A method and system for road surface identification based on sprung acceleration

Abstract

The application provides a road surface identification method and system based on unsprung acceleration, which comprises the following steps: collecting unsprung vertical acceleration signals and auxiliary signals such as vehicle speed and wheel speed, and obtaining effective acceleration sequences after pretreatment; calculating the root mean square value of acceleration and the energy characteristic value of the window in the adaptive time window; according to the corresponding relationship between the current vehicle speed and the energy threshold of the road surface grade, the road surface condition is divided into multiple grades and the road surface grade of the current window is determined; after median filtering and low-pass smoothing of the road surface grade of the current window, the output is sent to the CDC shock absorber controller. Through comprehensive pretreatment of the unsprung vertical acceleration signal, the influence of noise, missing values and outliers is effectively eliminated; the adaptive time window mechanism can quickly respond when the road surface suddenly changes and stably calculate the characteristic quantity when the road surface is stable. The adaptive time window mechanism can quickly respond when the road surface suddenly changes and stably calculate the characteristic quantity when the road surface is stable.

Description

Technical Field

[0001] This invention relates to the field of active suspension control technology for automobiles, and in particular to a road surface recognition method and system based on unsprung acceleration. Background Technology

[0002] Continuous Damping Control (CDC) shock absorbers can adjust damping force in real time according to road conditions, improving vehicle ride comfort and handling stability, and are a core actuator of modern intelligent chassis systems. Road surface recognition is a prerequisite for CDC shock absorber control, and its accuracy and real-time performance directly determine the damping adjustment effect. Existing road surface recognition solutions are mainly divided into two categories: one is a vision-based solution based on cameras or LiDAR, which has high hardware costs and its reliability drops significantly under complex lighting conditions such as rain, snow, fog, and night; the other is an inertial sensor solution based on the vehicle's vertical acceleration, which is greatly affected by changes in vehicle attitude such as acceleration, deceleration, steering, and pitch, and the signal has a significant response lag after being filtered by the suspension system, making it difficult to meet the real-time requirements of CDC shock absorbers for road surface information.

[0003] Compared to the above-mentioned methods, the road surface recognition method based on unsprung vertical acceleration has a faster response and is less affected by vehicle attitude. However, existing methods still have the following shortcomings: First, the traditional direct threshold method does not take into account the influence of vehicle speed on vibration energy, and the same road surface is easily misclassified as different levels at different vehicle speeds, resulting in poor recognition consistency. Second, the fixed time window is difficult to take into account both sudden changes in road surface conditions and steady-state conditions, and its ability to handle transient impact excitations such as speed bumps and potholes is insufficient. Short-duration high-amplitude pulses often cause the road surface level to be incorrectly raised, which in turn causes the CDC shock absorber to output excessive damping force, thus exacerbating the impact on the occupants. Third, existing methods lack sufficient cleaning and preprocessing of raw acceleration data, and sensor noise, missing values, outliers, and interference frequency components seriously affect recognition accuracy. Fourth, the system lacks fault tolerance mechanisms when sensors fail, resulting in insufficient robustness. Therefore, there is an urgent need for a real-time road surface level recognition method based on unsprung acceleration that integrates comprehensive data preprocessing, vehicle speed adaptive energy threshold, adaptive time window, transient impact shielding, and sensor fault tolerance. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of inconsistent recognition of vehicle speed influence in existing CDC shock absorber road surface recognition technology, easy misjudgment of special road conditions such as speed bumps, and poor adaptability of fixed time windows.

[0005] A first aspect of the present invention provides a road surface recognition method based on unsprung acceleration, comprising: The unsprung vertical acceleration signal and auxiliary signals such as vehicle speed and wheel speed are collected and preprocessed to obtain an effective acceleration sequence. The effective acceleration sequence is divided into frames using an adaptive time window, the root mean square of acceleration within the current window is calculated, and the energy of the current window is calculated based on the root mean square value of acceleration. Based on the correspondence between the current vehicle speed and the energy of the current window, the road surface grade of the current window is determined; After processing the road surface grade of the current window, the result is output to the CDC damper controller.

[0006] Furthermore, the adaptive time window sets a baseline window length, calculates the energy change rate of adjacent windows based on the current window energy and the previous window energy, and dynamically adjusts the next window length based on the energy change rate.

[0007] Furthermore, the current window energy: ; Where RMS is the root mean square of acceleration within the current window, and T is the duration of the current window.

[0008] Furthermore, based on the current vehicle speed and the pre-stored vehicle speed-energy threshold mapping table, multiple road surface level energy thresholds corresponding to the current vehicle speed are obtained, and the road surface is divided into multiple road surface levels based on the multiple road surface level energy thresholds. The current window energy is compared with the multiple road surface level energy thresholds to determine the road surface level of the current window.

[0009] Furthermore, after determining the road surface grade of the current window, the following is also included: If the current window energy is within the boundary region of the road surface grade energy threshold, then extract the time-domain auxiliary features and frequency-domain auxiliary features within the current window to perform a secondary confirmation of the road surface grade; The time-domain auxiliary features include peak acceleration, kurtosis, skewness, waveform factor, and impulse factor; The frequency domain auxiliary features include the main frequency energy ratio, specific frequency band energy, and power spectral density slope. Based on the combined results of the auxiliary features and energy features, the road surface grade of the current window is updated.

[0010] Furthermore, after determining the road surface grade of the current window, the method also includes special road condition identification: If the ratio of the energy characteristic value of the current window to the energy characteristic value of the previous window exceeds the preset range, and the peak instantaneous acceleration in the current window exceeds the preset threshold; Or when the standard deviation of wheel speed signal fluctuation exceeds a preset threshold; If the confidence level of a sudden change in the road surface ahead detected by the auxiliary sensor is greater than a preset threshold, it is determined to be a special road condition.

[0011] Furthermore, when a special road condition is determined, the currently output road surface level is the road surface level of the previous window; if it is not determined to be a special road condition, the currently output road surface level is the road surface level of the current window.

[0012] Furthermore, the process of processing the road surface grade of the current window includes: performing median filtering and first-order low-pass smoothing on the road surface grade of the current window in sequence, and sending the smoothed road surface grade to the CDC damper controller via the CAN bus. Before the output is sent to the CDC damper controller, the following steps are also included: real-time monitoring of the working status of each sensor, and if a sensor failure is detected, switching to the alternative signal or outputting a preset safety level and recording the corresponding fault code.

[0013] Furthermore, the preprocessing includes missing value imputation, based on... Outlier detection and removal of criteria or box plots, detrending, multi-level digital filtering, resampling to a unified time base, data standardization or normalization, or at least one of these.

[0014] A second aspect of the present invention provides a road surface recognition system based on unsprung acceleration, implementing the road surface recognition method based on unsprung acceleration as described in any of the preceding claims, comprising: The signal acquisition unit is used to acquire the unsprung acceleration signal; The signal processing unit is used to preprocess the unsprung acceleration signal, calculate the root mean square of acceleration within the current window using an adaptive time window, and calculate the energy of the current window based on the root mean square value of acceleration. The road surface grade determination unit determines the road surface grade of the current window based on the correspondence between the current vehicle speed and the energy of the current window. The output unit is used to process the road surface grade of the current window and output it to the CDC damper controller.

[0015] Compared to existing technologies, this invention offers at least the following advantages: By comprehensively preprocessing the unsprung vertical acceleration signal, the influence of noise, missing values, and outliers is effectively eliminated; an adaptive time window mechanism is employed, enabling rapid response during sudden road surface changes and stable calculation of characteristic quantities on steady-state road surfaces; an adaptive energy threshold mapping table indexed by vehicle speed is introduced, improving the impact of vehicle speed variations on vibration energy and ensuring consistent grade determination results for the same road surface at different vehicle speeds; median filtering and low-pass smoothing are sequentially applied to the output road surface grade, effectively suppressing grade jumps, thereby providing accurate and stable road surface grade information for the CDC damper controller and improving vehicle ride comfort and handling stability. 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.

[0017] Figure 1 This is a schematic diagram of the steps of a road surface recognition method based on unsprung acceleration in one embodiment of the present invention; Figure 2 This is a schematic diagram of a road surface recognition method based on unsprung acceleration in one embodiment of the present invention; Figure 3 This is a schematic diagram of the overall process of a road surface recognition method based on unsprung acceleration in one embodiment of the present invention; Figure 4 This is a flowchart illustrating the data preprocessing process in one embodiment of the present invention; Figure 5 This is a flowchart illustrating the multi-digital filter processing procedure in one embodiment of the present invention; Figure 6 This is a schematic diagram of the adaptive time window and feature extraction process in one embodiment of the present invention; Figure 7 This is a mapping diagram of vehicle speed-road level energy threshold in one embodiment of the present invention; Figure 8 This is a logical diagram illustrating special road condition identification in one embodiment of the present invention; Figure 9 This is a flowchart illustrating the processing of the Simulink top-level model and FPGA acceleration unit in one embodiment of the present invention.

[0018] Among them, 1-signal acquisition unit; 2-signal processing unit; 3-road surface grade determination unit; 4-output unit. Detailed Implementation

[0019] The present invention will now be described in more detail with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. It should be understood that those skilled in the art can modify the invention described herein while still achieving its advantageous effects. Therefore, the following description should be understood as being broadly known to those skilled in the art and is not intended to limit the invention.

[0020] 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.

[0021] The invention is described more specifically by way of example in the following paragraphs with reference to the accompanying drawings. The advantages and features of the invention will become clearer from the following description. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the invention.

[0022] Example 1 This embodiment provides a road surface recognition method based on unsprung acceleration. Please refer to [link / reference]. Figure 1 and Figure 3 ,include: The unsprung vertical acceleration signal and auxiliary signals such as vehicle speed and wheel speed are collected and preprocessed to obtain an effective acceleration sequence.

[0023] The effective acceleration sequence is divided into frames using an adaptive time window, the root mean square of acceleration within the current window is calculated, and the energy of the current window is calculated based on the root mean square value of acceleration.

[0024] Based on the correspondence between the current vehicle speed and the energy of the current window, the road surface grade of the current window is determined.

[0025] The road surface grade of the current window is processed and then output to the CDC damper controller.

[0026] In this embodiment, unsprung vertical acceleration signals of the four wheels are acquired in real time, along with auxiliary signals such as vehicle speed and wheel speed. The raw signals are preprocessed to obtain an effective acceleration sequence. An adaptive time window is used to slide and frame the effective acceleration sequence, and the root mean square (RMS) value of the unsprung vertical acceleration signal within the current window is calculated. The energy of the current window is then calculated based on the RMS value.

[0027] Based on the correspondence between the current vehicle speed and the current window energy, the current window energy is compared with the energy threshold of each level at the corresponding vehicle speed to determine the road surface level of the current window; finally, the road surface level is filtered and smoothed and then output to the CDC damper controller to drive the adaptive adjustment of the damping force.

[0028] Furthermore, the preprocessing includes missing value imputation, based on... Outlier detection and removal of criteria or box plots, detrending, multi-level digital filtering, resampling to a unified time base, data standardization or normalization, or at least one of these.

[0029] Specifically, the preprocessing step systematically cleans and organizes the raw acquired signals. To address the issue of occasional data loss from sensors, a missing value interpolation method is used to fill in the missing data positions, ensuring the signal sequence is complete. For abrupt changes introduced by sensor malfunctions or external electromagnetic interference, a method based on... Outlier detection methods, such as criteria or box plots, are used to identify and remove outliers to prevent them from contaminating subsequent feature calculation results. To address signal baseline drift during long-term vehicle operation, detrending processing is performed to eliminate low-frequency trend components. For high-frequency electrical noise and interference frequency components unrelated to road surface excitation, multi-stage digital filtering is used to shape the signal in the frequency domain, retaining effective frequency components relevant to road conditions. When the sampling frequencies or time bases of multiple sensor signals are inconsistent, each signal is resampled to align to a unified time base, ensuring time synchronization. Furthermore, signals are standardized or normalized according to actual needs to eliminate the impact of differences in sensor ranges on feature extraction. At least one of these preprocessing steps can be selected and combined according to the actual application scenario to provide a high-quality effective acceleration sequence for subsequent window energy feature calculation.

[0030] Furthermore, the adaptive time window sets a baseline window length, calculates the energy change rate of adjacent windows based on the current window energy and the previous window energy, and dynamically adjusts the next window length based on the energy change rate.

[0031] Furthermore, the window energy: .

[0032] Where RMS is the root mean square value of the unsprung vertical acceleration signal within the current window, and T is the duration of the current window.

[0033] In one embodiment, the adaptive time window mechanism uses a preset reference window length as the initial framing unit to perform sliding framing processing on the effective acceleration sequence. The root mean square (RMS) value of the unsprung vertical acceleration signal within each window is calculated, and the product of the square of the RMS and the current window duration T is used as the window energy value E. This incorporates both signal amplitude and window duration into the energy representation, making the energy characteristics under different window lengths comparable.

[0034] After each window's energy calculation is completed, the energy change rate of adjacent windows is obtained by comparing the difference between the current window's energy value and the previous window's energy value. This rate is used as the basis for dynamically adjusting the window length. When the rate of change exceeds a preset upper threshold, a sudden change in road conditions is determined, and the next window length is shortened to below the baseline length to improve time resolution and enable the system to respond quickly to changes in road conditions. When the rate of change is less than a preset lower threshold, the road surface is determined to be in a steady state, and the next window length is extended to above the baseline length to include more data samples and improve the statistical stability of feature calculations. When the rate of change is between the upper and lower thresholds, the current window length remains unchanged, thereby achieving automatic switching of the adaptive time window between sudden change conditions and steady-state conditions.

[0035] Furthermore, based on the current vehicle speed and the pre-stored vehicle speed-energy threshold mapping table, multiple road surface level energy thresholds corresponding to the current vehicle speed are obtained, and the road surface is divided into multiple road surface levels based on the multiple road surface level energy thresholds.

[0036] The current window energy is compared with the multiple road surface level energy thresholds to determine the road surface level of the current window.

[0037] In one possible embodiment of the present invention, the energy thresholds for each road surface level are generated through an offline calibration process. During the calibration phase, a large amount of real vehicle unsprung vertical acceleration data is collected under different known road surface types (such as good asphalt pavement, general road surface, slightly damaged road surface, severely damaged road surface, etc.) and multiple vehicle speed conditions. The window energy value for each condition is calculated, and the energy feature quantity samples at the same vehicle speed are clustered using the K-means clustering algorithm. The energy boundary values ​​between cluster centers are used as the energy thresholds for each level corresponding to that vehicle speed. This process is repeated for multiple discrete vehicle speed nodes, ultimately forming a threshold mapping table with vehicle speed as the index and energy thresholds for each level as the value. This mapping table is then embedded in the controller.

[0038] During the online recognition phase, the system reads the current vehicle speed, locates two adjacent speed nodes in the threshold mapping table, and calculates the energy threshold for each level corresponding to the current speed using linear interpolation. This eliminates abrupt threshold changes between discrete speed nodes, ensuring that the recognition result changes continuously with vehicle speed. After obtaining the interpolated threshold, the energy of the current window is compared with the thresholds for each level to determine the road level to which the current window belongs. Because the energy threshold adaptively adjusts with vehicle speed, consistent level determination results can be obtained for the same road surface at different driving speeds, effectively overcoming the problem of poor recognition consistency of the fixed threshold method under variable speed conditions.

[0039] Furthermore, after determining the road surface grade of the current window, the following is also included: If the current window energy is within the boundary region of the road surface grade energy threshold, then the time-domain auxiliary features and frequency-domain auxiliary features within the current window are extracted to perform a secondary confirmation of the road surface grade.

[0040] The time-domain auxiliary features include peak acceleration, kurtosis, skewness, waveform factor, and impulse factor.

[0041] The frequency domain auxiliary features include the main frequency energy ratio, specific frequency band energy, and power spectral density slope.

[0042] Based on the combined results of the auxiliary features and energy features, the road surface grade of the current window is updated.

[0043] In one embodiment, considering the decreased discriminative reliability of a single feature when the window energy value is near the boundary of adjacent level thresholds, the system introduces a secondary confirmation mechanism based on time-domain and frequency-domain auxiliary features. When the current window energy value is within the boundary region of the road surface level energy threshold, a secondary confirmation process is triggered to further extract time-domain and frequency-domain auxiliary features from the effective acceleration sequence within the current window.

[0044] In this embodiment, if the absolute value of the difference between the current window energy and a certain level threshold is less than the ratio of the level threshold to a preset proportion threshold, then the time-domain auxiliary features and frequency-domain auxiliary features within the current window are extracted to perform a secondary confirmation of the road surface level.

[0045] Time-domain auxiliary features include: peak acceleration, reflecting the maximum amplitude intensity of the signal within the window; kurtosis, reflecting the sharpness of the signal amplitude distribution, which is more sensitive to impact-type excitations; skewness, reflecting the symmetry of the signal amplitude distribution, used to distinguish between unidirectional impacts and bidirectional random excitations; waveform factor, reflecting the ratio of the signal's effective value to its mean, describing the overall shape of the waveform; and impulse factor, reflecting the ratio of the signal's peak value to its effective value, which has a strong ability to distinguish transient impacts. Frequency-domain auxiliary features include: dominant frequency energy proportion, reflecting the concentration of signal energy near the dominant frequency; specific frequency band energy, reflecting the energy distribution within a typical frequency range related to road surface type; and power spectral density slope, reflecting the trend of signal energy attenuation with frequency, which varies among different road surface types.

[0046] By combining the aforementioned time-domain and frequency-domain auxiliary features with the initial energy feature determination results, a weighted comprehensive determination of the road surface grade in the current window is made. If the auxiliary features are consistent with the initial determination results, the determination is confirmed. If there is a conflict, the road surface grade in the current window is updated based on the determination results of the auxiliary features, thereby effectively reducing the probability of misjudgment under boundary conditions and improving the overall recognition accuracy.

[0047] Furthermore, after determining the road surface grade of the current window, it also includes the identification of special road conditions: If the ratio of the energy characteristic value of the current window to the energy characteristic value of the previous window exceeds the preset range, and the peak instantaneous acceleration in the current window exceeds the preset threshold.

[0048] Or when the standard deviation of wheel speed signal fluctuation exceeds the preset threshold.

[0049] If the confidence level of a sudden change in the road surface ahead detected by the auxiliary sensor is greater than a preset threshold, it is determined to be a special road condition.

[0050] Furthermore, when a special road condition is determined, the currently output road surface level is the road surface level of the previous window; if it is not determined to be a special road condition, the currently output road surface level is the road surface level of the current window.

[0051] In one embodiment, after determining the road surface level for the current window, the system also executes a special road condition identification process to address the interference of transient impact excitations such as speed bumps and potholes on the road surface level output. The determination of special road conditions is based on the following three conditions; meeting any one of them constitutes a special road condition: First, the ratio of the energy value of the current window to the energy value of the previous window exceeds a preset normal range, and the peak instantaneous acceleration within the current window simultaneously exceeds a preset threshold. Both conditions being met simultaneously indicate an abnormal jump in window energy accompanied by a significant impact amplitude, exhibiting typical transient impact characteristics. Second, the standard deviation of wheel speed signal fluctuation exceeds a preset threshold, indicating that the wheels have passed over discrete obstacles such as bumps or depressions, resulting in significant wheel speed disturbances. Third, the confidence level of auxiliary sensors (such as forward-looking cameras or navigation systems) detecting a sudden change in the road surface ahead exceeds a preset threshold, indicating that the system has obtained prior information about the abnormal road surface ahead from other information sources.

[0052] When a special road condition is identified, the system maintains the road surface grade output from the previous window, rather than using the abnormally high grade from the current window due to transient impact. This prevents the CDC shock absorber from outputting excessive damping force due to short-term high-amplitude pulses, thus avoiding exacerbating the impact on occupants. When no special road condition is detected, the system normally uses the road surface grade from the current window as the output. This mechanism effectively shields against transient impact excitation, ensuring the rationality and stability of the road surface grade output under special road conditions.

[0053] Furthermore, the process of processing the road surface grade of the current window includes: performing median filtering and first-order low-pass smoothing on the road surface grade of the current window in sequence, and sending the smoothed road surface grade to the CDC damper controller via the CAN bus. Before the output is sent to the CDC damper controller, the following steps are also included: real-time monitoring of the working status of each sensor, and if a sensor failure is detected, switching to the alternative signal or outputting a preset safety level and recording the corresponding fault code.

[0054] Specifically, before the current road surface grade is output to the CDC damper controller, it undergoes two stages of post-processing: median filtering and first-order low-pass smoothing. Median filtering takes the road surface grades from several consecutive windows as input and outputs the median, effectively eliminating grade jumps caused by occasional misjudgments and strongly suppressing isolated abnormal grade points. Based on this, a first-order low-pass smoothing process is applied to the median-filtered grade sequence. By weighted fusion of the smoothing results from the current and previous moments, the road surface grade output transitions slowly over time, preventing severe fluctuations in the CDC damper's damping force caused by frequent switching between adjacent grades. The smoothed road surface grade obtained after these two stages of post-processing is sent to the CDC damper controller in real time via the CAN bus to drive adaptive adjustment of the damping force.

[0055] Furthermore, the system monitors the operating status of each sensor in real time during operation, including signal amplitude range checks, signal update frequency checks, and communication frame integrity verification. When a sensor failure is detected, the system switches to the corresponding alternative signal source according to a preset priority. For example, the average unsprung acceleration signal of the remaining wheels can be used to replace the failed channel signal, or the vehicle body acceleration signal can be used as a degraded alternative to ensure that the road surface recognition function can continue to operate even when a single sensor fails. If all available alternative signals do not meet the minimum quality requirements, the system outputs a preset safe road surface level, causing the CDC shock absorber to switch to a conservative medium damping mode to balance ride comfort and safety. At the same time, the corresponding fault code is recorded and reported through the diagnostic interface for subsequent maintenance and troubleshooting, thereby ensuring the basic functional availability and operational safety of the system under sensor failure conditions.

[0056] Example 2 This embodiment provides a road surface recognition system based on unsprung acceleration, implementing the road surface recognition method based on unsprung acceleration as described in Embodiment 1. Please refer to [the relevant documentation]. Figure 2 ,include: Signal acquisition unit 1 is used to acquire unsprung acceleration signals.

[0057] Signal processing unit 2 is used to preprocess the unsprung acceleration signal, calculate the root mean square of acceleration within the current window using an adaptive time window, and calculate the energy of the current window based on the root mean square value of acceleration.

[0058] Road surface grade determination unit 3 determines the road surface grade of the current window based on the correspondence between the current vehicle speed and the energy of the current window.

[0059] Output unit 4 is used to process the road surface grade of the current window and output it to the CDC damper controller.

[0060] This embodiment provides a road surface recognition system based on unsprung acceleration. The system adopts the adaptive road surface recognition method described in Embodiment 1, and consists of a complete road surface recognition processing link formed by the sequential cooperation of a signal acquisition unit 1, a signal processing unit 2, a road surface grade determination unit 3, and an output unit 4.

[0061] Signal acquisition unit 1 is responsible for acquiring the unsprung vertical acceleration signals of the four wheels in real time, and simultaneously acquiring auxiliary signals such as vehicle speed and wheel speed, providing raw data input for subsequent processing. Signal processing unit 2 receives the raw signals output by signal acquisition unit, and sequentially performs preprocessing operations such as missing value interpolation, outlier detection and removal, detrending, multi-level digital filtering, and resampling to obtain a clean and reliable effective acceleration sequence. Based on this, the sequence is framed using an adaptive time window, and the root mean square value and window energy value of each window are calculated. At the same time, the length of the next window is dynamically adjusted according to the energy change rate of adjacent windows. Road surface grade determination unit 3 receives the window energy value and current vehicle speed output by signal processing unit, obtains the energy threshold for each grade corresponding to the current vehicle speed by querying an adaptive energy threshold mapping table indexed by vehicle speed, compares the window energy value with the threshold to complete the initial classification of road surface grade, and triggers a secondary confirmation process based on time-domain and frequency-domain auxiliary features under boundary conditions. At the same time, it executes special road condition recognition logic to shield transient impact excitations such as speed bumps and potholes, and outputs the final road surface grade of the current window. Output unit 4 sequentially performs median filtering and first-order low-pass smoothing on the grade sequence output by the road surface grade determination unit to eliminate grade jumps and jitters. The smoothed road surface grade is then sent in real-time to the CDC damper controller via the CAN bus to drive adaptive adjustment of the damping force. Furthermore, this unit monitors the operating status of each sensor in real-time and automatically switches to a backup signal source or outputs a preset safety level when a sensor fails, ensuring continuous system availability and operational safety.

[0062] Example 3 The following detailed description of Embodiment 1 and Embodiment 2, in conjunction with the accompanying drawings and specific implementation methods, further illustrates these embodiments. This embodiment uses a Simulink and FPGA co-design platform, and all steps are described below in the order of data flow logic.

[0063] I. Hardware Configuration Four ADI ADXL355 unsprung accelerometers (MEMS capacitive, digital SPI output, 1kHz sampling rate) are used, installed at the steering knuckles of the left front, right front, left rear, and right rear wheels, respectively. Wheel speed, vehicle speed, and vehicle attitude signals (Bosch SMI130 six-axis IMU) are read via CAN-FD bus at sampling rates of 100 Hz, 100 Hz, and 200 Hz, respectively. The main control chip is a Xilinx Zynq UltraScale+MPSoC (ARM Cortex-A53+FPGA+NPU) with a main frequency of 200 MHz. The development environment is MATLAB / Simulink (2024b) and Vitis AI.

[0064] II. Raw Data Reception and Preprocessing An SPI interrupt is triggered every 1 ms to read four acceleration values ​​and store them in a circular buffer (depth 10000). CAN-FD messages are received at 100 Hz for wheel speed and vehicle speed, and IMU (Inertial Measurement Unit) data is received at 200 Hz.

[0065] Preprocessing flow (reference) Figure 4 It includes the following sub-steps: 2.1 Handling Missing Values If the current acceleration point is invalid, check the two valid points before and after it, and fill in the gap using linear interpolation:

[0066] Five consecutive invalid points indicate a sensor malfunction. The GPS PPS (Pulse Per Second) timestamp is used to align all signals.

[0067] 2.2 Outlier Detection The mean and standard deviation are calculated every 500 ms in a sliding window. Guidelines:

[0068] Outliers are replaced with the median of the window. If three outliers appear consecutively, the entire segment is marked as suspicious.

[0069] 2.3 Detrending Processing For each newly sampled acceleration, subtract the moving average of the most recent 100 ms (100 points):

[0070] 2.4 Multi-stage digital filtering (please refer to...) Figure 5 ) a_corr represents the acceleration signal after attitude compensation and detrending processing, i.e., the acceleration sequence after eliminating vehicle attitude changes and low-frequency drift components, which is used as the input for filtering.

[0071] Pass through in sequence: Low-pass filter: Butterworth second order, cutoff frequency 50 Hz, transfer function:

[0072] High-pass filter: Butterworth second order, cutoff frequency 0.5 Hz.

[0073] Adaptive notch filter: Real-time estimation of resonant frequency (20–30 Hz), transfer function:

[0074] The center frequency is updated every 2 seconds by peak detection using FFT (Fast Fourier Transform).

[0075] The parallel branch can be equipped with a median filter (window 5) to remove impulse noise.

[0076] a_filtered represents the effective acceleration signal output after complete filtering, that is, the pure acceleration sequence obtained after passing through low-pass filtering, high-pass filtering and adaptive notch filtering in sequence.

[0077] 2.5 Resampling and Time Base Alignment All signals (wheel speed, vehicle speed, IMU) are upsampled to 1 kHz using linear interpolation, aligned with the acceleration. For wheel speed signals with abrupt changes, a zero-order hold is used instead of interpolation.

[0078] 2.6 Data Standardization z-score normalization is used:

[0079] in and The data is updated every 60 seconds. Statistical parameters are established for each different wheel.

[0080] III. Adaptive Window Control (Please refer to) Figure 6 ) Define the energy of the k-th window as Rate of change of energy:

[0081] Exponentially weighted moving average smoothing is used:

[0082] Window length Determine according to Table 1: Table 1 Adaptive Window Length Adjustment Rules

[0083] The window sliding step size is fixed at 100 ms, and the overlap rate between adjacent windows varies with the window length (80% overlap at T=500ms, 67% overlap at T=300ms, and 87.5% overlap at T=800ms). The window length is configured in the FPGA's RMS calculation engine via registers.

[0084] IV. Feature Extraction 4.1 Root Mean Square and Energy For N sampling points within the window (N=1000×T):

[0085] 4.2 Temporal Auxiliary Features Calculate within the same window: Peak value: kurtosis: Skewness: Waveform factor: Kurtosis and skewness are calculated using a recursive online algorithm to avoid storing the entire window.

[0086] 4.3 Frequency Domain Auxiliary Features Perform an FFT on the windowed data after adding a Hanning window, and extract: Core frequency energy percentage: Frequency band energy: Calculate the energy percentage of the three frequency bands: 4-8 Hz, 8-12 Hz, and 12-20 Hz.

[0087] Power spectral density slope: for Linear fitting is performed, and the slope k reflects the road surface texture.

[0088] V. Threshold Table and Original Grade Determination (Please refer to) Figure 7 ) A pre-stored vehicle speed-energy threshold mapping table is shown in Table 2, with some data. This table was generated offline from real vehicle calibration data using K-means clustering.

[0089] Table 2 Vehicle Speed-Energy Threshold Mapping Table (Partial)

[0090] Please refer to Figure 7The initial road level determination includes obtaining multiple road surface level energy thresholds corresponding to the current vehicle speed based on the current vehicle speed and a pre-stored vehicle speed-energy threshold mapping table, and dividing the road surface into multiple road surface levels based on these energy thresholds. The current window energy is compared with the multiple road surface level energy thresholds to determine the road surface level of the current window.

[0091] In this embodiment, for any vehicle speed v, cubic spline interpolation is used to obtain Th1~Th5, corresponding to 6 levels. If If the path is not cleared, it enters the boundary region and uses auxiliary features for voting correction. Table 3 shows a typical correspondence between auxiliary features and road surface types: Table 3 Correspondence between auxiliary features and road surface types

[0092] Example of voting rules for the boundary region: If E is close to Th2 and K>3.5 and the 4-8 Hz frequency accounts for more than 30%, the level is upgraded; if K<3.0 and the frequency accounts for more than 35%, the level is downgraded.

[0093] VI. Multi-source special road condition identification (please refer to) Figure 8 ) Simultaneously assess four conditions: Condition A (Energy Mutation): If R < 0.33 or R > 3.0, it is true.

[0094] Condition B (Instantaneous Impact): .

[0095] Condition C (Wheel speed fluctuation): Standard deviation of wheel speed .

[0096] Condition D (Auxiliary Sensor): Optional camera or millimeter-wave radar detects a sudden change in road surface within 20 meters ahead, with a confidence level greater than 0.8.

[0097] Logical expression: mask= When mask=1, it is identified as a special road condition (speed bump, pothole, bridge joint, etc.), and the output level remains at the previous window value: ;otherwise: .

[0098] VII. Smooth Output and Fault Tolerance right Perform median filtering (window length 3). Definition of median filtering:

[0099] Then it is filtered by a first-order low-pass filter (time constant 0.2 seconds):

[0100] The final level is encapsulated as a CAN message (ID 0x310, data bytes 0-2 represent levels 0-7, 0 represents a fault) and sent to the CDC controller.

[0101] Fault diagnosis and fault tolerance: When the signal variance of a single sensor is abnormal, short-term replacement should be performed using the historical valid data of that sensor (the average value of the first 100 ms).

[0102] If the signal remains abnormal for 1 second, switch to the average value of other wheel signals.

[0103] When two or more sensors fail, the system enters a safe mode, outputs a default level of 3, and reduces the damping change rate of the CDC damper to 50% of the normal value.

[0104] The system will automatically exit safe mode once all sensors have recovered. All fault codes will be broadcast via CAN.

[0105] VIII. Experimental Verification and Principle Supplementation The verification was conducted by combining real vehicle data with digital twin simulation, and typical results are shown in the table below.

[0106] Table 4

[0107] Simulated failure of the left front sensor: The average value of the other three channels was automatically used, and the output was normal. After running federated learning for one week, the accuracy of unfamiliar city road recognition improved from 82% to 94%.

[0108] Explanation of core principles: The effective frequency range of unsprung acceleration is 0.5–50 Hz. A bandpass filter preserves this frequency band, while an adaptive notch filter removes the 20–30 Hz resonance. The amplitude-frequency response of the filter bank can be expressed as:

[0109] The sliding window overlap mechanism outputs the rank every 100 ms, with an update rate of 10 Hz. When the window length changes, the overlap ratio adapts to prevent abrupt feature changes.

[0110] Energy characteristics are used to achieve adaptive grading based on vehicle speed through a vehicle speed-energy threshold table.

[0111] Boundary auxiliary feature voting reduces the false positive rate by more than 50%. The recursive form of the kurtosis calculation formula is:

[0112] Please refer to Figure 9This diagram illustrates the Simulink top-level model and FPGA acceleration unit. Accel_FL / FR / RL / RR represent the unsprung vertical acceleration signals of the four wheels, corresponding to the front left, front right, rear left, and rear right, respectively; WheelSpeed ​​represents the wheel speed signal; CAN_Rx is the CAN bus receiver module; Sigma_Cala is the standard deviation calculation module; Energy is the window energy calculation module; and Camera_Radar represents the camera and radar used for target detection. The FPGA+ARM heterogeneous architecture uses the FPGA for parallel calculation of RMS and peak values, while the ARM executes the decision logic, resulting in a system latency of less than 10 ms.

[0113] Energy-saving mode: When the vehicle is on a flat highway and the energy efficiency rating is low... After 5 minutes, the sampling rate dropped to 200 Hz, the window remained fixed at 1000 ms, and power consumption decreased by approximately 70%; detection was achieved. Restore immediately.

[0114] Wheel speed fluctuation refinement: If the standard deviation of a single wheel exceeds 1.0 km / h and lasts for 200 ms, it is judged as a tire blowout or slippage, and the CDC damping is forcibly set to 80% of the maximum value, and broadcast to ESC (Electronic Stability Controller) via CAN.

[0115] Step size factor of adaptive notch filter Dynamically adjusts with signal-to-noise ratio: when SNR>20dB When SNR < 10dB .

[0116] For cost-sensitive vehicle models, only the left front wheel acceleration sensor can be used, eliminating the wheel speed fluctuation condition, auxiliary sensors, and federated learning. Preprocessing retains bandpass filtering and detrending, while the rest of the logic remains the same. This can run on an STM32F4 series MCU. In this case, the algorithm complexity is reduced by approximately 60%, while still maintaining a recognition accuracy of over 85%.

[0117] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A road surface recognition method based on unsprung acceleration, characterized in that, include: The unsprung acceleration signal is acquired and preprocessed to obtain an effective acceleration sequence; The effective acceleration sequence is divided into frames using an adaptive time window, the root mean square of acceleration within the current window is calculated, and the energy of the current window is calculated based on the root mean square of acceleration. Based on the correspondence between the current vehicle speed and the energy of the current window, the road surface grade of the current window is determined; The road surface grade of the current window is processed and then output to the CDC damper controller; The adaptive time window sets a baseline window length, calculates the energy change rate of adjacent windows based on the current window energy and the previous window energy, and dynamically adjusts the next window length based on the energy change rate. The current window energy: ; Where RMS is the root mean square of acceleration within the current window, and T is the duration of the current window.

2. The road surface recognition method based on unsprung acceleration as described in claim 1, characterized in that, Based on the current vehicle speed and the pre-stored vehicle speed-energy threshold mapping table, obtain multiple road surface level energy thresholds corresponding to the current vehicle speed, and divide the road surface into multiple road surface levels based on the multiple road surface level energy thresholds; The current window energy is compared with the multiple road surface level energy thresholds to determine the road surface level of the current window.

3. The road surface recognition method based on unsprung acceleration as described in claim 2, characterized in that, After determining the road surface grade of the current window, the following is also included: If the current window energy is within the boundary region of the road surface grade energy threshold, then extract the time-domain auxiliary features and frequency-domain auxiliary features within the current window to perform a secondary confirmation of the road surface grade; The time-domain auxiliary features include peak acceleration, kurtosis, skewness, waveform factor, and impulse factor; The frequency domain auxiliary features include the main frequency energy ratio, specific frequency band energy, and power spectral density slope. Based on the combined results of auxiliary features and energy features, the road surface grade of the current window is updated.

4. The road surface recognition method based on unsprung acceleration as described in claim 1, characterized in that, After determining the road surface grade of the current window, the method also includes special road condition identification: If the ratio of the energy characteristic value of the current window to the energy characteristic value of the previous window exceeds the preset range, and the peak instantaneous acceleration in the current window exceeds the preset threshold; Or when the standard deviation of wheel speed signal fluctuation exceeds a preset threshold; If the confidence level of a sudden change in the road surface ahead detected by the auxiliary sensor is greater than a preset threshold, it is determined to be a special road condition.

5. The road surface recognition method based on unsprung acceleration as described in claim 4, characterized in that, When a special road condition is identified, the road surface grade of the current window is the same as the road surface grade of the previous window; If the road condition is not determined to be a special road condition, output the road surface grade of the current window.

6. The road surface recognition method based on unsprung acceleration as described in claim 1, characterized in that, The processing of the road surface grade of the current window includes: performing median filtering and first-order low-pass smoothing on the road surface grade of the current window in sequence, and sending the smoothed road surface grade to the CDC damper controller via the CAN bus. Before the output is sent to the CDC damper controller, the following steps are also included: real-time monitoring of the working status of each sensor, and if a sensor failure is detected, switching to the alternative signal or outputting a preset safety level and recording the corresponding fault code.

7. The road surface recognition method based on unsprung acceleration as described in claim 1, characterized in that, The preprocessing includes missing value imputation, based on... Outlier detection and removal of criteria or box plots, detrending, multi-level digital filtering, resampling to a unified time base, data standardization or normalization, or at least one of these.

8. A road surface recognition system based on unsprung acceleration, implementing the road surface recognition method based on unsprung acceleration as described in any one of claims 1-7, characterized in that, include: The signal acquisition unit is used to acquire the unsprung acceleration signal; The signal processing unit is used to preprocess the unsprung acceleration signal, calculate the root mean square of acceleration within the current window using an adaptive time window, and calculate the energy of the current window based on the root mean square of acceleration. The road surface grade determination unit determines the road surface grade of the current window based on the correspondence between the current vehicle speed and the energy of the current window. The output unit is used to process the road surface grade of the current window and output it to the CDC damper controller.

Images