Optimization method and system for air spring

Abstract

The invention relates to the technical field of air suspensions, in particular to an optimization method and system of an air spring. The method comprises the following steps: acquiring multiple groups of dynamic response data of the air spring under different working conditions; establishing a physical model and a prediction model of the air spring according to the dynamic response data, and inputting rigidity and damping output by the prediction model into the physical model, according to a first error value between a total rigidity predicted value output by the physical model and a total rigidity true value of the air spring, a second error value between the theoretical value and the predicted value of the rigidity and a third error value between the theoretical value and the predicted value of the damping, back propagation is carried out on the prediction model, and the total rigidity of the air spring is obtained. Therefore, the prediction model is trained; and based on a trained prediction model, combining the physical model to predict the performance of the air spring. According to the method, the performance prediction precision of the air spring can be improved.

Description

Technical Field

[0001] The present invention generally relates to the technical field of air suspensions and more particularly to an optimization method and system for air springs. Background Art

[0002] With the development of my country's automobile industry and the improvement of people's living standards and needs, cars have become an indispensable tool in daily life, and the public's requirements for car power and comfort are also increasing. Among them, air suspension plays an important role in improving comfort. For example, the air spring in the air suspension can adaptively adjust the height of the car body to achieve the best driving experience.

[0003] Air springs are elastic elements that wrap high-pressure gas with rubber-cord airbags and upper and lower cover pistons. They mainly bear loads through internal air pressure and can provide a certain degree of damping characteristics. Air springs have broad development prospects. At present, when designing air springs for specific models, the performance of air springs is generally predicted through empirical formulas or finite element analysis to optimize their structures. However, both methods have certain limitations. The prediction accuracy of empirical formulas is not high, the calculation amount of finite element analysis is large and time-consuming, and the existing physical models are insufficient in dealing with nonlinear behaviors and complex working conditions of air springs. Summary of the invention

[0004] In order to solve the technical problem of low performance prediction accuracy of the above-mentioned air spring, the present invention provides solutions in the following aspects.

[0005] In a first aspect, the present invention provides an optimization method for an air spring, comprising: obtaining multiple sets of dynamic response data of the air spring under different working conditions, the dynamic response data including the excitation circular frequency, the pressure and volume of the gas in the air spring; establishing a physical model and a prediction model of the air spring according to the dynamic response data, the prediction model being obtained based on a neural network, inputting the stiffness and damping output by the prediction model into the physical model, and back-propagating the prediction model according to a first error value between a predicted total stiffness value output by the physical model and a true total stiffness value of the air spring, a second error value between a theoretical value and a predicted value of the stiffness, and a third error value between a theoretical value and a predicted value of the damping, thereby training the prediction model; and predicting the performance of the air spring based on the trained prediction model in combination with the physical model.

[0006] Furthermore, in the process of back-propagating the prediction model, it includes: back-propagating the prediction model according to the sum of the first error value, the second error value and the third error value.

[0007] Further, the first error value, the second error value and the third error value are all mean square error values.

[0008] Furthermore, the neural network is an LSTM network.

[0009] Furthermore, the physical model is:

[0010]

[0011] Where K(ω) represents the total stiffness of the air spring, k A represents the equivalent stiffness generated by the effective area, k1 represents the stiffness, c1 represents the damping, j represents the imaginary unit, and ω represents the excitation circular frequency.

[0012] Furthermore, the calculation expression of the theoretical value of the stiffness is as follows:

[0013]

[0014] Where k1 is the stiffness, P b0 is the initial pressure of the gas in the air spring, V b0 is the initial volume of the gas in the air spring, A eff is the effective area, and γ is the polynomial index.

[0015] Furthermore, the calculation expression of the theoretical value of the stiffness is as follows:

[0016]

[0017] Where k1 is the stiffness, P v0 is the initial pressure of the gas in the air spring, V b0 is the initial volume of the gas in the air spring, k is the coefficient, z m represents the intermediate coordinate displacement, A eff is the effective area, and γ is the polynomial index.

[0018] Furthermore, the calculation expression of the theoretical value of the damping is as follows:

[0019]

[0020] Where c1 is the damping, C v is the constant volume specific heat capacity of the gas in the air spring, K b is the equivalent heat exchange coefficient between the air spring and the outside world, m b0 is the initial mass of the gas in the air spring, A eff is the effective area, γ is the polynomial index, P b0represents the initial pressure of the gas in the air spring, V b0 represents the initial volume of the gas in the air spring.

[0021] Furthermore, the dynamic response data also includes the height, load, amplitude of the air spring and the temperature of the gas in the air spring.

[0022] In a second aspect, the present invention provides an air spring optimization system, comprising: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an air spring optimization method according to the first aspect is implemented.

[0023] The beneficial effects of the present invention are as follows: the prediction model of the present invention can obtain parameters that the physical model cannot directly measure when dealing with nonlinear behaviors and complex working conditions, thereby improving the accuracy of the air spring performance prediction, and being able to optimize the design of the air spring based on the parameters; further, by integrating the physical model and the prediction model, the prediction accuracy and generalization ability of the prediction model are improved, thereby improving the accuracy of the air spring performance prediction. In addition, the hybrid model obtained by integrating the physical model and the prediction model has strong applicability and is suitable for air springs of different types and under different working conditions, and can also be expanded to a hybrid model of a multi-chamber air spring. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0025] Figure 1 is a flow chart schematically illustrating an optimization method of an air spring according to an embodiment of the present invention;

[0026] Figure 2 is a flow chart schematically illustrating an MTS dynamometer test platform according to an embodiment of the present invention;

[0027] Figure 3 is a structural diagram schematically showing a single-chamber air spring according to an embodiment of the present invention;

[0028] Figure 4 is a framework diagram schematically illustrating a hybrid model according to an embodiment of the present invention;

[0029] Figure 5 Schematically shows the comparison between theoretical value and experimental value of the long-stroke gas frequency characteristic test according to an embodiment of the present invention Figure 1 .

[0030] Figure 6 Schematically shows the comparison between theoretical value and experimental value of the long-stroke gas frequency characteristic test according to an embodiment of the present invention Figure 2 .

[0031] Figure 7 is a structural diagram schematically showing an optimization system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0033] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0034] Figure 1 is a flow chart schematically illustrating an optimization method for an air spring according to an embodiment of the present invention.

[0035] In a first aspect, the present invention provides a method for optimizing an air spring, such as Figure 1 As shown, the method of the present invention comprises:

[0036] S101. Acquire multiple groups of dynamic response data of the air spring under different working conditions.

[0037] In one embodiment, the dynamic response data includes frequency, load, amplitude, height of the air spring, Calvin temperature of the gas in the air spring, pressure of the gas in the air spring, etc. Specifically, multiple sets of dynamic response data of the air spring under different working conditions can be obtained through the constructed MTS (Mechanical Testing & Simulation) dynamometer test platform. Through the dynamometer test, the frequency correlation of the air spring can be determined, and the frequency correlation mainly depends on the heat exchange and stiffness of the gas inside the air spring. The dynamometer test mainly includes two parts: a large-stroke gas frequency characteristic test and a small-stroke bladder skin dynamic stiffness test.

[0038] In one embodiment, the constructed MTS dynamometer test platform is as follows: Figure 2 As shown by Figure 2It can be seen that the MTS dynamometer test platform includes air springs, force sensors, hydraulic actuators, pressure sensors, high-pressure air pipes, pressure gauges and pressure reducing valves. Specifically, before the experiment, the air spring is inflated with a high-pressure air source. After the inflation is completed, the pressure is controlled by the pressure reducing valve, and the air spring is tested by using a strapping tape to limit the pipeline and pipe length. The experimental data can be read through the MTS sensor, including the force sensor value with a fixed sampling frequency, the displacement of the hydraulic actuator, the speed of the hydraulic actuator, time and other data. Among them, the data of the initial state of the air spring is shown in Table 1.

[0039] Table 1 Initial state of air spring

[0040]

[0041] In Table 1, T0 is the initial Calvin temperature of the gas in the air spring, A eff is the effective area, P b0 is the initial pressure of the gas in the air spring, V b0 is the initial volume of the gas in the air spring, f is the frequency (given by the MTS indicator), K b is the equivalent heat exchange coefficient between the air spring and the outside world, z0 is the initial vertical displacement of the air spring, k represents the coefficient, A is the amplitude (given by the corresponding sensor), C V is the specific heat at constant volume of the gas in the air spring.

[0042] It should be noted that the effective area is the cross-sectional area of ​​the air spring when the load-bearing force applied to the air spring is balanced with the air pressure inside the air spring.

[0043] When conducting a large-stroke gas frequency characteristic test, after adjusting the air spring to a balanced position (balance height and internal pressure), a sinusoidal excitation dynamometer test is performed. Specifically, the excitation parameters (amplitude and frequency) of the dynamometer test are set according to the preset values, and all frequency tests are performed for each amplitude to obtain the test results, which can be expressed as an air spring dynamometer hysteresis curve. It can be understood that different working conditions include different amplitudes and different frequencies.

[0044] S102. Establish a physical model and a prediction model of the air spring.

[0045] In one embodiment, a physical model of the air spring is established based on its structure and working principle. Specifically, the physical model of the air spring is:

[0046]

[0047] in,

[0048] Where K(ω) represents the total dynamic stiffness of the air spring, k a represents the equivalent stiffness generated by the effective area, k1 represents the stiffness generated by the gas in the air spring, c1 represents the damping generated by heat exchange, j represents the imaginary unit, ω represents the excitation circular frequency, P b0 is the initial pressure of the gas in the air spring, V b0 is the initial volume of the gas in the air spring, A eff is the effective area, γ is the polynomial index, C v is the constant volume specific heat capacity of the gas in the air spring, K b is the equivalent heat exchange coefficient between the air spring and the outside world, m b0 is the initial mass of the gas in the air spring, and z is the vertical displacement.

[0049] It should be noted that the true value of the total stiffness of the air spring is not obtained based on the above expression, but is obtained based on the hysteresis characteristic curve obtained from the MTS pilot test.

[0050] It should be noted that the excitation circular frequency is determined by the MTS dynamometer, and the equivalent stiffness generated by the effective area is obtained by the MTS dynamometer test. In addition, the calculation expression for the stiffness generated by the gas in the above air spring is based on the stiffness in the initial state and small amplitude (about ±5mm).

[0051] In another embodiment, under large amplitude conditions, due to the nonlinear change of the effective area and volume relative to the amplitude, its stiffness exhibits nonlinear characteristics. To this end, k1 can be expanded into a linear term (coefficient k) with respect to the amplitude using Taylor expansion to accurately express the dynamic stiffness of the air spring under large amplitude (greater than 10 mm). Specifically, under large amplitude conditions, the calculation expression for the stiffness generated by the gas in the air spring is as follows:

[0052]

[0053] In the formula, k1 represents the stiffness of the gas in the air bomb, P b0 Indicates the initial pressure of the gas in the air spring, V b0 represents the initial volume of the gas in the air spring, k represents the coefficient, z m Represents the displacement of the middle coordinate of the air spring.

[0054] It should be noted that the physical model of the present invention is applicable to a single-chamber air spring (without an additional air chamber), such as Figure 3As shown in the figure, the single-chamber air spring is mainly carried by an air bag wrapped with high-pressure gas. The upper end is connected to the cover plate and the lower end is attached to the piston. When the air spring adjusts its height, the rubber air bag and the piston surface curl, and its equivalent stiffness will also change. During the inflation process, the high-pressure gas enters the air spring from the gas tank or compressor through the pipeline; during the deflation process, the high-pressure gas inside the air bag will be directly discharged into the outside atmosphere through the control of the solenoid valve.

[0055] It should be noted that the above-mentioned pressure is absolute pressure, and it is stipulated that the gas mass flow rate is positive when it flows into the additional gas chamber, and the changes in gas pressure, temperature, mass and volume are positive when they increase; displacement and force are positive when they are upward.

[0056] In one embodiment, the prediction model can be obtained based on a neural network. In this embodiment, the neural network used is an LSTM network. Specifically, the LSTM network has 2 hidden layers, each of which has 256 neural units. The number of nodes in the input layer of the prediction model is the number of types of corresponding dynamic response data, which is 6 in this embodiment; the number of nodes in the output layer of the prediction model is 2, namely the stiffness generated by the gas in the air spring and the damping generated by heat exchange.

[0057] In one embodiment, the acquired dynamic response data is preprocessed, such as denoising, normalization, etc., to ensure the quality and consistency of the data, thereby improving the accuracy and stability of the prediction model.

[0058] S103: Fusing the physical model and the prediction model of the air spring to obtain a hybrid model.

[0059] In one embodiment, the framework of the hybrid model is as follows Figure 4 As shown. Figure 4 It can be seen that in the hybrid model, the dynamic response data (amplitude of the demonstration test) frequency Initial Load Initial pressure of gas in air spring Initial Calvin temperature of the gas in the air spring Initial height of air spring ) are input into the prediction model and the physical model respectively.

[0060] Furthermore, the stiffness generated by the gas in the air spring and the damping generated by the heat exchange output by the prediction model are input into the physical model, and then the physical model outputs a total stiffness prediction value based on the stiffness generated by the gas in the air spring, the damping generated by the heat exchange, the equivalent stiffness generated by the effective area and the excitation circular frequency.

[0061] In one embodiment, the prediction model is back-propagated and iterated according to the first error value between the predicted value of the total stiffness output by the physical model and the true value of the total stiffness of the air spring (obtained by the MTS device), the second error value between the theoretical value of the stiffness generated by the gas in the air spring and the predicted value (i.e., the stiffness output by the prediction model), and the third error value between the theoretical value of the damping generated by the heat exchange and the predicted value (i.e., the damping output by the prediction model). In this embodiment, the first error value, the second error value, and the third error value are summed, and the prediction model is back-propagated according to the summed total error value, so as to train the prediction model, that is, optimize the parameters in the prediction model so that it can obtain stiffness and damping more accurately, so that the prediction accuracy of the physical model under dynamic conditions is optimized.

[0062] When back-propagating the prediction model, compared with only considering the first error value, the present invention improves the accuracy and generalization ability of the hybrid model in predicting the performance of the air spring by comprehensively considering the first error value, the second error value and the third error value.

[0063] In one embodiment, the calculation expression of the theoretical value of the stiffness generated by the gas in the air spring is as follows:

[0064] or

[0065]

[0066] In one embodiment, the calculation expression of the theoretical value of the damping generated by heat exchange is as follows:

[0067]

[0068] In one embodiment, the first error value, the second error value and the third error value are all mean square error values ​​(MSE). In other optional embodiments, the error value may also be a mean absolute error value (MAE) or the like.

[0069] It can be understood that the stiffness and damping input into the physical model are parameters that cannot be directly measured. The present invention obtains the stiffness and damping through a neural network, which makes up for the shortcomings of the existing physical model in dealing with the nonlinear behavior and complex working conditions of the air spring, thereby being able to more accurately characterize the dynamic stiffness of the air spring.

[0070] The physical model in the hybrid model of the present invention has clear physical meaning and mathematical expression, and is simple in form. The hybrid model can reflect the frequency characteristics and hysteresis characteristics of the dynamic stiffness of the air spring under different working conditions. Among them, the prediction model effectively compensates for the shortcomings of the existing physical model in dealing with nonlinear behavior and complex working conditions, reduces the prediction error, and the prediction model obtained based on the neural network shortens the calculation time and cost, and improves the efficiency of the performance prediction of the air spring.

[0071] Furthermore, since the hybrid model of the present invention does not constrain the internal gas change process, it has strong universality, and the hybrid model of the present invention has strong scalability. For example, the prediction model can be expanded accordingly according to the structure and characteristics of the multi-chamber air spring (the prediction parameters output by the prediction model, in addition to the stiffness generated by the gas in the air spring and the damping generated by heat exchange, also include the equivalent stiffness and equivalent damping generated by the additional air chamber, etc.), and the physical model is also changed accordingly (when determining the predicted value of the total stiffness of the air spring, the stiffness and damping generated by the additional air chamber are added accordingly), thereby obtaining a hybrid model suitable for a multi-chamber air spring.

[0072] S104. Predict the performance of the air spring based on the hybrid model.

[0073] Specifically, the new dynamic response data is input into the trained prediction model, combined with the physical model, to obtain the dynamic stiffness prediction value corresponding to the dynamic response data, and then the performance of the air spring is determined based on the dynamic stiffness prediction value. Furthermore, the structure and material of the air spring (these parameters that affect the input hybrid model) can be adjusted according to the performance of the air spring to optimize the performance of the air spring (i.e., the total stiffness prediction value output by the hybrid model is optimal), thereby optimizing the air spring. The above process optimizes the air spring as a whole. Locally, the load-bearing capacity and elasticity of the air spring can be determined based on the stiffness output by the prediction model, and the shock absorbing effect and response characteristics of the air spring can be determined based on the damping output by the prediction model, so that the air spring can be optimized based on this information.

[0074] By predicting the damping and stiffness output by the model, the performance of the air spring can be determined in a targeted manner, and the air spring can be optimized accordingly.

[0075] In one embodiment, the hybrid model considering the hysteresis characteristics can well reflect the dynamic characteristics of the air spring, and the relative error with the experimental value is less than 0.7% under large stroke, which provides a good foundation for subsequent precise control of air suspension and vehicle simulation and experimental verification.

[0076] In one embodiment, the comparison results of the theoretical value and the experimental value of the long-stroke gas frequency characteristic test are as follows: Figure 5 and Figure 6 As shown, Figure 5 This is the comparison result when the excitation frequency is 0.02Hz. Figure 6 This is the comparison result when the excitation frequency is 0.2Hz. Figure 5 and Figure 6 It can be seen that the hybrid model described in the present invention can accurately describe the hysteresis characteristics and dynamic behavior of the air spring during movement, and the degree of consistency with the experiment is high, indicating that the hybrid model has high generalization ability and prediction accuracy.

[0077] Figure 7 Schematically shows a structural block diagram of an air spring optimization system according to this embodiment.

[0078] In a second aspect, the present invention also provides an optimization system for an air spring. Figure 7 As shown, the system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an air spring optimization method according to the first aspect of the present invention is implemented.

[0079] The system also includes other components well known to those skilled in the art, such as a communication interface, whose configuration and functions are known in the art and thus will not be described in detail here.

[0080] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that may be used to store the required information and may be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0081] In the description of this specification, "multiple groups" means at least two groups, such as two groups, three groups or more groups, etc., unless otherwise clearly and specifically defined.

[0082] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. An optimization method for an air spring, characterized in that: include: Acquire multiple sets of dynamic response data of the air spring under different working conditions, wherein the dynamic response data includes an excitation circle frequency, and a pressure and a volume of gas in the air spring; A physical model and a prediction model of the air spring are established according to the dynamic response data, wherein the prediction model is obtained based on a neural network, and the stiffness and damping output by the prediction model are input into the physical model. According to a first error value between a predicted total stiffness value output by the physical model and a true total stiffness value of the air spring, a second error value between a theoretical value and a predicted value of the stiffness, and a third error value between a theoretical value and a predicted value of the damping, the prediction model is back-propagated, thereby training the prediction model; Based on the trained prediction model and combined with the physical model, the performance of the air spring is predicted.

2. The air spring optimization method according to claim 1, characterized in that: The process of back-propagating the prediction model includes: back-propagating the prediction model according to the sum of the first error value, the second error value and the third error value.

3. The air spring optimization method according to claim 1, characterized in that: The first error value, the second error value and the third error value are all mean square error values.

4. The method for optimizing an air spring according to claim 1, characterized in that: The neural network is an LSTM network.

5. The method for optimizing an air spring according to claim 1, characterized in that: The physical model is: Where K(ω) represents the total stiffness of the air spring, k A represents the equivalent stiffness generated by the effective area, k1 represents the stiffness, c1 represents the damping, j represents the imaginary unit, and ω represents the excitation circular frequency.

6. The method for optimizing the design of an air spring according to claim 5, characterized in that: The calculation expression of the theoretical value of the stiffness is as follows: Where k1 is the stiffness, P b0 is the initial pressure of the gas in the air spring, V b0 is the initial volume of the gas in the air spring, A eff is the effective area, and γ is the polynomial index.

7. The method for optimizing an air spring according to claim 5, characterized in that: The calculation expression of the theoretical value of the stiffness is as follows: Where k1 is the stiffness, P b0 is the initial pressure of the gas in the air spring, V b0 is the initial volume of the gas in the air spring, k is the coefficient, z m represents the intermediate coordinate displacement, A eff is the effective area, and γ is the polynomial index.

8. The method for optimizing an air spring according to claim 5, characterized in that: The calculation expression of the theoretical value of the damping is as follows: Where c1 is the damping, C v is the constant volume specific heat capacity of the gas in the air spring, K b is the equivalent heat exchange coefficient between the air spring and the outside world, m b0 is the initial mass of the gas in the air spring, A eff is the effective area, γ is the polynomial index, P b0 Represents the initial pressure of the gas in the air spring, V b0 represents the initial volume of the gas in the air spring.

9. The method for optimizing the design of an air spring according to claim 1, characterized in that: The dynamic response data also includes the height, load, amplitude of the air spring and the temperature of the gas in the air spring.

10. An optimization design system for an air spring, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an optimization design method for an air spring according to any one of claims 1-9 is implemented.

Images