An AI-based shock working condition classification and damping matching control method and device

Abstract

The application discloses an AI-based impact working condition classification and damping matching control method and device, and relates to the technical field of vehicle suspension systems, wherein the method comprises the following steps: collecting road surface images in front of a vehicle, vertical vibration acceleration of a vehicle body at wheels, and wheel speed signals; image space features and signal time sequence features are respectively extracted through an embedded neural network, impact types and impact energy values are fused and output, and working condition labels are combined; corresponding damping control strategies and target damping parameters are matched according to the working condition labels; current instructions are generated before the wheels contact impact objects and are sent to shock absorber electromagnetic valve drivers, the damping of the shock absorber is adjusted to the target value in advance, and damping feedforward control of road surface impact is realized. Through the layered feedforward control of visual prediction and AI working condition classification, the application realizes adaptive regulation and control of damping.

Description

Technical Field

[0001] This invention relates to the field of vehicle suspension system technology, and in particular to an AI-based method and device for shock condition classification and damping matching control. Background Technology

[0002] As a core component of a semi-active suspension system, the continuously damped controlled shock absorber (CDC) directly affects the vehicle's ride comfort and handling stability through its real-time damping characteristics adjustment capability. Currently, the damping adjustment strategy of CDC shock absorbers mainly relies on vertical vibration signals collected by acceleration sensors located near the vehicle's center of gravity and suspension. A closed-loop feedback control algorithm is used to calculate changes in vehicle attitude in real time, and then output corresponding damping force adjustment commands.

[0003] However, this feedback control mode is inherently limited by the logic of "sensing the impact first, then adjusting," meaning the system can only trigger the adjustment of damping parameters after the wheels have actually rolled over the impact surface and the vehicle's vibration response has been captured by the acceleration sensor. This sequential "detection-calculation-execution" link inevitably introduces a time delay between the impact and damping intervention. Consequently, the shock absorber often only implements effective damping intervention after some of the impact energy has already been transferred to the vehicle body, making it difficult to provide optimal damping force immediately upon impact. This weakens the suspension system's ability to actively suppress sudden road surface impacts and limits the vehicle's shock comfort performance under complex road conditions.

[0004] On the other hand, road impact conditions exhibit significant diversity and time-varying characteristics, with fundamentally different impact mechanisms affecting the vehicle's vertical dynamic response. Typical conditions such as single large impacts, continuous high-frequency small impacts, and symmetrical impacts place varying performance demands on the suspension system's compression stroke control, energy dissipation rate, and vehicle attitude maintenance. Existing technologies generally lack the ability to accurately identify impact condition types, making it difficult to distinguish between different road excitation modes such as unilateral impacts, bilateral impacts, and continuous impacts. They typically employ a uniform damping adjustment logic to address various impact conditions. The direct consequence of this is that when facing continuous high-frequency small impacts, the control system is highly prone to misinterpreting multiple independent impacts as a single impact event or frequently triggered repetitive impact events. This causes the damping valve to endure high-frequency reciprocating commands in a short period, which not only accelerates the wear rate of solenoid valves and hydraulic components and shortens the lifespan of shock absorbers, but also leads to disordered suspension dynamic response due to frequent switching of damping force, weakening the shock absorber's effective dissipation of continuous vibration energy and making it impossible to match differentiated damping control strategies to the physical characteristics of various impacts.

[0005] In summary, existing damping control methods for CDC shock absorbers have significant shortcomings in terms of real-time impact condition perception, accuracy of impact type identification, and targeted damping strategy matching. They struggle to fully utilize the proactive information from visual perception to pre-adjust damping before an impact, and cannot implement dedicated damping control logic for different impact types, thus limiting the overall performance of semi-active suspension systems under complex road conditions. Therefore, there is an urgent need for a method that can pre-classify impact conditions before the wheel contacts the impactor and match a dedicated damping control strategy accordingly. This would overcome the lag bottleneck of traditional feedback control modes and improve impact comfort and system durability. Summary of the Invention

[0006] The main objective of this invention is to provide an AI-based method for classifying impact conditions and matching damping control.

[0007] Another objective of this invention is to propose an AI-based impact condition classification and damping matching control device.

[0008] The third objective of this invention is to provide an electronic device.

[0009] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above objectives, a first aspect of the present invention proposes an AI-based method for impact condition classification and damping matching control, comprising: Acquire images of the road surface in front of the vehicle, as well as vertical vibration acceleration and wheel speed signals of the vehicle body at the corresponding positions of each wheel; The embedded neural network processing unit extracts spatial features from the road surface image and extracts time-series features from the vertical vibration acceleration and wheel speed signals. Based on the extracted spatial and time-series features, the impact type and impact energy value are output respectively, and the two are combined into a working condition label. Based on the impact type in the working condition label, determine the corresponding damping control strategy, and match the corresponding target damping parameter under the damping control strategy based on the impact energy value. Before the wheel contacts the impacting object, a current command is generated based on the target damping parameters and sent to the solenoid valve driver of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, thus completing the damping feedforward control before the impact.

[0011] Optionally, acquire images of the road surface in front of the vehicle, as well as vertical vibration acceleration and wheel speed signals of the vehicle body at corresponding positions of each wheel, including: A forward-facing camera installed inside the vehicle's windshield captures images of the road surface ahead of the vehicle's direction of travel, including information on impact objects on the road ahead. Four acceleration sensors, each installed at a suspension mounting point near one of the four wheels, are used to collect the vertical vibration acceleration of the vehicle body at the corresponding position of each wheel. The vertical vibration acceleration is used to characterize the vibration response of the wheel to the vehicle body transmitted through the suspension when the wheel is impacted by the road surface. Four wheel speed sensors, each installed at the hub of one of the four wheels, collect wheel speed signals from each wheel. These wheel speed signals are used to help determine the vehicle's current speed and the impact energy that the wheel will soon experience.

[0012] Optionally, spatial feature extraction is performed on the road surface image using an embedded neural network processing unit, and time-series feature extraction is performed on the vertical vibration acceleration and wheel speed signals, including: The convolutional neural network model in the embedded neural network processing unit performs convolution operations on the road surface image to extract the spatial morphological features of the impactor in the road surface image, wherein the spatial morphological features include the outline, size and position information of the impactor relative to the vehicle. The long short-term memory network model in the embedded neural network processing unit is used to perform time series analysis on the vertical vibration acceleration and the wheel speed signal, and extract the time series variation characteristics of the vertical vibration acceleration and the wheel speed signal. The time series characteristics are used to characterize the dynamic response trend of the wheel under the impact of the road surface.

[0013] Optionally, the impact type and impact energy value are output based on the extracted spatial and time series features, respectively, and combined into a working condition label, including: The embedded neural network processing unit is constructed by connecting a convolutional neural network model and a long short-term memory network model in series. The spatial features and the time series features are respectively used as the outputs of the two models. After the embedded neural network processing unit fuses the two types of features, the impact type and the impact energy value are generated respectively. The generated impact type and impact energy value are combined together to form the working condition label.

[0014] Optionally, based on the impact type in the operating condition label, a corresponding damping control strategy is determined, and based on the impact energy value, a corresponding target damping parameter is matched under the damping control strategy, including: When the impact type is a unilateral impact, a high impedance strategy is adopted to control the unilateral shock absorber corresponding to the impact type to perform a high impedance action before the impact arrives, so as to resist suspension compression and reduce the transmission of impact to the vehicle body. When the impact type is a bilateral impact, a synchronous compression strategy is adopted, the front and rear axle shock absorbers are synchronously controlled to perform synchronous compression actions, and the vehicle body is kept in a horizontal posture when passing through the impact object by adjusting the front and rear axle damping ratio. When the impact type is continuous impact, an energy dissipation strategy is adopted to control the four-wheel shock absorbers to perform energy dissipation actions simultaneously, thereby attenuating continuous vibration by increasing the energy absorption in the compression stroke and rebound stroke of the shock absorbers. Based on the energy level to which the impact energy value belongs and in combination with the impact type, the corresponding target compression damping coefficient and target rebound damping coefficient are matched from the preset damping control map. Each map in the preset damping control map corresponds one-to-one with the combination of impact type and impact energy level.

[0015] Optionally, before the wheel contacts the impacting object, a current command is generated based on the target damping parameters and sent to the solenoid valve actuator of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, completing the damping feedforward control before the impact, including: Based on the pre-calibrated correspondence between current and damping coefficient, the target damping parameter is converted into a corresponding current command; The current command is sent to the solenoid valve driver of the corresponding shock absorber via the CAN bus; According to the received current command, each of the solenoid valve drivers adjusts the valve core opening of the corresponding solenoid valve, changes the cross-sectional area of ​​the damping channel inside the shock absorber, and makes the actual damping coefficient of the corresponding shock absorber reach the target damping value corresponding to the target damping parameter. Taking the moment when the front-view camera captures an image of the road surface containing an impact object as the starting moment, before the wheel contacts the impact object, the matching of the target damping parameters, the generation and transmission of the current command, and the adjustment of the solenoid valve opening are completed sequentially, so that the corresponding shock absorber has the target damping value at the moment the wheel contacts the impact object, thus completing the damping feedforward control before the impact.

[0016] To achieve the above objectives, a second aspect of the present invention provides an AI-based impact condition classification and damping matching control device, comprising: The multi-source sensing and acquisition module is used to acquire images of the road surface in front of the vehicle, as well as vertical vibration acceleration and wheel speed signals of the vehicle body at the corresponding positions of each wheel; The feature extraction and classification module is used to extract spatial features from the road surface image through an embedded neural network processing unit, and to extract time series features from the vertical vibration acceleration and wheel speed signals. Based on the extracted spatial features and time series features, it outputs the impact type and impact energy value respectively, and combines the two into a working condition label. The strategy matching parameter module is used to determine the corresponding damping control strategy based on the impact type in the working condition label, and to match the corresponding target damping parameter under the damping control strategy based on the impact energy value. The instruction generation and execution module is used to generate a current instruction based on the target damping parameter before the wheel contacts the impacting object, and send it to the solenoid valve driver of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, thereby completing the damping feedforward control before the impact.

[0017] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0018] To achieve the above objectives, a third aspect of this application provides an electronic device, including a processor and a memory; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing an AI-based impact condition classification and damping matching control method as described in the first aspect embodiment.

[0019] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements an AI-based shock condition classification and damping matching control method as described in the first aspect embodiment.

[0020] The embodiments of the present invention have the following beneficial effects: (1) By introducing visual perception information, the type of impact and the energy level are identified before the wheel comes into contact with the impacting object, and the feedforward pre-adjustment of the damping force is realized. This avoids the inherent delay of "adjusting after impact" in the traditional feedback control mode from the mechanism perspective, and significantly improves the response time of the suspension system to sudden impacts on the road surface.

[0021] (2) By classifying the impact conditions in detail, and matching high impedance strategy, synchronous compression strategy and energy dissipation strategy for single-sided impact, double-sided impact and continuous impact respectively, the damping control logic is adapted to the physical characteristics of the impact excitation. It can provide high damping force support to resist large compression of the suspension under single-sided large impact conditions, and maintain low compression damping and moderate rebound damping to maintain a soft road feel transmission under continuous shock conditions, effectively taking into account the comfort requirements under different impact scenarios.

[0022] (3) By accurately identifying continuous impact conditions through the neural network model, the high-frequency reciprocating switching of the damping valve caused by the inability to distinguish between single impact and continuous impact in the traditional scheme is avoided. This reduces the unnecessary operation frequency of the solenoid valve and hydraulic components, reduces valve body wear and energy loss, and improves the working durability and reliability of the system under harsh road conditions.

[0023] (4) A complete feedforward control link of "perception-classification-decision-execution" was constructed. The data flow between each level is clear and the timing is closely coordinated, ensuring an efficient closed loop from the output of the impact condition label to the damping force response. This provides an integrated control scheme for the semi-active suspension system that combines intelligent perception and differentiated execution capabilities. Attached Figure Description

[0024] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart of an AI-based impact condition classification and damping matching control method provided for an embodiment of the present invention; Figure 2 This is a structural diagram of an AI-based impact condition classification and damping matching control device provided in an embodiment of the present invention. Detailed Implementation

[0025] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] The following description, with reference to the accompanying drawings, describes an AI-based method and apparatus for impact condition classification and damping matching control according to an embodiment of the present invention.

[0028] Example 1 This invention provides an AI-based method for classifying impact conditions and matching damping control. Figure 1 This is a flowchart illustrating an AI-based impact condition classification and damping matching control method provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps: Step S1: Acquire images of the road surface in front of the vehicle, as well as the vertical vibration acceleration and wheel speed signals of the vehicle body at the corresponding positions of each wheel.

[0029] It should be understood that the signal acquisition in this embodiment is independently completed by the system perception layer. The perception layer provides all the raw input data for the entire control method and is the foundation for subsequent intelligent recognition and precise control. The perception layer hardware includes a forward-looking camera, four acceleration sensors, and four wheel speed sensors. Each sensor is installed according to preset position specifications to ensure the accuracy and relevance of the acquired data.

[0030] It should be noted that the forward-facing camera is fixedly installed inside the vehicle's windshield. Its core function is to collect real-time images of the road surface in front of the vehicle's direction of travel. The collected road surface images fully contain information on various impact objects on the road ahead, providing data support for subsequent image spatial feature extraction.

[0031] To further explain, the four acceleration sensors are installed at the suspension mounting points near the four wheels, and can independently collect the vertical vibration acceleration of the vehicle body at the corresponding position of each wheel. This vertical vibration acceleration parameter can accurately characterize the vibration response state of the wheel when it is impacted by the road surface, which is transmitted to the vehicle body through the suspension structure. It is the core dynamic data for judging the intensity of the road impact.

[0032] It should be noted that the four wheel speed sensors are installed one by one at the wheel hub of each of the four wheels to collect wheel speed signals in real time. These signals can not only help determine the vehicle's current real-time speed, but also, in combination with acceleration data, predict the magnitude of road impact energy that the wheels will experience, effectively improving the accuracy of subsequent impact energy identification.

[0033] Step S2: Spatial features are extracted from the road surface image by the embedded neural network processing unit, and time series features are extracted from the vertical vibration acceleration and wheel speed signals. The impact type and impact energy value are output according to the extracted spatial features and time series features, and the two are combined into a working condition label.

[0034] It should be understood that the signal processing and feature recognition work in this step is completed by the classification layer, which is integrated inside the vehicle chassis domain controller. The core computing carrier is an embedded neural network processing unit. This processing unit adopts a hybrid architecture of convolutional neural network and long short-term memory network in series, which can realize the synchronous parsing and feature fusion of multi-dimensional data.

[0035] It should be noted that the convolutional neural network (CNN) model and the long short-term memory (LSTM) network model have a clear division of labor and work together to adapt to the feature extraction needs of different types of collected data. The CNN model is specifically designed for processing road surface image data. Through pre-defined convolution operation rules, it analyzes the road surface image frame by frame, accurately extracting various spatial morphological features of impact objects in the image. Specifically, this includes key spatial information such as the object's outline size, shape, and relative distance to vehicles, providing image-based dimensional support for determining the impact type.

[0036] To further explain, the Long Short-Term Memory (LSTM) network model is mainly used to process dynamically changing time-series signals. It performs comprehensive time-series analysis on the collected vehicle vertical vibration acceleration signals and wheel speed signals, accurately extracting the time-series change patterns and dynamic response characteristics of the two types of signals during continuous driving. These time-series features can intuitively characterize the dynamic response trend of the wheel under different road surface impacts, providing time-series data support for the quantitative determination of impact energy.

[0037] It is worth mentioning that, in this embodiment, the embedded neural network processing unit deeply integrates the spatial features output by the convolutional neural network with the time series features output by the long short-term memory network. The feature fusion algorithm built into the model completes the matching and integration of the two types of features, and finally outputs accurate impact type and impact energy value respectively.

[0038] For example, the impact types identified in this application embodiment specifically include three standard working conditions: unilateral impact, bilateral impact, and continuous impact. The impact energy value is set to a quantitative value range of 0 to 100, and is divided into clear energy levels: 0 to 30 corresponds to low energy impact, 31 to 70 corresponds to medium energy impact, and 71 to 100 corresponds to high energy impact. Finally, the identified impact type and corresponding energy value are combined and encapsulated into a unified working condition label, which is transmitted to the subsequent decision-making layer to complete the control judgment.

[0039] It should be further noted that the hybrid neural network model used in this application has been trained and optimized with massive amounts of real vehicle data. The model training data comes from 100,000 kilometers of real vehicle road test data, covering various typical impact conditions such as speed bumps with a height of 2cm to 10cm, road potholes with a depth of 2cm to 8cm, gravel roads with a particle size of 1cm to 5cm, and Belgian roads. It is adapted to driving scenarios with different vehicle speeds and different road surfaces. The accuracy of the condition classification in the model output layer is no less than 95%, which fully ensures the reliability and stability of condition recognition.

[0040] Step S3: Determine the corresponding damping control strategy based on the impact type in the working condition label, and match the corresponding target damping parameter under the damping control strategy based on the impact energy value.

[0041] It should be understood that the strategy determination and parameter matching in this step are completed by the decision-making layer, which is also integrated into the chassis domain controller. It has multiple preset damping control maps, and each map corresponds one-to-one with the combination of impact type and impact energy level, which can realize the rapid and accurate matching of operating condition labels and control parameters.

[0042] It should be noted that the embodiments of this application are configured with dedicated damping control strategies for three types of impact conditions. The control logic and damping parameter range of different strategies are set differently to accurately adapt to the control requirements of different road impacts.

[0043] To further explain, when the impact type identified by the working condition label is a unilateral impact, the decision layer activates a high-impedance strategy. The core control logic is to specifically control the unilateral shock absorber corresponding to the impact. Before the road impact arrives, the high-impedance adjustment action is performed in advance. By increasing the damping force of the unilateral suspension to resist the limit compression stroke of the suspension, the transmission efficiency of the road impact to the vehicle body is effectively reduced, and the body sway caused by unilateral bumps is reduced.

[0044] It should be added that when the impact type identified by the working condition label is a bilateral impact, the decision-making level activates the synchronous compression strategy, and simultaneously adjusts the damping of the front and rear axle shock absorbers of the vehicle. By precisely adjusting the damping ratio parameters of the front and rear axles, the vehicle maintains a level posture when passing through impact road surfaces such as speed bumps and bilateral potholes, completely avoiding the excessive bouncing problem caused by the lag in the rear axle damping response, and improving the overall vehicle driving stability.

[0045] It is worth mentioning that when the impact type identified by the working condition label is continuous impact, the decision-making level activates the energy dissipation strategy, uniformly controls the four-wheel shock absorbers to perform damping adjustment synchronously, and continuously attenuates the continuous vibration energy brought by the road surface by expanding the energy absorption range of the shock absorber's compression stroke and rebound stroke. At the same time, it precisely controls the vibration recovery speed of the vehicle body to avoid continuous shaking of the vehicle body caused by continuous bumps.

[0046] Optionally, after determining the corresponding damping control strategy, the decision-maker retrieves a pre-matched damping control map based on the impact energy level in the operating condition label, accurately matching the corresponding target compression damping coefficient and target rebound damping coefficient. The specific parameter adaptation logic is as follows: combining single-sided, double-sided, and continuous impact types with low, medium, and high impact energy levels, specific map parameters are retrieved. For example, when a single-sided impact is combined with a high-energy operating condition, target parameters of 100% compression damping coefficient and 80% rebound damping coefficient are matched. All damping parameters have been calibrated and optimized on actual vehicles to adapt to the control requirements of various operating conditions.

[0047] Step S4: Before the wheel contacts the impacting object, a current command is generated according to the target damping parameter and sent to the solenoid valve driver of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, thus completing the damping feedforward control before the impact.

[0048] It should be understood that this step is the final execution stage of the entire control process. The corresponding execution layer hardware includes four continuously damped control shock absorbers. Each shock absorber integrates a controllable solenoid valve. The damping coefficient of the shock absorber can be changed by precisely adjusting the working current of the solenoid valve. Moreover, the correspondence between the current magnitude and the damping coefficient is pre-calibrated on the actual vehicle and stored in the decision layer to ensure the accuracy of parameter conversion.

[0049] It should be noted that the embodiments of this application adopt feedforward control logic, and rely on precise control timing to complete advance adjustment throughout the process. The moment when the camera captures the impact object on the road ahead is defined as time T0, and the moment when the wheel begins to contact the impact object on the road is defined as time T1. The time interval between T0 and T1 is determined by the real-time vehicle speed and the camera's pre-aiming distance, with a typical range of 100ms to 500ms, which provides sufficient response time for the pre-damping adjustment.

[0050] To further explain, the overall control timing has strict time logic: at time T0, the perception layer starts data acquisition and transmits it to the classification layer in real time. Within 50ms to 150ms after time T0, the classification layer completes all neural network inference operations and outputs complete operating condition labels to the decision layer. After receiving the operating condition labels, the decision layer can complete the Map lookup table matching and target damping parameter confirmation within 20ms to 50ms, and quickly convert it into the corresponding solenoid valve current command.

[0051] It should be further explained that the current command generated by the decision-making level is transmitted in real time to the solenoid valve driver of the corresponding shock absorber via the vehicle CAN bus. The driver precisely adjusts the opening of the solenoid valve core according to the received current command, thereby changing the effective cross-sectional area of ​​the damping channel inside the shock absorber. This achieves dynamic adjustment of the actual damping coefficient of the shock absorber, so that the actual damping parameters accurately match the target damping parameters.

[0052] It is worth mentioning that all damping adjustment actions are completed before time T1. Specifically, the opening of all solenoid valves and the damping coefficient are calibrated 20ms to 100ms before the wheel contacts the impact object. This ensures that the shock absorber has the preset target damping force at the moment the wheel contacts the impact object, thus fully realizing the damping feedforward control before the road impact arrives and suppressing the vibration impact of the road impact on the vehicle body from the source.

[0053] For example, by combining the parameter standards of different control strategies, the damping coefficient of the corresponding side shock absorber can be instantly adjusted to 80% to 100% of the maximum value under the high impedance strategy, maximizing the resistance to suspension compression shock; under the energy dissipation strategy, the shock absorber compression damping is maintained at 30% to 50% of the calibrated value and the rebound damping is maintained at 50% to 70% of the calibrated value, continuously dissipating continuous vibration energy and comprehensively optimizing the smoothness of the vehicle driving on complex road surfaces.

[0054] To facilitate the explanation of the technical solution of the present invention, the following description is based on three specific application embodiments.

[0055] First implementation method: When a vehicle enters a gravel road surface, it is subjected to a typical continuous impact driving condition. It should be understood that gravel roads have dense and irregular small bumps and depressions, which generate continuous, low-amplitude road impacts on the wheels, easily causing continuous vehicle vibration and affecting ride smoothness. In this embodiment, the vehicle's forward-facing camera captures real-time images of the road surface ahead, and the vision system identifies the road texture density in real-time. When the detected road texture density is greater than 10 per square meter, the current road surface is determined to be a high-frequency disturbed gravel road surface. It should be noted that the perception layer simultaneously collects four-wheel wheel speed signals and vehicle vertical vibration acceleration data in real-time, and transmits all perception data to the classification layer inside the chassis domain controller via an internal bus. Furthermore, the embedded neural network processing unit inside the classification layer combines the spatial features of the road image with temporal vibration and wheel speed features for fusion inference, stably outputting a continuous impact condition label, and simultaneously outputting an impact energy value of 65, corresponding to the medium-energy impact level preset in this application. It should be further explained that after receiving the operating condition label, the decision-making level automatically retrieves the energy dissipation strategy damping control map that matches the continuous impact and medium energy level. Based on the parameters specified in the map, the compression damping of the vehicle's four-wheel shock absorbers is uniformly set to 40% of the factory calibration value, and the rebound damping is uniformly set to 60% of the factory calibration value. It is worth mentioning that this damping parameter ratio can appropriately relax the suspension motion constraints, allowing the wheels to move up and down rapidly following the undulations of the gravel road surface. Through the dynamic travel of the suspension, the continuous vibration energy of the road surface is fully absorbed, while the rebound speed of the vehicle body is reasonably constrained, effectively suppressing continuous body shaking. Ultimately, this ensures that the overall vehicle body posture remains stable, significantly improving the comfort and stability of driving on gravel roads.

[0056] Second implementation method: The vehicle is traveling normally at 60 km / h when a 5cm deep pothole appears in front of one of its wheels, indicating an imminent high-amplitude impact. It should be understood that when one wheel rapidly enters a deep pothole, the suspension will experience a sudden and significant drop, easily triggering a severe impact that could strike the suspension's buffer structure, resulting in noticeable bumps, noises, and a strong impact sensation. In this embodiment, the perception layer uses a forward-facing camera to pre-aim at and identify the pothole, simultaneously collecting real-time wheel speed and vehicle acceleration data and uploading it to the classification layer. The classification layer uses a neural network model to quickly analyze spatial and temporal features, accurately outputting a unilateral impact condition label and identifying the current impact energy value as 85, corresponding to the high-energy impact level preset in this application. Furthermore, after receiving the condition label, the decision layer quickly completes strategy matching, activating a high-impedance control strategy. Combined with the system's preset control timing logic, all damping adjustments are completed 20ms before the wheel physically contacts the edge of the pothole. It should be further explained that, based on the pre-calibrated relationship between current and damping coefficient, the decision-makers adjusted the control current of the compression damping valve of the target-side shock absorber to its maximum value, instantly adjusting the damping coefficient of that side shock absorber to 95% of its maximum calibration value. It is worth noting that this instantaneous high damping force effectively restrains the suspension's instantaneous downward tendency, resists the suspension's extreme compression stroke, significantly reduces the transmission of road impact energy to the vehicle side, and effectively protects the suspension buffer blocks from damage caused by instantaneous impacts. In actual driving, the driver only feels slight road feedback, completely avoiding the severe impact sensation caused by deep potholes under traditional suspension control methods, and significantly improving driving stability under severe unilateral impact conditions.

[0057] Third implementation method: A vehicle traveling at a low speed of 30 km / h over a speed bump is a typical bilateral symmetrical impact condition. It should be understood that traditional passive suspension vehicles commonly experience asynchronous damping responses and inconsistent compression strokes between the front and rear axles when crossing speed bumps, easily leading to rear axle lag, body pitch, and noticeable secondary bumps. In this embodiment, the perception layer performs real-time image recognition of the speed bump ahead and acquires dynamic signals from all four wheels. The classification layer, after neural network inference calculations, stably outputs a bilateral impact condition label, identifying an impact energy value of 50, corresponding to the medium-energy impact level preset in this application. It should be noted that the decision layer precisely invokes the damping control Map corresponding to the synchronous compression control strategy for bilateral medium-energy impact conditions. Furthermore, the controller, based on preset calibration ratio parameters, sets the compression damping of the front axle shock absorber to 70% of the calibrated value and the compression damping of the rear axle shock absorber to 50% of the calibrated value. By precisely adjusting the front and rear axle damping ratio through differentiation, synchronous linkage of the front and rear axle suspension compression strokes is achieved. It should be further explained that this control method ensures that the suspension undulation rhythm is highly uniform as the front and rear axles pass over speed bumps in sequence, effectively suppressing vehicle pitch deviation and allowing the entire vehicle to maintain a level posture and smoothly pass over impact sections. It is worth mentioning that this implementation method completely solves the drawbacks of traditional vehicles' delayed rear axle rebound and excessive bouncing when passing over speed bumps, eliminating the discomfort caused by secondary bounces and significantly improving the overall vehicle smoothness when crossing bumps.

[0058] Example 2 This invention provides an AI-based impact condition classification and damping matching control system. The system adopts a hierarchical architecture design, which consists of four functional layers from top to bottom: perception layer, classification layer, decision layer, and execution layer. Each layer has a clear division of labor and a closed-loop data interaction logic. It can automatically identify and accurately classify road impact conditions and adaptively match and control the shock absorption damping by relying on intelligent neural network algorithms, effectively improving the ride comfort of vehicles on complex road surfaces.

[0059] It should be understood that the control system designed in this application is a four-layer integrated hardware and software collaborative architecture. The hardware and software at each layer cooperate with each other to complete the entire process of signal acquisition, feature analysis, working condition determination, strategy decision-making and damping execution in sequence. The specific composition and function of each layer are as follows.

[0060] The perception layer, as the system's data input source, is responsible for comprehensively collecting road surface image information and wheel dynamic response data during vehicle operation, providing raw data support for backend intelligent analysis. In this embodiment, the perception layer mainly includes a forward-facing camera, four sets of acceleration sensors, and four sets of wheel speed sensors. The forward-facing camera is fixedly installed inside the vehicle's windshield and can collect real-time images of the road surface in the vehicle's forward direction, clearly capturing the shape and position information of various impact objects on the road. It should be noted that the four acceleration sensors are respectively installed at the suspension mounting points near the four wheels, and can independently collect the vertical vibration acceleration of the vehicle body at the corresponding position of each wheel, accurately reflecting the vehicle body vibration state after the suspension is impacted by the road surface. Furthermore, the four wheel speed sensors are each mounted at one of the four wheel hubs, collecting the wheel speed signals of each wheel in real time. This can not only be used to determine the vehicle's real-time driving speed, but also, combined with acceleration data, to help predict the level of road impact energy that the wheels will soon experience.

[0061] The classification layer, integrated within the vehicle chassis domain controller, is the core computational unit for intelligent identification and condition classification. In this embodiment, the core hardware of the classification layer is an embedded neural network processing unit. This unit employs a hybrid architecture combining a convolutional neural network module and a long short-term memory network module, with each module adapting to the feature extraction needs of different data types. Notably, the convolutional neural network module specifically processes road image data collected by the perception layer to accurately extract spatial features such as the contour, size, and location of road impact objects. The long short-term memory network module focuses on processing temporal dynamic data, performing time-series analysis on vehicle vertical vibration acceleration signals and wheel speed signals to extract dynamic feature patterns of signal changes over time. It should be further noted that the overall configuration of the classification layer is to receive all collected data transmitted from the perception layer in real time. After feature fusion and intelligent inference through the dual network modules, it ultimately outputs standardized impact condition labels. The working condition label specifically includes two core parameters: impact type and impact energy value. The impact type covers three typical road working conditions: unilateral impact, bilateral impact, and continuous impact. The impact energy value is set to a continuous range of 0 to 100 and is divided into three energy levels: 0 to 30 for low energy impact, 31 to 70 for medium energy impact, and 71 to 100 for high energy impact, providing a standardized judgment basis for subsequent damping strategy matching.

[0062] The decision-making layer is also integrated within the chassis domain controller, undertaking the core functions of condition analysis, strategy retrieval, and parameter matching. It should be understood that the decision-making layer pre-stores multiple damping control maps, each corresponding one-to-one with combinations of different impact types and energy levels. After being calibrated and solidified on actual vehicles, it enables precise and rapid matching of condition labels and damping parameters. In this embodiment, the core working logic of the decision-making layer is to receive the condition labels output by the classification layer, identify the specific type and energy level of the current road impact, retrieve the corresponding preset damping control map, execute the adapted dedicated damping control strategy, and complete the precise matching of the target damping parameters.

[0063] The execution layer is the system's terminal actuator, responsible for receiving control commands from the decision-making layer and performing dynamic damping adjustment. In this embodiment, the execution layer includes four continuously damped control shock absorbers, corresponding to the four wheels of the vehicle, each with a built-in controllable solenoid valve. It should be noted that the damping coefficient of the shock absorber and the operating current of the solenoid valve have a preset correspondence. This correspondence is pre-stored within the decision-making layer after extensive calibration on actual vehicles. The decision-making layer can dynamically change the actual damping coefficient of the shock absorber by precisely adjusting the solenoid valve current, ultimately achieving adaptive damping control under different operating conditions.

[0064] It should be understood that this application designs exclusive damping control strategies for three typical road impact conditions, with each strategy parameter range being precisely controllable to adapt to the vibration reduction requirements of different impact scenarios. The specific control logic is as follows.

[0065] It should be noted that when a unilateral impact condition is identified, the decision-making level implements a high-impedance strategy, intervening and adjusting the damping coefficient of the corresponding shock absorber to 80% to 100% of its maximum damping value before the road impact contacts the wheel. By constraining the suspension compression travel limit with instantaneous high damping force, the system effectively resists the violent suspension compression caused by a unilateral road impact, significantly reducing the transmission of impact energy to the vehicle body and avoiding body swaying and impact noises caused by unilateral bumps.

[0066] To further explain, when a continuous impact condition is identified, the decision-making level activates an energy dissipation strategy, uniformly controlling the synchronous adjustment of damping parameters of the four-wheel shock absorbers. This maintains the shock absorber compression damping stably at 30% to 50% of the rated value, while simultaneously maintaining the rebound damping at 50% to 70% of the rated value. This parameter configuration can appropriately relax the suspension motion restrictions, continuously absorbing the energy of continuous road vibrations by increasing the dynamic motion travel of the suspension, while precisely constraining the rebound speed of the vehicle body, avoiding continuous vehicle body shaking caused by continuous minor impacts, and stabilizing the vehicle's driving posture.

[0067] It should be further explained that when a bilateral impact condition is identified, the decision-making team adopts a synchronous compression strategy, simultaneously adjusting the damping ratio of the front and rear axle shock absorbers. By optimizing the damping coefficient ratio of the front and rear axle shock absorbers, the compression stroke of the vehicle's front and rear suspensions is kept synchronized with the movement rhythm. This ensures that when the vehicle passes over speed bumps or other bilaterally symmetrical impact surfaces, the vehicle body maintains a level posture at all times, completely avoiding the problems of excessive rear axle bounce and body pitch sway caused by the lag in response of traditional suspensions.

[0068] It is worth mentioning that this application adopts feedforward closed-loop control logic, which has a complete data transmission link and strict timing control logic, and can realize early prediction and pre-adjustment before the impact arrives. The specific data flow and control timing are as follows.

[0069] In terms of data flow, the perception layer collects real-time images of the road surface ahead, four-channel vertical acceleration signals of the vehicle body, and four-channel wheel speed signals, which are then transmitted in real-time to the classification layer inside the chassis domain controller via the vehicle's internal bus. The classification layer relies on an embedded neural network processing unit to perform multi-dimensional data analysis. It performs convolutional operations on the road surface images using a convolutional neural network to extract the spatial features of the impact object and output the corresponding impact type label. Simultaneously, it analyzes and processes the acceleration and wheel speed time-series signals using a long short-term memory network, quantifying and outputting the corresponding impact energy value. The classification layer packages and integrates the identified impact type and impact energy value into a complete operating condition label, which is then transmitted to the decision layer via the internal bus. Upon receiving the operating condition label, the decision layer retrieves the uniquely corresponding preset damping control map based on the combination of impact type and energy level. It quickly looks up and matches the target compression damping coefficient and target rebound damping coefficient to obtain accurate values. For example, a single-sided impact combined with a high-energy operating condition can match target parameters of 100% compression damping coefficient and 80% rebound damping coefficient. Based on the pre-calibrated current-damping relationship, the decision-making level converts the target damping parameters into precise solenoid valve control current commands, which are then sent to the solenoid valve driver of the corresponding shock absorber via the vehicle CAN bus. By adjusting the valve core opening, the damping channel cross-sectional area is changed, ultimately causing the shock absorber to output the target damping force, thus completing a single closed-loop control.

[0070] Regarding the control timing, this application relies on the camera's pre-aiming function to achieve advanced control. A unified timeline is used to define the system's operating timing: the moment the forward-looking camera first captures the impact object ahead is defined as time T0, and the moment the wheel actually contacts the impact object on the road is defined as time T1. It should be noted that the time interval between time T0 and time T1 is determined by both the vehicle's real-time speed and the camera's pre-aiming distance, with a typical interval range of 100ms to 500ms, allowing sufficient response time for system pre-adjustment. To further explain, at time T0, the perception layer synchronously initiates full-dimensional data acquisition and uploads the acquired data to the classification layer in real time. Within a time window of 50ms to 150ms after time T0, the classification layer quickly completes all neural network inference operations, outputs complete operating condition labels, and transmits them to the decision layer. After receiving the operating condition labels, the decision layer can complete the Map lookup, target damping parameter matching, and current command conversion within 20ms to 50ms, and then issue control commands to the execution layer. Within 20ms to 100ms before time T1, the execution layer completes all solenoid valve current adjustments and damping coefficient calibration to ensure that the shock absorber has accurately reached the target damping value at the moment the wheel contacts the impact, thus fully realizing damping feedforward control based on road surface pre-aiming.

[0071] It should be understood that the neural network model used for condition identification in this application adopts a hybrid architecture combining convolutional neural networks and long short-term memory networks, taking into account both image spatial feature extraction and temporal signal feature analysis capabilities. The model's identification accuracy and condition adaptability have been iteratively optimized through massive amounts of real vehicle data. It should be noted that the training dataset for this model comes from 100,000 kilometers of full-coverage real-vehicle road test data, covering various typical adverse road surfaces and driving conditions, specifically including speed bumps with heights ranging from 2cm to 10cm, potholes with depths ranging from 2cm to 8cm, gravel roads with particle sizes ranging from 1cm to 5cm, and various impact road surface conditions such as Belgian roads at different driving speeds. Furthermore, after training and iteration with the aforementioned massive amounts of real-vehicle data across multiple scenarios and road conditions, the output layer of the neural network model in this application achieves a classification accuracy of no less than 95% for various impact conditions, ensuring the stability, accuracy, and real-time performance of condition identification during vehicle operation, and providing reliable algorithmic support for subsequent accurate matching of damping strategies.

[0072] Example 3 This invention provides an AI-based impact condition classification and damping matching control device. Figure 2 This is a schematic diagram of an AI-based impact condition classification and damping matching control device provided in an embodiment of the present invention. Figure 2 As shown, the device includes: The multi-source sensing and acquisition module 100 is used to acquire images of the road surface in front of the vehicle, as well as vertical vibration acceleration and wheel speed signals of the vehicle body at the corresponding positions of each wheel. The feature extraction and classification module 200 is used to extract spatial features from the road surface image through an embedded neural network processing unit, and to extract time series features from the vertical vibration acceleration and wheel speed signals. Based on the extracted spatial features and time series features, it outputs the impact type and impact energy value respectively, and combines the two into a working condition label. The strategy matching parameter module 300 is used to determine the corresponding damping control strategy based on the impact type in the working condition label, and to match the corresponding target damping parameter under the damping control strategy based on the impact energy value. The instruction generation and execution module 400 is used to generate a current instruction based on the target damping parameter before the wheel contacts the impacting object, and send it to the solenoid valve driver of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, thereby completing the damping feedforward control before the impact.

[0073] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0074] Example 4 To implement the methods of the above embodiments, the present invention also provides an electronic device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the methods described above.

[0075] Example 5 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0077] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0078] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. An AI-based method for classifying impact conditions and matching damping control, characterized in that, include: Acquire images of the road surface in front of the vehicle, as well as vertical vibration acceleration and wheel speed signals of the vehicle body at the corresponding positions of each wheel; The embedded neural network processing unit extracts spatial features from the road surface image and extracts time-series features from the vertical vibration acceleration and wheel speed signals. Based on the extracted spatial and time-series features, the impact type and impact energy value are output respectively, and the two are combined into a working condition label. Based on the impact type in the working condition label, determine the corresponding damping control strategy, and match the corresponding target damping parameter under the damping control strategy based on the impact energy value. Before the wheel contacts the impacting object, a current command is generated based on the target damping parameters and sent to the solenoid valve driver of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, thus completing the damping feedforward control before the impact.

2. The method according to claim 1, characterized in that, Acquire images of the road surface in front of the vehicle, as well as vertical vibration acceleration and wheel speed signals of the vehicle body at corresponding positions of each wheel, including: A forward-facing camera installed inside the vehicle's windshield captures images of the road surface ahead of the vehicle's direction of travel, including information on impact objects on the road ahead. Four acceleration sensors, each installed at a suspension mounting point near one of the four wheels, are used to collect the vertical vibration acceleration of the vehicle body at the corresponding position of each wheel. The vertical vibration acceleration is used to characterize the vibration response of the wheel to the vehicle body transmitted through the suspension when the wheel is impacted by the road surface. Four wheel speed sensors, each installed at the hub of one of the four wheels, collect wheel speed signals from each wheel. These wheel speed signals are used to help determine the vehicle's current speed and the impact energy that the wheel will soon experience.

3. The method according to claim 1, characterized in that, Spatial features are extracted from the road surface image using an embedded neural network processing unit, and time-series features are extracted from the vertical vibration acceleration and wheel speed signals, including: The convolutional neural network model in the embedded neural network processing unit performs convolution operations on the road surface image to extract the spatial morphological features of the impactor in the road surface image, wherein the spatial morphological features include the outline, size and position information of the impactor relative to the vehicle. The long short-term memory network model in the embedded neural network processing unit is used to perform time series analysis on the vertical vibration acceleration and the wheel speed signal, and extract the time series variation characteristics of the vertical vibration acceleration and the wheel speed signal. The time series characteristics are used to characterize the dynamic response trend of the wheel under the impact of the road surface.

4. The method according to claim 1, characterized in that, Based on the extracted spatial and temporal features, the impact type and impact energy value are output respectively, and these two are combined into a working condition label, including: The embedded neural network processing unit is constructed by connecting a convolutional neural network model and a long short-term memory network model in series. The spatial features and the time series features are respectively used as the outputs of the two models. After the embedded neural network processing unit fuses the two types of features, the impact type and the impact energy value are generated respectively. The generated impact type and impact energy value are combined together to form the working condition label.

5. The method according to claim 1, characterized in that, Based on the impact type in the operating condition label, a corresponding damping control strategy is determined, and based on the impact energy value, a corresponding target damping parameter is matched under the damping control strategy, including: When the impact type is a unilateral impact, a high impedance strategy is adopted to control the unilateral shock absorber corresponding to the impact type to perform a high impedance action before the impact arrives, so as to resist suspension compression and reduce the transmission of impact to the vehicle body. When the impact type is a bilateral impact, a synchronous compression strategy is adopted, the front and rear axle shock absorbers are synchronously controlled to perform synchronous compression actions, and the vehicle body is kept in a horizontal posture when passing through the impact object by adjusting the front and rear axle damping ratio. When the impact type is continuous impact, an energy dissipation strategy is adopted to control the four-wheel shock absorbers to perform energy dissipation actions simultaneously, thereby attenuating continuous vibration by increasing the energy absorption in the compression stroke and rebound stroke of the shock absorbers. Based on the energy level to which the impact energy value belongs and in combination with the impact type, the corresponding target compression damping coefficient and target rebound damping coefficient are matched from the preset damping control map. Each map in the preset damping control map corresponds one-to-one with the combination of impact type and impact energy level.

6. The method according to claim 1, characterized in that, Before the wheel contacts the impacting object, a current command is generated based on the target damping parameters and sent to the solenoid valve actuator of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, completing the damping feedforward control before the impact, including: Based on the pre-calibrated correspondence between current and damping coefficient, the target damping parameter is converted into a corresponding current command; The current command is sent to the solenoid valve driver of the corresponding shock absorber via the CAN bus; According to the received current command, each of the solenoid valve drivers adjusts the valve core opening of the corresponding solenoid valve, changes the cross-sectional area of ​​the damping channel inside the shock absorber, and makes the actual damping coefficient of the corresponding shock absorber reach the target damping value corresponding to the target damping parameter. Taking the moment when the front-view camera captures an image of the road surface containing an impact object as the starting moment, before the wheel contacts the impact object, the matching of the target damping parameters, the generation and transmission of the current command, and the adjustment of the solenoid valve opening are completed sequentially, so that the corresponding shock absorber has the target damping value at the moment the wheel contacts the impact object, thus completing the damping feedforward control before the impact.

7. An AI-based impact condition classification and damping matching control device, characterized in that, include: The multi-source sensing and acquisition module is used to acquire images of the road surface in front of the vehicle, as well as the vertical vibration acceleration and wheel speed signals of the vehicle body at the corresponding positions of each wheel; The feature extraction and classification module is used to extract spatial features from the road surface image through an embedded neural network processing unit, and to extract time series features from the vertical vibration acceleration and wheel speed signals. Based on the extracted spatial features and time series features, it outputs the impact type and impact energy value respectively, and combines the two into a working condition label. The strategy matching parameter module is used to determine the corresponding damping control strategy based on the impact type in the working condition label, and to match the corresponding target damping parameter under the damping control strategy based on the impact energy value. The instruction generation and execution module is used to generate a current instruction based on the target damping parameter before the wheel contacts the impacting object, and send it to the solenoid valve driver of the corresponding shock absorber to adjust the damping of the corresponding shock absorber to the target value, thereby completing the damping feedforward control before the impact.

8. The apparatus according to claim 7, characterized in that, The multi-source sensing and acquisition module is also used for: A forward-facing camera installed inside the vehicle's windshield captures images of the road surface ahead of the vehicle's direction of travel, including information on impact objects on the road ahead. Four acceleration sensors, each installed at a suspension mounting point near one of the four wheels, are used to collect the vertical vibration acceleration of the vehicle body at the corresponding position of each wheel. The vertical vibration acceleration is used to characterize the vibration response of the wheel to the vehicle body transmitted through the suspension when the wheel is impacted by the road surface. Four wheel speed sensors, each installed at the hub of one of the four wheels, collect wheel speed signals from each wheel. These wheel speed signals are used to help determine the vehicle's current speed and the impact energy that the wheel will soon experience.

9. An electronic device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method as described in any one of claims 1-6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images