster working status monitoring results.

10. An application of the roller coaster operation status monitoring method based on multi-source data fusion according to any one of claims 1 to 8, characterized in that, The method is applied to at least one of the following: online detection of roller coasters, auxiliary detection for periodic inspections of roller coasters, evaluation of the operational status of roller coasters, inversion of the geometric state of roller coaster tracks, and structural health monitoring of large-scale amusement facilities with track systems.

Images