Preview active suspension rigidity control method based on road surface grading and MAP mapping

Abstract

The invention is suitable for the technical field of vehicle dynamics modeling and control, and provides a preview active suspension stiffness control method based on road surface grading and MAP mapping, and the method comprises the steps: collecting a road image through a vehicle-mounted visual sensor, dividing regions, and recognizing the road surface unevenness grade of each region through a convolutional neural network; obtaining the vehicle speed; and according to the current pavement grade and the vehicle speed, a vehicle speed-pavement grade-rigidity mapping MAP table generated in advance is inquired, and the target suspension rigidity is rapidly obtained and adjusted. According to the invention, the dependence on a high-precision sensor and a high-computing-power processor is reduced, and low-cost and fast-response rigidity optimization control of the preview active suspension is realized.

Description

Technical Field

[0001] This invention belongs to the field of vehicle dynamics modeling and control technology, and in particular relates to a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping. Background Technology

[0002] The automotive suspension system is a key component affecting vehicle ride quality, and its development has evolved from passive suspension to semi-active suspension to active suspension. To further coordinate and improve ride comfort and handling stability, anti-suspension active suspension technology has emerged. This technology senses road information ahead and proactively adjusts suspension stiffness and damping in advance, thereby effectively improving vehicle dynamics. However, most existing anti-suspension active suspension systems rely on high-precision road geometry elevation information, requiring expensive road sensing equipment and high-performance processors, resulting in high system costs and limiting their widespread application in the mass market.

[0003] In existing technologies, some solutions have attempted to acquire road surface information through vision, navigation, and sensor fusion to achieve suspension adjustment. For example, some solutions use onboard cameras, accelerometers, and navigation modules to identify road information and adjust the suspension; others use precise measurements of road surface elevation data and complex calculations to distinguish road surface smoothness, thereby controlling suspension stiffness and damping; still others use binocular cameras to collect road surface elevation information and combine semantic segmentation and predictive control algorithms to regulate the suspension. While these methods have some effectiveness, they generally suffer from the following drawbacks: First, they are highly dependent on accurate road surface geometry information, placing stringent requirements on sensing hardware and perception algorithms, resulting in complex systems and high costs; second, they are mostly based on real-time processing of time-domain road surface information, heavily relying on the real-time performance of computing resources and failing to fully utilize the frequency domain characteristics of the road surface, thus limiting the efficiency and adaptability of the algorithms.

[0004] Therefore, it is necessary to develop a pre-aiming active suspension stiffness control method that has low dependence on road surface information, low computational load, and controllable hardware cost, so as to promote the widespread application of this technology. Summary of the Invention

[0005] The purpose of this invention is to provide a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping, aiming to solve the problems mentioned in the background art.

[0006] The present invention is implemented as follows: a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping includes the following steps:

[0007] Step 1: Collect road images and vehicle speed information in front of the vehicle using onboard vision sensors;

[0008] Step 2: Identify the road surface roughness level in different areas of the road image based on a convolutional neural network;

[0009] Step 3: Based on the identified road surface roughness level and vehicle speed information, query the preset vehicle speed-road surface roughness-stiffness mapping MAP table to obtain the corresponding target suspension stiffness;

[0010] Step 4: Adjust the vehicle's active suspension in real time according to the target suspension stiffness.

[0011] In a further technical solution, in step 1, the road image information format is RGB three-channel data format, and the resolution is set to [resolution value missing]. Pixel.

[0012] In a further technical solution, in step 2, the road image is divided into 9 regions in 3 rows and 3 columns, and the road surface unevenness level is identified for each region.

[0013] In a further technical solution, in step 2, the convolutional neural network includes a pre-trained EfficientNet-b2 network model and an output head; the output head sequentially includes a global pooling layer, a fully connected layer, and a Softmax activation layer.

[0014] A further technical solution is to classify the road surface roughness into four levels:

[0015] Class I pavement has a pavement roughness of less than or equal to 3.5 m / km, corresponding to Class A and Class B pavement in ISO standards;

[0016] Class II pavement has a pavement roughness greater than 3.5 m / km and less than or equal to 6 m / km, corresponding to Class C pavement in ISO standards;

[0017] Class III pavement has a pavement roughness greater than 6 m / km and less than or equal to 10 m / km, corresponding to Class D pavement in ISO standards.

[0018] Class IV pavement has a surface roughness greater than 10 m / km and less than or equal to 16 m / km, corresponding to Class E pavement in the ISO standard.

[0019] A further technical solution, the process of constructing the vehicle speed-road surface roughness-stiffness mapping (MAP) table, is as follows:

[0020] First, multiple working points are constructed based on multiple preset discrete road surface roughness levels and multiple discrete vehicle speeds.

[0021] Then, for each working condition, a time-domain stochastic excitation model is generated based on the road surface roughness parameters and vehicle speed corresponding to the road surface roughness level.

[0022] Subsequently, based on the dynamic model of the variable stiffness suspension including sprung mass, unsprung mass, variable stiffness suspension and vehicle quarter suspension, and with the weighted multi-objective comprehensive cost function that comprehensively optimizes ride comfort, handling stability and structural safety as the optimization objective, a genetic algorithm is used to optimize the stiffness value of the variable stiffness suspension to obtain the optimal stiffness value at this working condition.

[0023] Finally, the optimal stiffness values ​​for all operating conditions are stored in a structured manner to form the vehicle speed-road roughness-stiffness mapping MAP table.

[0024] A further technical solution, the specific steps for generating a time-domain stochastic excitation model are as follows:

[0025] First, determine the road surface roughness coefficient based on the road surface roughness grade. and reference space frequency ;

[0026] Then, spatial frequencies are constructed according to the formula. The power spectral density function of road surface displacement :

[0027]

[0028] in, It is the frequency index;

[0029] After that, based on vehicle speed According to the mapping relationship spatial frequency Convert to time frequency Based on the principle of energy conservation, the time-domain power spectral density is obtained. :

[0030]

[0031] Finally, the continuous power spectrum is discretized using spectral representation, and the number of discrete intervals is set. Calculate the spatial frequency interval. ;No. The discrete spatial frequency points are ; generated in Uniformly distributed random phase sequence within the interval This random phase sequence represents the random characteristics of road surface unevenness; by superimposing triangular series, a time-varying phase sequence is constructed. Time-domain random displacement excitation of changing road surface For each time step ,calculate:

[0032] .

[0033] A further technical solution is to construct the dynamic equations of the vehicle's quarter suspension dynamic model as follows:

[0034]

[0035]

[0036] in, For suspension stiffness; For the sprung mass, Let be the vertical displacement of the mass on the spring. Let be the vertical displacement of the unsprung mass. For unsprung mass, The road surface is subjected to random displacement excitation in the time domain; This is the equivalent damping coefficient of the suspension; This is the equivalent radial stiffness of the tire.

[0037] Further technical solutions include weighted multi-objective comprehensive cost functions. as follows:

[0038]

[0039] in, , and These are the weighting coefficients for each item; This is the root mean square value of the acceleration of the sprung mass. This represents the root mean square value of the suspension dynamic travel. This represents the root mean square value of the tire dynamic load.

[0040] The present invention provides a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping, which has the following advantages:

[0041] (1) Low hardware cost and computing power requirements: This invention only requires vehicle-mounted vision sensors to collect road surface images and uses lightweight convolutional neural networks to classify road surface levels. It does not require high-precision three-dimensional geometric information perception and complex real-time optimization calculations. In the online stage, only table lookup is required, which greatly reduces the dependence on expensive sensors and high-computing processors and is conducive to engineering promotion.

[0042] (2) Fast and reliable control response: By completing the complex multi-objective optimization calculation offline, the online control is simplified to efficient image classification and table lookup operations, with fast response speed and high determinism, which meets the real-time requirements of vehicle driving.

[0043] (3) Significantly optimized overall performance: Based on the road surface frequency domain characteristics and vehicle dynamics model, the MAP table obtained by offline global optimization through genetic algorithm can ensure that the suspension stiffness is in a better state of comprehensive balance between ride comfort, handling stability and safety under various combinations of vehicle speed and road conditions. Attached Figure Description

[0044] Figure 1 A functional architecture diagram of a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping provided in an embodiment of the present invention;

[0045] Figure 2 A flowchart of a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping is provided for an embodiment of the present invention;

[0046] Figure 3 This is an example result image of predicting road surface roughness level based on convolutional neural network;

[0047] Figure 4 This is an example result of the vehicle speed-road roughness level-stiffness mapping table obtained using a genetic algorithm;

[0048] Figure 5 Example diagrams showing the simulated speed and road conditions used;

[0049] Figure 6 Comparative results of indicators such as sprung mass acceleration power spectral density, sprung mass acceleration time-domain response, suspension dynamic travel time-domain response, and stiffness adjustment are provided.

[0050] Figure 7 The bar chart provides a comparative example of the root mean square values ​​of sprung mass acceleration, suspension dynamic travel, and tire dynamic load. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0053] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a pre-aiming active suspension stiffness control method based on road surface classification and MAP mapping, comprising the following steps:

[0054] Step 1: Acquire road images, vehicle speed, and other relevant status information using the vehicle-mounted vision sensor. The road image information is in RGB three-channel data format, and the resolution is set to [resolution value missing]. Pixel;

[0055] Step 2: Based on the convolutional neural network, identify the road surface unevenness level in different regions of the acquired road image. Here, the road image is divided into 9 regions (3 rows and 3 columns) for identification. The convolutional neural network part includes a pre-trained convolutional neural network model, specifically the Efficient-b2 network model, and an output head containing a three-layer network consisting of a global pooling layer, a fully connected layer, and a softmax activation layer.

[0056] The road surface grades and their corresponding road surface roughness grades are as follows: Grade I road surface roughness range is less than or equal to 3.5 m / km, corresponding to Grade A and Grade B road surfaces in the ISO standard; Grade II road surface roughness range is less than or equal to 6 m / km and greater than 3.5 m / km, corresponding to Grade C road surfaces in the ISO standard; Grade III road surface roughness range is less than or equal to 10 m / km and greater than 6 m / km, corresponding to Grade D road surfaces in the ISO standard; Grade IV road surface roughness range is less than or equal to 16 m / km and greater than 10 m / km, corresponding to Grade E road surfaces in the ISO standard.

[0057] like Figure 3 The table shows an example result of road surface unevenness prediction based on a convolutional neural network in this embodiment. The network parameters are shown in Table 1.

[0058] Table 1 Network Parameters

[0059] Parameter Name Parameter Value Model Architecture efficientnet-b2 Input Resolution 224,224 Whether to use pre-trained weights True Whether to freeze the backbone network True Batch Size 64 Optimizer Adam Learning Rate 1e-4 Number of Hidden Layer Nodes 256 Loss Function NLLLoss Regularization Rate 0.4

[0060] Step 3: Based on the obtained road surface roughness level and vehicle speed information, find the target suspension stiffness by referring to the preset vehicle speed-road surface roughness-stiffness mapping MAP table. The following focuses on explaining the process of constructing the preset vehicle speed-road surface roughness-stiffness mapping MAP table.

[0061] First, a time-domain stochastic excitation model is constructed considering the road surface roughness level and speed information, with a detailed introduction using a Class II roughness road surface as an example. Class II roughness road surface corresponds to Class C road surface in the ISO 8608 standard, and its roughness coefficient... m 3 Reference space frequency m -1 Frequency index Construct spatial frequency based on the formula The power spectral density function of road surface displacement :

[0062]

[0063] A time-space frequency mapping relationship is established using vehicle speed information, with vehicle speed as the basis. Calculations are performed based on km / h. spatial frequency Convert to time frequency Based on the principle of energy conservation, the time-domain power spectral density is obtained. :

[0064]

[0065] The continuous power spectrum is discretized using spectral representation, and the number of discrete intervals is set. Calculate the spatial frequency interval. . No. The discrete spatial frequency points are Generated in Uniformly distributed random phase sequence within the interval This random phase sequence represents the random characteristics of road surface unevenness. By superimposing triangular series, a time-varying phase sequence is constructed. Time-domain random displacement excitation of changing road surface Set the simulation duration. s, time step s, for each time step ,calculate:

[0066]

[0067] Similarly, by iterating through all vehicle speeds and road surface unevenness levels, the time-domain random excitation of the road surface under all working conditions is calculated.

[0068] Next, a dynamic model of the vehicle's quarter-suspension is established. The key physical quantities of the model are as follows: sprung mass is 277.5 kg; unsuspension mass is 35.7 kg; tire equivalent radial stiffness is 107200 N / m; and the suspension base damping coefficient is 2000 N·s / m. Based on Newton's second law, the dynamic equations of the system are constructed as follows:

[0069]

[0070]

[0071] in, For suspension stiffness, its physical constraint range is set to [15000, 80000] N / m; For the sprung mass, Let be the vertical displacement of the mass on the spring. Let be the vertical displacement of the unsprung mass. For unsprung mass, The road surface is subjected to random displacement excitation in the time domain; This is the equivalent damping coefficient of the suspension; This is the equivalent radial stiffness of the tire.

[0072] To determine the multi-objective optimization function and balance vehicle ride comfort, handling stability, and structural safety under different operating conditions, this embodiment constructs a weighted multi-objective comprehensive cost function. :

[0073]

[0074] in, , and These are the weighting coefficients for each item. , , ; This is the root mean square value of the acceleration of the sprung mass. This represents the root mean square value of the suspension dynamic travel. This represents the root mean square value of the tire dynamic load.

[0075] Genetic algorithm is used to measure suspension stiffness Perform optimization under all operating conditions. The specific process is as follows:

[0076] 1. Divide the vehicle driving conditions into a grid, where the road surface unevenness level is divided into level 1, level 2, level 3 and level 4, and the vehicle driving speed is divided into 20 km / h, 40 km / h, 60 km / h, 80 km / h, 100 km / h and 120 km / h.

[0077] 2. Population initialization: Set the population size to 50 and the maximum number of iterations to 50. Each individual represents a candidate stiffness value. The initial value is randomly generated within the range of [15000, 80000] N / m.

[0078] 3. For each discrete working condition (e.g., secondary road surface, 20 km / h), generate the corresponding time-domain random displacement excitation of the road surface using the method described in step 3. The candidate stiffness values ​​in the current population. Substituting the values ​​into the vehicle dynamics state-space model, a time-domain simulation with a duration of 20 s and a step size of 0.001 s was performed. Based on the state response output from the simulation, the aforementioned comprehensive cost function was calculated. The value of .

[0079] 4. The next generation of population is generated through selection, crossover, and mutation operations, so that... The value continuously decreases and converges. Among them, the crossover probability... The initial value of the mutation probability is The algorithm will automatically adjust according to the algebra, with a convergence accuracy of 1e-4. When the objective function value is less than this value, it will stop evolving.

[0080] 5. Output the optimal stiffness value that minimizes the overall cost under this specific working condition. .

[0081] 6. The optimal stiffness values ​​obtained from traversing all working points are stored in a structured manner, forming a two-dimensional mapping (MAP) table, such as... Figure 4 As shown.

[0082] Figure 5 The set speed variation and road grade variation curves are displayed. Under this condition, simulations are performed on four suspension control strategies: fixed stiffness and fixed damping, variable stiffness and fixed damping, fixed stiffness and variable damping, and variable stiffness and variable damping.

[0083] Figure 6 The results show a comparison of the dynamic responses of the four control strategies in the time and frequency domains. The figures include the power spectral density of sprung mass acceleration, the time-domain response of sprung mass acceleration, the time-domain response of suspension travel, and the stiffness adjustment process curves. Observing the power spectral density curves, it can be seen that compared to the traditional fixed stiffness control strategy, the control strategy incorporating a stiffness adjustment mechanism exhibits lower energy amplitudes in both the low-frequency resonance region and the high-frequency range, effectively suppressing vehicle body vibration. Simultaneously, the time-domain response results show that the variable stiffness strategy significantly reduces the fluctuation amplitude of suspension travel, verifying the significant advantage of this strategy in improving vehicle ride comfort.

[0084] Figure 7 The results of comparing four suspension control strategies on three key evaluation indicators were statistically analyzed: the root mean square value of sprung mass acceleration (reflecting comfort), the root mean square value of suspension dynamic travel (reflecting safety), and the root mean square value of tire dynamic load (reflecting handling stability). The bar chart data shows that, after using a low-cost visual road feature recognition-based stiffness adaptive adjustment strategy, especially when employing variable stiffness and variable damping coordinated control, the root mean square values ​​of all three key indicators exhibit a significant decreasing trend. This result confirms the effectiveness of the proposed method.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the stiffness of a pre-aiming active suspension based on road surface classification and MAP mapping, characterized in that, Includes the following steps: Step 1: Collect road images and vehicle speed information in front of the vehicle using onboard vision sensors; Step 2: Identify the road surface roughness level in different areas of the road image based on a convolutional neural network; Step 3: Based on the identified road surface roughness level and vehicle speed information, query the preset vehicle speed-road surface roughness-stiffness mapping MAP table to obtain the corresponding target suspension stiffness; Step 4: Adjust the vehicle's active suspension in real time according to the target suspension stiffness.

2. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 1, characterized in that, In step 1, the road image information is in RGB three-channel data format, and the resolution is set to [resolution value missing]. Pixel.

3. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 1, characterized in that, In step 2, the road image is divided into 9 regions in 3 rows and 3 columns, and the road surface unevenness level is identified for each region.

4. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 1, characterized in that, In step 2, the convolutional neural network includes a pre-trained EfficientNet-b2 network model and an output head; the output head sequentially includes a global pooling layer, a fully connected layer, and a Softmax activation layer.

5. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 3, characterized in that, The road surface roughness is classified into four levels: Class I pavement has a pavement roughness of less than or equal to 3.5 m / km, corresponding to Class A and Class B pavement in ISO standards; Class II pavement has a pavement roughness greater than 3.5 m / km and less than or equal to 6 m / km, corresponding to Class C pavement in ISO standards; Class III pavement has a pavement roughness greater than 6 m / km and less than or equal to 10 m / km, corresponding to Class D pavement in ISO standards. Class IV pavement has a surface roughness greater than 10 m / km and less than or equal to 16 m / km, corresponding to Class E pavement in the ISO standard.

6. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 1, characterized in that, The process of constructing the vehicle speed-road roughness-stiffness mapping (MAP) table is as follows: First, multiple working points are constructed based on multiple preset discrete road surface roughness levels and multiple discrete vehicle speeds. Then, for each working condition, a time-domain stochastic excitation model is generated based on the road surface roughness parameters and vehicle speed corresponding to the road surface roughness level. Subsequently, based on the dynamic model of the variable stiffness suspension including sprung mass, unsprung mass, variable stiffness suspension and vehicle quarter suspension, and with the weighted multi-objective comprehensive cost function that comprehensively optimizes ride comfort, handling stability and structural safety as the optimization objective, a genetic algorithm is used to optimize the stiffness value of the variable stiffness suspension to obtain the optimal stiffness value at this working condition. Finally, the optimal stiffness values ​​for all operating conditions are stored in a structured manner to form the vehicle speed-road roughness-stiffness mapping MAP table.

7. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 6, characterized in that, The specific steps for generating a time-domain stochastic excitation model are as follows: First, determine the road surface roughness coefficient based on the road surface roughness grade. and reference space frequency ; Then, spatial frequencies are constructed according to the formula. The power spectral density function of road surface displacement : in, It is the frequency index; After that, based on vehicle speed According to the mapping relationship spatial frequency Convert to time frequency Based on the principle of energy conservation, the time-domain power spectral density is obtained. : Finally, the continuous power spectrum is discretized using spectral representation, and the number of discrete intervals is set. ; Calculate spatial frequency interval ;No. The discrete spatial frequency points are ; generated in Uniformly distributed random phase sequence within the interval This random phase sequence represents the random characteristics of road surface unevenness; by superimposing triangular series, a time-varying phase sequence is constructed. Time-domain random displacement excitation of changing road surface For each time step ,calculate: 。 8. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 7, characterized in that, The dynamic equations for the vehicle's quarter suspension dynamic model are constructed as follows: in, For suspension stiffness; For the sprung mass, Let be the vertical displacement of the mass on the spring. Let be the vertical displacement of the unsprung mass. For unsprung mass, The road surface is subjected to random displacement excitation in the time domain; This is the equivalent damping coefficient of the suspension; This is the equivalent radial stiffness of the tire.

9. The anti-sighting active suspension stiffness control method based on road surface classification and MAP mapping according to claim 6, characterized in that, Weighted multi-objective comprehensive cost function as follows: in, , and These are the weighting coefficients for each item. This is the root mean square value of the acceleration of the sprung mass. This represents the root mean square value of the suspension dynamic travel. This represents the root mean square value of the tire dynamic load.

Images