A Three-Vehicle Cooperative Aircraft Transfer Method and System Based on Multi-Sensor Fusion

Abstract

This invention provides a three-trailer collaborative aircraft transfer method and system based on multi-sensor fusion. This invention relates to the field of aircraft ground support technology, specifically to a three-trailer collaborative aircraft transfer method based on multi-sensor fusion, applicable to short-haul aircraft ground transport to hangars. Through multi-sensor data fusion and multi-trailer collaborative control, it improves transfer accuracy, safety, and efficiency. The method specifically includes the following steps: system setup and sensor calibration; trailer attitude initialization and collaborative parameter configuration; real-time acquisition and preprocessing of multi-sensor data; multi-sensor data fusion and state estimation; dynamic path planning based on fused data; attitude error quantification and calibration based on complex statistics; multi-trailer collaborative driving control and traction distribution; real-time environmental monitoring and risk warning; precise docking at the hangar entrance and attitude fine-tuning; and parking and system review and optimization.

Description

Technical Field

[0001] Specifically, this invention relates to a three-vehicle collaborative aircraft transport method and system based on multi-sensor fusion. Background Technology

[0002] Aircraft ground transport to depots is a crucial link in the aviation support system, primarily used for short-distance relocation during aircraft maintenance, parking, and upkeep. Currently, the mainstream aircraft transport methods in the industry are mainly divided into three categories: single-trailer towing, dual-trailer symmetrical towing, and manual-assisted towing. Among these, single-trailer towing is the most widely used due to its simple structure and convenient operation, but it suffers from concentrated traction torque, poor turning flexibility, and high requirements for ground flatness, making it particularly suitable for transporting small and medium-sized aircraft. Dual-trailer symmetrical towing improves traction stability by symmetrically distributing two trailers on either side of the aircraft's center of gravity, but this method requires high precision in the synchronous control of the two trailers, making it prone to attitude deviation. Manual-assisted towing relies on operator experience to adjust the trailer's attitude, resulting in low efficiency and significant susceptibility to human factors, making safety difficult to guarantee. Regarding sensor applications, existing transport methods mostly use a single sensor for positioning and attitude detection; common sensors include Global Positioning System (GPS), Inertial Measurement Unit (IMU), and ultrasonic sensors. Among them, GPS positioning has high accuracy, but it is prone to signal loss or drift in scenarios such as around hangars or under the obstruction of tall buildings; IMU can collect attitude information such as acceleration and angular velocity of trailers in real time and is not affected by environmental obstruction, but there is cumulative error and the accuracy will decrease significantly with long-term use; ultrasonic sensors are mainly used for obstacle detection, but the detection range is limited and they are easily affected by environmental noise.

[0003] Existing aircraft transport methods face numerous technical bottlenecks in practical applications, specifically: First, insufficient accuracy in multi-trailer coordination. Current dual-trailer or multi-trailer coordination methods lack effective data fusion and synchronization control mechanisms. Position and attitude information between trailers cannot be shared and calibrated in real time, easily leading to problems such as asynchronous traction and attitude deviation, resulting in uneven stress on the aircraft and even damage to the fuselage structure. Second, poor reliability of sensor data. Single sensors, affected by their own characteristics and environmental factors, cannot stably output accurate positioning and attitude information. In complex ground environments (such as obstructions, turbulence, and electromagnetic interference), data distortion is prone to occur, affecting transport safety. Third, weak path planning and dynamic adjustment capabilities. Existing methods mostly adopt preset path planning modes, unable to dynamically adjust according to real-time environmental changes (such as sudden obstacles or changes in ground slope), and lack quantitative analysis and real-time calibration methods for attitude errors during aircraft transport. Fourth, it is difficult to balance transport efficiency and safety. To avoid attitude deviation and structural damage, existing methods mostly employ low-speed transport modes and require extensive manual monitoring, resulting in low transport efficiency. Furthermore, manual monitoring suffers from response delays, making it impossible to promptly mitigate potential risks. Fifth, there is a lack of standardized collaborative control procedures. Existing multi-trailer collaborative operations rely on operator experience, resulting in non-standardized procedures and significant differences in operation among different operators. This leads to unstable transport quality, making it difficult to meet the high-precision transport requirements of large aircraft. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a three-vehicle collaborative aircraft transfer method based on multi-sensor fusion, which can effectively solve the aforementioned problems.

[0005] To achieve the above requirements, the technical solution adopted by the present invention is as follows: A three-vehicle cooperative aircraft transfer method based on multi-sensor fusion is provided, which includes the following steps:

[0006] S1: Steps for system setup and sensor calibration;

[0007] S2: Steps for initializing trailer attitude and configuring collaborative parameters;

[0008] S3: Steps for real-time acquisition and preprocessing of multi-sensor data;

[0009] S4: Steps for multi-sensor data fusion and state estimation;

[0010] S5: Steps for performing dynamic path planning based on fused data;

[0011] S6: Steps for quantifying and calibrating attitude error based on complex statistics;

[0012] S7: Steps for controlling and distributing traction for multi-trailer cooperative driving;

[0013] S8: Steps for conducting real-time environmental monitoring and risk warning;

[0014] S9: Steps for precise docking and attitude fine-tuning at the hangar entrance;

[0015] S10: Steps for inbound storage and system review and optimization.

[0016] The advantages of this three-vehicle collaborative aircraft transfer method based on multi-sensor fusion are as follows:

[0017] 1. Improve the accuracy of multi-trailer coordination: Through multi-sensor fusion technology and a standardized ten-step coordination process, the three trailers are precisely synchronized and controlled. The attitude synchronization error is controlled within ≤1° and the position synchronization error is controlled within ≤8cm, which is significantly better than the existing dual-trailer coordination method (attitude synchronization error ≤3°, position synchronization error ≤20cm). This avoids uneven aircraft stress and attitude deviation caused by insufficient coordination accuracy, and reduces the risk of fuselage structural damage.

[0018] 2. Enhanced data reliability and environmental adaptability: Employing multi-sensor fusion technology combining GPS, IMU, LiDAR, ultrasonic sensors, and visual sensors, along with federated Kalman filtering and preprocessing algorithms, it overcomes the influence of single sensors on factors such as obstruction, noise, and electromagnetic interference. Even in complex ground environments (such as the area around hangars, obstruction by tall buildings, and bumpy ground), it can still output stable and accurate data, improving positioning accuracy to ±2cm and significantly enhancing environmental adaptability.

[0019] 3. Achieve dynamic path planning and precise attitude control: Introduce improved algorithms and dynamic window methods to achieve global optimization and local dynamic adjustment of the transfer path. Combine complex statistical methods such as Bayesian estimation and Bootstrap resampling to achieve precise quantification and real-time calibration of attitude error, ensuring that the aircraft travels smoothly along the optimal path and improving the docking accuracy to ≤3cm, meeting the high-precision transfer requirements of large aircraft.

[0020] 4. Improve transportation efficiency and safety: By using multi-trailer collaborative control and dynamic traction distribution, the driving speed is increased (by 30% compared to existing methods), and the transportation time is shortened; a multi-dimensional environmental monitoring and three-level risk warning mechanism is established to avoid risks such as collisions and connection failures in a timely manner. Combined with emergency braking measures, the transportation safety is significantly improved and the accident rate is reduced.

[0021] 5. Establish a closed-loop optimization mechanism: Through data storage and review optimization during the transfer process, continuously adjust system parameters and control strategies to improve system adaptability and stability. This allows the system to adapt to the transfer needs of different aircraft models (from small and medium-sized aircraft to large passenger aircraft), reduce reliance on operator experience, and achieve standardized and intelligent transfer. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, use the same reference numerals to denote the same or similar parts. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0023] Figure 1 A schematic flowchart of a three-vehicle cooperative aircraft transfer method based on multi-sensor fusion according to an embodiment of this application is shown. Detailed Implementation

[0024] To make the objectives, technical solutions and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and specific embodiments.

[0025] In the following description, references to "an embodiment," "an embodiment," "an example," "example," etc., indicate that the described embodiment or example may include a particular feature, structure, characteristic, property, element, or limitation, but not every embodiment or example necessarily includes that particular feature, structure, characteristic, property, element, or limitation. Furthermore, the repeated use of the phrase "an embodiment according to this application," while possibly referring to the same embodiment, does not necessarily refer to the same embodiment.

[0026] For simplicity, certain technical features known to those skilled in the art are omitted in the following description.

[0027] This invention provides a three-trailer coordinated aircraft transport method based on multi-sensor fusion. The three trailers are a lead trailer, a left auxiliary trailer, and a right auxiliary trailer, arranged in an isosceles triangle. The lead trailer is located directly in front of the aircraft's center of gravity, while the left and right auxiliary trailers are symmetrically distributed on either side of the aircraft's center of gravity. This method utilizes multi-sensor fusion technology based on GPS, IMU, lidar, ultrasonic sensors, and visual sensors. It achieves precise aircraft transport through ten logically connected steps, each step building upon the previous one and providing a foundation for the next. The specific steps are as follows:

[0028] Step 1: System Setup and Sensor Calibration

[0029] This step forms the foundation of the entire transport method. Its core is the construction and precise calibration of the multi-sensor system to ensure the initial reliability of the sensor data. First, sensor kits are installed on three trailers: each trailer is equipped with a high-precision GPS module (positioning accuracy ±1cm), an IMU module (sampling frequency 100Hz, capable of collecting three-axis acceleration, three-axis angular velocity, and three-axis magnetic field strength), a lidar (detection range 0.5-50m, angular resolution 0.1°), an ultrasonic sensor (detection range 0.1-5m, sampling frequency 50Hz), and a vision sensor (high-definition camera, 30fps frame rate, with target recognition function). Second, a data transmission and central control platform is built: a 5G+WiFi6 dual-mode communication module enables real-time transmission of sensor data from each trailer (transmission latency ≤20ms). The central control platform uses an industrial-grade embedded processor, running a Linux operating system and ROS (Robot Operating System), to receive, process, fuse, and output control commands. Finally, sensor calibration is performed: the GPS module uses differential positioning technology, combined with base station data, for static calibration to eliminate system errors; the IMU module uses a six-position calibration method to correct zero bias and scale coefficient errors, while also performing temperature compensation to reduce the impact of ambient temperature on measurement accuracy; the lidar and vision sensor use a hand-eye calibration method to establish the transformation relationship between the sensor coordinate system and the trailer body coordinate system; the ultrasonic sensor uses standard distance calibration to correct detection errors. After calibration, the validity of each sensor data is verified to ensure that the data acquisition accuracy meets the design requirements (positioning error ≤3cm, attitude angle error ≤0.5°). The core principle of this step is to eliminate sensor defects and installation errors through hardware adaptation and multi-dimensional calibration, providing high-quality raw data for subsequent data fusion. Inaccurate sensor calibration will lead to distortion of subsequent data fusion results, affecting the accuracy of collaborative control.

[0030] Step 2: Trailer attitude initialization and coordination parameter configuration

[0031] This step, based on the calibrated system built in Step 1, completes the attitude initialization and coordination parameter configuration of the three trailers, determining the initial position and relative relationship of each trailer. First, the three trailers are parked in a preset isosceles triangle layout: the lead trailer is positioned 5m directly in front of the aircraft's center of gravity, the left and right auxiliary trailers are positioned 3m to the left and 3m to the right of the aircraft's center of gravity respectively, with the angle between the lead trailer and the line connecting them to the lead trailer being 60°, ensuring a uniform distribution of traction force on the aircraft from the three trailers. Second, trailer attitude initialization is performed: the initial positioning coordinates of each trailer are collected using a GPS module, establishing a global coordinate system (with the hangar entrance center point as the origin, the X-axis along the hangar entrance direction, the Y-axis perpendicular to the X-axis, and the Z-axis perpendicular to the ground); the initial attitude angles (roll angle, pitch angle, and yaw angle) of each trailer are collected using an IMU module, combined with aircraft fuselage markers captured by a visual sensor, to correct the relative attitude between the trailers and the aircraft, ensuring precise docking of the trailer towing interface with the aircraft towing point (docking error ≤ 5cm). Finally, configure the collaborative control parameters: Based on parameters such as the model, weight, and center of gravity of the transport aircraft, determine the traction torque distribution ratio of each trailer (the lead trailer bears 60% of the traction force, and the left and right auxiliary trailers each bear 20%); set collaborative control thresholds, including position synchronization error threshold (≤8cm), attitude synchronization error threshold (≤1°), and speed synchronization error threshold (≤0.2m / s); configure the data fusion frequency (100Hz) and control command output frequency (50Hz). The core principle of this step is to determine the initial state of each trailer through initialization and clarify the collaborative control standard through parameter configuration, providing a benchmark for subsequent synchronous operation of multiple trailers. If the initial attitude deviation is too large or the collaborative parameter configuration is unreasonable, it will lead to problems such as uneven traction force distribution and attitude deviation during subsequent collaborative control.

[0032] Step 3: Real-time acquisition and preprocessing of multi-sensor data

[0033] This step builds upon the initialization state in step two, enabling real-time acquisition and preprocessing of multi-sensor data to remove invalid data and improve data quality. First, each sensor is activated for real-time data acquisition: the GPS module acquires the global coordinates (X, Y, Z) of each trailer in real time; the IMU module acquires the triaxial acceleration (ax, ay, az), triaxial angular velocity (ωx, ωy, ωz), and triaxial magnetic field strength (Bx, By, Bz) of each trailer in real time; the lidar acquires surrounding environmental information (obstacle location, ground flatness); the ultrasonic sensor detects the close proximity of the trailers to the aircraft and surrounding obstacles in real time; and the visual sensor captures real-time images of the aircraft fuselage and surrounding environment, identifying fuselage markers and potential obstacles. Secondly, data preprocessing is performed: GPS and IMU data are smoothed using a moving average filter to eliminate high-frequency noise (filter window size set to 5); GPS positioning data is initially corrected using a Kalman filter to reduce signal drift (state equation: X(k)=X(k-1)+B*u(k)+w(k), observation equation: Z(k)=H*X(k)+v(k)); LiDAR and ultrasonic sensor data are threshold-filtered to remove abnormal data (distance threshold set to 0.1-50m); visual sensor images are grayscaled, edge-detected, and target-recognized to extract effective feature information. Finally, the preprocessed data is encapsulated with a unified timestamp (based on the central control platform clock) and transmitted to the data fusion module. The core principle of this step is to improve the reliability and consistency of the original data through filtering, screening, and time synchronization, providing high-quality input for subsequent data fusion. Insufficient data preprocessing can lead to deviations in the fusion results, affecting subsequent path planning and attitude control.

[0034] Step 4: Multi-sensor data fusion and state estimation

[0035] This step, based on the multi-sensor data preprocessed in step three, employs a federated Kalman filter algorithm to achieve data fusion and complete the state estimation of each trailer and aircraft. First, a multi-sensor fusion model is constructed: GPS positioning data, IMU attitude data, LiDAR environmental data, ultrasonic sensor distance data, and visual sensor feature data are used as fusion inputs to establish a state vector X=[x,y,z,φ,θ,ψ,vx,vy,vz] (where x,y,z are position coordinates, φ,θ,ψ are attitude angles, and vx,vy,vz are velocities). Secondly, federated Kalman filtering is used for data fusion: the sensor systems of the three trailers are divided into three local filtering subsystems. Each subsystem is responsible for processing its own sensor data and obtaining local state estimates through local Kalman filtering. The central filtering subsystem receives the state estimates and covariance matrices of the three local subsystems and uses an adaptive weight allocation algorithm (weights are dynamically adjusted according to the reliability of each subsystem's data; reliability is calculated through data variance, with smaller variance resulting in larger weights) for global fusion to obtain the globally optimal state estimate, including the precise position, attitude, and speed of each trailer, as well as the overall state of the aircraft (center of gravity position, attitude angles). Finally, the validity of the fusion result is verified: the covariance matrix of the fusion result is calculated. If the covariance value is less than a preset threshold (≤0.01), the fusion result is considered valid; if the covariance value is greater than the threshold, the process returns to step three for data acquisition and preprocessing again, while adjusting the filtering parameters. The core principle of this step is to leverage the redundancy and complementarity of multi-sensor data, and achieve data fusion through federated Kalman filtering to improve the accuracy and reliability of state estimation, overcome the limitations of a single sensor, and provide accurate state information for subsequent path planning and collaborative control. Insufficient data fusion accuracy can lead to path planning deviations and attitude control errors.

[0036] Step 5: Dynamic path planning based on fused data

[0037] This step, based on the globally optimal state estimate obtained in step four, combines the hangar location and real-time environmental information to achieve dynamic path planning. First, the starting and ending points of the target path are determined: the starting point is the current aircraft's center of gravity, and the ending point is the preset parking location within the hangar (determined based on the aircraft model and hangar layout). Simultaneously, path constraints are defined, including turning radius constraints (determined based on the aircraft's minimum turning radius, ≥15m), ground slope constraints (≤5°), and obstacle avoidance constraints (distance from obstacles ≥2m). Second, an improved algorithm is used for initial path planning: the global coordinate system is divided into a 1m×1m grid, with each grid marked as passable, impassable (obstacle area), or semi-passable (slope exceeding limit area). Using path length, number of turns, and ground flatness as cost functions (cost function f(n)=g(n)+h(n), where g(n) is the actual cost from the starting point to the current node, and h(n) is the estimated cost from the current node to the ending point), the initial optimal path is searched. Finally, dynamic path adjustment is performed: real-time environmental change information (such as sudden obstacles or changes in ground slope) is received from LiDAR and vision sensors. If an impassable area is detected on the path, local path replanning is immediately initiated, using a dynamic window method to adjust the local path (the window size is dynamically adjusted according to speed and steering angle) to ensure that the path always meets the constraints. The core principle of this step is to combine accurate state information and real-time environmental information, and achieve global path planning and local adjustment through improved algorithms and the dynamic window method, ensuring the optimality and safety of the transport path, and providing a path basis for subsequent multi-trailer collaborative driving. If the path planning is unreasonable, it will lead to trailer driving conflicts, attitude deviations, or collisions with obstacles.

[0038] Step Six: Attitude Error Quantification and Calibration Based on Complex Statistics

[0039] This step, based on the path planned in step five and the state estimation results in step four, quantifies and calibrates the attitude errors of the trailer and aircraft in real time. First, an attitude error model is constructed: based on the target attitude planned by the path ( , , Using φ as a reference, the deviation between the actual attitude (φ,θ,ψ) and the target attitude is taken as the attitude error. Assume the attitude error follows a normal distribution N(μ,σ²), where μ is the mean error and σ² is the variance. Next, Bayesian estimation is used to quantify the attitude error: a Bayesian estimation model is established, incorporating prior information (obtained from statistical analysis of historical attitude error data). , The posterior probability distribution is calculated by combining the likelihood information (attitude error sample data obtained from current multi-sensor fusion) with the Bayesian estimation formula as follows:

[0040] ;

[0041] in, For the currently collected n attitude error samples (n=50). Let p(μ,σ²) be the likelihood function, representing the likelihood probability following a normal distribution. Let p(μ,σ²) be the prior probability distribution, using a conjugate prior distribution (normal-inverse gamma distribution). Posterior estimates of the mean error μ and variance σ² are obtained through Bayesian estimation, quantifying the attitude error. To improve estimation accuracy, a Bootstrap resampling method is used to resample the sample data (resampling times B=1000), calculating the error estimate for each resampling, ultimately obtaining the confidence interval for the error estimate (confidence level set to 95%). Finally, real-time calibration is performed based on the quantified attitude error: if the mean error μ is within the confidence interval and its absolute value is less than the attitude synchronization error threshold (≤1°), fine-tuning is performed by adjusting the steering angle and driving speed of each trailer; if the mean error μ exceeds the threshold, the sensor data fusion results and path planning rationality are re-examined, and the traction torque distribution ratio is adjusted to ensure the attitude error is controlled within the allowable range. The meanings of the symbols in this step are as follows: , , φ, θ, ψ represent the target roll angle, pitch angle, and heading angle; φ, θ, ψ represent the actual roll angle, pitch angle, and heading angle; Δφ, Δθ, Δψ represent the errors in roll angle, pitch angle, and heading angle; μ represents the mean error; σ² represents the variance of the error. These are attitude error samples; This is the posterior probability distribution; Let be the likelihood function; p(μ,σ²) be the prior probability distribution; and B be the number of Bootstrap resampling iterations. The core principle of this step is to achieve accurate quantification of attitude error by combining Bayesian estimation with Bootstrap resampling, providing a quantitative basis for attitude calibration and overcoming the shortcomings of existing methods that cannot accurately quantify errors. If the error quantification is inaccurate, the calibration measures will be insufficiently targeted and unable to effectively eliminate attitude deviations.

[0042] Step 7: Multi-trailer cooperative driving control and traction distribution

[0043] This step, based on the attitude calibration results from step six and the dynamic path from step five, achieves coordinated driving control and dynamic traction distribution for the three trailers. First, a coordinated driving control strategy is formulated: the central control platform outputs target speed and steering angle commands to the three trailers based on the planned path and attitude calibration results; the lead trailer, acting as the main control unit, sends its own driving status (speed, attitude, and position) to the left and right auxiliary trailers in real time. The auxiliary trailers adjust their own driving status according to the lead trailer's status and the central command, ensuring synchronization among the three. A PID control algorithm is used to adjust the driving speed and steering angle of each trailer. For speed PID control, the proportional coefficient Kp = 5.0, integral coefficient Ki = 0.1, and derivative coefficient Kd = 0.5; for steering PID control, the proportional coefficient Kp = 3.0, integral coefficient Ki = 0.05, and derivative coefficient Kd = 0.2. Second, dynamic traction distribution is performed: based on the attitude error quantified in step six and the real-time aircraft status, a fuzzy control algorithm is used to adjust the traction torque distribution ratio of each trailer. The inputs to the fuzzy control are attitude errors Δφ, Δθ, and Δψ, and the aircraft's center of gravity offset ΔG. The output is the traction torque adjustment coefficient for each trailer. If a left-side attitude deviation occurs (Δφ>0), the traction torque of the right-side auxiliary trailer is increased, and the traction torque of the left-side auxiliary trailer is decreased. If a heading angle deviation occurs (Δψ>0), the traction torque of the left-side auxiliary trailer is increased, and the steering angle of the lead trailer is adjusted. Finally, the driving status and traction output of each trailer are monitored in real time. If the speed synchronization error exceeds the threshold (≤0.2m / s) or the traction distribution is unbalanced (traction deviation of a single trailer ≥10%), the PID parameters and fuzzy control rules are immediately adjusted. The core principle of this step is to achieve synchronous driving and dynamic traction distribution of multiple trailers by combining PID control and fuzzy control, ensuring that the aircraft travels smoothly along the planned path. Insufficient coordination control precision can lead to asynchronous trailer driving, causing aircraft attitude deviation or uneven structural stress.

[0044] Step 8: Real-time environmental monitoring and risk early warning

[0045] This step builds upon the cooperative driving control in step seven, combining multi-sensor fusion environmental data to achieve real-time environmental monitoring and risk warning, ensuring safe transport. First, a multi-dimensional environmental monitoring model is constructed: LiDAR and visual sensors monitor obstacles (such as equipment, personnel, and buildings) around the transport path in real time, calculating the distance and relative speed between obstacles and the trailer / aircraft to assess collision risk; ultrasonic sensors monitor the connection status between the trailer and aircraft in real time, detecting changes in the distance at the connection point to assess connection stability; and IMU and LiDAR are combined to monitor ground flatness in real time, calculating ground slope and bumpiness to assess driving stability. Second, a risk warning mechanism is established: three levels of risk warning thresholds are set: Level 1 (low risk): obstacle distance ≥ 5m, ground slope ≤ 3°, connection distance change ≤ 2cm; Level 2 (medium risk): 3m ≤ obstacle distance < 5m, 3° < ground slope ≤ 5°, 2cm < connection distance change ≤ 4cm; Level 3 (high risk): obstacle distance < 3m, ground slope > 5°, connection distance change > 4cm. Based on monitoring data and early warning thresholds, a fuzzy comprehensive evaluation method is used to calculate the risk level (evaluation indicators include collision risk, connection stability risk, and driving stability risk, with weights of 0.5, 0.3, and 0.2, respectively). Finally, corresponding measures are taken for different risk levels: For a Level 1 warning, normal driving is maintained with continuous monitoring; for a Level 2 warning, driving speed is reduced (to 70% of the original speed), and the route is adjusted to avoid the risk; for a Level 3 warning, emergency braking is immediately triggered to stop the transfer, and the transfer is restarted after the risk is eliminated. The core principle of this step is to promptly identify potential risks and take countermeasures through multi-dimensional environmental monitoring and graded early warning, avoiding safety accidents caused by collisions, connection failures, or ground bumps. If environmental monitoring is not timely or the early warning mechanism is unreasonable, risks cannot be avoided in time, affecting the safety of the transfer.

[0046] Step Nine: Precise docking and attitude fine-tuning at the hangar entrance

[0047] This step builds upon the safe operating conditions established in step eight. When the aircraft approaches the hangar entrance (≤20m away), the positioning and attitude control accuracy is enhanced to achieve precise docking at the hangar entrance. First, a high-precision positioning mode is activated: the update frequency of the GPS differential positioning base station data is increased (from 10Hz to 20Hz), and combined with close-range precise measurement data from LiDAR and visual sensors, a fusion positioning algorithm is used to improve positioning accuracy (positioning error ≤2cm). The normal operating speed mode is then deactivated, and a low-speed precision operating mode (operating speed ≤0.5m / s) is activated to reduce the impact of operating speed on attitude control. Secondly, fine-tuning of the hangar entrance attitude is performed: using the hangar entrance centerline as a reference, visual sensors are used to identify the hangar entrance marking line and the aircraft fuselage marking points, calculating the attitude deviations between the aircraft and the hangar entrance (lateral deviation Δx, heading angle deviation Δψ). A model predictive control algorithm (prediction step size N=5, control step size M=2) is used to precisely fine-tune the steering angles and travel speeds of the three trailers, ensuring that the aircraft fuselage is parallel to the hangar entrance centerline (heading angle deviation Δψ≤0.3°) and lateral deviation Δx≤3cm. Finally, real-time feedback of the docking status is provided: ultrasonic sensors monitor the distance between the aircraft and both sides of the hangar entrance in real time, ensuring that the distance difference between the two sides is ≤2cm; the central control platform displays the docking progress and attitude deviations in real time. If the deviation exceeds the threshold, fine-tuning is immediately stopped, and the adjustment parameters are recalculated. The core principle of this step is to achieve precise docking between the aircraft and the hangar entrance through high-precision positioning and model predictive control, overcoming the problems of limited space and high positioning difficulty in the hangar entrance area, and providing a precise attitude basis for subsequent parking. If the docking accuracy is insufficient, the aircraft may not be able to enter the hangar smoothly or may collide with the hangar entrance.

[0048] Step 10: Warehouse Storage and System Review and Optimization

[0049] This step, building upon the precise docking in step nine, completes the aircraft's parking and reviews the entire transfer process to improve subsequent transfer efficiency and accuracy. First, parking: Following the pre-set parking location within the hangar, three trailers are controlled to move synchronously at low speeds (≤0.3m / s) to tow the aircraft to the target parking position. LiDAR and visual sensors monitor the distance between the aircraft and hangar equipment and walls in real time, ensuring the parking distance meets safety requirements (≥1.5m from surrounding equipment, ≥2m from walls). Upon reaching the target location, each trailer stops towing, releasing the towing interface to complete parking. Second, data storage and review: The central control platform stores raw sensor data, preprocessed data, fusion results, path planning data, control commands, attitude error data, and risk warning information throughout the transfer process, establishing a transfer database. Data analysis algorithms are used to review the transfer process, calculating key indicators including positioning accuracy, attitude synchronization accuracy, travel speed, transfer time, and the number of risk warnings. Finally, parameter optimization: Based on the review results, the gradient descent algorithm was used to adjust sensor calibration parameters, data fusion filtering parameters, PID control parameters, and path planning cost function weights. For problems encountered during the transfer process (such as excessive attitude deviation and frequent risk warnings), the collaborative control strategy and warning thresholds were optimized. The optimized parameters and strategies were updated to the system to improve the accuracy and efficiency of subsequent transfers. The core principle of this step is to complete the transfer task through storage and parking, and to continuously improve system performance through review and optimization, forming a closed-loop optimization mechanism. This overcomes the shortcomings of existing methods that lack continuous optimization capabilities. If the review and optimization are insufficient, the system performance will not improve and will be unable to adapt to the transfer needs of different aircraft models and complex environments.

[0050] Core Innovation Points

[0051] This invention addresses the problems in existing aircraft ground transport technologies, such as insufficient accuracy of multi-trailer coordination, poor reliability of sensor data, weak path planning and dynamic adjustment capabilities, difficulty in balancing transport efficiency and safety, and lack of standardized collaborative control processes.

[0052] This invention achieves accurate acquisition and fusion of positioning, attitude, and environmental information by constructing a multi-sensor fusion data processing module.

[0053] A standardized ten-step collaborative control process is designed to ensure synchronization and attitude consistency among trailers; complex statistical methods are introduced to quantify attitude errors and achieve real-time calibration.

[0054] By combining dynamic path planning algorithms, the adaptability to complex environments is improved, ultimately enabling efficient, accurate, and safe transport of aircraft into the depot.

[0055] The present invention will be further described in detail below with reference to specific embodiments. This embodiment takes the ground transportation of a large passenger aircraft (fuselage length 70m, weight 200t) into a warehouse as the application scenario. Three trailers are electric towing trailers (rated towing force 50t / unit). The sensors are high-precision industrial-grade products (GPS module is Trimble R10, IMU module is ADIADIS16488, lidar is Velodyne VLP-16, ultrasonic sensor is SensComp600, and vision sensor is Baslerac A2500-14uc).

[0056] The implementation steps are as follows:

[0057] 1. System setup and sensor calibration: Sensor kits and communication modules were installed on three trailers to build a central control platform; GPS was calibrated using differential positioning technology, IMU was calibrated using the six-position calibration method, and LiDAR and vision sensors were calibrated using the hand-eye calibration method. Ultrasonic sensors were calibrated using standard distances. After calibration, the positioning error was ≤3cm and the attitude angle error was ≤0.5°.

[0058] 2. Trailer attitude initialization and coordination parameter configuration: Park the three trailers in an isosceles triangle layout, with the lead trailer positioned 5m directly in front of the aircraft's center of gravity and the left and right auxiliary trailers symmetrically distributed 3m to either side of the center of gravity; collect initial positioning and attitude data, correct relative attitudes, and ensure precise docking of the traction interface (error ≤ 5cm); configure the traction torque distribution ratio (60% for the lead trailer and 20% for each of the left and right trailers), and the synchronization error thresholds (position ≤ 8cm, attitude ≤ 1°, speed ≤ 0.2m / s).

[0059] 3. Real-time acquisition and preprocessing of multi-sensor data: The sensor acquisition data is initiated, and GPS and IMU data are processed by moving average filtering and Kalman filtering. LiDAR and ultrasonic sensor data are processed by threshold filtering. Image preprocessing is performed to extract visual features, and data is packaged and transmitted according to a unified timestamp.

[0060] 4. Multi-sensor data fusion and state estimation: Federated Kalman filtering is used to fuse multi-sensor data. Three local subsystems process their own data, and the central subsystem adaptively weights and fuses the data to obtain a global state estimate with a covariance ≤ 0.01, ensuring accurate state estimation.

[0061] 5. Dynamic path planning based on fused data: Taking the hangar entrance as the destination, the path constraints are determined, and an improved algorithm is used to plan the initial path. When the lidar detects a sudden obstacle, the local path is adjusted using the dynamic window method.

[0062] 6. Attitude error quantification and calibration based on complex statistics: 50 attitude error samples were collected, and Bayesian estimation combined with Bootstrap resampling (1000 resampling times) was used to quantify the error to obtain a 95% confidence interval. When the error exceeds the threshold, the steering and speed calibration are adjusted.

[0063] 7. Multi-trailer cooperative driving control and traction distribution: PID control adjusts speed and steering, fuzzy control dynamically distributes traction, real-time monitoring of synchronization, and speed synchronization error ≤0.2m / s.

[0064] 8. Real-time environmental monitoring and risk warning: Multi-sensor environmental monitoring, fuzzy comprehensive evaluation of risk level, reduction to 70% of the original speed when a level 2 warning occurs, and emergency braking when a level 3 warning occurs.

[0065] 9. Precise docking and attitude fine-tuning at the hangar entrance: High-precision positioning is initiated when the distance to the hangar entrance is 20m, the speed is reduced to 0.5m / s, and the model predictive control fine-tunes the attitude, with a heading angle deviation of ≤0.3° and a lateral movement deviation of ≤3cm.

[0066] 10. Storage and System Review and Optimization: Simultaneously tow the aircraft to the target parking location at low speed, store the transfer data, review and calculate key indicators, optimize parameters using the gradient descent algorithm, and update the system strategy.

[0067] In this embodiment, the aircraft transfer time is reduced by 30% compared to the existing method, the positioning accuracy is improved to ±2cm, the attitude synchronization error is ≤1°, no risk warnings or safety accidents occur, and efficient, accurate and safe warehouse transfer is achieved.

[0068] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the claims.

Claims

1. A three-vehicle collaborative aircraft transfer method based on multi-sensor fusion, characterized in that, Includes the following steps: S1: Steps for system setup and sensor calibration; S2: Steps for initializing trailer attitude and configuring collaborative parameters; S3: Steps for real-time acquisition and preprocessing of multi-sensor data; S4: Steps for multi-sensor data fusion and state estimation; S5: Steps for performing dynamic path planning based on fused data; S6: Steps for quantifying and calibrating attitude error based on complex statistics; S7: Steps for controlling and distributing traction for multi-trailer cooperative driving; S8: Steps for conducting real-time environmental monitoring and risk warning; S9: Steps for precise docking and attitude fine-tuning at the hangar entrance; S10: Steps for inbound storage and system review and optimization.

2. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S1 specifically includes: First, sensor kits were installed on three trailers: each trailer was equipped with a high-precision GPS module, IMU module, LiDAR, ultrasonic sensor, and vision sensor. Second, a data transmission and central control platform was built: a 5G+WiFi6 dual-mode communication module was used to achieve real-time transmission of sensor data from each trailer. The central control platform used an industrial-grade embedded processor, running Linux and ROS, to receive, process, fuse, and output control commands. Finally, sensor calibration was performed: the GPS module used differential positioning technology, combined with base station data, for static calibration to eliminate system errors; the IMU module used a six-position calibration method to correct zero bias and scale coefficient errors, while also performing temperature compensation to reduce the impact of ambient temperature on measurement accuracy; the LiDAR and vision sensors used a hand-eye calibration method to establish the transformation relationship between the sensor coordinate system and the trailer body coordinate system; the ultrasonic sensor used standard distance calibration to correct detection errors. After calibration, the validity of each sensor's data was verified to ensure that the data acquisition accuracy met design requirements.

3. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S2 specifically includes: First, the three trailers are parked in a pre-defined isosceles triangle layout: the lead trailer is positioned 5m directly in front of the aircraft's center of gravity, the left and right auxiliary trailers are positioned 3m to the left and 3m to the right of the aircraft's center of gravity respectively, with the angle between the lead trailer and the line connecting them to the lead trailer being 60°, ensuring that the traction force of the three trailers on the aircraft is evenly distributed. Second, the trailer attitude is initialized: the initial positioning coordinates of each trailer are collected using a GPS module to establish a global coordinate system, with the center point of the hangar entrance as the origin, the X-axis along the direction of the hangar entrance, the Y-axis perpendicular to the X-axis, and the Z-axis perpendicular to the ground; the IMU module is used to collect the initial positioning coordinates of each trailer. The initial attitude angle of the trailer, combined with the aircraft fuselage markers captured by the visual sensor, is used to correct the relative attitude between the trailer and the aircraft, ensuring precise docking between the trailer's towing interface and the aircraft's towing point. Finally, collaborative control parameters are configured: based on parameters such as the model, weight, and center of gravity of the transported aircraft, the traction torque distribution ratio of each trailer is determined, with the lead trailer bearing 60% of the traction force and the left and right auxiliary trailers each bearing 20%; collaborative control thresholds are set, including position synchronization error threshold, attitude synchronization error threshold, and speed synchronization error threshold; and the data fusion frequency and control command output frequency are configured.

4. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S3 specifically includes: First, all sensors are activated for real-time data acquisition: the GPS module acquires the global coordinates (X, Y, Z) of each trailer in real time; the IMU module acquires the triaxial accelerations (ax, ay, az), triaxial angular velocities (ωx, ωy, ωz), and triaxial magnetic field strengths (Bx, By, Bz) of each trailer in real time; the lidar acquires information about the surrounding environment, including obstacle locations and ground flatness; the ultrasonic sensor detects the proximity of the trailers to the aircraft and surrounding obstacles in real time; and the vision sensor captures real-time images of the aircraft fuselage and its surrounding environment, identifying fuselage markers and potential obstacles. Second, data preprocessing is performed: a moving average filtering method is used to process the GPS and IMU data. The data is smoothed to eliminate high-frequency noise, with a filter window size of 5. Kalman filtering is used to initially correct GPS positioning data, reducing signal drift. The state equation is X(k) = X(k-1) + B*u(k) + w(k), and the observation equation is Z(k) = H*X(k) + v(k). Threshold filtering is applied to the lidar and ultrasonic sensor data to remove abnormal data, with a distance threshold set to 0.1-50m. The visual sensor images are then subjected to grayscale conversion, edge detection, and target recognition to extract effective feature information. Finally, the preprocessed data is encapsulated with a unified timestamp and transmitted to the data fusion module.

5. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S4 specifically includes: First, a multi-sensor fusion model is constructed: GPS positioning data, IMU attitude data, LiDAR environmental data, ultrasonic sensor distance data, and visual sensor feature data are used as fusion inputs to establish a state vector X=[x,y,z,φ,θ,ψ,vx,vy,vz], where x,y,z are position coordinates, φ,θ,ψ are attitude angles, and vx,vy,vz are velocities. Second, federated Kalman filtering is used for data fusion: the sensor systems of the three trailers are divided into three local filtering subsystems, each responsible for processing its own sensor data. Local state estimates are obtained through local Kalman filtering; a central filtering subsystem... The system receives the state estimates and covariance matrices of three local subsystems. An adaptive weight allocation algorithm is used, with weights dynamically adjusted based on the data reliability of each subsystem. Reliability is calculated through data variance; the smaller the variance, the greater the weight. Global fusion is then performed to obtain the globally optimal state estimate, including the precise position, attitude, and speed of each trailer, as well as the overall state of the aircraft, including its center of gravity position and attitude angles. Finally, the validity of the fusion result is verified: the covariance matrix of the fusion result is calculated. If the covariance value is less than a preset threshold, the fusion result is considered valid; if the covariance value is greater than the threshold, the process returns to step three to re-acquire and preprocess the data, while simultaneously adjusting the filtering parameters.

6. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S5 specifically includes: First, determine the starting and ending points of the target path: the starting point is the current center of gravity of the aircraft, and the ending point is the preset parking position within the hangar. Simultaneously, define the path constraints, including turning radius constraints, ground slope constraints, and obstacle avoidance constraints. Second, use an improved algorithm for initial path planning: divide the global coordinate system into a 1m×1m grid, with each grid marked as passable, impassable, or partially passable. Use path length, number of turns, and ground smoothness as cost functions (cost function f(n) = g(n) + h(n), where g(n) is the actual cost from the starting point to the current node, and h(n) is the estimated cost from the current node to the ending point) to search for the initial optimal path. Finally, perform dynamic path adjustment: receive environmental change information from LiDAR and visual sensors in real time. If an impassable area is detected on the path, immediately initiate local path replanning, using a dynamic window method to adjust the local path.

7. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S6 specifically includes: First, construct the attitude error model: based on the target attitude of the path planning. , , Using the actual attitude, φ, θ, ψ as a reference, the deviation between the actual attitude and the target attitude is taken as the attitude error. , , Assuming the attitude error follows a normal distribution N(μ,σ²), where μ is the mean error and σ² is the variance, a Bayesian estimation method is used to quantify the attitude error: a Bayesian estimation model is established, which combines prior information and statistically obtained historical attitude error data. , Combined with likelihood information and the attitude error sample data obtained from multi-sensor fusion, the posterior probability distribution is calculated, and the Bayesian estimation formula is as follows: ； in, For the currently collected n attitude error samples (n=50). Let p(μ,σ²) be the likelihood function, representing the likelihood probability following a normal distribution. The prior probability distribution is p(μ,σ²), using a conjugate prior distribution. Posterior estimates of the mean error μ and variance σ² are obtained through Bayesian estimation, quantifying the attitude error. To improve estimation accuracy, a Bootstrap resampling method is used to resample the sample data, calculating the error estimate for each resampling to obtain the confidence interval for the error estimate. Finally, real-time calibration is performed based on the quantified attitude error: if the mean error μ is within the confidence interval and its absolute value is less than the attitude synchronization error threshold, fine-tuning is performed by adjusting the steering angle and driving speed of each trailer; if the mean error μ exceeds the threshold, the sensor data fusion results and path planning rationality are re-examined, and the traction torque distribution ratio is adjusted to ensure the attitude error is controlled within the allowable range. The meanings of the symbols are as follows: , , φ, θ, ψ represent the target roll angle, pitch angle, and heading angle; φ, θ, ψ represent the actual roll angle, pitch angle, and heading angle; Δφ, Δθ, Δψ represent the errors in roll angle, pitch angle, and heading angle; μ represents the mean error; σ² represents the variance of the error. These are attitude error samples; This is the posterior probability distribution; is the likelihood function; p(μ,σ²) is the prior probability distribution; B is the number of Bootstrap resampling attempts.

8. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S7 specifically includes: First, a cooperative driving control strategy is formulated: the central control platform outputs target driving speed and steering angle commands to the three trailers based on the planned path and attitude calibration results; the lead trailer, acting as the main control unit, sends its own driving status to the left and right auxiliary trailers in real time. The auxiliary trailers adjust their own driving status according to the lead trailer's status and the central command, ensuring synchronization among the three. A PID control algorithm is used to adjust the driving speed and steering angle of each trailer. The proportional coefficient Kp=5.0, integral coefficient Ki=0.1, and derivative coefficient Kd=0.5 for speed PID control; the proportional coefficient Kp=3.0, integral coefficient Ki=0.05, and derivative coefficient Kd=0.2 for steering PID control. Second, dynamic traction force distribution is performed: based on... The quantified attitude error and real-time aircraft status are used to adjust the traction torque distribution ratio of each trailer using a fuzzy control algorithm. The inputs to the fuzzy control are the attitude errors Δφ, Δθ, and Δψ, and the aircraft's center of gravity offset ΔG. The output is the traction torque adjustment coefficient of each trailer. If the left attitude deviation Δφ > 0, the traction torque of the right auxiliary trailer is increased, and the traction torque of the left auxiliary trailer is decreased. If the heading angle deviation Δψ > 0, the traction torque of the left auxiliary trailer is increased, and the steering angle of the lead trailer is adjusted. Finally, the driving status and traction output of each trailer are monitored in real time. If the speed synchronization error exceeds the threshold ≤ 0.2 m / s or the traction force distribution is unbalanced and the traction force deviation of a single trailer is ≥ 10%, the PID parameters and fuzzy control rules are immediately adjusted.

9. The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion according to claim 1, characterized in that, Step S8 specifically includes: First, a multi-dimensional environmental monitoring model is constructed: LiDAR and visual sensors are used to monitor obstacles around the transport path in real time, calculating the distance and relative speed between obstacles and the trailer and aircraft to assess collision risk; ultrasonic sensors are used to monitor the connection status between the trailer and aircraft in real time, detecting changes in the distance at the connection point to assess connection stability; and an IMU combined with LiDAR is used to monitor ground flatness in real time, calculating ground slope and bumpiness to assess driving stability. Second, a risk warning mechanism is established: a three-level risk warning threshold is set. Level 1 warning: obstacle distance ≥ 5m, ground slope ≤ 3°, connection distance change ≤ 2cm; Level 2 warning: obstacle distance ≤ 3m. <5m, 3° < ground slope ≤ 5°, 2cm < connection distance change ≤ 4cm; Level 3 warning: obstacle distance < 3m, ground slope > 5°, connection distance change > 4cm. Based on monitoring data and warning thresholds, the fuzzy comprehensive evaluation method is used to calculate the risk level. The evaluation indicators include collision risk, connection stability risk, and driving stability risk, with weights of 0.5, 0.3, and 0.2, respectively. Finally, countermeasures are taken for different risk levels: during Level 1 warning, maintain normal driving and continue monitoring; during Level 2 warning, reduce driving speed and adjust the route to avoid risks; during Level 3 warning, immediately trigger emergency braking, stop transport, and restart after the risk is eliminated. Step S9 specifically includes: First, activate the high-precision positioning mode: increase the update frequency of GPS differential positioning base station data, combine close-range precise measurement data from LiDAR and visual sensors, and use a fusion positioning algorithm to improve positioning accuracy; disable the normal driving speed mode and activate the low-speed precision driving mode to reduce the impact of driving speed on attitude control. Second, perform fine-tuning of the hangar entrance attitude: using the hangar entrance centerline as a reference, identify the hangar entrance marking line and the aircraft fuselage marking points through visual sensors, and calculate the attitude deviation between the aircraft and the hangar entrance; use model predictive control algorithms to precisely fine-tune the steering angle and driving speed of the three trailers to ensure that the aircraft fuselage is parallel to the hangar entrance centerline and the lateral deviation Δx ≤ 3cm. Finally, provide real-time feedback on the docking status: monitor the distance between the aircraft and both sides of the hangar entrance in real time through ultrasonic sensors to ensure that the distance difference between the two sides is ≤ 2cm; the central control platform displays the docking progress and attitude deviation in real time. If the deviation exceeds the threshold, the fine-tuning is immediately stopped and the adjustment parameters are recalculated. Step S10 specifically includes: First, parking: Following the pre-set parking positions within the hangar, three trailers are controlled to move synchronously at low speeds, towing the aircraft to the target parking location. LiDAR and visual sensors monitor the distance between the aircraft and hangar equipment and walls in real time to ensure the parking distance meets safety requirements. Upon reaching the target location, each trailer stops towing, releases the towing interface, and parking is complete. Second, data storage and review: The central control platform stores the entire transfer process's raw sensor data, preprocessed data, fusion results, path planning data, control commands, attitude error data, and risk warning information, establishing a transfer database. Data analysis algorithms are used to review the transfer process, calculating key indicators, including positioning accuracy, attitude synchronization accuracy, etc. The process includes considering factors such as driving speed, transfer time, and the number of risk warnings. Finally, parameter optimization is performed: based on the review results, the gradient descent algorithm is used to adjust sensor calibration parameters, data fusion filtering parameters, PID control parameters, and path planning cost function weights; for problems encountered during the transfer process, the collaborative control strategy and warning thresholds are optimized; the optimized parameters and strategies are updated to the system to improve the accuracy and efficiency of subsequent transfers. The core principle of this step is to complete the transfer task by parking in the warehouse, and to continuously improve system performance through review and optimization, forming a closed-loop optimization mechanism. This overcomes the shortcomings of existing methods that lack continuous optimization capabilities. If the review and optimization are insufficient, the system performance will not be able to improve and will be difficult to adapt to the transfer needs of different aircraft models and complex environments.

10. A three-vehicle collaborative aircraft transfer system based on multi-sensor fusion, characterized in that, The three-vehicle collaborative aircraft transfer method based on multi-sensor fusion as described in any one of claims 1 to 9 is used.

Images