Vibration active control method based on active learning

Through the active vibration control method based on active learning, a quadratic objective function and self-immune disturbance controller are constructed, and the control parameter design domain is adaptively reduced, which solves the multi-channel control coupling problem, and realizes efficient control of flexible structure vibration, which is suitable for Aerospace and robotics fields.

CN118915839APending Publication Date: 2024-11-08DALIAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410967368.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In complex systems, multi-channel control coupling leads to complex adjustment of control parameters, making it difficult to quickly and accurately determine the control parameters of high-dimensional coupling systems, especially in the vibration control of flexible structures in the fields of aerospace and robotics, vibrations caused by external airflow or space environments. Difficult to effectively suppress.

Method used

Using the vibration active control method based on active learning, the optimal control parameters are gradually determined by constructing a quadratic objective function, designing self-immune controller, building an initial Kriging agent model, adaptive reduction control parameter design domain and point sampling. , realize the optimal control strategy for vibration structure.

Benefits of technology

It significantly improves the efficiency of control parameters optimization, reduces the dependence on the dynamic characteristics of the initial control parameters selection, and provides an efficient, adaptive and dynamic update control strategy suitable for multi-channel control parameter tuning, especially in aerospace and The robot field effectively suppresses structural vibration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915839A_ABST
    Figure CN118915839A_ABST
Patent Text Reader

Abstract

The invention discloses a vibration active control method based on active learning. The method comprises the following steps: constructing a quadratic objective function of control consumption and vibration response of a vibration structure; designing a vibration controller of the vibration structure based on an active disturbance rejection control algorithm; acquiring a control instruction of the vibration controller and a vibration response under the corresponding instruction, and constructing an initial Kriging agent model of the vibration structure; determining a new control parameter group according to a design domain adaptive reduction method and a point adding sampling criterion of the vibration parameters; and performing vibration control training on the vibration structure based on the new control parameter group to obtain a trained Kriging agent model, and determining an optimal control parameter of the vibration structure based on the trained Kriging agent model and the minimum quadratic objective function, namely an optimal control strategy of the vibration structure to realize active vibration control of the vibration structure. The method can be expanded and applied to rapid control strategy solution of control problems or planning problems with multiple control parameters in many fields.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of structural vibration control and relates to an active vibration control method based on active learning. Background Art

[0002] Based on the inverse piezoelectric effect of piezoelectric materials, intelligent piezoelectric structures can realize highly reliable, lightweight, and low-power vibration control systems. Intelligent piezoelectric vibration control has become a very promising and active field in vibration control research and applications. With the continuous expansion of the size of intelligent piezoelectric structures and the increasingly severe service environment, the control concept of using multi-channel piezoelectric actuators for distributed control has gradually developed, such as aircraft tail vibration control, satellite solar panel vibration control, rotating machinery vibration suppression, etc.

[0003] With the continuous increase in the number of control channels and structural flexibility, the high-dimensionality problem faced by the controller and the resulting uncertainty and multi-channel control coupling problems make the setting of a large number of control parameters extremely complicated. The vibration suppression effect is most directly related to the selection of control parameters. Therefore, how to quickly and accurately determine the control parameters of high-dimensional coupling systems has become an urgent problem to be solved.

[0004] Lightweight, integrated, large-sized flexible structures are widely used in aerospace and robotics. With the further use of intelligent materials and structures, the size of typical structures such as wings and satellite solar panels has become larger and more flexible, resulting in a further reduction in their natural frequency and making them more susceptible to vibration. At the same time, the problems of attached material nonlinearity and geometric nonlinearity, as well as the need for multi-channel control of large-size structural vibration control, further aggravate the difficulty of vibration control of the above structures. In the aviation field, the wing structure needs to maintain a stable shape during flight to ensure aerodynamic performance. However, due to the influence of external airflow, these structures may vibrate, affecting flight stability and structural durability. In the aerospace field, solar panels on satellites are subject to various tiny vibrations in the space environment, which may reduce the power generation efficiency of the solar panels and shorten their service life. Summary of the invention

[0005] In order to solve the problem that it is difficult to adjust a large number of control parameters for multi-channel control coupling of complex systems, the technical solution adopted by the present invention is: a vibration active control method based on active learning, comprising the following steps:

[0006] Constructing quadratic objective functions of control consumption and vibration response of vibrating structures;

[0007] Design of vibration controller for vibrating structures based on active disturbance rejection control algorithm;

[0008] Obtain the control instructions of the vibration controller and the vibration response under the corresponding instructions, and build an initial Kriging proxy model of the vibration structure;

[0009] Determine a new control parameter group according to the design domain adaptive reduction method of control parameters and the point-added sampling criterion;

[0010] Based on the new control parameter group, vibration control training is performed on the vibrating structure to obtain a trained Kriging proxy model;

[0011] Based on the trained Kriging proxy model and the minimum quadratic objective function, the optimal control parameters of the vibrating structure, that is, the optimal control strategy of the vibrating structure, are determined to achieve active vibration control of the vibrating structure.

[0012] Further: the expression of the design domain adaptive reduction method is:

[0013]

[0014] Among them, LB represents the lower boundary of the control parameter design domain, UB represents the upper boundary of the control parameter design domain, Best_Set represents the current optimal control parameter group, and k is the design domain reduction coefficient.

[0015] Further: the sampling criteria include one of the following:

[0016] EI points criteria;

[0017] Or MSP point addition criteria;

[0018] Or MSE point addition criteria;

[0019] Or PI plus point criteria.

[0020] Furthermore: the vibration structure adopts a flexible structure with a frequency ≤ 500 Hz.

[0021] Further: the vibration controller designed based on the active disturbance rejection control algorithm is expressed as:

[0022]

[0023] Among them, v is the signal value to be tracked, e is the tracking error, u0 is the virtual control amount, u is the final control output, k p , k d To control the gain, according to the control bandwidth ω c Determine, denoted as k p =ω c 2 , k d =2ω c .

[0024] Further: the control parameter: control bandwidth ω c , observation bandwidth ω0 and control gain b0.

[0025] Further: the process of determining a new control parameter group according to the design domain adaptive reduction method of the control parameters and the point sampling criterion is as follows:

[0026] Based on the global search, the upper and lower boundaries of the current design space are determined based on the current optimal control parameter group and vibration control parameters.

[0027] The upper and lower boundaries of the new design space of vibration control parameters are calculated by using the adaptive reduction strategy of the design domain.

[0028] After the design space of vibration control parameters is reduced, new sample points, i.e., new control parameter groups, are selected in the new design space according to the adopted point-adding sampling criterion.

[0029] The present invention provides an active vibration control method based on active learning, which is a new data-driven active learning optimization control method, and is proposed based on the Kriging proxy model, the self-disturbance rejection control algorithm, the design domain adaptive reduction strategy and the point sampling criterion. The method has the following advantages:

[0030] The present invention aims to meet the higher requirements for multi-modal vibration control problems caused by complex excitations and efficient multi-channel control parameter adaptive setting. This method can significantly improve the efficiency of control parameter optimization while ensuring the vibration control effect. In addition, the design domain can be adaptively adjusted to reduce the dependence on the dynamic characteristics required for the initial control parameter selection in the early stage. The main contribution of the present invention is to provide an efficient, adaptive and dynamically updated strategy for the multi-channel control parameter setting of flexible structures. The present invention is also applicable to the control parameter setting of other multi-channel control coupling systems. It is particularly suitable for rapid engineering implementation in the fields of aerospace engineering, robotics, etc. The present application can update the control parameters through an active learning algorithm to effectively suppress structural vibrations.

[0031] The parameter training process of the embodiment of the present invention is fast and efficient, can adaptively reduce the design variable space, reduce the requirements for the system dynamic characteristics, require fewer training times, and is suitable for multi-channel vibration control of linear and nonlinear systems, especially when combined with distributed piezoelectric actuators. It is particularly suitable for low-frequency vibration control of large-sized flexible structures with vibration frequencies within 500 Hz. The invention can also be expanded to apply to the rapid control strategy solution of control problems or planning problems with multiple control parameters in many fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0033] Figure 1 It is a design framework of a vibration active control method based on active learning provided by an embodiment of the present invention;

[0034] Figure 2 is a schematic diagram of a new control parameter group determination process provided by an embodiment of the present invention;

[0035] Figure 3 is a step diagram of a vibration active control method based on active learning provided by an embodiment of the present invention;

[0036] Figure 4 It is an intelligent flexible large-span board provided by an embodiment of the present invention;

[0037] Figure 5 This is the objective function training process provided by the embodiment of the present invention;

[0038] Figure 6 is the optimal value of the objective function in the training process provided by the embodiment of the present invention;

[0039] Figure 7 is the vibration displacement time domain response under the optimal control strategy provided by the embodiment of the present invention;

[0040] Figure 8 It is the vibration displacement frequency domain response under the optimal control strategy provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0041] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Figure 1 It is a design framework of a vibration active control method based on active learning provided by an embodiment of the present invention;

[0044] A vibration active control method based on active learning comprises the following steps:

[0045] S1: Constructing a quadratic objective function of the control consumption and vibration response of the vibrating structure;

[0046] S2: Design a vibration controller for the vibrating structure based on the ADRC algorithm;

[0047] S3: Obtaining the control instructions of the vibration controller and the vibration response under the corresponding instructions, and constructing an initial Kriging proxy model of the vibration structure;

[0048] S4: Determine a new control parameter group according to the design domain adaptive reduction method of the control parameters and the point sampling criterion;

[0049] S5: Performing vibration control training on the vibrating structure based on the new control parameter group to obtain a trained Kriging proxy model;

[0050] S6: Based on the trained Kriging proxy model and the minimum quadratic objective function, the optimal control parameters of the vibrating structure, that is, the optimal control strategy of the vibrating structure, are determined to achieve active vibration control of the vibrating structure.

[0051] Steps S1 / S2 / S3 / S4 / S5 / S6 are executed in sequence;

[0052] The quadratic objective function expression of the control consumption and vibration response of the vibration structure is:

[0053]

[0054] Where J represents the quadratic objective function; x represents the structural vibration displacement response; u represents the control voltage; t c Indicates the time point when active control starts; t f Represents a c The specific time point after that is used to calculate the objective function; Q is the vibration amplitude weight coefficient, and R is the energy consumption weight coefficient. represents the active vibration control [t c t f ]The vibration displacement response size in the time period, the second term represents the active vibration control [t c t f ] control voltage consumption during the time period.

[0055] The vibration controller designed based on the active disturbance rejection control algorithm is expressed as:

[0056]

[0057] Among them, v is the signal value to be tracked, e is the tracking error, u0 is the virtual control amount, u is the final control output, k p , k d To control the gain, the control bandwidth ω can be used c Determine, denoted as k p =ω c 2 , k d =2ω c ;

[0058] The ADRC consists of three core parts: tracking differentiator (TD), extended state observer (ESO) and state error feedback control law (SEF).

[0059] Determine the three control parameters that need to be adjusted for the ADRC: control bandwidth ω c , observation bandwidth ω0 and control gain b0.

[0060] Furthermore, obtaining the control instructions of the vibration control law and the vibration response under the corresponding instructions, and constructing the initial Kriging proxy model of the vibration structure includes the following steps:

[0061] S31: Determine the initial control parameter group using modern experimental design methods; the modern experimental design methods include quasi-Monte Carlo method, quasi-Monte Carlo method, Latin hypercube, orthogonal experimental design (OAD), Hamersley sequence sampling method (HSS), or improved forms of the above specific methods.

[0062] S32: Performing off-line or on-line structural vibration control training / experiments based on the initial control parameter group;

[0063] Construct a structural vibration control simulation or experimental system based on an ADRC and set the three control parameters (control bandwidth ω c , observation bandwidth ω0 and control gain b0) are input into the ADRC to perform vibration control simulation or experiment.

[0064] S33: Calculate the objective function corresponding to the control parameter group;

[0065] According to the voltage control signal input in the simulation or experiment and the vibration displacement response of the vibration structure, the corresponding objective function value is calculated according to the quadratic objective function expression.

[0066] S34: Construct an initial Kriging proxy model based on the initial control parameter group and the obtained objective function. The expression can be:

[0067] J(P)=f T(P)β+z(P)

[0068] Among them, J(P) is the approximate fit of the objective function, f T (P) is a function of any linear or nonlinear design variable, providing a global approximation in the design space, f T (P)β forms the regression term and z(P) is the random distribution error.

[0069] The design domain adaptive reduction method expression is:

[0070]

[0071] Among them, LB represents the lower boundary of the control parameter design domain, UB represents the upper boundary of the control parameter design domain, Best_Set represents the current optimal control parameter group, and k is the design domain reduction coefficient. The design domain adaptive reduction strategy can adaptively adjust the subsequent design space according to the current optimal value obtained in the previous training, and finally gradually reduce the design space to the vicinity of the global optimal value. The design domain adaptive reduction strategy improves the training efficiency and reduces the accuracy requirements for the dynamic characteristics of the controlled object and the initial design domain.

[0072] The sampling criteria include one of the following:

[0073] EI points criteria;

[0074] Or MSP point addition criteria;

[0075] Or MSE point addition criteria;

[0076] Or PI point addition criteria;

[0077] Or other improved forms based on the above-mentioned point-adding criteria.

[0078] Reference Figure 2 The process of determining a new control parameter group according to the design domain adaptive reduction strategy of the control parameters and the point sampling criterion is as follows:

[0079] On the basis of global search, based on the current optimal control parameter group and the current upper and lower boundaries of the design space of the vibration control parameters, the current optimal control parameter group is obtained according to the current training data, and the control parameter group corresponding to the minimum quadratic objective function in all current training data is the current optimal control parameter group.

[0080] The upper and lower boundaries of the new design space of vibration control parameters are calculated according to the design domain adaptive reduction strategy;

[0081] After reducing the design space, new sample points, i.e., new control parameter groups, are selected in the new design space according to the adopted point-adding sampling criteria (for example, if the EI point-adding criteria is adopted, the new sample points need to satisfy the maximization of the difference between the control parameter group corresponding to the minimum value of the current quadratic objective function and the newly found control parameter group).

[0082] By repeating the above steps, the design domain is continuously narrowed down to the vicinity of the optimal control parameter group.

[0083] Further, vibration control training is performed on the vibration structure based on the new control parameter group, including the following steps:

[0084] S51: Calculating the quadratic objective function value based on the active vibration control result;

[0085] S52: Update the Kriging proxy model based on the new control parameter set and objective function value.

[0086] Furthermore, based on the trained Kriging proxy model and the minimum quadratic objective function, the optimal control parameters of the vibrating structure, that is, the optimal control strategy of the vibrating structure, are determined as follows:

[0087] Repeat the process of determining a new control parameter group according to the design domain adaptive reduction strategy and the point-added sampling criterion;

[0088] Repeat the vibration control training of the controlled structure based on the newly added control parameter group until the convergence condition is reached (the total number of training times reaches the set total number n_set, or no better solution is found after n_find consecutive training times, where n_set and n_find are pre-set times), and determine the optimal control strategy.

[0089] Embodiment 1:

[0090] The technical effects of the embodiments of the present invention are further described below in conjunction with specific application scenarios:

[0091] 1. Scenario description:

[0092] A smart flexible long-span board with four piezoelectric actuators, referring to Figure 3 , 4 M8557-P1 piezoelectric fiber actuators (MFC) are equidistantly pasted on the flexible large-span plate, and the base structure is a flexible large-span aluminum plate of 1105mm*76mm*1.14mm. MFC can excite the structure to produce vibration or suppress structural vibration. The control goal of this application case is to suppress the vibration of the base structure to the greatest extent by designing the self-disturbance control parameters of the corresponding channel of MFC. The first three natural frequencies of the intelligent piezoelectric structure are 0.77Hz, 4.8Hz, and 13.5Hz.

[0093] The application case specifically considers the problem of multi-channel vibration control with the goal of minimizing the vibration amplitude at the measurement point. In order to highlight the applicability of the present invention to complex dynamic structures, the application case uses the second MFC to excite the structural vibration. The excitation signal is the aliased signal of the first three natural frequencies to excite the multi-modal vibration of the base structure. For example, the excitation signal in the application case is

[0094] s excite =0.15sin(0.77*2π)+0.3*sin(4.8*2π)+sin(13.5*2π)

[0095] Another three MFCs are used for vibration control to demonstrate the applicability of the present invention to the multi-channel coupled vibration problem. Therefore, the control parameters that need to be adjusted are 9 control parameters of the three active disturbance rejection controllers.

[0096] 2. Implementation process and effect:

[0097] Three piezoelectric actuators correspond to three ADRCs. Therefore, the control parameter group that needs to be adjusted is

[0098]

[0099] The initial design domain selection is:

[0100]

[0101] Select the design domain reduction factor k = 2;

[0102] According to the design domain adaptive reduction strategy, the new design domain boundary is

[0103]

[0104] Among them, Best_Set i is the control parameter group corresponding to the minimum quadratic objective function in the previous i training data. i is the upper boundary of the design space during the i-th training, LB i is the lower boundary of the design space during the i-th training.

[0105] The EI plus point criterion is used to obtain a new set of control parameters in the new design domain.

[0106] E[I(Control_Set)]=E[max(f min -Y,0)]

[0107] Among them, I(Control_Set) is the improvement of the quadratic objective function, which can be defined as

[0108]

[0109] The new control parameter set can be determined as

[0110] Control_Set i+1 = argmax p (E[I(Control_Set i )])

[0111] LB i+1 ≤Control_Set i ≤UB i+1

[0112] The application case uses input and output data for iterative training, and obtains the optimal control parameters through active learning, ultimately achieving the best vibration suppression effect. The input data is 9 control parameters, and the output data is the objective function calculated based on the control voltage consumption and vibration amplitude response. Figure 4 First, the initial control parameter group is selected according to the modern experimental design method. In this embodiment, the Latin hypercube sampling method is selected to select the initial control parameter group. The initial Kriging proxy model is obtained through offline or online training. Then, a new control parameter group is added according to the adaptive dynamic reduction strategy and the EI point addition criterion, and the Kriging proxy model is updated. In the application case, the total duration of each training simulation is 80s. After 40s, the vibration controller is turned on and the cost function is calculated based on the control output and vibration response. The initial training number is set to 90 times. When the number of new training times reaches 100 or no better strategy is found for 30 consecutive times, the active learning process is completed.

[0113] Figure 5 , Figure 6 The active learning process of the application case is shown. It can be seen that the total number of training times is less than 150. Under the optimal control strategy obtained by active learning, the structural vibration control effect is as follows Figure 7 As shown in the figure, it can be seen that the structural vibration is significantly reduced after the vibration control system is turned on. Figure 8 The frequency domain analysis results of the structural response are shown. It can be seen that the first three vibration modes are almost completely suppressed under the action of the controller, and the amplitudes are all reduced to less than 0.1 mm. It can be seen from the embodiment of multi-modal vibration control of the piezoelectric cantilever plate that the active vibration control method based on active learning proposed in the embodiment of the present invention can achieve the best vibration control effect by quickly realizing the control parameter setting.

[0114] This patent application can be used to actively control the vibration of solar panels to ensure that they can work stably in various space environments. In the field of robotics, components such as flexible robotic arms or sensors need to suppress vibrations caused by external interference or their own motion when performing precise operations. The control method provided by this patent can adjust the control strategy in real time to adapt to the dynamic changes of the robot in a complex environment and improve the accuracy and reliability of the operation.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vibration active control method based on active learning, characterized in that: The following steps are involved: Constructing quadratic objective functions of control consumption and vibration response of vibrating structures; Design of vibration controller for vibrating structures based on active disturbance rejection control algorithm; Obtain the control instructions of the vibration controller and the vibration response under the corresponding instructions, and build an initial Kriging proxy model of the vibration structure; Determine a new control parameter group according to the design domain adaptive reduction method of control parameters and the point-added sampling criterion; Based on the new control parameter group, vibration control training is performed on the vibrating structure to obtain a trained Kriging proxy model; Based on the trained Kriging proxy model and the minimum quadratic objective function, the optimal control parameters of the vibrating structure, that is, the optimal control strategy of the vibrating structure, are determined to achieve active vibration control of the vibrating structure.

2. The active vibration control method based on active learning according to claim 1, characterized in that: The design domain adaptive reduction method expression is: Among them, LB represents the lower boundary of the control parameter design domain, UB represents the upper boundary of the control parameter design domain, Best_Set represents the current optimal control parameter group, and k is the design domain reduction coefficient.

3. The active vibration control method based on active learning according to claim 1, characterized in that: The sampling criteria include one of the following: EI points criteria; Or MSP point addition criteria; Or MSE point addition criteria; Or PI plus point criteria.

4. The active vibration control method based on active learning according to claim 1, characterized in that: The vibration structure adopts a flexible structure with a frequency of ≤500 Hz.

5. The active vibration control method based on active learning according to claim 1, characterized in that: The vibration controller designed based on the active disturbance rejection control algorithm is expressed as: Among them, v is the signal value to be tracked, e is the tracking error, u0 is the virtual control amount, u is the final control output, k p , k d To control the gain, according to the control bandwidth ω c Determine, denoted as k p =ω c 2 , k d =2ω c .

6. The active vibration control method based on active learning according to claim 1, characterized in that: The control parameter: control bandwidth ω c , observation bandwidth ω0 and control gain b0.

7. The active vibration control method based on active learning according to claim 1, characterized in that: The process of determining a new control parameter group according to the design domain adaptive reduction method of the control parameters and the point sampling criterion is as follows: Based on the global search, the upper and lower boundaries of the current design space are determined based on the current optimal control parameter group and vibration control parameters. The upper and lower boundaries of the new design space of vibration control parameters are calculated by using the adaptive reduction strategy of the design domain. After the design space of vibration control parameters is reduced, new sample points, i.e., new control parameter groups, are selected in the new design space according to the adopted point-adding sampling criterion.