Anti-carsickness intelligent damping control system

Abstract

The invention relates to the technical field of vehicle damping control, and discloses an anti-carsickness intelligent damping control system. A multi-dimensional sensing array of the system collects vehicle driving states, passenger physiological indexes and environment disturbance data in real time, and a time-space aligned multi-dimensional driving data set is generated through timestamp alignment and space coordinate synchronization; performing feature decoupling reconstruction on the data set by a data separation engine, and separating out a main feature matrix and an abnormal disturbance identification set; the feature screening unit screens a core feature sequence in combination with a vehicle kinetic parameter library; a damping strategy generator generates a multi-stage cooperative damping control strategy according to the multi-stage cooperative damping control strategy; the abnormal feedback regulator analyzes the abnormal disturbance identification set and triggers strategy dynamic tuning; and the control instruction dispatcher converts the tuned strategy into a driving signal and issues the driving signal to the suspension execution mechanism cluster. The system can comprehensively sense multi-dimensional information, dynamically adjust a damping strategy, effectively relieve the carsickness phenomenon and improve the riding comfort of passengers.

Description

Technical Field

[0001] This invention relates to the field of vehicle shock absorption control technology, specifically to an intelligent shock absorption control system for preventing motion sickness. Background Technology

[0002] Motion sickness is a common problem affecting passenger experience during vehicle operation, and its occurrence is closely related to changes in vehicle driving conditions, environmental disturbances, and the physiological state of passengers. Currently, most vehicle damping systems on the market use passive damping structures. These structures can only buffer road bumps within a fixed frequency range through preset mechanical damping characteristics, and cannot adjust the damping strategy according to real-time changes in vehicle driving conditions and passenger physiological needs. When a vehicle encounters irregular bumps on unpaved roads, or experiences frequent starts, stops, and turns in urban areas, the passive damping system struggles to respond quickly to these dynamic driving scenarios, resulting in significant vibrations and changes in the vehicle's posture. This unstable driving state causes continuous and irregular stimulation to the occupants' inner ear vestibular system, leading to motion sickness symptoms such as dizziness and nausea. These discomforts are particularly pronounced in the elderly and children, whose vestibular functions are more sensitive. While some vehicles are equipped with active damping systems, most rely solely on a few vehicle parameters, such as vehicle speed and suspension travel, for control, lacking real-time monitoring and feedback of occupant physiological indicators. These systems cannot accurately identify whether occupants are experiencing pre-motion sickness symptoms or discomfort; they can only adjust damping parameters according to fixed control logic, making it difficult to achieve personalized damping control based on individual occupant differences. Furthermore, existing active damping systems exhibit a lag in adjusting their control strategies when faced with sudden road disturbances, failing to promptly suppress sudden, severe vibrations of the vehicle body, still causing significant physiological stimulation to occupants and failing to fundamentally solve the problem of motion sickness. As consumers increasingly demand higher levels of vehicle comfort, the shortcomings of traditional shock absorption systems in terms of adaptability, personalization, and response speed are becoming more and more apparent. There is an urgent need for an intelligent control system that can comprehensively sense the vehicle's driving status, occupant physiological indicators, and environmental disturbances, and can quickly and dynamically adjust the shock absorption strategy to effectively alleviate or even eliminate motion sickness and improve the occupant's riding experience. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent anti-motion sickness shock absorption control system to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides an intelligent motion sickness damping control system, the system comprising: A multi-dimensional sensor array is used to collect vehicle driving status data, occupant physiological index data and environmental disturbance data in real time. The vehicle driving status data, occupant physiological index data and environmental disturbance data are time-stamp aligned and spatial coordinate synchronized to generate a spatiotemporally aligned multi-dimensional driving dataset. A data separation engine is used to perform feature decoupling and reconstruction processing on the spatiotemporally aligned multidimensional driving dataset, separating the main feature matrix for shock absorption control and the abnormal disturbance identifier set for feedback tuning; The feature filtering unit is used to perform multi-dimensional feature correlation analysis based on the main feature matrix and combined with a pre-set vehicle dynamics parameter library to filter out the core feature sequence for shock reduction decision-making. A vibration reduction strategy generator is used to generate a multi-level collaborative vibration reduction control strategy based on the core feature sequence. An abnormal feedback regulator is used to parse the abnormal disturbance identifier set, identify sudden disturbance events in vehicle driving status data or abnormal fluctuation patterns in occupant physiological index data, and trigger dynamic tuning of the shock absorption control strategy. The control command distributor is used to convert the tuned damping control strategy into drive signals and send them to the suspension actuator cluster.

[0005] Preferably, the data separation engine includes: The matrix transformation module is used to convert the spatiotemporally aligned multidimensional driving dataset into a normalized feature tensor; The tensor decomposition module is used to perform rank-constrained decomposition on the standardized feature tensor to obtain the steady-state feature basis and transient disturbance components. The steady-state feature basis constitutes the principal feature matrix, and the transient disturbance components constitute the abnormal disturbance identifier set.

[0006] Preferably, the feature filtering unit includes: The association map construction module is used to extract vehicle longitudinal acceleration features, lateral sway angle features, vertical vibration spectrum features, occupant heart rate variation features and eye movement trajectory features from the main feature matrix to construct a multimodal feature association map; The entropy calculation module is used to calculate the dynamic association entropy of each feature based on the multimodal feature association map; The sequence generation module is used to filter the top K features in descending order based on the dynamic association entropy and generate the core feature sequence.

[0007] Preferably, the vibration damping strategy generator includes: The strategy mapping module is used to match the core feature sequence with a pre-set vehicle dynamics parameter library to map out the basic damping control curve. The multi-level optimization module is used to perform hierarchical optimization decomposition of the basic damping control curve based on the vehicle suspension topology, and generate the multi-level coordinated damping control strategy that includes the main suspension level control strategy and the auxiliary suspension level control strategy.

[0008] Preferably, the anomaly feedback regulator includes: The disturbance pattern recognition module is used to perform time-frequency feature analysis on the transient disturbance components and identify sudden road impact mode, continuous resonance mode or occupant physiological instability mode. The tuning parameter generation module is used to retrieve a preset tuning rule library based on the identified disturbance mode and generate suspension stiffness correction coefficients or damping compensation gradients. The strategy recalculation module is used to inject the suspension stiffness correction coefficient or damping compensation gradient into the multi-level collaborative damping control strategy for real-time recalculation.

[0009] Preferably, the control command distributor includes: The signal conversion module is used to convert the recalculated multi-level collaborative vibration reduction control strategy into a pulse width modulation signal; The partition distribution module is used to distribute the pulse width modulation signal to the corresponding main suspension controller or auxiliary suspension controller according to the position code of the suspension actuator.

[0010] Preferably, the system further includes: A cloud-based coprocessor is used to receive cross-vehicle cooperation identifiers in the transient disturbance components and to schedule the damping feature matrices of adjacent vehicles from the cloud. A coupling strategy optimizer is used to generate coupling constraint terms based on the damping feature matrices of the adjacent vehicles, and to perform collaborative optimization on the multi-level collaborative damping control strategy.

[0011] Preferably, the system further includes: The feature tracing module is used to trace back the associated feature vectors in the historical driving dataset when the transient disturbance component exceeds a preset threshold. The disturbance clustering module is used to cluster and group the associated feature vectors according to the vehicle operating condition labels to generate equipment-level disturbance clusters and environment-level disturbance clusters. The device-level disturbance clusters are input to the suspended health diagnostic unit, and the environmental-level disturbance clusters are input to the environmental adaptation learning unit.

[0012] Preferably, the system further includes: Edge computing partitions contain isolated real-time processing areas and non-volatile storage areas; The real-time processing area runs the data separation engine and feature filtering unit, while the non-volatile storage area stores the preset vehicle dynamics parameter library and tuning rule library.

[0013] Preferably, the system further includes: A secure isolation repeater is used to establish an encrypted channel between the edge computing partition and the cloud coprocessor, and to perform runtime verification on the issued coupling constraints.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This intelligent anti-motion sickness shock absorption control system uses a multi-dimensional sensor array to collect real-time data on vehicle driving status, occupant physiological indicators, and environmental disturbances. It also aligns timestamps and synchronizes spatial coordinates to generate a spatiotemporally aligned multi-dimensional driving dataset. Compared to traditional shock absorption systems that rely on sensing only a single or a few vehicle parameters, this system can more comprehensively and accurately capture various key information that affects the occupant's riding experience and motion sickness, providing a richer and more reliable data foundation for the formulation of subsequent shock absorption control strategies. The data separation engine performs feature decoupling and reconstruction on the spatiotemporally aligned multidimensional driving dataset, separating the main feature matrix for shock absorption control and the abnormal disturbance identifier set for feedback tuning. This processing method can effectively filter out the core data features directly related to shock absorption control, while accurately identifying abnormal disturbance information in the dataset, avoiding interference from irrelevant data on shock absorption control decisions, improving the targeting and efficiency of data processing, and enabling subsequent feature selection and strategy generation stages to focus more on key control elements, reduce redundant calculations, and ensure the high efficiency of system operation. The feature selection unit, based on the main feature matrix and combined with a pre-set vehicle dynamics parameter library, performs multi-dimensional feature correlation analysis to select the core feature sequence for shock absorption decision-making. This process fully integrates professional knowledge of vehicle dynamics, and can accurately extract key features that play a decisive role in shock absorption effect from complex multi-dimensional data features. This ensures that the generated shock absorption decision basis conforms to the mechanical laws of vehicle driving, making the subsequently generated shock absorption control strategy more scientific and reasonable, and avoiding the problem of poor shock absorption effect caused by improper feature selection. The damping strategy generator generates multi-level collaborative damping control strategies based on core feature sequences. Compared with the fixed and single control mode of traditional damping systems, the multi-level collaborative control strategy can formulate differentiated damping schemes for different driving scenarios and different physiological states of passengers, realize fine control of vehicle vibration and attitude, effectively suppress vehicle movement that may cause motion sickness in different driving scenarios, and better adapt to diverse driving needs and passenger states. The abnormal feedback regulator identifies sudden disturbance events in vehicle driving status data or abnormal fluctuation patterns in occupant physiological index data by parsing the abnormal disturbance identifier set, and triggers dynamic tuning of the damping control strategy. This real-time feedback tuning mechanism can quickly respond to sudden road disturbances or changes in occupant physiological state, and adjust the damping control strategy in a timely manner to avoid severe vehicle vibration or increased occupant discomfort caused by control strategy lag. It ensures that the system can still maintain good damping effect under various emergency situations and effectively alleviate occupant motion sickness symptoms. The control command distributor converts the tuned damping control strategy into drive signals and sends them to the suspension actuator cluster, ensuring accurate transmission and rapid execution of control commands. This allows the suspension actuators to respond promptly to the adjusted strategy, quickly adjusting the damping characteristics and support force of the suspension to rapidly suppress vehicle vibration, stabilize vehicle posture, and reduce stimulation to the occupant's vestibular system. Through multi-faceted synergy, this effectively alleviates motion sickness, significantly improves passenger comfort, and adapts to the physiological differences of different occupants, meeting consumers' high-quality demands for a superior vehicle riding experience. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the anti-motion sickness intelligent shock absorption control system described in this invention; Figure 2 A flowchart illustrating the working principle of a data separation engine; Figure 3 A flowchart illustrating the working principle of the vibration reduction strategy generator; Figure 4 This is a flowchart illustrating the working principle of a system containing feature tracing and perturbation clustering modules. Detailed Implementation

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Please see Figure 1 The present invention provides an intelligent shock absorption control system for preventing motion sickness. The system includes: a multi-dimensional sensor array, a data separation engine, a feature filtering unit, a shock absorption strategy generator, an abnormal feedback regulator, and a control command distributor.

[0018] A multi-dimensional sensor array is deployed in the vehicle chassis, seats, and occupant wearable devices to collect real-time vehicle driving status data such as longitudinal acceleration, lateral angular velocity, and vertical vibration frequency; physiological indicators such as occupant heart rate, skin conductance, and nystagmus frequency; and environmental disturbance data such as road slope and air turbulence intensity. All data are timestamped and synchronized using a high-precision clock source and spatial coordinates based on the vehicle coordinate system, generating a spatiotemporally aligned multi-dimensional driving dataset. The data separation engine receives this dataset and separates the main feature matrix and the abnormal disturbance identifier set through matrix transformation and tensor decomposition. The feature filtering unit performs correlation analysis based on the main feature matrix and a pre-set vehicle dynamics parameter library to filter out the core feature sequence. The damping strategy generator maps the core feature sequence to a multi-level collaborative damping control strategy, including control parameters for the main suspension and auxiliary suspension levels. The abnormal feedback regulator parses the abnormal disturbance identifier set, identifies sudden disturbance events or physiological abnormal fluctuations, and triggers dynamic tuning of the damping control strategy. The control command distributor converts the tuned strategy into drive signals and distributes them to the cluster of suspended actuators.

[0019] Example 1: See Figure 2 The multidimensional driving dataset is input into the matrix transformation module of the data separation engine. This module receives heterogeneous data streams from the multidimensional sensor array, including time-series signals of vehicle longitudinal acceleration, angular velocity samples of lateral sway angle, triaxial spectral data of vertical vibration, RR interval sequences of occupant heart rate, admittance change curves of skin conductance, angular displacement trajectories of nystagmus, inclination measurements of road slope, and pressure fluctuation readings of air turbulence. The matrix transformation module first performs Z-score standardization on all numerical features to eliminate dimensional differences: it calculates the mean and standard deviation within a sliding time window for each feature data, subtracts the mean from the original data, and divides by the standard deviation. For categorical features (such as road type encoding), one-hot encoding is used to convert them into binary vectors. The standardized data are arranged in timestamp order to form a three-dimensional standardized feature tensor, whose dimensions are time step, number of feature channels, and spatial sampling point number, respectively.

[0020] The tensor decomposition module applies rank constraint processing based on Tucker decomposition to the standardized feature tensors. The rank constraint value of the core tensor is set to R, which is dynamically adjusted based on the correlation analysis of the feature dimensions of historical driving data. The decomposition process first initializes the factor matrix and then iteratively optimizes it using alternating least squares to minimize the Frobenius norm error between the original and reconstructed tensors. After decomposition, three factor matrices and one core tensor are obtained. The steady-state feature basis is reconstructed from the product of the core tensor and the factor matrices. Its data characteristics are characterized by low-frequency trend components: the smooth curve of vehicle longitudinal acceleration reflects a constant-speed driving state; the baseline fluctuation of the occupant heart rate variability coefficient represents the resting physiological state; and the road slope in the environmental disturbance data shows a slow changing trend. The output of this basis is a two-dimensional principal feature matrix, where the row vectors correspond to the time step, and the column vectors contain the fused steady-state feature channels.

[0021] Transient disturbance components are obtained by calculating the residuals between the original and reconstructed tensors. The residual tensors are then thresholded to remove noise before extracting significant transient components: millisecond-level impact pulses in the longitudinal acceleration residual correspond to wheel impacts; high-frequency resonance clusters in the vertical vibration spectrum residual reflect suspension resonance; spikes in the heart rate RR interval residual mark sudden sympathetic activation; and abnormal angular displacements in the eye movement trajectory residual characterize vestibular system dysfunction. These components are encoded into an anomalous disturbance identifier set, each identifier containing a four-tuple structure: disturbance type encoding, timestamp start position, spatial coordinate index, and intensity amplitude. The main feature matrix is ​​transmitted to the feature filtering unit, and the anomalous disturbance identifier set is transmitted to the anomalous feedback regulator.

[0022] During the matrix transformation phase, timestamp alignment uses GPS-synchronized clock signals, and spatial coordinate synchronization establishes a right-handed coordinate system with the vehicle's center of mass as the origin. In the standardization process, the sliding time window length is set to 200 milliseconds, and the window step size is 50 milliseconds to ensure a balance between real-time performance and continuity. The uniquely coded road surface types include six predefined categories such as asphalt, gravel, and snow / ice, identified through the fusion of data from the vehicle's camera and radar.

[0023] The rank constraint value R of tensor decomposition was determined through offline training: historical data of typical driving conditions (urban roads, highways, mountain curves) were collected, the mutual information matrix between each feature channel was calculated, and the rank corresponding to the sum of singular values ​​of the mutual information matrix accounting for 95% of the energy was taken as the baseline value of R. During online operation, the R value adaptively expanded according to the number of feature dimensions of the current driving environment, and the expansion coefficient was adjusted by the visibility and humidity parameters in the environmental disturbance data. The factor matrix was initialized using the stochastic SVD method to accelerate convergence, and the iteration termination condition was set to the reconstruction error change rate being less than 0.1% or reaching the maximum number of iterations of 50.

[0024] The residual filtering employs a dual-threshold mechanism: first, the root mean square (RMS) value of each channel of the residual tensor is calculated, and residuals below 30% of the RMS value are considered noise and set to zero; second, morphological filtering is applied to the remaining residuals to remove isolated pulses with a duration of less than 5 milliseconds. The intensity amplitude of the abnormal disturbance marker is calculated using the normalized energy integration method: over the disturbance duration, the residual signal is squared and integrated, then divided by the time window length. The marker for sudden road impact patterns is supplemented with spatial coordinate indices, and the suspension zone of the impact source is determined using wheel position sensors.

[0025] A smoothing constraint is introduced during the reconstruction of the steady-state feature basis: second-order difference regularization is applied to the time dimension factor matrix of the core tensor to suppress high-frequency jitter in the basis. Channel fusion is performed before the output of the principal feature matrix, reducing the dimensionality of vehicle motion-related longitudinal / lateral / vertical features to three channels through principal component analysis, while physiological features retain independent channels. The abnormal disturbance identifier set is stored using incremental compression; consecutive disturbance events at the same spatial location are merged into a single entry, and the event duration and maximum amplitude are recorded.

[0026] This implementation leverages the mathematical properties of tensor decomposition to achieve physical separation of steady-state control characteristics from transient disturbance characteristics while maintaining the spatiotemporal correlation of driving data. The construction of standardized feature tensors ensures the comparability of heterogeneous data within a unified mathematical space; rank-constrained decomposition extracts the essential structure of the data through the low-rank assumption; and residual analysis focuses on the nonlinear response components of the system. The principal feature matrix provides stable input for vibration reduction control, while the abnormal disturbance identifier set establishes a precise triggering mechanism for dynamic tuning.

[0027] Example 2: See Figure 3 The main feature matrix is ​​input to the association graph construction module of the feature filtering unit. This module extracts five types of feature vectors from the matrix: the vehicle longitudinal acceleration feature is obtained by low-pass filtering of the time-domain signal collected by the chassis accelerometer; the lateral roll angle feature is calculated by integrating the roll angular velocity measured by the gyroscope; the vertical vibration spectrum feature is the fundamental frequency amplitude extracted by fast Fourier transform of the triaxial data collected by the seat accelerometer; the occupant heart rate variability feature is the standard deviation calculated from the RR interval sequence recorded by the electrocardiogram monitoring device; and the eye movement trajectory feature is the angular velocity curve obtained by transforming the pupil position coordinates collected by the infrared eye tracker. After these feature vectors are aligned by timestamps, they are input into the graph convolutional network to construct a multimodal feature association graph. Each node in the graph corresponds to a feature vector, and the node attributes include feature type and time window statistics; the edge weights between nodes are determined by calculating the Pearson correlation coefficient of feature pairs within a sliding time window, with the window length set to 300 milliseconds and the step size to 50 milliseconds. Longitudinal acceleration and lateral sway angle form a strong negative correlation (weight -0.82), vertical vibration spectrum and heart rate variability establish a positive correlation (weight 0.76), and eye movement trajectory and lateral sway angle show a nonlinear correlation (weight 0.68).

[0028] The entropy calculation module performs dynamic correlation entropy analysis based on the multimodal feature correlation graph. For each feature node, its information entropy is calculated over 20 consecutive time windows: first, the numerical distribution histogram of the feature within the time window is statistically analyzed, and the Shannon entropy value is calculated based on the probability distribution. Simultaneously, the joint conditional entropy of this feature relative to other features is calculated: the top three features with the weights of the connected edges are selected as conditional variables, and the conditional probability distribution is calculated using kernel density estimation. The dynamic correlation entropy is ultimately represented by the weighted sum of information entropy and conditional entropy, with the weighting coefficients determined based on the feature sensitivity table in the vehicle dynamics parameter library. Under emergency braking conditions, the information entropy of the longitudinal acceleration feature surges to 2.8 bits, and its joint conditional entropy relative to the lateral sway angle reaches 1.5 bits, resulting in a combined dynamic correlation entropy value of 4.3 bits; during constant speed cruising, the dynamic correlation entropy of the vertical vibration spectrum remains at 0.9 bits.

[0029] The sequence generation module receives the dynamic correlation entropy sequence of all features, sorts it in descending order, and processes it according to preset filtering rules. The feature dimension table stored in the vehicle dynamics parameter library specifies that the first 3 features are selected for urban road conditions and the first 5 features are selected for highway conditions. A redundancy suppression mechanism is introduced in the filtering process—if the entropy value difference between adjacent features is less than 0.3 bits, they are merged into a feature group. The final generated core feature sequence adopts a three-dimensional data structure: timestamp, feature type encoding, and feature quantization value. For example, when driving on a bumpy road, the sequence includes three features: vertical vibration spectrum (amplitude 0.8g), heart rate variability (standard deviation 35ms), and eye movement trajectory (angular velocity 42deg / s).

[0030] The core feature sequence is input into the strategy mapping module of the damping strategy generator. This module retrieves a damping reference curve library from the vehicle dynamics parameter library. The library contains control curves for 120 standard operating conditions, each defined by a suspension displacement-damping force relationship matrix. The matching process employs a dynamic time warping algorithm: nonlinearly aligning the core feature sequence with the feature templates corresponding to the reference curves, and calculating the minimum path cumulative distance. When the vehicle passes through an S-shaped curve, the lateral sway angle (peak value 15 degrees) in the core feature sequence achieves a 95% match with the "high-speed cornering" template (feature template ID#07) in the reference curve library, mapping out the basic damping control curve. This curve includes camber compensation parameters (front wheels -1.2 degrees, rear wheels +0.8 degrees) and roll stiffness parameters (increased by 25%).

[0031] The multi-level optimization module receives the basic damping control curves and performs hierarchical decomposition based on the vehicle's suspension topology. The main suspension-level control strategy is designed independently for each of the four wheels: the stiffness parameters from the basic curves are input into the LQR optimizer, and the optimal damping force sequence is solved by combining sprung mass, unsprung mass, and suspension geometry parameters. For example, the front left wheel suspension generates a damping force curve under braking conditions: an initial segment of 500N (compression stroke), a peak segment of 1200N (rebound stroke), and a decay segment of 800N (steady-state stroke). The auxiliary suspension-level control strategy is designed for the active stabilizer bar: the compensation parameters from the basic curves are input into the PID optimizer, and the front stabilizer bar torque (linearly increasing from 0-120Nm) and the rear stabilizer bar dynamics (step change from 50-200N) are generated by combining vehicle attitude sensor data. The final multi-level collaborative damping control strategy adopts a hierarchical command structure: the top layer is the main suspension-level damping force matrix, and the bottom layer is the auxiliary suspension-level dynamics vector, with the two synchronously coupled via timestamps.

[0032] In the feature association graph update mechanism, edge weights are recalculated every 100 milliseconds, and node attributes are refreshed every 200 milliseconds. Dynamic association entropy calculation uses a sliding window recursive update to avoid redundant calculations. When generating the core feature sequence, the selected features are normalized: each feature value is mapped to the [0,1] interval, and the normalization baseline value comes from the feature extremum table in the vehicle dynamics parameter library. The strategy mapping module sets a matching confidence threshold (80%). When the threshold is lower than 80%, a multi-template weighted fusion mechanism is activated, for example, fusing the "high-speed cornering" template (weight 0.6) with the "lane change and overtaking" template (weight 0.4) to generate a new curve. The state weight matrix in LQR optimization is dynamically adjusted according to the driving mode: in sport mode, the roll rate weight increases by 3 times, and in comfort mode, the vertical acceleration weight decreases by 50%.

[0033] This implementation reveals the coupling relationship between vehicle motion and physiological response through multimodal feature correlation mapping, quantifies the contribution of features to system uncertainty through dynamic correlation entropy, and focuses on the most critical control variables through the screening of core feature sequences. Strategy mapping transforms complex feature combinations into physically executable control curves, while hierarchical optimization decomposition adapts the control strategy to the dynamic characteristics of different suspension actuators. LQR control at the main suspension level ensures global system optimality, while PID optimization at the auxiliary suspension level achieves rapid local response; the two work together to improve the accuracy and robustness of damping control.

[0034] Example 3: See Figure 4The disturbance pattern recognition module of the abnormal feedback regulator receives transient disturbance components from the data separation engine. This component contains a residual signal in time-series form, sampled at 1 kHz, with its duration dynamically varying according to the disturbance event. The module first performs time-frequency feature analysis on the input signal: short-time Fourier transform is used to analyze the frequency domain characteristics, with a Hanning window selected as the window function, a window length of 256 points, and an overlap rate of 50%, generating a spectrum; simultaneously, continuous wavelet transform is performed, with Morlet wavelets selected as the mother wavelet, and the scale parameter 'a' ranging from 1 to 100, generating a time-frequency distribution matrix. The identification of sudden road impact patterns is based on the high-frequency energy surge characteristics: within the 800 Hz-1 kHz frequency band, if the energy amplitude increases beyond a threshold within 5 milliseconds... ( , If the wavelet coefficients exhibit local maxima in the time domain (where the standard deviation is the background noise), it is labeled as an impact event. Detection of continuous resonance modes focuses on sustained energy peaks in specific frequency bands: within the 100-200Hz band, if the energy consistently exceeds 2.5 times the baseline level and lasts for more than 50 milliseconds, and the wavelet coefficients exhibit periodic fluctuations, it is determined to be a resonance event. Identification of occupant physiological instability patterns is achieved through heart rate coefficient of variation analysis: the residual heart rate signal is extracted from the transient component, and its coefficient of variation is calculated. If it exceeds the threshold of 0.35 for three consecutive sampling cycles, and is accompanied by high-frequency jitter in the eye movement signal, it is classified as a physiological instability event.

[0035] All identified events are encoded into a structured data format, including event type, start timestamp, duration, peak intensity, main frequency band, and spatial location information. The tuning parameter generation module retrieves a pre-set tuning rule base based on the identification results. This rule base is stored in a hierarchical structure: the first layer is a disturbance type index, the second layer is vehicle operating parameters (vehicle speed, load, road type), and the third layer is a tuning parameter mapping table. For sudden road impact modes, the suspension stiffness correction coefficient table is queried to obtain a correction function based on impact intensity and duration.

[0036] , in: The adjusted suspension stiffness coefficient. As the reference stiffness value, Peak impact strength (unit: g). The duration of the impact (unit: ms). The intensity sensitivity factor (value range 0.1-0.3) This is the duration weighting coefficient (range 0.05-0.2). For continuous resonant modes, consult the damping compensation gradient table to obtain the compensation amount based on the resonant frequency and energy level: , in: Damping compensation (unit: Ns / m). This is the reference damping value. The amplitude of the resonant energy. As the baseline energy value, The resonant frequency (unit: Hz) The system's inherent frequency, and This represents the frequency-dependent adjustment coefficient. For occupant physiological instability modes, a parameter attenuation strategy is adopted: by querying the physiological instability rule table to obtain the control parameter attenuation curve, the suspension actuation intensity is reduced exponentially.

[0037] The strategy recalculation module receives tuning parameters and updates the strategy in real time. This module maintains the currently effective multi-level collaborative damping control strategy; its data structure includes a primary suspension level parameter matrix and an auxiliary suspension level parameter vector. (Regarding stiffness correction coefficients...) The stiffness parameters in the main suspension stage control strategy are updated using linear interpolation: the stiffness values ​​in the current strategy are read. Calculate the new value ,in This is the interpolation rate factor (default value 0.8). For damping compensation... The damping parameters are updated using gradient descent: the learning rate η = 0.05, and the optimized damping value is obtained through three iterations. For the decay strategy triggered by physiological instability, first-order inertial decay is implemented with a time constant τ = 200 ms. ,in Let be the intensity of the action at time t. These are the initial values. All parameter updates are completed within strict time constraints, with an end-to-end latency of no more than 20 milliseconds from disturbance identification to policy recalculation.

[0038] The signal conversion module of the control command distributor converts the recalculated strategy into drive signals. This module uses pulse width modulation (PWM) technology to discretize continuous control parameters into square wave signals with adjustable duty cycles. The conversion process is based on a pre-calibrated mapping relationship: the suspension stiffness parameter K is mapped to the pulse width. ,in and These are the minimum and maximum pulse widths (typical values ​​of 1ms and 5ms). and This represents the range of stiffness parameters. The damping parameter C is mapped to the pulse frequency. ,in Taking the reference frequency (100 Hz) as the basis and κ as the scaling factor (0.8 Hz·m / Ns), the partitioned distribution module distributes signals according to the position encoding of the suspension actuator: The main suspension controller (near the wheel) receives the pulse width modulation signal and transmits it through the CAN bus. The data frame includes the controller ID, pulse width value, and effective timestamp; the auxiliary suspension controller (in the vehicle body stability system) receives the pulse frequency modulation signal and transmits it through the FlexRay bus. The data frame includes the frequency value, phase offset, and duration. Each controller address corresponds strictly to the physical position. The ID of the front left wheel controller is 0x10, the front right wheel ID is 0x11, the rear left wheel ID is 0x12, the rear right wheel ID is 0x13, and the active stabilizer bar controller ID is 0x20.

[0039] The essential characteristics of the disturbance event are accurately captured through time-frequency analysis. The tuning rule library provides parameter mapping based on the physical model, and the strategy recalculation ensures the real-time adaptability of the control system. The signal conversion and partitioned distribution mechanism transforms the digital control strategy into the precise actions of the physical actuator, forming a closed-loop response from disturbance perception to execution control. While ensuring real-time performance, the entire process maintains the stability and reliability of the control system.

[0040] Example 4: The cloud collaborative processor receives the cross-vehicle collaboration identifier in the transient disturbance component through the vehicle-mounted V2X communication module. These identifiers include the vehicle's longitude and latitude coordinates (in WGS84 format), the current velocity vector, the acceleration amplitude, and the disturbance type encoding. When the vehicle is traveling at 85 km / h on the highway and a continuous resonance mode (encoded RES_102) is detected, a collaboration request frame is immediately generated, including the vehicle's position (longitude 116.46°, latitude 39.92°), velocity vector (23.6 m / s eastward, 0.3 m / s northward), and resonance frequency 123 Hz. This request is broadcast to the cloud server through 5G-V2X. The server retrieves adjacent vehicles within a radius of 500 meters and filters out three vehicles with shock absorption collaboration capabilities (license plate numbers Beijing A12345, Beijing B67890, Beijing C24680). The cloud extracts the compressed version of the main feature matrix of these vehicles in the last 5 minutes from the historical database, including 12-dimensional features such as the smoothed value of the longitudinal acceleration, the amplitude of the vertical vibration fundamental frequency, and the suspension actuation status flag. The data is transmitted to the requesting vehicle in binary stream format, and the transmission delay is controlled within 35 milliseconds.

[0041] After receiving the damping feature matrices of neighboring vehicles, the coupling strategy optimizer first performs time alignment and coordinate transformation. All vehicle data are uniformly transformed to a coordinate system with the vehicle's center of mass as the origin, and the time axis is synchronized with the GPS clock. Then, coupling constraints are constructed: the difference in vertical vibration phase between the vehicle and each neighboring vehicle is calculated to generate phase synchronization constraint boundaries; the vibration energy transfer relationship is analyzed to establish amplitude coordination constraint ranges. The optimization objective is set to minimize the vibration transmission difference between vehicles, and a distributed optimization algorithm is used for iterative calculation. For phase synchronization constraints, the vertical vibration phase difference between the vehicle and neighboring vehicles must be kept within ±15 degrees; for amplitude coordination constraints, the vibration amplitude of the vehicle must not exceed ±20% of the average amplitude of neighboring vehicles. After three rounds of iterative calculation, the final set of coupling constraints is generated, including phase adjustment, amplitude scaling factor, and time offset parameters.

[0042] The feature tracing module continuously monitors the amplitude changes of transient disturbance components. When the vertical vibration residual amplitude exceeds the threshold of 0.4g for five consecutive sampling cycles, the historical data backtracking mechanism is triggered. The system automatically loads the multi-dimensional driving dataset from the past 30 seconds and extracts feature vectors related to the current disturbance, including 18 dimensions such as road surface roughness index, suspension hydraulic pressure value, vehicle body vertical acceleration spectrum characteristics, and occupant heart rate variability coefficient. The backtracking time window is dynamically adjusted according to the duration of the disturbance, and can trace back to historical data up to 120 seconds ago.

[0043] The perturbation clustering module receives the associated feature vectors obtained from backtracking and first adds a vehicle condition label to each data point. The label includes three types of information: driving mode, road surface type, and load status. An improved K-means clustering algorithm is used: when initializing cluster centers, data points with significant differences in operating conditions are prioritized; a weighted Euclidean distance metric is used, assigning higher weights to vehicle motion-related features; the number of clusters is dynamically adjusted during iteration, allowing a maximum of 5 clusters to be generated. The clustering results generate two types of perturbation clusters: equipment-level perturbation clusters and environmental-level perturbation clusters. Equipment-level perturbation clusters mainly include suspension system wear characteristics and sensor drift characteristics; environmental-level perturbation clusters include road surface unevenness characteristics, climate perturbation characteristics, and traffic flow characteristics.

[0044] Table 1: Distribution of Perturbation Clustering Results

[0045] Referring to Table 1, equipment-level disturbance clusters are output to the suspension health diagnosis unit. This unit uses a deep learning-based predictive maintenance algorithm to analyze the performance degradation trend of the suspension system and generate a maintenance recommendation report. Environmental-level disturbance clusters are output to the environmental adaptation learning unit. This unit uses an incremental learning mechanism to continuously update the road feature library and climate disturbance model, optimizing the environmental adaptability of the shock absorption control system.

[0046] Throughout the implementation process, data transmission employs an end-to-end encryption mechanism, and the cloud-based coprocessor uses the national cryptographic algorithm SM4 to encrypt interactive data. After the coupling constraint terms are generated, they need to be verified by the vehicle controller at runtime to check whether the constraint boundaries are within the vehicle's physical limits, avoiding the generation of infeasible control commands. Historical data backtracking uses a circular buffer storage mechanism, with the latest data overwriting the oldest data to ensure storage efficiency. The clustering algorithm retrains the cluster centers every 24 hours to adapt to the slow changes in the vehicle system. Analysis results of device-level disturbance clusters generate a health status report weekly, while the learning model for environmental-level disturbance clusters is fully updated monthly. Inter-vehicle collaborative perception expands the system's cognitive range regarding environmental disturbances, and coupling optimization improves the shock absorption effect during multi-vehicle cooperative driving. The historical data tracing mechanism provides contextual information for disturbance analysis, while intelligent clustering accurately distinguishes between internal system device problems and external environmental factors.

[0047] Example 5: The edge computing partition adopts a heterogeneous computing architecture to achieve physical isolation between the real-time processing area and the non-volatile storage area. The real-time processing area is built based on programmable logic devices, configured with a dual-core processing unit and a dedicated hardware accelerator. The data separation engine operates through a pipeline mechanism: the input multidimensional driving dataset enters the first-level cache, the matrix transformation module calls the normalization processing coprocessor to perform sliding window normalization on the acceleration data, perform baseline drift correction on the physiological data, and write the processed data to the second-level cache. The tensor decomposition module activates the tensor operation accelerator, loads the preset rank constraint parameters, and completes feature basis extraction through the parallel matrix multiplication unit. The feature selection unit is deployed on the second core, the association graph construction module calls the graph computing engine to generate a multimodal feature association graph in a dedicated memory area; the entropy calculation module uses the hardware entropy calculation unit to complete the real-time update of dynamic association entropy. The data processing latency of the entire real-time processing area is controlled within 8 milliseconds, and the clock frequency is stabilized at 450MHz.

[0048] The non-volatile storage area uses a flash memory array with power-loss protection and is divided into a parameter library and a rule library. The parameter library stores vehicle dynamics parameter datasets, including structured records of suspension geometry parameters: front suspension kingpin caster angle range, rear suspension trailing arm mounting angle, stabilizer bar torsional stiffness coefficient, etc.; the mass distribution matrix is ​​dynamically stored according to vehicle configuration, including the three-dimensional moment of inertia of sprung mass and the wheel center coordinates of unsprung mass; the damping response curve library stores force-displacement characteristic curves under different operating conditions, with each curve containing 256 sampling points. The rule library stores a tuning rule mapping table using a hierarchical index structure: the first-level index is categorized by disturbance type, the second-level index is categorized by vehicle speed, and the third-level index stores specific tuning parameters. All data undergoes ECC verification before writing, parity checking is performed during reading, and a pre-write logging mechanism is used to prevent damage during power outages.

[0049] The secure isolation repeater establishes a bidirectional encrypted channel between the edge computing partition and the cloud coprocessor. During channel initialization, bidirectional authentication is performed: the edge generates an authentication request based on the vehicle's VIN code, and the cloud returns a digital certificate; both parties exchange session keys using the national cryptographic algorithm SM2. Data transmission employs a hybrid encryption mechanism: control commands are encrypted using the SM4 algorithm with a 256-bit key; data packets are appended with a message authentication code based on the SM3 algorithm. The encrypted channel maintains a heartbeat detection mechanism, exchanging timestamp verification packets every 200 milliseconds, and automatically disconnecting after a 300-millisecond timeout.

[0050] Runtime verification is performed on the coupling constraints issued from the cloud. The verification process includes three levels: format verification checks whether the data packet structure conforms to the predefined protocol specification; signature verification uses a pre-set cloud public key to verify the digital signature; and physical constraint verification checks whether the parameter values ​​are within the vehicle's safety range. For the phase adjustment amount, it is verified whether it is within ±30 degrees; for the amplitude scaling factor, it is checked whether it is within the range of 0.5 to 2.0; the timing offset parameter must meet the requirement that the difference in delay between the front and rear wheels is less than 50 milliseconds. A three-level response is triggered when verification fails: for minor anomalies, the log is recorded and the default value is used; for moderate anomalies, a retransmission request is sent to the cloud; for severe anomalies, a safe mode is initiated and cloud control is isolated.

[0051] The fault recovery mechanism of the real-time processing area adopts a dual-backup design. In the event of a primary processing unit failure, the backup unit takes over within 5 milliseconds and loads the most recent valid state snapshot from the non-volatile storage area. Snapshots are updated every 50 milliseconds, containing intermediate tensor results from the data separation engine and the latest correlation entropy value from the feature filtering unit. The wear leveling algorithm in the non-volatile storage area dynamically adjusts storage locations, and flash memory blocks with more than 100,000 write cycles are automatically marked as read-only. A secure isolation repeater maintains the session state machine, recording the encryption parameters of the last 10 communications, supporting rapid session recovery after connection interruption.

[0052] This implementation ensures the real-time performance of critical control processes through hardware-level isolation, while non-volatile storage design guarantees the persistent availability of the parameter library. Encrypted channels and runtime verification mechanisms construct a defense-in-depth system, maintaining the integrity and security of the local control system while supporting cloud-based collaboration. Fault recovery strategies minimize the risk of system outages, creating a highly reliable in-vehicle edge computing environment.

[0053] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent motion sickness damping control system, characterized in that, include: A multi-dimensional sensor array is used to collect vehicle driving status data, occupant physiological index data and environmental disturbance data in real time. The vehicle driving status data, occupant physiological index data and environmental disturbance data are time-stamp aligned and spatial coordinate synchronized to generate a spatiotemporally aligned multi-dimensional driving dataset. A data separation engine is used to perform feature decoupling and reconstruction processing on the spatiotemporally aligned multidimensional driving dataset, separating the main feature matrix for shock absorption control and the abnormal disturbance identifier set for feedback tuning; The feature filtering unit is used to perform multi-dimensional feature correlation analysis based on the main feature matrix and combined with a pre-set vehicle dynamics parameter library to filter out the core feature sequence for shock reduction decision-making. A vibration reduction strategy generator is used to generate a multi-level collaborative vibration reduction control strategy based on the core feature sequence. An abnormal feedback regulator is used to parse the abnormal disturbance identifier set, identify sudden disturbance events in vehicle driving status data or abnormal fluctuation patterns in occupant physiological index data, and trigger dynamic tuning of the shock absorption control strategy. The control command distributor is used to convert the tuned damping control strategy into drive signals and send them to the suspension actuator cluster.

2. The anti-motion sickness intelligent shock absorption control system according to claim 1, characterized in that, The data separation engine includes: The matrix transformation module is used to convert the spatiotemporally aligned multidimensional driving dataset into a normalized feature tensor; The tensor decomposition module is used to perform rank-constrained decomposition on the standardized feature tensor to obtain the steady-state feature basis and transient disturbance components. The steady-state feature basis constitutes the principal feature matrix, and the transient disturbance components constitute the abnormal disturbance identifier set.

3. The anti-motion sickness intelligent shock absorption control system according to claim 2, characterized in that, The feature filtering unit includes: The association map construction module is used to extract vehicle longitudinal acceleration features, lateral sway angle features, vertical vibration spectrum features, occupant heart rate variation features and eye movement trajectory features from the main feature matrix to construct a multimodal feature association map; The entropy calculation module is used to calculate the dynamic association entropy of each feature based on the multimodal feature association map; The sequence generation module is used to filter the top K features in descending order based on the dynamic association entropy and generate the core feature sequence.

4. The anti-motion sickness intelligent shock absorption control system according to claim 3, characterized in that, The vibration reduction strategy generator includes: The strategy mapping module is used to match the core feature sequence with a pre-set vehicle dynamics parameter library to map out the basic damping control curve. The multi-level optimization module is used to perform hierarchical optimization decomposition of the basic damping control curve based on the vehicle suspension topology, and generate the multi-level coordinated damping control strategy that includes the main suspension level control strategy and the auxiliary suspension level control strategy.

5. The anti-motion sickness intelligent shock absorption control system according to claim 4, characterized in that, The abnormal feedback regulator includes: The disturbance pattern recognition module is used to perform time-frequency feature analysis on the transient disturbance components and identify sudden road impact mode, continuous resonance mode or occupant physiological instability mode. The tuning parameter generation module is used to retrieve a preset tuning rule library based on the identified disturbance mode and generate suspension stiffness correction coefficients or damping compensation gradients. The strategy recalculation module is used to inject the suspension stiffness correction coefficient or damping compensation gradient into the multi-level collaborative damping control strategy for real-time recalculation.

6. The anti-motion sickness intelligent shock absorption control system according to claim 5, characterized in that, The control command distributor includes: The signal conversion module is used to convert the recalculated multi-level collaborative vibration reduction control strategy into a pulse width modulation signal; The partition distribution module is used to distribute the pulse width modulation signal to the corresponding main suspension controller or auxiliary suspension controller according to the position code of the suspension actuator.

7. The anti-motion sickness intelligent shock absorption control system according to claim 2, characterized in that, The system also includes: A cloud-based coprocessor is used to receive cross-vehicle cooperation identifiers in the transient disturbance components and to schedule the damping feature matrices of adjacent vehicles from the cloud. A coupling strategy optimizer is used to generate coupling constraint terms based on the damping feature matrices of the adjacent vehicles, and to perform collaborative optimization on the multi-level collaborative damping control strategy.

8. The anti-motion sickness intelligent shock absorption control system according to claim 7, characterized in that, The system also includes: The feature tracing module is used to trace back the associated feature vectors in the historical driving dataset when the transient disturbance component exceeds a preset threshold. The disturbance clustering module is used to cluster and group the associated feature vectors according to the vehicle operating condition labels to generate equipment-level disturbance clusters and environment-level disturbance clusters. The device-level disturbance clusters are input to the suspended health diagnostic unit, and the environmental-level disturbance clusters are input to the environmental adaptation learning unit.

9. The anti-motion sickness intelligent shock absorption control system according to claim 1, characterized in that, The system also includes: Edge computing partitions contain isolated real-time processing areas and non-volatile storage areas; The real-time processing area runs the data separation engine and feature filtering unit, while the non-volatile storage area stores the preset vehicle dynamics parameter library and tuning rule library.

10. The anti-motion sickness intelligent shock absorption control system according to claim 9, characterized in that, The system also includes: A secure isolation repeater is used to establish an encrypted channel between the edge computing partition and the cloud coprocessor, and to perform runtime verification on the issued coupling constraints.

Images