A smart system for semi-trailer drivers to actively improve their driving safety

Abstract

This application relates to the field of new energy vehicle technology and discloses an intelligent system for actively improving driver safety in semi-trailers. The invention establishes a seven-dimensional state representation system covering the complete motion state of the tractor and trailer, making the folding angle change rate the core variable for driving safety assessment. It constructs a six-axis independent adhesion coefficient calculation channel that integrates the motion constraints of the articulated body and the differences in inter-axle adhesion, eliminating the estimation blind spot caused by the lack of slip signals on the trailer axles. It forms a folding evolution prediction mechanism embedded in the non-uniform distribution of inter-axle adhesion, taking into account the calculation accuracy and real-time requirements of high-risk areas. It establishes a cross-vehicle intention response phase analysis and individualized dynamic threshold judgment system, realizing accurate identification of driver expectation failure. It builds a dual-core driving closed loop of physical environment cognition and human behavior cognition, enabling the intervention strategy to continuously evolve with the operating data.

Description

Technical Field

[0001] This application relates to the field of new energy vehicle technology, and in particular to an intelligent system for actively improving driver safety in semi-trailers. Background Technology

[0002] With the deep integration of new energy vehicle technology and intelligent transportation systems, new energy semi-trailers, as articulated vehicles formed by the tractor and trailer connected by a saddle hinge, have a folding angle that exhibits strong nonlinear characteristics under the combined effects of braking, steering, and crosswinds, making it a core state variable that restricts driving safety.

[0003] In existing technologies, the driving safety monitoring of new energy semi-trailers mainly adopts the architecture and algorithms of passenger car ADAS systems. For example, Chinese patent application CN111688707A proposes a road adhesion coefficient estimation method that integrates vision and dynamics, achieving adhesion coefficient estimation through the fusion of road image classification and tire dynamic information; Chinese patent application CN113460056B discloses a vehicle road adhesion coefficient estimation method based on Kalman filtering and least squares, effectively reducing the uncertainty of tire force observation values. Although these methods have a certain accuracy in passenger car scenarios, they use a single-vehicle six-degree-of-freedom model and do not include the folding angle and its rate of change in the state observation dimension, failing to fully characterize the seven-dimensional motion state of the articulated body, resulting in a significant phase deviation between the folding trend prediction and the actual vehicle response. With the deepening research on articulated body dynamics, Chinese patent application CN119305567A points out that existing vehicle road adhesion coefficient estimation methods suffer from insufficient estimation accuracy under different driving conditions. Specifically, in the semi-trailer scenario, existing systems mostly employ fixed threshold warning mechanisms, failing to dynamically adjust intervention strategies based on the phase difference between driver intent and trailer response. Furthermore, they lack overcorrection identification methods based on the relative motion analysis of the tractor and trailer, and have not established a compensation channel to structurally inject the spatial non-uniformity of the inter-axle adhesion coefficient into the folding prediction model. This results in insufficient accuracy in critical period folding determination and the inability to quantify and absorb road surface adhesion differences within the system. At the perception architecture level, existing systems lack means for independent observation of the trailer axle road surface state, making it impossible to guarantee safety when facing conditions where inter-axle adhesion differences exceed the sampling coverage of the drive axle. More critically, existing technologies treat state observation, trend prediction, and intervention execution in isolation, failing to establish a two-way reinforcement closed loop between seven-dimensional state perception, folding trend prediction, and hierarchical collaborative intervention. Consequently, the governance results cannot be fed back to optimize the core dynamic model. Summary of the Invention

[0004] This application proposes an intelligent system for actively enhancing driver safety in semi-trailers. It possesses the advantages of precise perception of articulated body dynamics and cross-vehicle interpretation of driver intent, addressing issues arising from the transplantation of existing passenger vehicle ADAS architectures to semi-trailer scenarios, such as missing dimensions of articulated body motion state observation, blind spots in inter-axle adhesion coefficient estimation, mismatch in folding trend prediction models, failure to identify driver expectation failures, and rigid open-loop intervention strategies. To achieve the above objectives, this application adopts the following technical solution: A state perception module constructs a seven-dimensional state vector for the tractor and trailer; an adhesion decoupling module establishes articulated body constraint equations and uses weighted least squares to solve for the six-axle independent adhesion coefficients; a trend prediction module embeds inter-axle adhesion differences to construct a folding evolution prediction equation; an intent interpretation module uses a dynamic time warping algorithm to calculate cross-vehicle phase differences to determine expectation failures; and a collaborative intervention module constructs a two-dimensional coupled decision matrix to achieve closed-loop feedback, forming a complete enhancement loop of perception, cognition, decision-making, execution, and evolution.

[0005] To achieve the above objectives, this application adopts the following technical solution: an intelligent system for actively improving driver safety in semi-trailers, comprising: a state perception module, an attachment decoupling module, a trend prediction module, an intent parsing module, and a collaborative intervention module; State perception module: Collects the three-degree-of-freedom motion state of the tractor through the inertial measurement unit of the tractor, collects the three-degree-of-freedom motion state of the trailer through the inertial measurement unit of the trailer, and collects the folding angle and its rate of change through the saddle hinge point angle encoder, and outputs the seven-dimensional state data synchronously according to a unified timestamp. The attachment decoupling module receives the seven-dimensional state data output by the state perception module, establishes the relative motion constraint equations between the tractor and the trailer, uses the folding angle and its rate of change as the coupling constraint conditions of the articulated body, constructs the overdetermined observation equation set of independent attachment coefficients for each axis, solves the six-axis independent attachment coefficients, and outputs the attachment space distribution matrix. Trend prediction module: Receives seven-dimensional state data from the state perception module and the attachment space distribution matrix from the attachment decoupling module, constructs the folding evolution prediction equation of the articulated body dynamics model, substitutes the independent attachment coefficients of each axis into the yaw moment calculation of the tractor and trailer, predicts the folding angle trajectory at future moments, and calculates the folding risk probability. Intent parsing module: Real-time acquisition of driver steering input signal as intent signal, reception of trailer yaw rate data as response signal, calculation of phase difference and correlation coefficient between intent signal and response signal, determination of expected failure state, identification of panic overcorrection mode, and two-dimensional coupling decision-making between parsing results and folding risk level of trend prediction module; Collaborative Intervention Module: Receives the risk level of the folding angle from the trend prediction module and the driver status determination from the intent analysis module, constructs a two-dimensional coupled decision matrix of risk level and driver status, triggers differentiated intervention strategies, monitors the trend of folding angle changes and driver operation response after intervention, and feeds back the evaluation results to the attachment decoupling module and the intent analysis module.

[0006] Furthermore, the state awareness module includes a multi-source acquisition unit and a timing alignment unit; The multi-source acquisition unit acquires the longitudinal acceleration, lateral acceleration, and yaw rate of the tractor vehicle's inertial measurement unit, and simultaneously acquires the longitudinal acceleration, lateral acceleration, and yaw rate of the trailer vehicle's inertial measurement unit. The folding angle and its rate of change are obtained through the saddle hinge point angle encoder, and the three-degree-of-freedom motion state of the tractor vehicle and trailer vehicle are calculated to generate the original state data. The timing alignment unit receives the original state data, interpolates to compensate for the sampling frequency differences between the inertial measurement units and angle encoders of the tractor and trailer, corrects for the transmission delay, and fuses the data into a seven-dimensional state vector according to a unified timestamp to ensure the timing consistency of each degree of freedom data and outputs it.

[0007] Furthermore, the attachment decoupling module includes a constraint construction unit and a distributed solution unit; The constraint construction unit receives seven-dimensional state data, extracts the acceleration and yaw rate of the inertial measurement units of the tractor and trailer, establishes the relative motion constraint equations at the hinge point where the displacement is continuous and the velocity is equal, uses the folding angle and its rate of change as the coupling constraint conditions of the hinge body, constructs and outputs the observation equation set of independent adhesion coefficients of each axis. The distributed solution unit receives the observation equation set, introduces online identification parameters of tire lateral stiffness and vertical load, compensates for the dynamic transfer of axle load under braking and steering conditions, uses the weighted least squares method to solve the six-axle independent adhesion coefficient, performs time-domain consistency verification and outlier filtering on the solution results, and outputs the adhesion space distribution matrix.

[0008] Furthermore, the constraint building unit includes motion constraint units and coupled equation units; Motion constraint unit: Receives the seven-dimensional state vector, extracts the longitudinal acceleration, lateral acceleration, and yaw rate of the inertial measurement units of the tractor and trailer, establishes the relative motion constraint equations at the hinge point where the displacement is continuous and the velocity is equal, and generates a constraint parameter set that includes the kinematic relationship between the tractor and trailer and the continuity of the acceleration at the hinge point. Coupled equation unit: Receives the constraint parameter set, takes the folding angle and its rate of change as the coupling constraint condition of the articulated body, establishes the dynamic correlation between the folding angle acceleration and the difference in yaw moment between the tractor and the trailer, introduces the equivalent rotational inertia parameter of the articulated body, and constructs a set of observation equations for the independent adhesion coefficients of each axis that includes the difference in adhesion between axes.

[0009] Furthermore, the distributed solution unit includes a load compensation unit and an attachment solution unit; Load compensation unit: Receives the observation equation set, extracts tire lateral stiffness and vertical load online identification parameters, calculates the dynamic transfer load of each axle under braking and steering conditions based on the vertical acceleration response of the tractor and trailer inertial measurement units, performs real-time compensation and correction of the vertical load, and outputs the compensated axle load parameters. The attachment calculation unit receives the compensated axle load parameters, substitutes them into the observation equations, and uses the weighted least squares method to calculate the six-axis independent attachment coefficients. It performs time-domain consistency verification and outlier filtering on the calculation results, generates an attachment spatial distribution matrix containing the differences in inter-axis attachment, and outputs it to the trend prediction module.

[0010] Furthermore, the trend prediction module includes an evolution modeling unit and a risk projection unit; Evolutionary modeling unit: acquires seven-dimensional state data and attachment space distribution matrix, extracts independent attachment coefficients and folding angle change rates of each axis, establishes dynamic balance relationship between the difference in yaw moment between tractor and trailer and folding angle acceleration, embeds the inter-axle attachment difference into the articulated body dynamics model, constructs and outputs folding evolution prediction equations that reflect the inter-axle attachment difference. Risk simulation unit: Receives the folding evolution prediction equation, takes the current seven-dimensional state as the initial value, uses an adaptive numerical integration algorithm to solve the time-domain evolution sequence of folding angular displacement and angular velocity at future moments, compares the predicted trajectory with the safety threshold point by point, calculates the probability of folding angle exceeding the limit, and classifies the risk level according to the magnitude and duration of the exceedance, and outputs the risk level and the predicted trajectory.

[0011] Furthermore, the intent parsing module includes an intent parsing unit and a behavior determination unit; Intent parsing unit: Real-time acquisition of driver steering input signal as intent signal, synchronous reception of trailer yaw rate data as response signal, interpolation resampling and phase delay compensation of the two, dynamic time warping and phase coupling analysis algorithm to calculate the phase difference and correlation coefficient of cross vehicle body intent signal and response signal, dynamically adjust adaptive threshold based on phase feature baseline based on historical driving data, determine expected failure state, generate parsing result containing phase feature quantity and failure mark and output it; Behavior determination unit: Receives the analysis results, monitors the driver's steering correction behavior after the expected result is lost, extracts the steering wheel angle change rate and correction range, establishes a correction behavior evaluation index based on historical control data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction range exceeds the normal range, couples the phase analysis results with the overcorrection identification results and the fold risk level in a two-dimensional decision-making process, and generates a judgment command that integrates human behavior and vehicle risk.

[0012] Furthermore, the intent parsing unit includes a signal calibration unit and a failure determination unit; Signal calibration unit: Real-time acquisition of driver steering input signal and marking it as intention signal, synchronous reception of trailer yaw rate data and marking it as response signal, interpolation resampling and phase delay compensation of the two, elimination of timing deviation caused by sampling frequency difference and transmission delay, and generation of timing-aligned intention response signal pair after frequency domain filtering; The failure determination unit receives the time-aligned intention-response signal pair, calculates the phase difference and correlation coefficient between the intention signal and the response signal, dynamically adjusts the adaptive threshold based on the phase feature baseline of historical driving data, determines the expected failure state when the phase difference exceeds the threshold or the correlation coefficient is lower than the boundary, and extracts the phase feature quantity to generate the failure identification resolution result.

[0013] Furthermore, the behavior determination unit includes a correction and recognition unit and a coupled decision-making unit; Correction identification unit: Receives the analysis results output by the failure determination unit, monitors the driver's steering correction behavior after the expected failure, extracts the steering wheel angle change rate and correction range, establishes a correction behavior evaluation index based on historical control data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction range exceeds the normal range, and generates overcorrection identification results. Coupled Decision Unit: Receives the overcorrection identification result, synchronously acquires the folded risk level output by the trend prediction module, establishes a two-dimensional coupled decision matrix of human behavior state and vehicle risk level, performs joint judgment based on the coupling weight of risk level and behavior state, and generates a judgment instruction that integrates human behavior and vehicle risk.

[0014] Furthermore, the collaborative intervention module includes a decision-triggered unit and an effect evaluation unit; Decision triggering unit: acquires the risk level and driver status judgment, establishes a two-dimensional coupled decision matrix of risk level and driver status, triggers differentiated intervention strategies such as sound and light reminders, steering assist gain limitation or independent braking force distribution of trailer electronic braking system based on the matrix element mapping relationship, generates intervention commands and outputs them; Effectiveness evaluation unit: Monitors the trend of folding angle change and driver operation response after intervention, evaluates the effectiveness of intervention measures on folding angle convergence and driver behavior recovery, feeds the evaluation results back to the attachment decoupling module to optimize the attachment coefficient identification parameters, and simultaneously feeds them back to the intent parsing module to adjust the phase feature baseline based on historical driving data.

[0015] The beneficial effects of this invention are as follows: This application provides an intelligent system for proactively improving driver safety in semi-trailers. It establishes a seven-dimensional state representation system covering the complete motion state of the tractor and trailer, making the folding angle change rate a core variable for driver safety assessment. It constructs a six-axis independent adhesion coefficient calculation channel that integrates articulated body motion constraints and inter-axle adhesion differences, eliminating estimation blind spots caused by the lack of trailer axle slip signals. It forms a folding evolution prediction mechanism embedded with non-uniform distribution of inter-axle adhesion, balancing computational accuracy and real-time requirements in high-risk areas. It establishes a cross-vehicle intention response phase analysis and individualized dynamic threshold determination system, achieving accurate identification of driver expectation failures. It constructs a dual-core driving closed loop of physical environment cognition and human behavior cognition, enabling intervention strategies to continuously evolve with operational data. This achieves a three-dimensional capability leap from single-vehicle protection to articulated body collaborative protection, from single-point average estimation to inter-axle distribution decoupling, and from fixed threshold open-loop triggering to dynamic closed-loop self-evolution, significantly improving the accuracy of folding risk prediction and the adaptability of intervention strategies under complex working conditions. Attached Figure Description

[0016] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort: Figure 1 This is the overall system flowchart of this application.

[0017] Figure 2 This is an extended diagram of the attached decoupling module in this application.

[0018] Figure 3 This is an extended diagram of the intent parsing module in this application. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1, as Figure 1 An intelligent system for actively improving driver safety in semi-trailers, characterized in that it includes: a state perception module, an attachment decoupling module, a trend prediction module, an intent parsing module, and a collaborative intervention module; State perception module: Collects the three-degree-of-freedom motion state of the tractor through the inertial measurement unit of the tractor, collects the three-degree-of-freedom motion state of the trailer through the inertial measurement unit of the trailer, and collects the folding angle and its rate of change through the saddle hinge point angle encoder, and outputs the seven-dimensional state data synchronously according to a unified timestamp. The attachment decoupling module receives the seven-dimensional state data output by the state perception module, establishes the relative motion constraint equations between the tractor and the trailer, uses the folding angle and its rate of change as the coupling constraint conditions of the articulated body, constructs the overdetermined observation equation set of independent attachment coefficients for each axis, solves the six-axis independent attachment coefficients, and outputs the attachment space distribution matrix. Trend prediction module: Receives seven-dimensional state data from the state perception module and the attachment space distribution matrix from the attachment decoupling module, constructs the folding evolution prediction equation of the articulated body dynamics model, substitutes the independent attachment coefficients of each axis into the yaw moment calculation of the tractor and trailer, predicts the folding angle trajectory at future moments, and calculates the folding risk probability. Intent parsing module: Real-time acquisition of driver steering input signal as intent signal, reception of trailer yaw rate data as response signal, calculation of phase difference and correlation coefficient between intent signal and response signal, determination of expected failure state, identification of panic overcorrection mode, and two-dimensional coupling decision-making between parsing results and folding risk level of trend prediction module; Collaborative Intervention Module: Receives the risk level of the folding angle from the trend prediction module and the driver status determination from the intent analysis module, constructs a two-dimensional coupled decision matrix of risk level and driver status, triggers differentiated intervention strategies, monitors the trend of folding angle changes and driver operation response after intervention, and feeds back the evaluation results to the attachment decoupling module and the intent analysis module.

[0021] In this embodiment, the state perception module, through the coordinated configuration of the inertial measurement units of the tractor and trailer, and the saddle hinge point angle encoder, overcomes the observation limitations of the traditional six-degree-of-freedom model of a single passenger vehicle, establishing a seven-dimensional state perception system covering the three degrees of freedom of the tractor, the three degrees of freedom of the trailer, and the rate of change of the folding angle. This module completes the real-time acquisition and time-series alignment of the complete motion state of the articulated body, enabling the folding angle and its rate of change to be included as correlated variables in driver safety assessment for the first time. This provides a high-dimensional data foundation for subsequent decoupling of inter-axle attachment and prediction of folding trends, fundamentally eliminating the blind spot in folding monitoring caused by the lack of state observation dimensions.

[0022] The attachment decoupling module establishes relative motion constraint equations between the tractor and trailer, and uses the folding angle and its rate of change as coupling constraints for the articulated body, overcoming the limitations of existing technologies that rely on wheel speed differences to indirectly estimate the adhesion coefficient. This module completes the overdetermined observation and calculation of the six-axle independent adhesion coefficients, solving the estimation blind zone problem caused by the sparse drive axles of the semi-trailer and the lack of slip signals on the trailer axles, and achieving online quantification of the spatial non-uniformity of the inter-axle adhesion coefficients. This eliminates the prediction failure risk based on the single-point average adhesion assumption, reducing the folding prediction error from a significant deviation in existing technologies to a controllable range, and providing an accurate physical environment understanding basis for folding evolution prediction.

[0023] The trend prediction module constructs a dynamic model of the articulated body that integrates differences in inter-axle adhesion. It substitutes the independent adhesion coefficients of each axle into the calculation of the yaw moment of the tractor and trailer, overcoming the limitations of traditional single-vehicle dynamic models in articulated vehicles. This module completes the task of predicting the folding angle trajectory at future moments and calculating the risk probability, advancing folding trend prediction from macroscopic average estimation to the microscopic level of inter-axle decoupling. This enables early perception of the folding angle evolution trend, providing drivers with ample reaction and handling time, and effectively avoiding warning delays or misjudgments caused by mismatch between the prediction model and actual vehicle dynamics.

[0024] The intent parsing module, by acquiring the driver's steering input signal and the trailer's yaw rate response signal in real time and calculating their phase difference and correlation coefficient, overcomes the fragmented architecture of existing systems that only monitor vehicle status while ignoring the coupling of human factors and behavior. This module completes the tasks of determining the driver's expected failure state and recognizing overcorrection patterns, transforming driver behavior from a black-box input into quantifiable phase characteristics. This enables accurate identification of the failure of mental models based on passenger vehicle experience and subsequent panic corrections, providing a human factor cognitive dimension for risk decision-making and effectively filling the assessment gap between human-caused risk and vehicle dynamics risk.

[0025] The collaborative intervention module constructs a two-dimensional coupled decision matrix between risk level and driver state, and feeds back the evaluation results to the attachment decoupling module and the intent parsing module, breaking through the rigid strategies of traditional single-dimensional threshold triggering and fixed-level early warning. This module completes the tasks of triggering differentiated intervention strategies and evaluating their effects, realizing a dual-core driven closed loop of physical environment cognition and human behavior cognition. This allows intervention measures to be dynamically adjusted according to the coupling relationship between vehicle risk and driver state, and to continuously optimize the attachment coefficient identification parameters and phase thresholds through feedback, forming a complete reinforcement loop of perception, cognition, decision-making, execution, and evolution.

[0026] Existing technologies directly transplant the passenger car single-vehicle monitoring architecture to the semi-trailer scenario, describe vehicle motion with a single rigid body dynamics model, characterize road surface conditions with a single-point average adhesion coefficient, and trigger warnings with a fixed threshold. Vehicle condition monitoring and driver behavior assessment are disconnected from each other, and there is a lack of feedback pathway between warning intervention and condition perception, resulting in a systemic blind spot in articulated vehicle folding dynamics monitoring.

[0027] The aforementioned technical solution achieves fundamental improvements on three levels: First, it expands the state observation object from a single rigid body to the articulated body composed of the tractor and trailer, establishing a seven-dimensional state representation system. This makes the rate of change of the folding angle the core variable for driving safety assessment, achieving an architectural upgrade from single-vehicle protection to articulated body collaborative protection. Second, it advances the estimation of the adhesion coefficient from single-point averaging to decoupling the spatial distribution between axles, and advances the assessment of driver behavior from black-box input to intention response phase analysis, enabling vehicle dynamics risk and human-caused behavioral risk to form a coupled decision-making process, achieving a paradigm shift from single-dimensional vehicle monitoring to human-vehicle collaborative cognition. Third, it advances early warning intervention from fixed threshold open-loop triggering to differentiated closed-loop feedback, enabling the intervention assessment results to back-optimize the physical model identification parameters and behavioral baseline thresholds, achieving a leap from static rules to dynamic self-evolutionary capabilities.

[0028] This technical solution eliminates the structural mismatch of the existing architecture in the semi-trailer scenario, establishes a collaborative safety protection system covering the physical environment, vehicle dynamics and driver behavior, significantly improves the accuracy of folding risk prediction and the adaptability of intervention strategies under complex working conditions, and provides an implementable technical path for the intelligent upgrade of the active safety system of new energy semi-trailers.

[0029] Example 2, as Figure 1 The state awareness module includes a multi-source acquisition unit and a timing alignment unit; Multi-source acquisition unit: Acquires longitudinal acceleration, lateral acceleration and yaw rate from the inertial measurement unit of the tractor vehicle, and simultaneously acquires longitudinal acceleration, lateral acceleration and yaw rate from the inertial measurement unit of the trailer vehicle. It obtains the folding angle and its rate of change through the saddle hinge point angle encoder, calculates the three-degree-of-freedom motion state of the tractor vehicle and trailer vehicle, and generates raw state data. Timing Alignment Unit: Receives raw state data, interpolates to compensate for the sampling frequency differences between the tractor and trailer inertial measurement units and angle encoders, corrects for transmission delays, and fuses the data into a seven-dimensional state vector using a unified timestamp to ensure the timing consistency of each degree of freedom data before outputting it.

[0030] In this embodiment, the multi-source acquisition unit, by configuring the tractor inertial measurement unit, the trailer inertial measurement unit, and the saddle hinge point angle encoder, synchronously acquires the three-degree-of-freedom motion state of the tractor, the three-degree-of-freedom motion state of the trailer, and the rate of change of the folding angle. This realizes the generation of the original state data of the seven-dimensional motion information of the articulated body, completes the task of expanding the observation dimension from the single-vehicle six-degree-of-freedom model to the complete state representation of the articulated body, and achieves the purpose of eliminating the blind spot in the observation of trailer motion state and folding dynamics information, and providing a complete data source for subsequent decoupling of inter-axle attachment.

[0031] The timing alignment unit receives the original state data, interpolates to compensate for differences in sampling frequencies, corrects for transmission delays, and fuses it into a seven-dimensional state vector according to a unified timestamp. This achieves time synchronization and quality calibration of multi-source heterogeneous data, completes the task of aligning the output of data of each degree of freedom under a unified timing reference, and achieves the purpose of eliminating phase distortion caused by asynchronous sampling and transmission delay, and providing timing-consistent input conditions for the overdetermined observation equations of the attachment decoupling module.

[0032] Existing technologies directly transplant the single-vehicle monitoring architecture of passenger cars to the semi-trailer scenario. Typically, only a single inertial measurement unit (IMU) is deployed to collect the tractor's motion state, neglecting the independent motion and folding angle information of the trailer. Furthermore, the lack of a time-series alignment mechanism for multi-source data leads to trailer response lag and loss of folding dynamics information, resulting in insufficient state perception dimensions and inconsistent data quality. The multi-source acquisition unit and time-series alignment unit, through coordinated configuration of multi-source acquisition and time-series alignment, extend the state perception object from a single rigid body to the articulated body composed of the tractor and trailer, establishing a seven-dimensional state representation system. The multi-source acquisition unit, by configuring the tractor's IMU, trailer's IMU, and the saddle articulation point angle encoder, simultaneously collects the motion state and folding angle change rate of the tractor and trailer, achieving an architectural upgrade from a single-vehicle six-degree-of-freedom model to a seven-dimensional state observation of the articulated body. This eliminates the monitoring blind spots caused by the black box nature of trailer motion and the lack of folding angle information in the traditional architecture. The timing alignment unit achieves accurate fusion of multi-source heterogeneous data under a unified timestamp by interpolating and compensating for sampling frequency differences and correcting for transmission delays. This eliminates phase distortion and calculation deviations caused by data timing mismatch. Together, these two components enable the state perception module to overcome the structural mismatch of a single passenger car model in a semi-trailer scenario. It establishes a complete observation system covering the tractor, trailer, and folding angle, providing a dimensionally complete and temporally consistent data foundation for subsequent inter-axle attachment decoupling estimation and folding trend prediction. This significantly improves the accuracy and reliability of state observation in the active safety system of articulated vehicles.

[0033] Example 3, as Figure 1 The attachment decoupling module includes constraint construction units and distributed solution units; Constraint Construction Unit: Receives seven-dimensional state data, extracts the acceleration and yaw rate of the inertial measurement units of the tractor and trailer, establishes relative motion constraint equations at the hinge point where the displacement is continuous and the velocity is equal, uses the folding angle and its rate of change as the coupling constraint conditions of the hinge body, constructs and outputs the observation equation set of independent adhesion coefficients of each axis. Distributed solution unit: Receives the observation equation set, introduces online identification parameters of tire lateral stiffness and vertical load, compensates for the dynamic transfer of axle load under braking and steering conditions, uses the weighted least squares method to solve the six-axle independent adhesion coefficient, performs time-domain consistency verification and outlier filtering on the solution results, and outputs the adhesion space distribution moment.

[0034] The six-axis independent adhesion coefficients are calculated using the following formulas;

[0035] In the formula: A is a six-axis independent adhesion coefficient estimation vector with a dimension of 6×1. A is the coefficient matrix of the overdetermined observation equation system with a dimension of m×6, where m is the number of observation equations and m>6. The matrix elements are generated by kinematic constraint transformation of the acceleration, yaw rate, and folding angle change rate of the tractor and trailer inertial measurement units. b is the observation vector with a dimension of m×1. Its elements are the constant terms of the equilibrium equations of longitudinal and lateral forces of the tires on each axle. It is obtained by multiplying the acceleration measurement values ​​of the tractor and trailer inertial measurement units by the equivalent mass of each axle, and then subtracting the product of the tire lateral stiffness and slip angle. No additional tire force sensors are required; it only relies on the output of the inertial measurement units. W is a diagonal weighted matrix with a dimension of m×m. The diagonal elements are the normalized values ​​of the vertical load signal-to-noise ratio of each axle, and the off-diagonal elements are zero.

[0036] The weighting matrix W is constructed based on the signal-to-noise ratio of the vertical loads along each axis. The diagonal elements are the normalized values ​​of the signal-to-noise ratio of each axis. The vertical loads with higher signal-to-noise ratios correspond to larger weighting coefficients, while those with lower signal-to-noise ratios correspond to smaller weighting coefficients. This ensures that the high-confidence observation equations dominate the solution process, while the low-confidence observation equations are adaptively suppressed.

[0037] The vertical load signal-to-noise ratio (SNR) is estimated using the Welch periodogram method: the vertical acceleration signals output by the tractor inertial measurement unit and the trailer inertial measurement unit are segmented with a window length of 1024 sampling points and a 50% overlap rate. After applying a Hanning window to each segment, a fast Fourier transform is performed to calculate the power spectral density. The signal power is the integrated power in the 0.5Hz to 10Hz frequency band, and the noise power is the integrated power in the 20Hz to 50Hz frequency band. The ratio of the two is the SNR.

[0038] When the signal-to-noise ratio is below 3dB, the weighting coefficient of this axis is set to 0.1; when the signal-to-noise ratio is above 20dB, the weighting coefficient of this axis is set to 1.0; when it is in between, linear interpolation is used.

[0039] In this embodiment: the constraint construction unit receives a seven-dimensional state vector, extracts the acceleration and yaw rate of the inertial measurement units of the tractor and trailer, establishes a relative motion constraint equation at the articulation point where the displacement is continuous and the velocity is equal, and embeds the folding angle and its rate of change as the coupling constraint condition of the articulation body into the dynamic relationship, constructs the observation equation set of independent adhesion coefficients of each axle, realizes the decoupling of the relative motion constraint of the articulation body from the single vehicle model, completes the equation construction task from the single vehicle architecture of passenger car to the articulation constraint architecture of semi-trailer, and achieves the purpose of providing a mathematical basis including the inter-axle coupling relationship for the calculation of the six-axle independent adhesion coefficient.

[0040] The distributed solution unit receives the observation equation set, introduces online identification parameters of tire lateral stiffness and vertical load, and performs real-time compensation for the dynamic transfer of axle load under braking and steering conditions. It uses the weighted least squares method to solve the independent adhesion coefficients of the six axles, and performs time-domain consistency verification and outlier filtering on the solution results. This realizes the online quantization of the spatial distribution of the adhesion coefficients of each axle, and completes the task of transforming the estimation paradigm from single-point average estimation to independent decoupling between axles. It achieves the goal of eliminating the estimation blind zone caused by the sparse drive axle and the lack of slip signal of the trailer axle, and outputting a time-consistent and robust adhesion spatial distribution matrix.

[0041] In existing technologies, active safety systems for semi-trailers generally adopt the single-vehicle monitoring architecture of passenger cars, indirectly estimating the single-point average adhesion coefficient based on wheel speed differences. This neglects the constraints of articulated body motion and the differences in axle adhesion, and employs a fixed load assumption under braking and steering conditions. This results in no effective slip signal from the trailer axles and uncompensated dynamic axle load transfer, leading to significant blind spots and systematic biases in adhesion coefficient estimation. The constraint construction unit and distributed solution unit, through the synergy of constraint construction and distributed solution, extend the adhesion coefficient estimation object from the rigid body of a single vehicle to the articulated body composed of the tractor and trailer. The constraint construction unit establishes relative motion constraints at the articulated points and uses the rate of change of the folding angle as a coupling condition, achieving explicit modeling of the dynamic constraints of the articulated body and eliminating the lack of kinematic correlation caused by the use of a single-vehicle model in existing technologies. The distributed solution unit introduces online identification parameters and compensates for dynamic axle load transfer, achieving the calculation and verification of independent adhesion coefficients for six axles, eliminating the ambiguity of inter-axle differences caused by the fixed load assumption and single-point average estimation. The two work together to enable the attachment decoupling module to overcome the structural mismatch of passenger car architecture in semi-trailer scenarios, establish a distribution estimation system that integrates the differences in inter-axle attachment, significantly improve the accuracy and reliability of attachment coefficient estimation under complex working conditions such as low attachment, uneven road surfaces, and braking and steering, and provide a physical environment cognitive basis for predicting folding trends by eliminating systematic biases.

[0042] Example 4, as Figure 2 The constraint building unit includes motion constraint units and coupled equation units; Motion constraint unit: Receives seven-dimensional state data, extracts longitudinal acceleration, lateral acceleration, and yaw rate from the inertial measurement units of the tractor and trailer, establishes relative motion constraint equations at the articulation point where displacement is continuous and velocity is equal, generates and outputs a set of constraint parameters that includes the kinematic relationship between the tractor and trailer and the continuity of acceleration at the articulation point. Coupled Equation Unit: Receives the constraint parameter set, takes the folding angle and its rate of change as the coupling constraint condition of the articulated body, establishes the dynamic correlation between the folding angle acceleration and the difference in yaw moment between the tractor and the trailer, introduces the equivalent rotational inertia parameter of the articulated body, constructs and outputs the independent adhesion coefficient observation equation set for each axis that includes the difference in inter-axis adhesion.

[0043] In this embodiment, the motion constraint unit extracts the longitudinal acceleration, lateral acceleration, and yaw rate from the inertial measurement units of the tractor and trailer, and establishes a triple kinematic constraint equation at the articulation point that ensures continuous displacement, equal velocity, and continuous acceleration. This generates a constraint parameter set that includes the kinematic relationship between the tractor and trailer and the continuity of acceleration at the articulation point. This achieves explicit modeling of the kinematic constraints of the articulated body, completes the transformation from the rigid body assumption of a single vehicle to the relative motion constraints of the articulated body, and achieves the goal of providing basic parameters containing kinematic relationships and acceleration continuity for the subsequent construction of coupling equations.

[0044] The coupled equation unit receives a set of constraint parameters, uses the folding angle and its rate of change as coupling constraints for the articulated body, establishes a dynamic correlation between the folding angle acceleration and the difference in yaw moment between the tractor and trailer, introduces the equivalent rotational inertia parameter of the articulated body, and constructs a set of independent adhesion coefficient observation equations for each axis that includes the differences in inter-axle adhesion. This achieves a deep integration of articulated body dynamic constraints and adhesion estimation, completes the task of constructing coupled equations for articulated body folding dynamics from single-vehicle moment balance, and achieves the goal of enabling the solution of six-axis independent adhesion coefficients to have physical constraints of the articulated body and providing an overdetermined observation basis.

[0045] Existing technologies use a single-vehicle model for passenger cars, considering only the six degrees of freedom of the vehicle's motion. They fail to establish kinematic constraints at the articulation point between the tractor and trailer, and do not incorporate the rate of change of the folding angle into the dynamic correlation, resulting in a lack of physical basis for the adhesion coefficient estimation. The motion constraint unit and coupling equation unit, through a progressive construction of motion constraints and coupling equations, extend the constraint conditions from the assumption of a rigid body on a single vehicle to a multi-body system of articulated bodies. The motion constraint unit, by establishing relative motion constraint equations at the articulation point that ensure continuous displacement, equal velocity, and continuous acceleration, makes the kinematic correlation between the tractor and trailer explicit, eliminating the constraint gaps caused by the fragmented treatment of the tractor and trailer's motion states in existing technologies. The coupling equation unit, by using the rate of change of the folding angle as a coupling constraint condition for the articulated body, establishes a dynamic correlation between the folding angle acceleration and the difference in yaw moment, achieving a deep integration of folding dynamics and adhesion estimation, eliminating the bias in existing technologies that do not consider the impact of folding angle evolution on adhesion calculation. The synergy of these two elements enables the constraint construction unit to overcome the limitations of the single-vehicle model for passenger cars in the semi-trailer scenario, establishes a complete equation framework that includes kinematic constraints and dynamic coupling, provides the distributed solution unit with an overdetermined set of observation equations based on the physical basis of articulated bodies, significantly improves the structural rationality and physical consistency of the six-axis independent adhesion coefficient solution, and lays the model foundation for eliminating the blind spot in the estimation of inter-axle adhesion differences.

[0046] Example 5, as Figure 2 The distributed solution unit includes a load compensation unit and an attachment solution unit; Load compensation unit: Receives the observation equation set, extracts the tire lateral stiffness and vertical load online identification parameters, calculates the dynamic transfer load of each axle under braking and steering conditions based on the vertical acceleration response of the tractor and trailer inertial measurement units, performs real-time compensation and correction of the vertical load, and outputs the compensated axle load parameters. The attachment calculation unit receives the compensated axle load parameters, substitutes them into the observation equations, and uses the weighted least squares method to calculate the six-axis independent attachment coefficients. It performs time-domain consistency verification and outlier filtering on the calculation results, generates an attachment spatial distribution matrix containing the differences in inter-axis attachment, and outputs it to the trend prediction module.

[0047] In this embodiment: the load compensation unit receives the observation equation set, extracts the tire lateral stiffness and vertical load online identification parameters, calculates the dynamic transfer load of each axle under braking and steering conditions based on the vertical acceleration response of the tractor and trailer inertial measurement units, and performs real-time compensation and correction of the vertical load. This realizes the transformation of the vertical load of each axle from fixed calibration to dynamic identification, completes the calculation task of axle load redistribution under combined braking and steering conditions, and achieves the purpose of eliminating the adhesion limit calculation deviation caused by the fixed load assumption and providing dynamic axle load input for adhesion solution.

[0048] The attachment calculation unit receives the compensated axle load parameters, substitutes them into the observation equations, and uses the weighted least squares method to calculate the six-axle independent attachment coefficients. It performs time-domain consistency verification and outlier filtering on the calculation results, realizing a breakthrough in the estimation paradigm from sparse sampling of the drive axle to independent decoupling of all axles. It completes the online quantization and quality verification of the spatial distribution of the six-axle attachment coefficients, achieving the goal of eliminating the estimation blind zone caused by the lack of slip signal of the trailer axle and outputting a time-consistent and robust attachment spatial distribution matrix.

[0049] In existing technologies, semi-trailer active safety systems adopt passenger car architecture, relying on wheel speed differences to indirectly estimate the adhesion coefficient. In a six-axle layout, only the drive axle provides effective slip signals, while the trailer axle, lacking driving torque, completely loses its estimation capability. Furthermore, the use of a fixed load assumption fails to consider the dynamic transfer of axle load under braking and steering conditions, leading to information gaps and systematic biases in adhesion coefficient estimation. The load compensation unit and adhesion calculation unit, through the synergy of load compensation and adhesion calculation, advance adhesion estimation from a single-point average to a decoupled, distributed approach between axles. The load compensation unit, by introducing tire lateral stiffness and vertical load online identification parameters and calculating the dynamically transferred load based on the vertical acceleration response, achieves a correction of the axle load from a fixed-caliber dynamic compensation, eliminating the adhesion limit deviation caused by the failure to account for axle load redistribution under braking and steering conditions. The adhesion calculation unit achieves decoupled calculation of the six-axle independent adhesion coefficients by substituting the compensated axle load into the observation equations and employing weighted least squares. This eliminates the estimation blind spots caused by the sparse drive axle and the lack of slip signals on the trailer axles, and ensures output quality through temporal consistency verification and outlier filtering. Together, these features enable the distributed calculation unit to overcome the structural limitations of passenger vehicle wheel speed difference estimation in multi-axle semi-trailer scenarios, establishing a spatial distribution estimation system for independent adhesion coefficients covering all six axles. This significantly improves the estimation accuracy and reliability under complex conditions such as low adhesion, uneven road surfaces, and braking / steering, providing a physical environment cognitive basis for predicting folding trends by eliminating ambiguity in axle differences.

[0050] Example 6, as Figure 1 The trend prediction module includes an evolution modeling unit and a risk projection unit; Evolutionary modeling unit: acquires seven-dimensional state data and attachment space distribution matrix, extracts independent attachment coefficients and folding angle change rates of each axle, establishes dynamic balance relationship between the difference in yaw moment between tractor and trailer and folding angle acceleration, embeds the inter-axle attachment difference into the articulated body dynamics model, and constructs folding evolution prediction equations that reflect the inter-axle attachment difference. Risk simulation unit: Receives the folding evolution prediction equation, performs numerical integration with the current seven-dimensional state as the initial value, solves the time-domain evolution sequence of folding angular displacement and angular velocity at future moments, compares the predicted trajectory with the safety threshold point by point, calculates the probability of exceeding the folding angle limit, classifies the risk level according to the magnitude and duration of the exceedance, and outputs the risk level and predicted trajectory.

[0051] The step size adjustment for the adaptive numerical integration algorithm is calculated using the following formula:

[0052] In the formula, For the next integration step, This is the current integration step size. The folding angle safety threshold, and These are the approximate folding angles calculated using the fifth-order and fourth-order Runge-Kutta methods, respectively. This is the preset tolerance coefficient; The step size adjustment rule is as follows: when the folding angular velocity is close to the safety threshold, the error ratio increases, causing the step size to automatically shrink to the lower limit to ensure the calculation accuracy in high-risk areas; when the folding angular velocity is far from the safety threshold, the error ratio decreases, causing the step size to expand to the upper limit to reduce the computational load. Simultaneously, based on the spatial distribution differences of the independent adhesion coefficients of each axis, dynamic weights are assigned to the torque terms of each axis in the differential equation system. The weights of torque terms corresponding to regions with abrupt changes in adhesion coefficients are increased, while the weights are decreased in smooth regions, ensuring that the integration process prioritizes responses to regions with drastic changes in adhesion conditions. Folding angle safety threshold From static limit value With dynamic correction coefficient Multiplying them together yields the result, i.e. .in The dynamic correction coefficient was obtained through mechanical simulation calibration based on the structural strength of the saddle hinge point, trailer center of gravity height, and wheelbase parameters. Typical values ​​range from 12 to 15 degrees, and specific values ​​are stored in the semi-trailer structural parameter database, retrieved by vehicle model index. ,in The current folding angle change rate is preferentially extracted directly from the seven-dimensional state vector output by the timing alignment unit of the state perception module; when this data is unavailable due to transmission interruption, an alternative path is activated, which is obtained by first-order differential calculation from the output of the saddle hinge point angle encoder, and a data quality degradation flag is triggered. This represents the maximum allowable rate of change of the folding angle for the structure, typically 30 degrees per second.

[0053] In this embodiment: the evolution modeling unit obtains the seven-dimensional state vector and the attachment space distribution matrix, extracts the independent attachment coefficients and folding angle change rates of each axle, establishes the dynamic balance relationship between the difference in yaw moment between the tractor and the trailer and the folding angle acceleration, embeds the inter-axle attachment difference into the articulated body dynamics model, constructs the folding evolution prediction equation reflecting the inter-axle attachment difference, realizes the transformation from the single-vehicle dynamics model to the articulated body coupled dynamics model, completes the task of modeling the folding angle evolution law of the non-uniform distribution of inter-axle attachment, and achieves the goal of making the folding prediction have the physical basis of the articulated body and eliminating the model mismatch caused by the single-point average attachment assumption.

[0054] The risk simulation unit receives the folding evolution prediction equation, uses the current seven-dimensional state as the initial value, and employs an adaptive numerical integration algorithm to solve the time-domain evolution sequence of folding angular displacement and angular velocity at future moments. It compares the predicted trajectory with the safety threshold point by point, calculates the probability of exceeding the folding angle limit, and classifies the risk level according to the magnitude and duration of the exceedance. This achieves the time-domain quantification and risk classification of the folding angle evolution trend, completes the task of simulating the folding dynamic behavior from the current state to future moments, and achieves the goal of providing drivers with sufficient reaction time and accurate risk level judgment.

[0055] Existing technologies use a single-vehicle dynamics model for passenger cars, substituting a single-point average adhesion coefficient into fixed equations. This neglects the articulated coupling between the tractor and trailer, as well as differences in inter-axle adhesion, leading to significant discrepancies between folding predictions and actual vehicle responses, resulting in delayed warnings or frequent misjudgments. The evolutionary modeling unit and risk extrapolation unit, through progressive collaboration between evolutionary modeling and risk extrapolation, expand the prediction object from a single rigid vehicle to an articulated system. The evolutionary modeling unit establishes a dynamic balance between the yaw moment difference between the tractor and trailer and the folding angular acceleration, embedding inter-axle adhesion differences into the articulated system dynamics model. This achieves a deep match between the prediction equations and the physical properties of the articulated system, eliminating directional misjudgments caused by the single-point average adhesion assumption. The risk extrapolation unit employs an adaptive numerical integration algorithm to achieve temporal quantization and precise grading of the predicted trajectory. This algorithm, through the variable step-size mechanism of the Runge-Kutta-Fehlberg method, automatically improves computational accuracy in high-risk areas and reduces computational load in low-risk areas, balancing prediction accuracy and real-time requirements. By employing a dynamic weighting mechanism for the torque term, the integration process prioritizes responses to regions of drastic change in attachment conditions, eliminating prediction biases caused by the failure to dynamically incorporate inter-axle attachment differences. This allows the trend prediction module to overcome the limitations of single-vehicle models in semi-trailer scenarios, establishing a folding evolution prediction system that integrates inter-axle attachment differences, significantly improving the physical consistency and timeliness of folding risk prediction under complex operating conditions.

[0056] Example 7, as Figure 1 The intent parsing module includes an intent parsing unit and a behavior determination unit; Intent parsing unit: Real-time acquisition of driver steering input signal as intent signal, synchronous reception of trailer yaw rate data as response signal, interpolation resampling and phase delay compensation of the two, calculation of phase difference and correlation coefficient, dynamic adjustment of adaptive threshold based on phase feature baseline based on historical driving data, determination of expected failure state, and generation of parsing result containing phase feature quantity and failure indicator; Behavior determination unit: Receives the analysis results, monitors the driver's steering correction behavior after the expected result is lost, extracts the steering wheel angle change rate and correction range, establishes a correction behavior evaluation index based on historical control data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction range exceeds the normal range, couples the phase analysis results with the overcorrection identification results and the fold risk level in a two-dimensional decision-making process, and generates a judgment command that integrates human behavior and vehicle risk.

[0057] The cumulative distance of the dynamic time warping and phase coupling analysis algorithm is calculated using the following formula:

[0058] In the formula, The element in the i-th row and j-th column of the cumulative distance matrix. The Euclidean distance between the i-th sampling point of the intent signal and the j-th sampling point of the response signal is taken as the cumulative distance, which is the sum of the minimum value of the three adjacent elements on the left, top, and upper left and the current local distance. The intent signal comes from the steering wheel angle sensor and the response signal comes from the trailer inertial measurement unit.

[0059] The algorithm flow is as follows: The driver's steering input signal is used as the reference sequence, and the trailer yaw rate response signal is used as the test sequence. The Euclidean distance between each point in the two sequences is calculated, and a cumulative distance matrix is ​​constructed based on the aforementioned cumulative distance formula. A dynamic programming method is used to find the optimal bending path in the cumulative distance matrix that minimizes the global cumulative distance. This path allows for nonlinear scaling of the time axis, adapting to the variable time delay characteristics between driver operation and trailer response. Path constraints include monotonicity and continuity constraints, ensuring that the time axis does not regress and that adjacent path points remain continuous. The two sequences are nonlinearly aligned according to the optimal bending path to eliminate timing deviations caused by sampling frequency differences, transmission delays, and lag in the dynamics of hinge folding. The two aligned sequences are then processed separately... A Hilbert transform is performed to extract the instantaneous phase sequence, and the cross-vehicle phase difference sequence is calculated. Based on the phase characteristic baseline of the same driver in historical driving data, a sliding window statistical method is used to calculate the deviation of the current phase difference from the baseline. The length of the sliding window is dynamically adjusted according to the current vehicle speed. The window is shortened at high speeds to improve response speed, and extended at low speeds to improve stability. An adaptive threshold is dynamically adjusted based on the current vehicle speed, road adhesion coefficient, and folding angle change rate. The threshold is tightened at high speeds, low adhesion, and large folding angle change rate, and widened at low speeds, high adhesion, and small folding angle change rate. When the phase deviation exceeds the adaptive threshold, the expected failure state is determined, and the Pearson correlation coefficient of the two sequences after normalization and alignment is calculated simultaneously as an auxiliary confirmation basis.

[0060] When the historical driving data storage unit is used for the first time or when the data volume is insufficient, a group statistical default value is used as the initial phase characteristic baseline. This default value is obtained by fitting the phase difference data of the driver under standard test conditions to a Gaussian distribution, with the mean used as the initial baseline value and the standard deviation used as the initial threshold width. As data from the same driver accumulates, the group default value is gradually replaced by individual data, with the replacement weight being the ratio of the individual data volume to the total data volume.

[0061] Adaptive threshold .

[0062] in The baseline threshold is determined by the standard deviation of historical driving data; Vehicle speed correction factor .

[0063] Where v is the current vehicle speed. The reference speed is 60 km / h; Adhesion correction factor

[0064] in The six-axis average adhesion coefficient, The reference adhesion coefficient is 0.7; Folding Angle Correction Factor

[0065] in The reference folding angle change rate is 5 degrees / second; The baseline value of the correction coefficient is 1.0, corresponding to the uncorrected state under standard operating conditions. When any correction coefficient is greater than 1.2, it is set to 1.2. This upper limit corresponds to a physical boundary where the threshold is relaxed by a maximum of 20% under extreme operating conditions to prevent over-relaxation from causing missed judgments. When any correction coefficient is less than 0.8, it is set to 0.8. This lower limit corresponds to a physical boundary where the threshold is tightened by a maximum of 20% under extreme operating conditions to prevent over-tightening from causing misjudgments. Different vehicle models are fine-tuned according to wheelbase classification, and the fine-tuning range does not exceed ±0.05.

[0066] In this embodiment, the intent parsing unit acquires the driver's steering input signal in real time and simultaneously receives the trailer's yaw rate data. It employs dynamic time warping and phase coupling analysis algorithms to interpolate, resample, and compensate for phase delays, calculating the cross-vehicle phase difference and correlation coefficient. Based on a phase characteristic baseline derived from historical driving data, it dynamically adjusts an adaptive threshold to determine the expected failure state, generating a parsing result containing phase feature quantities and a failure indicator. This unit completes the task of extracting the temporal correlation features between the driver's steering intent and the trailer's actual response, transforming driver behavior from a black-box input into quantifiable phase features. This achieves the goal of identifying the failure of the mental model based on passenger vehicle experience and eliminating misjudgments caused by temporal deviations.

[0067] The behavior determination unit receives the analysis results and monitors the driver's steering correction behavior after the expected outcome is missed. It extracts the steering wheel angle change rate and correction magnitude, establishes a correction behavior evaluation index based on historical handling data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction magnitude exceeds the normal range, and performs a two-dimensional coupling decision-making process with the phase analysis results, overcorrection identification results, and risk level assessment. This generates a judgment command that integrates human behavior and vehicle risk. This unit completes the tasks of extracting driver correction behavior features and identifying overcorrection patterns, jointly assessing human behavior risk and vehicle dynamics risk, and achieving the goal of providing the collaborative intervention module with a comprehensive judgment basis that integrates human factors and vehicle status.

[0068] Existing technologies, based on passenger vehicle monitoring architecture, rely solely on facial image recognition to identify fatigue or distraction. They fail to establish a cross-vehicle phase correlation between driver steering intent and trailer response, thus failing to identify unintended consequences and subsequent panic corrections caused by trailer response lag. The intent parsing unit and behavior determination unit, through progressive collaboration between intent parsing and behavior determination, extend driver behavior assessment from single visual recognition to phase analysis of steering input and trailer response. The intent parsing unit employs dynamic time warping and phase coupling analysis algorithms to calculate cross-vehicle phase differences and correlation coefficients, dynamically adjusting adaptive thresholds to achieve accurate determination of unintended consequences, eliminating the blind spots in behavior recognition caused by the lack of monitoring of cross-vehicle phase relationships in existing technologies. The nonlinear time axis alignment characteristic of dynamic time warping eliminates phase calculation distortion caused by the time lag of articulated bodies, while the instantaneous phase extraction characteristic of Hilbert transform enables accurate quantification of cross-vehicle phase differences. Sliding window statistics and adaptive threshold mechanisms achieve individualized dynamic determination. The behavior determination unit, by establishing modified behavior evaluation indicators and identifying over-correction patterns, achieves a two-dimensional coupled assessment of human behavior risk and vehicle dynamics risk, eliminating the limitations of existing technologies that treat vehicle state and driver behavior separately. The synergy between these two components enables the intent parsing module to overcome the bottleneck of human factor monitoring in passenger vehicle architectures, establishing a phase parsing system that integrates steering intent and trailer response. This significantly improves the accuracy and reliability of driver behavior risk identification under complex conditions, providing a comprehensive judgment basis for the collaborative intervention module to eliminate the black box effect of human behavior.

[0069] Example 8, as Figure 3 The intent parsing unit includes a signal calibration unit and a failure determination unit; Signal calibration unit: Real-time acquisition of driver steering input signal and marking it as intention signal, synchronous reception of trailer yaw rate data and marking it as response signal, interpolation resampling and phase delay compensation of the two, elimination of timing deviation caused by sampling frequency difference and transmission delay, generation of timing-aligned intention response signal pair after frequency domain filtering and output; The failure determination unit receives time-aligned intention-response signal pairs, calculates the phase difference and correlation coefficient between the intention signal and the response signal, dynamically adjusts the adaptive threshold based on the phase feature baseline of historical driving data, determines the expected failure state when the phase difference exceeds the threshold or the correlation coefficient is lower than the boundary, extracts phase feature quantities to generate failure identification resolution results, and outputs them to the behavior determination unit.

[0070] In this embodiment, the signal calibration unit acquires the driver's steering input signal in real time and marks it as the intention signal, and simultaneously receives the trailer yaw rate data and marks it as the response signal. It then performs interpolation resampling and phase delay compensation on both signals to eliminate timing deviations caused by sampling frequency differences and transmission delays. Finally, it generates a timing-aligned intention-response signal pair through frequency domain filtering. This unit completes the task of unifying the time base and calibrating the signal quality of heterogeneous data across vehicle bodies, placing the driver's steering input and trailer dynamic response under the same time coordinates. This achieves the goal of eliminating phase calculation distortion caused by asynchronous sampling and transmission delays, and providing a timing-consistent input condition for failure determination.

[0071] The expected failure determination unit receives time-aligned intention-response signal pairs, calculates the phase difference and correlation coefficient between the intention and response signals, and dynamically adjusts an adaptive threshold based on a phase feature baseline derived from historical driving data. When the phase difference exceeds the threshold or the correlation coefficient falls below the boundary, the expected failure state is determined, and phase feature quantities are extracted to generate a failure identifier resolution result. This unit completes the quantitative evaluation and individualized threshold determination of the temporal correlation characteristics between the driver's steering intention and the trailer's response. It transforms a fixed threshold based on group statistics into a dynamic boundary adapted to individual driving habits, achieving the goal of accurately identifying expected failures and reducing misjudgments and omissions caused by fixed thresholds.

[0072] Existing technologies, based on passenger vehicle architecture, rely solely on facial image recognition to identify fatigue or distraction. They fail to establish a cross-vehicle phase correlation between driver steering intent and trailer yaw response, neglecting timing deviations caused by sampling frequency differences and transmission delays between the tractor and trailer. Furthermore, they lack a phase baseline based on individual driving habits, and the use of a uniform fixed threshold results in insufficient accuracy in predicting expected failures. The signal calibration unit and the failure determination unit, through progressive collaboration in signal calibration and failure determination, extend intent analysis from fragmented state monitoring to cross-vehicle temporal correlation analysis. The signal calibration unit eliminates timing deviations through interpolation resampling and phase delay compensation, establishing a precise time alignment between the intent signal and the response signal. The failure determination unit, by introducing a phase feature baseline based on historical driving data and dynamically adjusting adaptive thresholds, achieves individualized and accurate determination of expected failure states. The synergy of these two elements enables the intent analysis unit to overcome the monitoring limitations of not considering cross-vehicle timing and individual differences, and establishes an analysis system that integrates timing alignment and individualized thresholds. This significantly improves the accuracy and reliability of expectation failure determination and provides an analytical basis for the behavior determination unit to eliminate timing deviations and misjudgments due to individual differences.

[0073] Example 9, as Figure 3 The behavior determination unit includes a correction and recognition unit and a coupled decision unit; Correction identification unit: Receives the analysis results output by the failure determination unit, monitors the driver's steering correction behavior after the expected failure, extracts the steering wheel angle change rate and correction range, establishes a correction behavior evaluation index based on historical control data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction range exceeds the normal range, and generates overcorrection identification results. Coupled Decision Unit: Receives the overcorrection identification result, synchronously acquires the folded risk level output by the trend prediction module, establishes a two-dimensional coupled decision matrix of human behavior state and vehicle risk level, performs joint judgment based on the coupling weight of risk level and behavior state, and generates a judgment instruction that integrates human behavior and vehicle risk.

[0074] In this embodiment: the correction identification unit receives the analysis results output by the failure determination unit, monitors the driver's steering correction behavior after the expected deviation, extracts the steering wheel angle change rate and correction magnitude, establishes a correction behavior evaluation index based on historical handling data, identifies overcorrection patterns where the steering angle suddenly reverses and the correction magnitude exceeds the normal range, and generates an overcorrection identification result. This unit completes the task of individualized extraction of driver correction behavior characteristics and overcorrection pattern recognition after the expected deviation, transforming unobservable driving operations into quantifiable evaluation indicators, achieving the goal of accurately identifying overcorrection behavior and reducing the error of a single threshold judgment.

[0075] The coupled decision-making unit receives the overcorrection identification results and simultaneously acquires the folded risk level output by the trend prediction module. It then establishes a two-dimensional coupled decision matrix of human behavior status and vehicle risk level. Based on the coupling weights of risk level and behavior status, it performs a joint judgment, generating a judgment instruction that integrates human behavior and vehicle risk. This unit completes the matrix-style coupled assessment task of human behavior status and vehicle risk level, extending single-dimensional judgment to two-factor joint decision-making, thus achieving the goal of enabling intervention strategies to have both behavioral and vehicle risk considerations.

[0076] In existing technologies, the active safety system for semi-trailers adopts the architecture of passenger cars, relying solely on facial image recognition to identify fatigue or distraction. It lacks a mechanism for evaluating driver corrective behavior after anticipated failures and fails to jointly assess human behavior with vehicle risk levels. This results in a disconnect between driver behavioral risk and vehicle dynamics risk, leading to a lack of targeted intervention strategies. The Correction Recognition Unit and the Coupled Decision Unit, through progressive collaboration between correction recognition and coupled decision-making, extend behavioral judgment from single-state recognition to the quantification of corrective behavior and risk-coupled decision-making. The Correction Recognition Unit extracts the steering wheel angle change rate and correction magnitude and establishes evaluation indicators based on historical handling data, achieving individualized quantification of corrective behavior after anticipated failures and accurate identification of overcorrection patterns. The Coupled Decision Unit establishes a two-dimensional coupled decision matrix and performs joint judgments based on coupling weights, achieving a deep integration of human behavior and vehicle risk levels. The synergy between the two enables the behavior determination unit to overcome the limitations of the existing technology that separates the assessment of human behavior and vehicle risk, and establishes a determination system that integrates the quantitative analysis of corrective behavior with risk-coupled decision-making. This significantly improves the pertinence and accuracy of intervention strategies and provides a comprehensive determination basis for the collaborative intervention module to eliminate the blind spots of single-dimensional decision-making.

[0077] Example 10, as follows Figure 1 The collaborative intervention module includes a decision triggering unit and an effect evaluation unit; Decision triggering unit: acquires the risk level and driver status judgment, establishes a two-dimensional coupled decision matrix of risk level and driver status, triggers differentiated intervention strategies such as sound and light reminders, steering assist gain limitation or independent braking force distribution of trailer electronic braking system based on the matrix element mapping relationship, generates intervention commands and outputs them; Effectiveness evaluation unit: Monitors the trend of folding angle change and driver operation response after intervention, evaluates the effectiveness of intervention measures on folding angle convergence and driver behavior recovery, feeds the evaluation results back to the attachment decoupling module to optimize the attachment coefficient identification parameters, and simultaneously feeds them back to the intent parsing module to adjust the phase feature baseline based on historical driving data.

[0078] In this embodiment, the decision triggering unit obtains the risk level and driver status determination, establishes a two-dimensional coupled decision matrix between the risk level and driver status, and triggers differentiated intervention strategies such as audible and visual reminders, steering assist gain limitation, or independent braking force distribution of the trailer electronic braking system based on the mapping relationship of matrix elements. This realizes the transformation of intervention strategies from fixed thresholds to dynamic coupling, and completes the task of triggering precise intervention measures based on the joint determination results of vehicle risk and driver status. This achieves the goal of matching the intervention intensity with the risk level and behavioral status and avoiding over-intervention or under-intervention.

[0079] The effect evaluation unit monitors the trend of folding angle changes and driver operation response after intervention to evaluate the effectiveness of intervention measures on folding angle convergence and driver behavior recovery. The evaluation results are fed back to the attachment decoupling module to optimize the attachment coefficient identification parameters and simultaneously fed back to the intent parsing module to adjust the phase feature baseline. This achieves closed-loop evaluation of intervention effect and adaptive optimization of system parameters, completing the architectural transformation from open-loop early warning to closed-loop self-evolution. This enables the system to continuously absorb operational data and improve subsequent prediction and recognition accuracy.

[0080] In existing technologies, most active safety systems for semi-trailers use fixed thresholds to trigger warnings. These intervention strategies are rigid and lack linkage with driver status. Furthermore, the effectiveness of the intervention is not tracked and evaluated after the warning is issued, and no feedback channel for optimizing the core model based on the evaluation results is established. This results in the system's performance failing to continuously improve with operation. The decision triggering unit and the effect evaluation unit, through the collaboration of decision triggering and effect evaluation, extend intervention execution from open-loop triggering to closed-loop feedback. The decision triggering unit establishes a two-dimensional coupled decision matrix between risk level and driver status, and triggers differentiated intervention strategies based on the mapping relationship between matrix elements. This achieves precise matching of intervention intensity with risk level and behavioral status, eliminating over-intervention or under-intervention caused by fixed thresholds. The effect evaluation unit monitors the convergence trend of the folding angle after intervention and the driver's behavioral recovery, feeding the evaluation results back to the attachment decoupling module and the intent parsing module. This achieves adaptive optimization of the system's core parameters, eliminating the performance stagnation caused by the lack of a feedback loop in existing technologies. The synergy between the two enables the collaborative intervention module to overcome the architectural limitations of fixed threshold open-loop early warning, and establish a dynamic coupling and closed-loop self-evolutionary intervention system, which significantly improves the adaptability of intervention strategies under complex working conditions and the reliability of long-term system operation.

[0081] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent system for actively improving driver safety in semi-trailers, characterized in that: include: The module includes a state awareness module, an attachment decoupling module, a trend prediction module, an intent parsing module, and a collaborative intervention module. State perception module: Collects the three-degree-of-freedom motion state of the tractor through the inertial measurement unit of the tractor, collects the three-degree-of-freedom motion state of the trailer through the inertial measurement unit of the trailer, and collects the folding angle and its rate of change through the saddle hinge point angle encoder, and outputs the seven-dimensional state data synchronously according to a unified timestamp. The attachment decoupling module receives the seven-dimensional state data output by the state perception module, establishes the relative motion constraint equations between the tractor and the trailer, uses the folding angle and its rate of change as the coupling constraint conditions of the articulated body, constructs the overdetermined observation equation set of independent attachment coefficients for each axis, solves the six-axis independent attachment coefficients, and outputs the attachment space distribution matrix. Trend prediction module: Receives seven-dimensional state data from the state perception module and the attachment space distribution matrix from the attachment decoupling module, constructs the folding evolution prediction equation of the articulated body dynamics model, substitutes the independent attachment coefficients of each axis into the yaw moment calculation of the tractor and trailer, predicts the folding angle trajectory at future moments, and calculates the folding risk probability. Intent parsing module: Real-time acquisition of driver steering input signal as intent signal, reception of trailer yaw rate data as response signal, calculation of phase difference and correlation coefficient between intent signal and response signal, determination of expected failure state, identification of panic overcorrection mode, and two-dimensional coupling decision-making between parsing results and folding risk level of trend prediction module; Collaborative Intervention Module: Receives the risk level of the folding angle from the trend prediction module and the driver status determination from the intent analysis module, constructs a two-dimensional coupled decision matrix of risk level and driver status, triggers differentiated intervention strategies, monitors the trend of folding angle changes and driver operation response after intervention, and feeds back the evaluation results to the attachment decoupling module and the intent analysis module.

2. The intelligent system for actively improving driver safety in semi-trailers according to claim 1, characterized in that, The state awareness module includes a multi-source acquisition unit and a timing alignment unit; The multi-source acquisition unit acquires the longitudinal acceleration, lateral acceleration, and yaw rate of the tractor vehicle's inertial measurement unit, and simultaneously acquires the longitudinal acceleration, lateral acceleration, and yaw rate of the trailer vehicle's inertial measurement unit. The folding angle and its rate of change are obtained through the saddle hinge point angle encoder, and the three-degree-of-freedom motion state of the tractor vehicle and trailer vehicle are calculated to generate the original state data. The timing alignment unit receives the original state data, interpolates to compensate for the sampling frequency differences between the inertial measurement units and angle encoders of the tractor and trailer, corrects for the transmission delay, and fuses the data into a seven-dimensional state vector according to a unified timestamp to ensure the timing consistency of each degree of freedom data and outputs it.

3. The intelligent system for actively improving driver safety in semi-trailers according to claim 1, characterized in that, The attachment decoupling module includes a constraint construction unit and a distributed solution unit; The constraint construction unit receives seven-dimensional state data, extracts the acceleration and yaw rate of the inertial measurement units of the tractor and trailer, establishes the relative motion constraint equations at the hinge point where the displacement is continuous and the velocity is equal, uses the folding angle and its rate of change as the coupling constraint conditions of the hinge body, constructs and outputs the observation equation set of independent adhesion coefficients of each axis. The distributed solution unit receives the observation equation set, introduces online identification parameters of tire lateral stiffness and vertical load, compensates for the dynamic transfer of axle load under braking and steering conditions, uses the weighted least squares method to solve the six-axle independent adhesion coefficient, performs time-domain consistency verification and outlier filtering on the solution results, and outputs the adhesion space distribution matrix.

4. The intelligent system for actively improving driver safety in semi-trailers according to claim 3, characterized in that, The constraint building unit includes motion constraint units and coupled equation units; Motion constraint unit: Receives the seven-dimensional state vector, extracts the longitudinal acceleration, lateral acceleration, and yaw rate of the inertial measurement units of the tractor and trailer, establishes the relative motion constraint equations at the hinge point where the displacement is continuous and the velocity is equal, and generates a constraint parameter set that includes the kinematic relationship between the tractor and trailer and the continuity of the acceleration at the hinge point. Coupled equation unit: Receives the constraint parameter set, takes the folding angle and its rate of change as the coupling constraint condition of the articulated body, establishes the dynamic correlation between the folding angle acceleration and the difference in yaw moment between the tractor and the trailer, introduces the equivalent rotational inertia parameter of the articulated body, and constructs a set of observation equations for the independent adhesion coefficients of each axis that includes the difference in adhesion between axes.

5. The intelligent system for actively improving driver safety in semi-trailers according to claim 3, characterized in that, The distributed solution unit includes a load compensation unit and an attachment solution unit; Load compensation unit: Receives the observation equation set, extracts tire lateral stiffness and vertical load online identification parameters, calculates the dynamic transfer load of each axle under braking and steering conditions based on the vertical acceleration response of the tractor and trailer inertial measurement units, performs real-time compensation and correction of the vertical load, and outputs the compensated axle load parameters. The attachment calculation unit receives the compensated axle load parameters, substitutes them into the observation equations, and uses the weighted least squares method to calculate the six-axis independent attachment coefficients. It performs time-domain consistency verification and outlier filtering on the calculation results, generates an attachment spatial distribution matrix containing the differences in inter-axis attachment, and outputs it to the trend prediction module.

6. The intelligent system for actively improving driver safety in semi-trailers according to claim 1, characterized in that, The trend prediction module includes an evolution modeling unit and a risk projection unit; Evolutionary modeling unit: acquires seven-dimensional state data and attachment space distribution matrix, extracts independent attachment coefficients and folding angle change rates of each axis, establishes dynamic balance relationship between the difference in yaw moment between tractor and trailer and folding angle acceleration, embeds the inter-axle attachment difference into the articulated body dynamics model, constructs and outputs folding evolution prediction equations that reflect the inter-axle attachment difference. Risk simulation unit: Receives the folding evolution prediction equation, takes the current seven-dimensional state as the initial value, uses an adaptive numerical integration algorithm to solve the time-domain evolution sequence of folding angular displacement and angular velocity at future moments, compares the predicted trajectory with the safety threshold point by point, calculates the probability of folding angle exceeding the limit, and classifies the risk level according to the magnitude and duration of the exceedance, and outputs the risk level and the predicted trajectory.

7. The intelligent system for actively improving driver safety in semi-trailers according to claim 1, characterized in that, The intent parsing module includes an intent parsing unit and a behavior determination unit; Intent parsing unit: Real-time acquisition of driver steering input signal as intent signal, synchronous reception of trailer yaw rate data as response signal, interpolation resampling and phase delay compensation of the two, dynamic time warping and phase coupling analysis algorithm to calculate the phase difference and correlation coefficient of cross vehicle body intent signal and response signal, dynamically adjust adaptive threshold based on phase feature baseline based on historical driving data, determine expected failure state, generate parsing result containing phase feature quantity and failure mark and output it; Behavior determination unit: Receives the analysis results, monitors the driver's steering correction behavior after the expected result is lost, extracts the steering wheel angle change rate and correction range, establishes a correction behavior evaluation index based on historical control data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction range exceeds the normal range, couples the phase analysis results with the overcorrection identification results and the fold risk level in a two-dimensional decision-making process, and generates a judgment command that integrates human behavior and vehicle risk.

8. The intelligent system for actively improving driver safety in semi-trailers according to claim 7, characterized in that, The intent parsing unit includes a signal calibration unit and a failure determination unit; Signal calibration unit: Real-time acquisition of driver steering input signal and marking it as intention signal, synchronous reception of trailer yaw rate data and marking it as response signal, interpolation resampling and phase delay compensation of the two, elimination of timing deviation caused by sampling frequency difference and transmission delay, and generation of timing-aligned intention response signal pair after frequency domain filtering; The failure determination unit receives the time-aligned intention-response signal pair, calculates the phase difference and correlation coefficient between the intention signal and the response signal, dynamically adjusts the adaptive threshold based on the phase feature baseline of historical driving data, determines the expected failure state when the phase difference exceeds the threshold or the correlation coefficient is lower than the boundary, and extracts the phase feature quantity to generate the failure identification resolution result.

9. The intelligent system for actively improving driver safety in semi-trailers according to claim 7, characterized in that, The behavior determination unit includes a correction and recognition unit and a coupled decision-making unit; Correction identification unit: Receives the analysis results output by the failure determination unit, monitors the driver's steering correction behavior after the expected failure, extracts the steering wheel angle change rate and correction range, establishes a correction behavior evaluation index based on historical control data, identifies overcorrection patterns where the steering angle reverses abruptly and the correction range exceeds the normal range, and generates overcorrection identification results. Coupled Decision Unit: Receives the overcorrection identification result, synchronously acquires the folded risk level output by the trend prediction module, establishes a two-dimensional coupled decision matrix of human behavior state and vehicle risk level, performs joint judgment based on the coupling weight of risk level and behavior state, and generates a judgment instruction that integrates human behavior and vehicle risk.

10. The intelligent system for actively improving driver safety in semi-trailers according to claim 1, characterized in that, The collaborative intervention module includes a decision triggering unit and an effect evaluation unit; Decision triggering unit: acquires the risk level and driver status judgment, establishes a two-dimensional coupled decision matrix of risk level and driver status, triggers differentiated intervention strategies such as sound and light reminders, steering assist gain limitation or independent braking force distribution of trailer electronic braking system based on the matrix element mapping relationship, generates intervention commands and outputs them; Effectiveness evaluation unit: Monitors the trend of folding angle change and driver operation response after intervention, evaluates the effectiveness of intervention measures on folding angle convergence and driver behavior recovery, feeds the evaluation results back to the attachment decoupling module to optimize the attachment coefficient identification parameters, and simultaneously feeds them back to the intent parsing module to adjust the phase feature baseline based on historical driving data.

Images