An iterative learning based wind tunnel model vibration suppression system control parameter adaptive adjustment method

Abstract

The present application belongs to the field of aircraft wind tunnel test measurement and control technology, and discloses a wind tunnel model vibration suppression system control parameter adaptive adjustment method based on iterative learning. The steps of the method are as follows: firstly, an iterative learning optimization framework is constructed, and a performance index function is designed; secondly, the damping ratio of the wind tunnel model vibration suppression system is identified online based on the half-power bandwidth method; finally, an iterative update law of the controller parameters in the wind tunnel model vibration suppression system based on data driving is designed, the gradient of the performance index to the controller parameters is estimated by the difference method, and the gradient descent method is used to automatically adjust the control parameters to minimize the performance index. The present application does not depend on the accurate model of the controlled object, takes the clear physical index as the optimization target, realizes the adaptive and efficient setting of the control parameters, and significantly improves the vibration suppression effect and efficiency of the wind tunnel test.

Description

Technical Field

[0001] This invention belongs to the field of wind tunnel test and control technology for aircraft, and relates to an adaptive adjustment method for control parameters of a wind tunnel model vibration suppression system based on iterative learning. Background Technology

[0002] In the development of aircraft, wind tunnel testing is a crucial step in acquiring aerodynamic data, and the quality of this data directly impacts the aircraft's design and performance evaluation. However, during wind tunnel testing, the interaction between high-speed airflow and the model can induce varying degrees of vibration in the model and its support system. This vibration not only interferes with the accuracy of force and pressure measurements, leading to measurement errors, but in severe cases, it can also threaten the safety of the model and wind tunnel equipment. This is particularly true for flexible models such as high-aspect-ratio wings, where aeroelastic effects are more pronounced, making vibration problems especially prominent.

[0003] To effectively suppress model vibration during wind tunnel testing, active vibration suppression technology is widely used. The basic principle of this technology is to use sensors (such as accelerometers) to sense model vibration in real time, then a controller processes the sensor signals and drives actuators (such as piezoelectric ceramics) to apply a reverse control force, thereby increasing system damping and achieving vibration suppression. The performance of the vibration suppression system depends primarily on the tuning quality of the controller parameters (such as the proportional, integral, and derivative coefficients in PID control).

[0004] Currently, there are several main methods for parameter tuning of wind tunnel vibration suppression systems: one is the manual "trial and error" method that relies on engineers' experience. This method is simple and direct, but inefficient, and the tuning results are heavily dependent on personal experience, making it difficult to guarantee optimal performance; the other is the offline simulation calculation method based on the accurate mathematical model of the controlled object. This method requires the establishment of an accurate system mathematical model during the design phase. However, the wind tunnel test environment is complex, and the dynamic characteristics of the model will change in real time with factors such as airflow velocity, angle of attack, and sideslip angle. In addition, there are many nonlinear elements that are difficult to model accurately (such as actuator saturation, connection gaps, etc.), making it difficult for controllers based on fixed parameters to maintain optimal vibration suppression performance throughout the entire test process.

[0005] In the existing technology, some adaptive control methods have been proposed to address the problems mentioned above, but most of them have certain limitations. For example, Model Reference Adaptive Control (MRAC) requires a known model structure of the system; Self-Tuning Control (STC) requires complex online parameter identification; and some modern intelligent control algorithms, such as fuzzy control and neural network control, although not dependent on an accurate model, usually require expert knowledge to design a rule base or a large amount of training data, and the correlation between controller parameters and actual physical indicators (such as damping ratio) is not direct, and the physical meaning is unclear.

[0006] A key characteristic of wind tunnel testing is its ability to maintain a stable excitation environment for extended periods, allowing for testing under specific conditions for considerable durations. This characteristic provides a natural advantage for applying Iterative Learning Control (ILC). ILC is a learning algorithm that progressively improves the performance of a control system through continuous testing. It utilizes information from previous tests to refine the control strategy for the current test, making it particularly suitable for the dynamic processes inherent in wind tunnels, which can be run multiple times under stable conditions. However, traditional ILC methods often directly optimize the control input signal rather than the controller parameters. For applications such as wind tunnel model vibration suppression, where the system's dynamic characteristics may change, directly optimizing the controller parameters is more practically meaningful.

[0007] Therefore, developing an adaptive adjustment method that does not rely on an accurate model of the controlled object, can automatically and efficiently tune control parameters based on experimental data, and uses clearly defined physical indicators as optimization targets, has become an urgent need to improve the quality and efficiency of wind tunnel test data. This invention addresses this need by proposing an adaptive adjustment method for control parameters of a wind tunnel model vibration suppression system based on iterative learning. Summary of the Invention

[0008] The main technical problem addressed by this invention is to overcome the shortcomings of existing methods for tuning control parameters of wind tunnel model vibration suppression systems. It provides an adaptive adjustment method for control parameters of wind tunnel model vibration suppression systems based on iterative learning. This method fully utilizes the repeatability of wind tunnel tests, taking the improvement of the closed-loop system damping ratio as a clear optimization objective. Through data-driven methods, it automatically and iteratively optimizes the controller parameters, achieving online self-improvement of vibration suppression performance. This significantly reduces reliance on manual experience and ensures excellent and consistent vibration suppression effects under different test conditions.

[0009] The technical solution of the present invention:

[0010] An adaptive adjustment method for control parameters of a wind tunnel model vibration suppression system based on iterative learning is proposed. The core idea is to formalize the controller parameter tuning process as an iterative optimization problem, gradually approximating the optimal parameters through multiple experiments. First, an iterative learning optimization framework is constructed, the optimization objective is determined, and a performance index function is established. Second, the damping ratio of the wind tunnel model vibration suppression system is calculated in real time based on the half-power bandwidth method. Then, a data-driven iterative update law for the controller parameters in the wind tunnel model vibration suppression system is designed, forming a control parameter adjustment strategy.

[0011] The specific steps are as follows:

[0012] Step 1: Construct an iterative learning optimization framework and design performance metric functions;

[0013] Iterative learning optimization framework: The adaptive adjustment process of controller parameters in a wind tunnel model vibration suppression system is constructed as a closed-loop optimization problem driven by iteration and with physical performance indicators as the optimization objective. First, the controller parameter vector θ in the wind tunnel model vibration suppression system to be adjusted is defined; then, a performance index function for quantitatively evaluating the vibration suppression effect is designed. Finally, an iterative update law for controller parameters in a data-driven wind tunnel model vibration suppression system is established for automatic optimization.

[0014] The objective of the aforementioned closed-loop optimization problem is to find the optimal solution for the controller parameter vector θ, such that the performance index J(θ) of the wind tunnel model vibration suppression system is minimized under excitation signals (e.g., 5-50Hz sinusoidal sweep) covering the key modes (e.g., first and second modes) of the system. The function for this performance index is constructed as follows:

[0015]

[0016] Among them, superscript This indicates the k-th iteration. Let J(θ) represent the performance index value corresponding to the k-th iteration; For damping ratio optimization; This is a vibration amplitude constraint term; For parameter variation regularization;

[0017] Damping ratio optimization term Defined as:

[0018]

[0019] in, The damping ratio of the m-th mode identified online in the k-th iteration; The target damping ratio for the m-th mode is preset according to the test requirements; α is the weighting coefficient; M is the total number of modes;

[0020] Vibration amplitude constraint Defined as:

[0021]

[0022] in, Let N be the discrete acceleration signal sequence acquired in the k-th iteration, where N is the number of samples in the discrete acceleration signal sequence, and β is the weighting coefficient.

[0023] Parameter variation regularization term Defined as:

[0024]

[0025] in, These are the weighting coefficients;

[0026] Step 2: Identify the damping ratio of the wind tunnel model vibration suppression system online based on the half-power bandwidth method;

[0027] The vibration acceleration signal of the wind tunnel model vibration suppression system is acquired in real time. The power spectral density of the vibration acceleration signal is calculated by fast Fourier transform to obtain the power spectral density curve, and the resonant frequency f of the dominant mode of the wind tunnel model vibration suppression system is identified. n And the corresponding peak value; on both sides of the peak value corresponding to the resonance frequency, find the frequency points f1 and f2 on the power spectral density curve where the power value drops to half of the above peak value; then the damping ratio ζ of the dominant mode of the wind tunnel model vibration suppression system is as follows:

[0028]

[0029] in, For half-power bandwidth, satisfy In actual online processing, parabolic interpolation or Sinc interpolation is used to process the power spectral density curve.

[0030] Step 3: Design the iterative update law for controller parameters in the data-driven wind tunnel model vibration suppression system;

[0031] Based on performance metrics An iterative learning control method is adopted, which updates the rules iteratively and finds ways to reduce performance indicators based on historical experimental data. The adjustment direction of the controller parameters in the wind tunnel model vibration suppression system is as follows: The update of the controller parameters in the wind tunnel model vibration suppression system follows the iterative format:

[0032]

[0033] in, and These are the controller parameter vectors for the (k+1)th and kth iterations, respectively. This represents the parameter update amount calculated in the k-th iteration;

[0034] Performance indicators Approximate negative gradient determination of controller parameters in a wind tunnel model vibration damping system:

[0035]

[0036] in, To learn the gain, control the update step size; The performance metric at the k-th iteration Compared to the controller parameter vector θ in the wind tunnel model vibration suppression system (k) gradient The estimated value;

[0037] Since the model of the wind tunnel vibration suppression system is unknown, the analytical gradient... Since the gradient cannot be obtained directly, a difference method based on experimental data is used to obtain the gradient. The specific steps for numerical estimation are as follows:

[0038] a. In the k-th iteration, the controller parameter vector θ in the k-th iteration (k) Wind tunnel tests were conducted, and the results were calculated based on the performance index function defined in the first step. ;

[0039] b. The controller parameter vector θ for the k-th iteration (k) Apply a pre-defined controller parameter vector θ with a disturbance amplitude equal to that of the k-th iteration. (k) The controller parameters θ in the wind tunnel model vibration suppression system are obtained by applying a disturbance of 1% to 5% δθ. (k) +δθ;

[0040] c. Controller parameter θ in the wind tunnel model vibration suppression system (k) The experiment was conducted again at +δθ, and the new performance index values ​​were calculated. ;

[0041] d. Estimate the gradient direction using the first-order forward difference formula:

[0042]

[0043] The beneficial effect of this invention is that it provides an adaptive adjustment method for control parameters of a wind tunnel model vibration suppression system based on iterative learning. Compared with the prior art, this invention has the following significant advantages:

[0044] 1. Clear Objective and Physical Meaning: Using the "damping ratio," a physical quantity that directly reflects the vibration attenuation capability of the wind tunnel model's vibration suppression system, as the core optimization objective is superior to simply minimizing the vibration amplitude, making the optimization process more scientific and directional. As an inherent characteristic of the wind tunnel model's vibration suppression system, the damping ratio has a clear correlation with the parameters of the controller in the system, which is beneficial for the convergence of the optimization algorithm.

[0045] 2. Data-driven, not reliant on precise mathematical models: Employing iterative learning and differential gradient estimation, this invention is entirely data-driven, avoiding the difficulty of establishing complex and precise mathematical models of the controlled object, thus demonstrating strong practicality. This feature makes the invention particularly suitable for complex environments such as wind tunnel tests, where precise modeling is difficult.

[0046] 3. Strong Adaptability: It can adapt to changes in the dynamic characteristics of the wind tunnel model vibration suppression system caused by variations in wind tunnel model attitude and flow velocity during wind tunnel testing. Through iterative learning, it automatically adjusts parameters to maintain excellent vibration suppression performance. Compared with traditional fixed-parameter controllers, it significantly expands the effective working range of the vibration suppression system.

[0047] 4. High degree of automation, reducing reliance on expert experience: The method automates the control parameter tuning process, reduces operational difficulty, and improves the efficiency and consistency of wind tunnel testing. Even non-control experts can achieve good vibration suppression results using this invention.

[0048] 5. Improved Convergence Stability: The regularization term introduced in the performance metrics effectively prevents divergence risks during parameter updates, ensuring the stability and convergence of the learning process. The gradient descent-based update law possesses favorable mathematical properties, guaranteeing convergence to a local optimum under reasonable conditions.

[0049] 6. High computational efficiency and good real-time performance: The half-power bandwidth method and gradient estimation method adopted have moderate computational loads, requiring no complex numerical optimization process, and can meet the real-time requirements of wind tunnel tests. The entire algorithm can be implemented on conventional test and control systems without additional hardware investment. Attached Figure Description

[0050] Figure 1 This is a schematic diagram illustrating the optimization process of an adaptive adjustment method for control parameters of a wind tunnel model vibration suppression system based on iterative learning, according to the present invention.

[0051] Figure 2 This is a diagram of an experimental setup for adaptive adjustment of control parameters in a wind tunnel model vibration suppression system, according to one embodiment of the present invention.

[0052] Figure 3 These are experimental data curves for adaptive adjustment of control parameters in a wind tunnel model vibration suppression system according to one embodiment of the present invention, wherein (a) is the damping ratio iteration optimization curve, (b) is the parameter P iteration optimization curve, and (c) is the parameter D iteration optimization curve.

[0053] Figure 2 The components include: 1. Host computer, 2. Data acquisition card and controller, 3. Piezoelectric amplifier, 4. Piezoelectric ceramic, 5. Wind tunnel model vibration damping device, and 6. Accelerometer. Detailed Implementation

[0054] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0055] Example

[0056] Taking a ground-based simulated wind tunnel experiment of a high aspect ratio aircraft model as the application scenario, the experimental setup is as follows: Figure 2 As shown, the algorithm is implemented in the LabVIEW programming environment. The experimental system mainly includes: a wind tunnel model vibration damping device, piezoelectric ceramics and piezoelectric amplifiers, an accelerometer, a data acquisition card, and a controller. The wind tunnel model vibration damping system controller adopts a PD (proportional-derivative) form, and the parameter vector to be optimized is θ = [K p , K d ] T K p For proportional gain, K d This is the differential gain.

[0057] Step 1: Initialize the vibration damping system of the wind tunnel model;

[0058] Set initial controller parameters, such as θ (0) = [K p =1, K d =0] T Set the target damping ratio ζ. target = 0.2 (this value is a reasonable target determined based on similar wind tunnel test experience), and set the learning rate η=0.01, with 30 iterations. The weight coefficients are set to α=0.6, β=0.3, and γ=0.1.

[0059] Step 2: Conduct experiments and calculate performance indicators;

[0060] For the k-th test (k starts from 0), a sinusoidal sweep signal with a frequency range of 5-50 Hz and an amplitude of 1g is applied to simulate airflow excitation in the wind tunnel. The vibration acceleration signal of the model is collected, with a sampling frequency of 1 kHz and a collection time of 10 seconds to ensure that the complete frequency response characteristics can be captured.

[0061] Real-time spectrum analysis is performed on the acquired signal, and the dominant mode damping ratio ζ of the current closed-loop system is calculated using the half-power bandwidth method according to equation (5). (k) The specific process is as follows: perform FFT transformation on the acceleration signal to obtain the power spectral density curve, identify the dominant resonance peak, find the frequency corresponding to the half-power point, and calculate the damping ratio.

[0062] Next, the system performance index under the current control parameters is calculated. The damping ratio optimization term can be calculated using equation (2). The vibration amplitude constraint term can be calculated using equation (3). The parameter variation regularization term can be calculated using equation (4). Finally, the performance index of the wind tunnel model vibration suppression system is calculated using equation (1).

[0063] Step 3: Iterate through experiments to calculate the gradient until convergence;

[0064] To estimate the gradient of the performance index with respect to the controller parameters, the control parameter θ is... (k) Apply a pre-defined controller parameter vector θ with a disturbance amplitude equal to that of the k-th iteration. (k) The 1% to 5% of the disturbance δθ (such as ΔK) p =0.1, keep K d (If the value remains unchanged), repeat the calculation process in step two to obtain the change in performance indicators. Thus, the gradient estimate is obtained. The control parameters are updated using formulas (6), (7), and (8).

[0065] Repeat the above steps until the preset convergence condition is met. This refers to the performance metrics. The change is less than a preset threshold (e.g., 10). −4 The iterative optimization process terminates when either condition is met, either by the number of iterations reaching a preset maximum value (e.g., 30 times) or by the number of iterations reaching a preset maximum value. The optimal control parameters are then obtained when either condition is met.

[0066] In this embodiment, the performance metrics essentially converged after approximately 12 iterations. The final controller parameters obtained are K. p =15.5, K d =0.15, the system damping ratio increased from the initial 0.08 to 0.18, approaching the target value of 0.2. The vibration amplitude decreased by more than 60% compared to the initial parameters, and the vibration suppression effect was significantly improved. The experimental data curves are shown below. Figure 3 As shown.

[0067] This invention proposes an adaptive adjustment method for control parameters of a wind tunnel model vibration suppression system based on iterative learning. By constructing an iterative learning optimization framework, establishing a performance index function, calculating the damping ratio in real time using the half-power bandwidth method, defining each performance index term, and designing an iterative learning parameter update law, the method achieves automatic iterative optimization of control parameters. Compared to traditional methods, this invention reduces reliance on expert experience during controller parameter setting and effectively prevents divergence risks during parameter updates, improving the efficiency and safety of wind tunnel testing. It features concise and reliable calculation formulas, a simple and clear solution process, and low computational complexity, meeting the real-time requirements of model vibration suppression in wind tunnel tests. It has broad application prospects and significant practical value.

Claims

1. A method for adaptive adjustment of control parameters of a wind tunnel model vibration suppression system based on iterative learning, characterized in that, The steps are as follows: Step 1: Construct an iterative learning optimization framework and design performance metric functions; Iterative learning optimization framework: The adaptive adjustment process of controller parameters in the wind tunnel model vibration suppression system is constructed as a closed-loop optimization problem driven by iteration and with physical performance indicators as the optimization objective; first, the controller parameter vector in the wind tunnel model vibration suppression system to be adjusted is defined. θ Redesign the performance index function for quantitatively evaluating vibration suppression effect. Finally, an iterative update law for controller parameters in a data-driven wind tunnel model vibration suppression system is established for automatic optimization. The goal of the aforementioned closed-loop optimization problem is to find the controller parameter vector. θ The optimal solution is found to ensure that, under excitation signals covering the key modes of the wind tunnel model vibration suppression system, the performance indicators of the wind tunnel model vibration suppression system are [values ​​missing]. J ( θ To minimize this performance metric, the function is constructed as follows: Among them, superscript ( k ) indicates the first k iteration Indicates the first k Performance metrics for the next iteration J ( θ )value; For damping ratio optimization; This is a vibration amplitude constraint term; For parameter variation regularization; Damping ratio optimization term Defined as: in, For the first k The damping ratio of the m-th mode identified online in the next iteration; To the pre-set first according to the test requirements m Target damping ratio for first-order modes; α These are the weighting coefficients; M The total number of modes; Vibration amplitude constraint Defined as: in, For the first k The discrete acceleration signal sequence acquired in the next iteration N The number of samples in the discrete acceleration signal sequence. β These are the weighting coefficients; Parameter variation regularization term Defined as: in, γ These are the weighting coefficients; Step 2: Identify the damping ratio of the wind tunnel model vibration suppression system online based on the half-power bandwidth method; The vibration acceleration signal of the wind tunnel model vibration suppression system is acquired in real time. The power spectral density of the vibration acceleration signal is calculated by fast Fourier transform, and the power spectral density curve is obtained to identify the resonant frequency of the dominant mode of the wind tunnel model vibration suppression system. f n And the corresponding peak value; on both sides of the peak value corresponding to the resonance frequency, find the frequency points on the power spectral density curve where the power value drops to half of the aforementioned peak value. f 1 and f 2 Then the damping ratio of the dominant mode of the vibration suppression system in this wind tunnel model is... ζ as follows: in, For half-power bandwidth, satisfy In actual online processing, parabolic interpolation or Sinc interpolation is used to process the power spectral density curve. Step 3: Design the iterative update law for controller parameters in the data-driven wind tunnel model vibration suppression system; Based on performance metrics An iterative learning control method is adopted, which updates the rules iteratively and finds ways to reduce performance indicators based on historical experimental data. The adjustment direction of the controller parameters in the wind tunnel model vibration suppression system is as follows: The update of the controller parameters in the wind tunnel model vibration suppression system follows the iterative format: in, and The first k +1st time and the first k The controller parameter vector for the next iteration. For the first k The parameter update amount calculated in the next iteration; Performance indicators Approximate negative gradient determination of controller parameters in a wind tunnel model vibration damping system: Where η is the learning gain, which controls the update step size; It is in the k Performance metrics at the next iteration Compared to the controller parameter vector in the wind tunnel model vibration suppression system θ (k) gradient The estimated value; Since the model of the wind tunnel vibration suppression system is unknown, the analytical gradient... Since the gradient cannot be obtained directly, a difference method based on experimental data is used to obtain the gradient. The specific steps for numerical estimation are as follows: a. in the first k In the nth iteration, at the nth k The controller parameter vector of the next iteration θ (k) Wind tunnel tests were conducted, and the results were calculated based on the performance index function defined in the first step. ; b. Regarding the first k The controller parameter vector of the next iteration θ (k) Apply a pre-set disturbance with an amplitude of the first... k The controller parameter vector of the next iteration θ (k) 1% to 5% of the disturbance δ θ The controller parameters of the vibration suppression system in the wind tunnel model after disturbance are obtained. θ (k) +δ θ ; c. Controller parameters in the wind tunnel model vibration suppression system θ (k) +δ θ The experiment was conducted again, and new performance index values ​​were calculated. ; d. Estimate the gradient direction using the first-order forward difference formula: 。

Images