Automobile air suspension automatic control method and system based on multi-modal data recognition

Abstract

The invention relates to the technical field of suspension control, and provides an automobile air suspension automatic control method and system based on multi-modal data recognition, the method is applied to an air suspension, and the method comprises the steps that at the first moment, first data collected by a multi-modal collector of a vehicle is obtained; wherein the first data is target road condition information determined by a multi-mode collector on the next driving road condition of the current vehicle; and according to the first data, a suspension scene at a second moment after the first moment and braking parameters corresponding to the suspension scene are determined. The method is used for automatically recognizing an automatic control scene of the air suspension, automatic braking operation is achieved under the automatic control scene, and a vehicle body is kept stable.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of suspension control, in particular to an automobile air suspension automatic control method and system based on multi-modal data recognition. BACKGROUND

[0002] At present, more and more vehicles are equipped with air suspension systems, which realize vehicle height adjustment by inflating and deflating air springs, so as to balance comfort and passability, and to keep the vehicle body comfortable during driving through automatic control.

[0003] However, the existing air suspension automatic control still mainly relies on manual control, and different scenes such as water crossing scene, bumpy road scene, turning road scene, charging scene, and refueling scene are not distinguished. Therefore, the air suspension may produce misoperation, for example, in the water crossing scene, the control scheme for the road scene is still used, and the air suspension is raised too slowly or is raised after entering the water. SUMMARY

[0004] The present application provides an automobile air suspension automatic control method and system based on multi-modal data recognition, which is used to automatically identify the automatic control scene of the air suspension, and realizes brake operation under the automatic control scene.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme:

[0006] In a first aspect, the present application provides an automobile air suspension automatic control method based on multi-modal data recognition, applied to an air suspension, the method comprising:

[0007] At a first time, first data collected by a multi-modal collector of a vehicle is obtained; wherein the first data is target road condition information of the next driving road condition determined by the multi-modal collector for the current vehicle;

[0008] According to the first data, the suspension scene at a second time after the first time is determined, and the brake parameter corresponding to the suspension scene.

[0009] In combination with the first aspect, the multi-modal collector comprises a configuration configured to collect visual data, sensing data and environmental data;

[0010] The visual data includes a road surface semantic segmentation image of the next driving road condition.

[0011] The sensing data includes vehicle body height data, water level data, acceleration sensing data, and temperature sensing data of the current vehicle in the current driving road condition and the next driving road condition.

[0012] The environmental data includes weather data and driving scene data of the current vehicle.

[0013] In combination with the first aspect, the first time point, according to the environment data, a braking scenario of the air suspension under a current driving road condition is determined;

[0014] And, according to the visual data, target road condition information of the air suspension under the braking scenario is determined;

[0015] And, according to the sensing data, a braking behavior interval of the air suspension under the target road condition information is determined; wherein, the braking behavior interval includes a vehicle body height interval, a suspension stiffness interval, a damping coefficient interval, and a posture balance interval.

[0016] In combination with the first aspect, the first data collection includes:

[0017] According to the driving scenario data, a braking behavior of a next driving road condition is determined;

[0018] According to the braking behavior and road condition features to be avoided in the road semantic segmentation image, target road condition information conforming to air suspension adjustment on the next driving road condition is determined.

[0019] In combination with the first aspect, the determination of the suspension scenario at the second time point after the first time point according to the first data includes:

[0020] The first data is input into a pre-trained scenario classification model; wherein, the scenario classification model adopts a CNN-LSTM fusion architecture, which is used to extract spatial features of multi-modal data and capture road condition time sequence variation rules;

[0021] According to a scenario probability distribution output by the scenario classification model, the suspension scenario at the second time point is determined.

[0022] In combination with the first aspect, when the suspension scenario at the second time point is determined,

[0023] According to the driving state and road condition features of the current vehicle at the first time point, a second road condition determination is made; wherein,

[0024] When the current vehicle is in a stable driving state and the road condition features show that the road surface is flat, the suspension scenario is divided into a regular comfort scenario;

[0025] If the current vehicle has a steering trend and the road condition features show that the road surface has a lateral inclination, the suspension scenario is adjusted to a steering stability scenario;

[0026] If the first data shows that there is a sudden obstacle in front and the vehicle has a deceleration intention, the suspension scenario is switched to an emergency buffer scenario.

[0027] In combination with the first aspect, after the suspension scenario at the second time point is determined, the method further includes:

[0028] Match and verify the real-time road condition information of the current vehicle at the second time with the target road condition information in the first data; wherein,

[0029] If the matching degree reaches the preset standard, maintain the original suspension scene and the corresponding brake parameter;

[0030] If the matching degree is lower than the standard, correct the suspension scene according to the real-time road condition information, and adjust the brake parameter synchronously.

[0031] If the road condition has a frequent change trend, the verification interval is shortened.

[0032] In combination with the first aspect, the determination of the brake parameter corresponding to the suspension scene comprises:

[0033] Extract the second road condition feature directly associated with the suspension scene in the first data, and perform parameter matching with the expected driving posture of the vehicle in the suspension scene; wherein,

[0034] According to the parameter matching, associate the change rate of the second road condition feature with the duration of the suspension scene to determine the brake parameter.

[0035] In combination with the first aspect, after the determination of the brake parameter corresponding to the suspension scene, the method further comprises:

[0036] Simulate the driving state of the current vehicle entering the suspension scene at the second time, and detect whether the brake parameter will cause abnormal stress of the air suspension component or imbalance of the vehicle body posture;

[0037] If the test result shows that a parameter setting in the brake parameter causes excessive stretching of the suspension spring or overloading of the shock absorber, fine-tune the brake parameter;

[0038] If the parameter setting makes the vehicle body stable in the simulation scene and the component stress is within a safe range, confirm that the brake parameter is effective.

[0039] In a second aspect, the application provides an automatic control system for an automobile air suspension based on multi-modal data recognition, comprising:

[0040] A multi-modal data acquisition module is configured to acquire first data collected by a multi-modal collector of a vehicle at a first time; wherein the first data is target road condition information determined by the multi-modal collector for the current vehicle on the next driving road condition;

[0041] An automatic control module is configured to determine a suspension scene at a second time after the first time and a brake parameter corresponding to the suspension scene according to the first data.

[0042] The application has the following beneficial effects:

[0043] The application acquires the information of the next driving road condition in advance through a multi-modal collector during the braking process of the electric vehicle, completes data acquisition and scene analysis before the vehicle drives to the target road condition, and then sets the air suspension braking parameters of the electric vehicle on the target road condition in advance through the cooperation of the air suspension and the braking system, thereby improving the comfort of the driver and the passengers.

[0044] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application. The objects and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims.

[0045] The technical solutions of the present application will be further described in detail below with the help of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0046] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate the present application, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation on the present application.

[0047] In the drawings:

[0048] Figure 1 The method flow chart of the automobile air suspension automatic control method based on multi-modal data recognition in the embodiment of the present application is shown in the figure.

[0049] Figure 2 The system composition diagram of the automobile air suspension automatic control method system based on multi-modal data recognition in the embodiment of the present application is shown in the figure.

[0050] Figure 3 The control system schematic diagram of the electric vehicle in the embodiment of the present application is shown in the figure.

[0051] Figure 4 The target road condition information determination process diagram in the embodiment of the present application is shown in the figure.

[0052] Figure 5 The suspension scene adjustment determination diagram in the embodiment of the present application is shown in the figure.

[0053] Figure 6 The braking defect detection flow chart in the embodiment of the present application is shown in the figure.

[0054] Reference signs:

[0055] 10 is an electric vehicle, 100 is a sensor system, 110 is a visual perception subsystem, 120 is a sensor network, 130 is an environment perception subsystem, 200 is an air suspension, 210 is an air spring electromagnetic valve, 220 is a damping adjuster, 230 is a guide mechanism, 240 is a double-fan air compressor, 300 is a vehicle control center, 310 is an ECU, 320 is an ESP / ABS, 330 is a vehicle-mounted computing platform, 400 is a bus, 410 is a CANFD, 420 is an Ethernet, 430 is a LIN bus, 500 is a positioning and navigation system, 510 is Beidou, 520 is GPS, 530 is an IMU, and 600 is a power management system. DETAILED DESCRIPTION

[0056] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, which should be understood as merely illustrative and explanatory, and not as limiting the present application.

[0057] In the process of driving an electric vehicle, in order to ensure that the driver and passengers have a good driving experience, the existing air suspension 200 has good handling performance when driving on standard roads during braking. However, for road sections with dense gravel, road sections with continuous deceleration strips, road sections with high-speed turns, or road sections with sudden changes to steep slopes or deep pits, the traditional air handling method has certain shortcomings.

[0058] In road sections with dense gravel:

[0059] The air suspension 200 can determine the vibration of the vehicle in real time through a height sensor and an accelerometer, and continuously adjust the adjustable damping system through a CCD to achieve dynamic adaptation of the damping force, reduce the continuous bumping intensity of the electric vehicle, or use the pre-raised suspension height used in high-end cars.

[0060] However, the continuous adjustment of the adjustable damping system by the CCD has a 0.3-0.5 second lag relative to the actual impact, and cannot avoid or determine the distribution of continuous bumps and bump locations. In addition, the pre-raised suspension height lacks scene specificity for actual bumping sections, and the suspension is adjusted too high or the suspension is not adjusted enough. In the case of too high, the comfort of the ride will deteriorate, there will be obvious shock, and the center of gravity will be offset, which will easily cause the vehicle to slide. In the case of insufficient adjustment, when the air suspension 200 turns, the vehicle body tilts, which can cause the vehicle to drift, reduce the handling, and produce a boat-like rocking phenomenon.

[0061] In road sections with high-speed turns:

[0062] The Air Suspension 200 uses a lateral acceleration sensor to trigger unilateral suspension height adjustment or adds an anti-roll bar to suppress body roll. However, the Air Suspension 200 and braking system are controlled independently. During cornering, the braking force distribution is not coordinated, leading to brake dive and sideslip. Adding an anti-roll bar would increase the weight of the electric vehicle, and the passive adjustment mechanism would be unable to adapt to the dual limitations of high-speed cornering on slippery surfaces.

[0063] On sections of road that suddenly become steep slopes or deep pits:

[0064] The air suspension 200 adjusts the vehicle height based on signals from distance sensors, or through driver input in combination with different modes. However, the sensors can only detect height changes; in complex road conditions with sudden road changes, this can cause the suspension to bottom out or fail to filter out bumps.

[0065] To address the aforementioned issues, this application provides an automatic control method for an automotive air suspension 200 based on multimodal data recognition. During the braking process of an electric vehicle, information about the next road condition is acquired in advance using a multimodal data acquisition device. Before the vehicle reaches the target road condition, data acquisition and scenario analysis are completed. Then, through the cooperation of the air suspension 200 and the braking system, the braking parameters of the air suspension 200 on the target road condition are set in advance, improving the comfort of the driver and passengers.

[0066] The solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0067] Example 1:

[0068] See Figure 3 , Figure 3 This is a schematic diagram of the control system of an electric vehicle 10 provided in an embodiment of this application. Figure 3 As shown, the electric vehicle 10 includes: a sensor system 100, an air suspension 200, a vehicle control system 300, a bus 400, a positioning and navigation system 500, and a power management system 600.

[0069] The electric vehicle 10 collects multi-modal data, i.e., target vehicle condition information, including current driving road condition information and target road condition information, through the sensor system 100 during driving, and transmits the multi-modal data to the vehicle general control 300 through the bus 400. The vehicle general control 300 receives the collected multi-modal data and receives the positioning and navigation data uploaded by the positioning and navigation system 500 (mainly the coordinate data of Beidou 510, GPS 520 and the inertial data of IMU 530). Overall calculation and analysis are performed to determine the suspension scene at the next moment and the brake control parameters of the air suspension 200 at the next moment. Then when driving to the target road condition, the brake control parameters can be adapted to the target road condition through the pre-set brake control parameters.

[0070] It can be understood that the electric vehicle 10 of the present application can be a truck, an off-road vehicle, a passenger bus, etc., but the present application is mainly applied to the electric vehicle 10, which is more in line with the brake rate and efficiency of the present application during brake control.

[0071] Embodiment 2:

[0072] Referring to Figure 1 , Figure 1 The method flowchart of the automatic control method of the automobile air suspension 200 for multi-modal data recognition. The automatic control process of the air suspension 200 of the electric vehicle 10 provided in the embodiments of the present application during driving.

[0073] Step S100: acquiring first data collected by a multi-modal collector of a vehicle at a first moment; wherein the first data is target road condition information of the vehicle at the next driving road condition determined by the multi-modal collector;

[0074] The multi-modal collector belongs to the sensor system 100 and is used to connect the visual perception subsystem 110, the environment perception subsystem 130 and the sensor network 120 of the electric vehicle 10, collect multi-modal data, and determine the first data in the multi-modal data.

[0075] During driving of the electric vehicle, the vehicle-mounted computing platform 330 of the vehicle general control 300 sets the current vehicle driving moment as the first moment, and the road at the future driving time as the road at the second moment, and then acquires the multi-modal data of the vehicle multi-modal collector through the sensor system 100. The multi-modal data includes image data collected by the front camera of the visual perception subsystem 110 of the vehicle and point cloud data of the laser radar of the vehicle, which is used to determine the target road condition information of the subsequent driving road, determine whether the target road condition is a normal road, a bumpy road, an emergency braking slope or an accident road, or a slippery road.

[0076] Step S200: determining the suspension scene at the second moment after the first moment according to the first data, and the brake parameters corresponding to the suspension scene.

[0077] The second time is set by the vehicle-mounted computing platform 330 of the whole vehicle control 300, and the target road condition in which the electric vehicle 10 travels at the second time is calculated and determined, and the wheel driving torque of the electric vehicle 10 in the target road condition at the second time is determined by the ESP.

[0078] The damping coefficient of the damping adjuster 220 of the air suspension 200 corresponding to the suspension scene is set, and the brake coefficient of the guiding mechanism 230 of the air suspension 200 controlling the lifting or lowering is set.

[0079] In an embodiment, the damping coefficient of the air suspension 200 is set to offset the centrifugal force, improve the shock absorption effect, and suppress the head nodding effect of the vehicle body.

[0080] In an embodiment, the brake coefficient of the air suspension 200 is controlled to lift or lower, so as to ensure that the vehicle is always in a balanced state, prevent the vehicle body from rolling and heaving, optimize the braking distance, and balance the grip force, and ensure the driving and riding comfort.

[0081] In an embodiment, the first time and the second time of the present application are repeatedly cycled, and each time is calculated. Therefore, it can be ensured that in the continuous driving process, the brake parameters of the air suspension 200 at the next time are predicted at each time, so as to improve the driving comfort.

[0082] In an embodiment, the front camera is used to collect visual data, and the vehicle side camera can also assist the front camera of the vehicle to collect road surface data. The laser radar of the vehicle synchronously collects road surface data, and can realize higher-precision environmental data sensing, and can synchronously measure the actual time and distance of the vehicle driving to the target road condition. The collected road surface data is subjected to spatio-temporal correlation of multi-modal data by a data fusion algorithm (such as CNN-LSTM), and the output structured road condition classification result determines the deceleration zone, the bumpy road, the straight road, etc., and can also divide the gas station, the parking garage, and the muddy road of the irregular road, etc.

[0083] In an embodiment, the first data collection is the driving data of the untravelled road section, mainly through the laser radar and the front camera to collect the road surface information. In special weather, for example, in rainy, snowy, slippery road sections, windy weather, and foggy weather, the target road condition information needs to be determined in combination with the sensors of the vehicle. For example, in wet and rainy weather or road section, the slip condition of the current vehicle in the current driving state needs to be combined with the adhesion coefficient to the target road condition corresponding to the first data. For foggy and windy weather, the data of the laser radar and the front camera can complement each other, eliminate the problem of contrast reduction in foggy weather, and the problem of detection reduction caused by insufficient cleanliness of the laser radar in foggy weather.

[0084] In an embodiment, the second time is the driving time when the current vehicle reaches the target road surface corresponding to the target road condition information determined at the first time. The pre-trained CNN-LSTM model built in the vehicle-mounted computing platform 330 of the whole vehicle control 300 is used to calculate the best air suspension 200 braking control scenario that the current vehicle can achieve under the target road condition to achieve the smooth comfort of the current vehicle reaching the target. The CNN-LSTM model will also automatically determine the best air suspension 200 braking control parameter suitable for the driver according to the driving data in the actual driving process, and continuously learn and iterate. The CNN-LSTM model recognition technology matches the suspension scenario under the target road condition, i.e., the air suspension 200 braking control scenario, with the low-speed bumping, high-speed straight, curve driving, or water driving scene in the preset scene library, triggers the corresponding control strategy, and determines the air suspension 200 braking parameter. Through the adjustable parameter set adapted by the air suspension 200, including air spring air pressure, shock absorber damping coefficient, and brake pressure distribution. Through parameter collaborative optimization, comfort and handling are balanced. For example, increasing air spring air pressure can reduce vehicle body vertical displacement, and increasing damping coefficient can quickly attenuate vibration.

[0085] In an embodiment, for the high-speed curve driving scene, the front camera of the vehicle recognizes the curve curvature and the marking line of the road, the laser radar detects the radius of the curve, and the high-precision map built in the car determines the curve slope data. In the first time, the specific curve data of the curve driving scene is recognized within 100 meters or 200 meters before the curve driving scene, and the high-speed curve driving scene in the second time is determined. High-speed curve driving scene requires high lateral acceleration, air suspension 200 needs to reduce vehicle body height, and front axle suspension damping coefficient needs to be increased, so as to suppress vehicle body roll. At the same time, through ESP cooperation, the inside wheel brake pressure is pre-adjusted to offset the centrifugal force, realize more stable curve driving, and air suspension 200 does not need to be adjusted after entering the curve.

[0086] In one embodiment, for bumpy road scenarios, the vehicle's front-facing camera identifies uneven road surfaces such as gravel or multiple speed bumps. High-density radar measures the height of the road surface bumps and the high-frequency vibrations of the wheel speed sensors to determine the target road condition information of the bumpy section at the first moment. This allows for the determination of the bumpy road scenario at the second moment, further classifying it as moderate, high, or slight bumps. The damping of the air suspension 200 is then adjusted to a low-damping state in soft-damping mode, while the air springs of the air suspension 200 are inflated to improve shock absorption. A longitudinal acceleration sensor predicts vehicle body undulations, allowing for advance adjustment of the front and rear axle suspension travel difference to prevent frequent nose-diving. Furthermore, upon entering a bumpy road, continuous dynamic braking of the air suspension 200, with continuous raising and lowering of the air suspension 200 under high-frequency vibrations, improves driving comfort. This eliminates the need for dynamic inflation and deflation of the air springs based on vehicle height changes after entering a bumpy road, resulting in lower vibration acceleration.

[0087] In one embodiment, for slippery road surfaces, the vehicle's front-facing camera identifies the rainy environment, lidar identifies the thickness of the water film on the road surface, and combined with wheel speed sensors, determines the road adhesion coefficient under slippery conditions, thus determining the target road conditions at the first moment, and further determining the slippery road surface scenario and wading road surface at the second moment. At the second moment, for slippery or wading roads, the vehicle height is increased to raise the ground clearance, preventing airflow interference from the underside, while balancing the grip of the air suspension 200 and adjusting the suspension stiffness. Furthermore, through ESP coordination, the driving torque of the inner wheels is reduced in advance, and the damping difference between the left and right sides is adjusted by the air suspension 200 to suppress oversteer. This allows for real-time correction via wheel speed difference and yaw rate after entering a slippery road surface.

[0088] Example 3:

[0089] Combination Figure 3 The sensor system 100 in this application demonstrates that the multimodal data acquisition device of this application can actively predict the effect of shortening the response time of the air suspension 200 in the braking control of the air suspension 200.

[0090] During the operation of the electric vehicle 10, a multimodal data acquisition unit collects visual data, sensor data, and environmental data. The visual perception subsystem 110 collects visual data, including camera and lidar data, to determine target road condition information. The sensor network 120 collects sensor data, primarily vehicle-specific data, such as real-time height and damping coefficient of the air suspension 200, braking parameters, vehicle acceleration, wheel speed, and stability. The environmental perception subsystem 130 collects weather data and map data of the road surface, used to determine road conditions at both the first and second time points, thus establishing a dual-time-domain road condition model.

[0091] The visual data includes a road surface semantic segmentation image of the next driving road condition. The visual data is processed by a deep learning model pre-trained on road surface sample data based on a Transformer architecture, to realize semantic segmentation and feature extraction of multi-modal data, and to output specific labels of the suspension scene in real time.

[0092] The sensor data includes vehicle body height data, water level data, acceleration sensor data, and temperature sensor data of the current vehicle in the current driving road condition and the next driving road condition.

[0093] The environmental data includes weather data and driving scene data of the current vehicle.

[0094] The sensor data and the environmental data are also subjected to semantic segmentation and feature extraction by a deep learning model pre-trained on sensor data samples and environmental data samples based on a Transformer architecture. On the basis of outputting specific labels of the suspension scene, real-time braking parameters of the current vehicle are added, so as to actively predict target road condition information of the specific road surface of the driving scene of the current vehicle at the second time. The data transmission is transmitted to the whole vehicle general control 300 through the CANFD 410 high-speed bus, and then the whole vehicle general control 300 synchronizes the control instructions to the ECU 310 (the core module of the automotive electronic control system, which realizes real-time monitoring and control of specific vehicle functions by integrating microprocessors, sensor interfaces, and actuator driving circuits) of the air suspension 200, the electronic stability system ESP 320 (a vehicle dynamic control system that prevents instability such as vehicle skidding and fishtailing by coordinated braking and engine torque adjustment), and the powertrain, to realize collaborative control of the suspension and the chassis system, and to provide the air suspension 200 with road condition data identified in advance for automatic control. The road condition data identified in advance determines the response time of the vehicle in advance according to the vehicle speed and the time during which the vehicle travels on the road surface corresponding to the target road condition at the second time, thereby determining the response time of the air suspension 200, which also indicates the time at which the air suspension 200 is braked in advance, so that the air suspension 200 can be realized.

[0095] In one embodiment, the visual data is obtained by high-definition cameras installed on the front and sides of the vehicle. The high-definition cameras are used to capture road surface images in front of and on both sides of the vehicle driving path, and generate road surface semantic segmentation images after processing by the image segmentation algorithm built-in the high-definition cameras. The visual data is transmitted at high speed through the Ethernet 420 of the bus 400 to the vehicle-mounted computing platform 330 (a high-performance central computing system for intelligent vehicles, which integrates multi-domain controller functions, supports multi-sensor fusion, complex algorithm operation, and whole vehicle level collaborative control).

[0096] In an embodiment, the role of the road semantic segmentation image is to divide the driving road surface of the current vehicle at the second time into different semantic categories. In the specific division process, it includes normal driving area, deceleration zone, manhole cover area, water accumulation area, and sidewalk edge, etc. Each category is distinguished by different pixel identifiers to improve recognition accuracy. The road semantic segmentation image can clearly identify the location and range of various features on the road surface. For example, the pixel set of the deceleration zone can display the image features corresponding to the length, width and protruding height of the deceleration zone; the water accumulation area is different from other areas through the color depth and reflective characteristics of the pixels, and can determine whether there is a special structure on the road surface that affects the suspension.

[0097] In an embodiment, the sensor data is obtained by the sensor network 120 of the current vehicle through various types of sensors on the vehicle, which obtains rich vehicle and road condition information. The body height data is collected by the displacement sensor installed on the suspension system, which can determine the relative height of the current vehicle body and the ground, as well as the change trend of the body height in the current driving road condition and the next driving road condition. When the vehicle is about to enter the concave road surface, the pre-descending trend of the body height will be captured in advance; the water level data is obtained by the liquid level sensor installed at the bottom of the vehicle, which can determine the depth of the road water, and can accurately reflect the impact range of the water on the bottom of the vehicle when passing through the water accumulation section; the acceleration sensor data comes from the accelerometer on the vehicle body and suspension components, which can determine and record the acceleration changes in the longitudinal, lateral and vertical directions during the vehicle driving. The change data can determine the degree of vehicle bumping and the inclination trend of the vehicle body; the temperature sensor data is collected by the temperature sensor distributed near the key components of the suspension and the tire, which is used to monitor the temperature state of the components during work to prevent high temperature from affecting the performance of the suspension. Part of the sensor data is transmitted to the vehicle computing platform 330 and ECU 310 through the CANFD 410 (which is an upgraded version of the traditional CAN bus protocol, aiming to solve the limitations of traditional CAN in data transmission rate and payload length, suitable for automotive electronics, industrial control and other scenes with high real-time and bandwidth requirements); the other part is transmitted to the vehicle computing platform 330 and ECU 310 through the LIN bus 430.

[0098] In an embodiment, the environmental data is collected by an environmental perception device outside the vehicle and a vehicle positioning system. The weather data includes the type of precipitation, the intensity of precipitation, the level of wind force, and the air visibility of the current vehicle on the current driving section, which is directly collected by a meteorological sensor on the top of the vehicle or received by the cloud through the vehicle, and the real-time broadcast information of the local weather station is obtained through cloud networking. For example, the degree of wetness and slipperiness of the road surface that may be caused by rainwater soaking is determined by the intensity of precipitation; the stability of the current vehicle when driving at high speed is determined by the level of wind force; the type of road (such as urban road, highway, and rural road), road speed limit information, and environmental features such as whether there are schools or residential areas around the current section are determined by the driving scene data in combination with the geographical position information and map data of the vehicle positioning system. Different road types are directly related to the maintenance condition of the road surface and the expected speed of the vehicle, and the surrounding environmental features affect the possible driving strategy of the vehicle and are related to the control of the air suspension 200.

[0099] In an embodiment, during the data collection process, the multi-modal collector performs preliminary noise reduction processing on various types of data. Before generating a semantic segmentation image, the visual data will be filtered to remove noise points and disturbed pixels caused by changes in light; the sensor data will be smoothed to eliminate transient abnormal readings caused by sensor vibration; and the environmental data will be cross-verified to ensure the accuracy of the weather data and driving scene data.

[0100] Embodiment 4:

[0101] In combination with Figure 3 , it is explained that the multi-modal collector of the present application improves the accuracy of scene recognition in the active prediction process through the ECU 310 and the vehicle computing platform 330 in the brake control of the air suspension 200, so that the brake parameters of the air suspension 200 are more accurate.

[0102] At the first time, according to the environmental data, the brake scene of the air suspension 200 under the current driving condition is determined;

[0103] During the driving process of the electric vehicle 10, based on the weather conditions, map data, traffic representation information, and other environmental data at the first time (current time), the preliminary brake scene of the current vehicle is determined to avoid single data identification of the scene and improve the accuracy.

[0104] In addition, according to the visual data, the target road condition information of the air suspension 200 in the brake scene is determined;

[0105] After the braking scene is determined, the specific road condition information of the air suspension 200 at the second time can be determined in combination with the semantic segmentation results of the camera and the lidar, and the braking mode switching (gear) can be realized. In the case of combining the road surface conditions, the target road condition information is determined. Because the non-air suspension 200 braking parameters of the vehicle are introduced, the accuracy is higher when the air suspension 200 braking parameters are set.

[0106] In addition, according to the sensing data, the braking behavior interval of the air suspension 200 under the target road condition information is determined; wherein the braking behavior interval includes the body height interval, the suspension stiffness interval, the damping coefficient interval, and the attitude balance interval.

[0107] After the target road condition is determined, the braking behavior interval is calculated in combination with the sensing data, i.e., the height, acceleration, temperature, and other sensing data of the vehicle body, to realize the upper and lower threshold values of the body height, stiffness, damping, and attitude balance, determine the specific control parameters of the air suspension 200 on the target road condition, and prevent lag control. If the air suspension 200 is directly regulated and controlled without combining the above data, it will also cause the vehicle body attitude to be unstable.

[0108] In one embodiment, for the high-speed over-bend scene, the environmental data can determine the curvature of the bend, and the navigation can prompt the risk of the bend (easy to fall rocks, and there may be bumps), in combination with the road surface data, to determine that the suspension scene at the second time is a high-speed over-bend scene, to integrate the possible risks in the high-speed over-bend scene, and to determine the target road condition information as a high-speed over-bend scene + rock bump scene (when there is water, the wet and slippery water scene), to generate a target road condition information of a complex driving scene, and then to determine the braking parameter interval of the air suspension 200 according to the sensing data in setting the air suspension 200 braking parameters of the current vehicle in combination with the sensing data (vehicle speed, steering angle, suspension roll), so as to ensure comfort while preventing air suspension 200 response delay or control parameters of other systems of the vehicle body and the air suspension 200 from not realizing cross-system cooperation (mainly sensing parameters), such as not considering the suspension roll, directly regulating and controlling the air suspension 200, causing the vehicle to roll, and causing the vehicle to roll along the roll angle in the high-speed over-bend and wet road surface scene. The braking behavior interval is the body height interval, the suspension stiffness interval, the damping coefficient interval, and the attitude balance interval, which ensures the handling while suppressing nodding and body shaking.

[0109] Embodiment 5:

[0110] In combination with Figure 2 and Figure 4 , Figure 4The target road condition information determination process diagram illustrates that the multi-modal collector of the present application actively avoids adverse road conditions in the brake control of the air suspension 200 through the ECU 310 and the vehicle-mounted computing platform 330 after the driving scene is pre-judged, so that the air suspension 200 does not appear to be greatly worn out when it is automatically controlled.

[0111] During the driving process of the electric vehicle 10, the multi-modal collector will also:

[0112] According to the driving scene data, the brake behavior of the next driving road condition is determined;

[0113] Based on high-precision map, traffic signs, and vehicle speed and other environmental data, the first time the brake behavior of the next stage is pre-judged to determine whether emergency braking, slow deceleration, and constant cruise are needed.

[0114] According to the brake behavior and the road surface semantic segmentation image, the target road condition information that meets the adjustment of the air suspension 200 on the next driving road condition is determined.

[0115] Through road surface semantic segmentation of the camera and the laser radar, the road condition features (pits, manhole covers, gravel, water accumulation) that need to be avoided are identified, and the relative distance, lane position, size, danger level, and other data of the road condition features that need to be avoided are output. The next driving road surface that the air suspension 200 of the current vehicle can adapt to and the next road surface that cannot be driven are determined, so as to determine the target trajectory of the next road surface and send it to the car system. Whether it is intelligent driving or driver driving, it will avoid roads and road surfaces that may cause great wear to the air suspension 200 and exceed the brake interval of the air suspension 200.

[0116] In one embodiment, for a bumpy road with gravel, the first data determines which gravel in the driving road can be controlled by the automatic brake control of the air suspension 200 to achieve a smooth running road. And identify which gravel in the driving road is too large to exceed the brake interval of the air suspension 200, even braking will cause the vehicle to be unstable, and forced automatic control will cause the air suspension 200 to wear out. When identifying which gravel in the driving road meets the brake behavior interval of the air suspension 200, the suspension control strategy information of "right side suspension raised by 8mm (left side remains unchanged), slight roll (1°) formed to reduce right side tire impact, meets air suspension 200 brake interval" will be generated for the vehicle height. For the damping coefficient, the suspension control strategy information of "right side front and rear wheel damping temporarily adjusted to 0.6 (left side 0.4) to buffer the vertical impact of gravel" will be generated. For EPS coordination, the suspension control strategy information of "EPS system assists small steering (steering angle 3°) to ensure that the wheels avoid the area with large gravel" will be generated.

[0117] Embodiment 6:

[0118] In combination Figure 2 In one possible embodiment, in the brake control of the air suspension 200 described in the present application, in the identification of the suspension scene by the ECU 310 and the vehicle-mounted computing platform 330, the identification of the target road surface will output the suspension scene that meets the continuous control of the air suspension 200 according to the change rule of the road condition in time sequence. In the generation of the second time suspension scene and the brake parameter, the air suspension 200 will not appear instantaneous and large amplitude adjustment, and the corresponding air suspension 200 brake parameter.

[0119] In the process of driving the electric vehicle 10, the suspension scene of the second time after the first time is determined according to the first data, specifically including:

[0120] The first data is input into the pre-trained scene classification model; wherein the scene classification model adopts a CNN-LSTM fusion architecture for extracting spatial features of multi-modal data and capturing the change rule of road condition in time sequence;

[0121] The road surface semantic segmentation image, the vehicle first time sensing data, such as vehicle body height, acceleration time sequence data, and environment data are input into the scene classification model of the CNN-LSTM fusion architecture through the same format, to extract spatial features in the visual data, such as road cracks, obstacle outlines, and lane line curvature; extract the time sequence change rule of the sensing data, such as vehicle body height fluctuation frequency and acceleration peak interval; then realize the weighted fusion of spatial mechanism and time sequence features through the attention mechanism, and output the multi-modal joint feature vector.

[0122] According to the scene probability distribution output by the scene classification model, the suspension scene of the second time is determined.

[0123] The multi-modal joint feature vector is output through a fully connected layer to output the distribution probability of the suspension scene, that is, the best scene of continuous control. In the case that the adjustment amplitude of the air suspension 200 is not large, the scene with the highest probability is selected as the target scene of the second time, so that the air suspension 200 will not appear instantaneous and large amplitude brake.

[0124] In one embodiment, when the first data is input into the pre-trained scene classification model to determine the suspension scene of the second time, the CNN-LSTM fusion architecture of the scene classification model includes a convolutional neural network layer and a long short-term memory network layer, which cooperates through a feature mapping channel.

[0125] The convolutional neural network layer extracts spatial features of multi-modal information in the first data.

[0126] In an embodiment, for image data, if there is road surface texture and obstacles in the image data, the correlation pattern between pixels is captured by multi-layer convolution, and a spatial feature map that can show the distribution of rough road surface area and the contour features of obstacles is generated; for environmental perception data, if there is precipitation intensity distribution, the difference in environmental parameters in different regions is calculated by convolution kernel sliding, and the precipitation distribution is converted into a feature vector with spatial dimension.

[0127] In an embodiment, the long short-term memory network layer is used to capture the time sequence variation law of the road condition. By receiving the spatial feature sequence output by the convolutional neural network layer, if it is found that there are multiple frames of continuous collection of road condition information in the first data, the long short-term memory network can identify the time sequence features such as the gradual change from gentle to severe of the road surface undulation and the motion trajectory from the appearance to the approach of the obstacle, and determine the brake interval of the air suspension 200 through these features, thereby generating the corresponding air suspension 200 brake mode.

[0128] In an embodiment, according to the air suspension 200 brake mode, the air suspension 200 generates phased air suspension 200 brake parameters during the driving process of the current vehicle. For example, the damping coefficient is 0.6 at the first time, and the damping coefficient is 0.1 at the subsequent road surface, such as the fifth time, to ensure the stability of the vehicle. The damping coefficient needs to be 0.3 at the second time to maintain stability. The damping coefficient needs to be 0.4 at the third time to maintain the stability of the vehicle body. The damping coefficient needs to be 0.3 at the fourth time to maintain stability. The suspension scene is set so that the damping coefficient is 0.4 at the second time, the damping coefficient is 0.4 at the third time, and the damping coefficient is 0.3 at the fourth time. Through the time sequence variation law of the road condition, the damping parameter of the air suspension 200 can be adjusted with low amplitude under the condition of maximum probability, so that the air suspension 200 will not be severely worn and the program error caused by instantaneous braking.

[0129] Embodiment 7:

[0130] In combination Figure 2 and Figure 5 , Figure 5 is a determination graph for suspension scene adjustment. In a possible embodiment, it is explained how the air suspension 200 of the present application controls the air suspension 200 to be in the most suitable suspension scene through the ECU 310 and the vehicle-mounted computing platform 330 when determining the suspension scene, so as to ensure the adjustment of the air suspension 200 and be ahead of the working condition change. Even if it is not the best air suspension 200 brake parameter, it will also ensure the comfort of the driver and will not miss or misjudge.

[0131] In the process of driving the electric vehicle 10, when determining the suspension scene at the second time:

[0132] According to the driving state of the current vehicle at the first moment and the road condition characteristics, a second road condition determination is performed;

[0133] The driving state is the driving state parameter of the first moment collected by the vehicle body sensor, including the current driving stability, steering tendency and acceleration or deceleration intention of the vehicle, and is aimed at automatic driving operation. Through visual semantic segmentation and laser radar point cloud, the physical characteristics of the road surface are extracted to determine the flatness, lateral inclination and obstacles of the road surface. The suspension scene is comprehensively judged, and in this application, the manipulation scenes such as the conventional comfort scene, the steering stability scene and the emergency buffer scene are mainly judged. In the manipulation scene, the actual road surface and site scene in the driving environment also need to be combined.

[0134] When the current vehicle is in a stable driving state and the road condition characteristics show that the road surface is flat, the suspension scene is divided into a conventional comfort scene;

[0135] If the current vehicle has a steering tendency and the road condition characteristics show that the road surface has a lateral inclination, the suspension scene is adjusted to a steering stability scene;

[0136] If the first data shows that there is a sudden obstacle in front of the vehicle and the vehicle has a deceleration intention, the suspension scene is switched to an emergency buffer scene.

[0137] Then, according to the specific road condition characteristics, the suspension scene is determined, and by combining the vehicle's own state with the road condition characteristics, the division of the suspension scene is more in line with the dynamic needs in actual driving, avoiding scene misjudgment caused by relying on single information.

[0138] In one embodiment, in the conventional comfort scene, the stiffness mode of the air suspension 200 is low stiffness, which preferentially filters the slight bumps of the road surface, the damping coefficient controls the vertical vibration frequency of the vehicle body to be controlled in the human comfort interval (set by the driver) of 1.2Hz, and the vehicle body height is set to the standard height, balancing the handling and aerodynamics. Not limited to the fixed parameter output in the comfort mode, the brake parameters of the air suspension 200 cannot be fine-tuned. At the same time, for the steering stability scene, ESP can be combined to realize vehicle body posture control. When encountering obstacles, the emergency buffer scene can also be used to set the height adjustment of the super-high air suspension 200 to prevent the ground from being scratched, and there is no need to trigger the pre-set physical conditions and fixed gears.

[0139] Embodiment 8:

[0140] In combination Figure 2 In one possible embodiment, when determining the suspension scene of the air suspension 200 of the present application, if there is a deviation between the scene prediction and the actual road condition, the ECU 310 and the vehicle-mounted computing platform 330 can continuously fine-tune through matching verification.

[0141] The electric vehicle 10 is in the process of driving, and the second time determines the suspension scene:

[0142] The real-time road condition information of the current vehicle at the second time is matched and verified with the target road condition information in the first data; wherein,

[0143] The target road condition information is the target road condition at the second time predicted by the multi-modal data at the first time, and the corresponding suspension scene and brake parameters (stiffness, damping, etc.) are generated; when the vehicle drives to the second time, the current road condition data (such as the actual bend curvature, road slope, obstacle position) is collected in real time through the laser radar / camera; in the matching verification, through the pre-selected similarity algorithm and other similarity algorithms, the real-time road condition data at the second time and the target road condition information at the first time are matched and compared to determine whether the preset standard is reached, so as to prevent errors in the adjustment of the air suspension 200.

[0144] If the matching degree of the two reaches the preset standard, the original suspension scene and the corresponding brake parameters are maintained;

[0145] When the data of the two are consistent, the suspension scene determined in advance can be maintained without adjustment.

[0146] If the matching degree is lower than the preset standard, the suspension scene is corrected according to the real-time road condition information, and the brake parameters are adjusted synchronously.

[0147] If the matching degree indicates that the two are quite different, the suspension scene will also be corrected synchronously according to the real-time road condition information, and the brake parameters will be adjusted synchronously to achieve dynamic synchronous compensation.

[0148] If the road condition changes frequently, the verification interval will be shortened.

[0149] If the road condition changes frequently, the verification time will be shortened to improve the response sensitivity of the air suspension 200 adjustment.

[0150] In one embodiment, for high-speed bends, there is a pre-judgment deviation, for example, the road surface information is the same, and there is water or wet slippery emergency bend on the road surface, which will adjust the brake parameters of the air suspension 200 at the current time while correcting the suspension scene, to improve the real-time grip. For continuous bumping scenes, the verification interval is adjusted to ensure comfort.

[0151] Embodiment 9:

[0152] In combination Figure 2 In one possible embodiment, the air suspension 200 of the present application is explained when determining the brake parameters, and the ECU 310 and the vehicle-mounted computing platform 330 shorten the cooperative response parameters of the brake parameters and the suspension scene.

[0153] The electric vehicle 10 determines the brake parameters corresponding to the suspension scene during driving, and further comprises:

[0154] The second road condition features directly related to the suspension scene in the first data are extracted and matched with the expected driving posture of the vehicle in the suspension scene; wherein,

[0155] The second road condition features are extracted in a targeted manner, and the key road condition features strongly related to the suspension scene, such as road flatness, adhesion coefficient, obstacle density, feature change rate and scene duration, are screened in the first data. Then the mapping relationship between the suspension scene and the expected driving posture of the vehicle is constructed to realize the correlation mapping of the high-speed cruising scene, the high-speed curve scene and the wet road scene, convert the second road condition features into the adjustment parameters of the air suspension 200, and set the constraint conditions of the brake parameters.

[0156] According to the parameter matching, the change rate of the second road condition features is associated with the duration of the suspension scene to determine the brake parameters. In the process of parameter matching, the speed change rate of the second road condition features and the duration of the suspension scene are associated to generate short-term high-frequency change scenes and slowly changing scenes, balance the brake efficiency and thermal decay, and also avoid the jolt of the vehicle body.

[0157] In one embodiment, when determining the brake parameters corresponding to the suspension scene, the current load distribution state of the vehicle and the dynamic change features of the road condition in the first data are considered. The load distribution state of the vehicle is obtained by a pressure sensing device at the bottom of the vehicle body, which can determine the load difference of the front and rear axles and the left and right sides. The area with heavier load will generate greater pressure on the suspension, thereby affecting the adaptability of the brake parameters; the dynamic change features of the road condition are extracted from the first data, including the continuous frequency of road undulation, the steepness of slope change and the suddenness of obstacle appearance.

[0158] If the current vehicle load is evenly distributed and the road condition changes gently, the adjustment range of the brake parameters is kept within the normal range to maintain the stable response of the air suspension 200;

[0159] If the current vehicle has obvious load bias and the road condition changes dramatically, for example: the road surface with large high-low difference appears continuously, the adjustment of the stiffness coefficient in the brake parameters will be inclined to the side with heavier load, and the change rate of the damping coefficient will be correspondingly accelerated to ensure that the air suspension 200 can also make the vehicle body stable when bearing uneven pressure.

[0160] In an embodiment, by combining the load distribution with the dynamic changes of the road conditions, in the process of setting the brake parameters of the suspension scene, it is determined not only to fit the characteristics of the road conditions itself, but also to adapt to the load difference of the vehicle itself, to avoid the response imbalance of the brake parameters in different parts due to uneven load. Improve the overall control effect of the air suspension 200 under complex driving conditions. After the brake parameters are determined, small corrections are made according to the real-time changes of the load distribution. When the load temporarily changes due to the movement of the passengers or the shaking of the goods, the parameters can be adaptively fine-tuned within the preset range to quickly adapt to such short-term load fluctuations and ensure the continuity of the suspension performance.

[0161] Embodiment 10:

[0162] In combination Figure 2 And Figure 6 , Figure 6 The flowchart for detecting brake defects In a possible embodiment, it is explained how the ECU 310 and the vehicle-mounted computing platform 330 eliminate potential risks in the air suspension 200 braking process after determining the brake parameters of the air suspension 200 of the present application.

[0163] After determining the brake parameters corresponding to the suspension scene, the electric vehicle 10 in the process of driving, the specific operation is:

[0164] Step S301: Simulate the driving state when the current vehicle enters the suspension scene at the second time, and detect whether the brake parameters will cause abnormal stress of the air suspension components or imbalance of the vehicle body posture;

[0165] Based on the ADAMS multi-body dynamics model, a virtual vehicle prototype is constructed, and the road surface distance, vehicle speed, steering angle, road surface condition and other suspension scene parameters at the second time are input, and the brake force distribution and deceleration and other brake parameters to be verified are fused to realize driving simulation and judge whether there is a potential risk simulation situation that the components appear abnormal stress or the vehicle body posture is imbalanced. The component stress is reflected in the cases that the air spring air bag pressure exceeds the threshold, the shock absorber piston rod strain is too large, and the lower arm connecting bolt torque exceeds the limit value.

[0166] Step S302: If the test result shows that a parameter setting in the brake parameter causes the suspension spring to be stretched too much or the shock absorber to be overloaded, the brake parameter is fine-tuned; if the stress is excessive or the posture is imbalanced, the brake parameter is fine-tuned through the particle swarm optimization algorithm (PSO); that is, a single brake parameter is fine-tuned. Improve the component reliability of the air suspension 200 in the brake control process, so that the air suspension 200 will not be abnormal due to component failure.

[0167] Step S303: If the parameter setting makes the vehicle body stable in the simulation scene and the component stress is within the safe range, the brake parameter is confirmed to be effective.

[0168] In an embodiment, for example, when detecting whether the brake parameter will cause abnormal stress of the air suspension component or imbalance of the vehicle body posture, if the front axle air spring air bag pressure peak value exceeds the safety threshold value, the air spring stress is excessive, the local strain concentration of the piston rod is present (conventional detection, unable to determine the local strain force concentration), and there is a risk of fracture, the brake force distribution front axle is reduced, the deceleration is reduced, the rear axle air spring pre-tightening force is synchronously increased through the double-fan air compressor 240 and the air spring solenoid valve 210, while preventing the local strain concentration of the piston rod and the occurrence of fracture, a certain degree of driving comfort is ensured, and the cause of the abnormality is found. For the shock absorber overload (the shock absorber overload is not found in advance, only continuous jolting exists, and the air suspension 200 has been adjusted to the brake parameter corresponding to the suspension scene), it is not necessary to repeatedly replace the shock absorber with different damping coefficients, and it is determined whether the shock absorber is abnormal. When the suspension scene is met and the target brake parameter is used, if the vehicle still has unstable brake abnormalities. However, by simulating the driving state at the second time and under the condition of fine adjustment, the jolting caused by the shock absorber overload is solved, and it can be determined that the shock absorber is abnormal.

[0169] In another embodiment of the present application, an automobile air suspension automatic control system based on multi-modal data recognition is also provided. It can be understood that all related contents of each step involved in the above method embodiments can be cited into the embodiments of the automobile air suspension automatic control system, and the embodiments of the present application will not be repeated here.

[0170] Referring to Figure 2 The present application also provides an automobile air suspension automatic control system based on multi-modal data recognition, comprising: a multi-modal data acquisition module, configured to acquire first data collected by a multi-modal collector of a vehicle at a first time; wherein the first data is target road condition information of the current vehicle on a next driving road condition determined by the multi-modal collector; and an automatic control module, configured to determine a suspension scene at a second time after the first time and brake parameters corresponding to the suspension scene according to the first data.

[0171] It can be understood that all related contents of each step involved in the above method embodiments can be cited into the embodiments of the automobile air suspension automatic control system, and the embodiments of the present application will not be repeated here.

[0172] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for automatic control of an automobile air suspension based on multi-modal data recognition, applied to an air suspension, characterized in that, The method comprises: At a first time, acquiring first data collected by a multi-modal collector of a vehicle; wherein the first data is target road condition information of the current vehicle on the next driving road condition determined by the multi-modal collector; According to the first data, determining the suspension scene at a second time after the first time, and the brake parameters corresponding to the suspension scene.

2. The automatic control method of an automobile air suspension based on multi-modal data recognition according to claim 1, characterized in that, The multi-modal collector comprises a configuration configured to collect visual data, sensor data and environmental data; Wherein, the visual data includes a road surface semantic segmentation image of the next driving road condition; The sensor data includes the vehicle body height data, water level data, acceleration sensor data and temperature sensor data of the current vehicle on the current driving road condition and the next driving road condition; The environmental data includes weather data and driving scene data of the current vehicle.

3. The automatic control method of an automobile air suspension based on multi-modal data recognition according to claim 2, characterized in that, At the first time, according to the environmental data, determine the brake scene of the air suspension on the current driving road condition; And, according to the visual data, determine the target road condition information of the air suspension in the brake scene; And, according to the sensor data, determine the brake behavior interval of the air suspension under the target road condition information; wherein, the brake behavior interval includes the vehicle body height interval, the suspension stiffness interval, the damping coefficient interval and the attitude balance interval.

4. The automatic control method of an automobile air suspension based on multi-modal data recognition according to claim 2, characterized in that, When the first data is collected, it includes: According to the driving scene data, determine the brake behavior of the next driving road condition; According to the brake behavior and the road condition features to be avoided in the road surface semantic segmentation image, determine the target road condition information of the next driving road condition that meets the air suspension adjustment.

5. The automatic control method of an air suspension of an automobile based on multi-modal data recognition according to claim 1, characterized in that, According to the first data, determining the suspension scene at a second time after the first time, includes: Input the first data into a pre-trained scene classification model; wherein, the scene classification model adopts a CNN-LSTM fusion architecture, which is used to extract the spatial features of multi-modal data and capture the time sequence variation law of the road condition; According to the scene probability distribution output by the scene classification model, determine the suspension scene at the second time.

6. The automatic control method of an automobile air suspension based on multi-modal data recognition according to claim 1, characterized in that, When determining the suspension scene at the second time: According to the driving state and road condition features of the current vehicle at the first time, make a second road condition judgment; wherein, When the current vehicle is in a stable driving state and the road condition features show that the road surface is flat, the suspension scene is divided into a regular comfort scene; If the current vehicle has a turning trend and the road condition features show that the road surface has a lateral inclination, the suspension scene is adjusted to a turning stability scene; If the first data shows that there is a sudden obstacle in front and the vehicle has a deceleration intention, the suspension scene is switched to an emergency buffer scene.

7. The automatic control method of an automotive air suspension based on multi-modal data recognition according to claim 1, wherein, After determining the suspension scene at the second time, it also includes: Match and verify the real-time road condition information of the current vehicle when it is driving to the second time with the target road condition information in the first data; wherein, If the matching degree of the two reaches the preset standard, maintain the original suspension scene and the corresponding brake parameters; If the matching degree is lower than the standard, correct the suspension scene according to the real-time road condition information and adjust the brake parameters synchronously; If the road condition has a frequent change trend, the verification interval is shortened.

8. The automatic control method of an automotive air suspension based on multi-modal data recognition according to claim 1, wherein, The determination of the brake parameters corresponding to the suspension scene includes: Extract the second road condition features in the first data that are directly related to the suspension scene, and perform parameter matching with the expected driving attitude of the vehicle in the suspension scene; wherein, According to parameter matching, the change rate of the second road feature is associated with the duration of the suspension scene to determine the brake parameter.

9. The automatic control method of an air suspension of an automobile based on multi-modal data recognition according to claim 1, wherein, After determining the brake parameter corresponding to the suspension scene, the method further includes: Simulating the driving state of the current vehicle entering the suspension scene at the second time, detecting whether the brake parameter will cause abnormal stress of the air suspension component or imbalance of the vehicle body posture; If the test result shows that a parameter setting in the brake parameter causes excessive stretching of the suspension spring or overloading of the shock absorber, the brake parameter is fine-tuned; If the parameter setting keeps the vehicle body stable in the simulated scene and the component stress is within a safe range, the brake parameter is confirmed to be effective.

10. An automatic control system for a vehicle air suspension based on multi-modal data recognition, characterized in that, The method includes: A multi-modal data acquisition module is configured to acquire first data collected by a multi-modal collector of a vehicle at a first time; wherein the first data is target road condition information determined by the multi-modal collector for the current vehicle on the next driving road condition; An automatic control module is configured to determine, according to the first data, a suspension scene at a second time after the first time, and a brake parameter corresponding to the suspension scene.

Images