1. Introduction
With the rapid development of modern industry and the progress of engineering technology, the number of industrial products used in daily life has gradually increased, such as vehicles, aircraft, and ships. The negative impact of vibrations should be emphatically considered in the design of such products. System structure vibration has become a hot issue in today’s society [
1,
2]. The commonly used control methods are divided into three main categories: passive control, active control, and integrated active and passive control [
3,
4,
5]. A common passive control structure is passive constrained layer damping (PCLD) [
6]. As shown in
Figure 1, when a PCLD structure is excited by the outside world, the damping layer itself strains to dissipate the energy generated by the vibration into thermal energy, thus achieving passive control [
7]. This method is low-cost, stable, and effective for high-frequency vibration areas. However, it is not effective for low-frequency vibration areas and cannot be controlled in time when the external environment changes [
8]. The advent of active control makes up for this deficiency by using intelligent materials such as piezoelectric ceramics as a piezoelectric layer to be pasted onto the surface, enabling the interconversion of electrical and kinetic energy through the piezoelectric effect, effectively enabling the control of low-frequency vibrations [
9,
10]. The disadvantages are the significant reduction in effectiveness when the controller fails, the poor suppression of high-frequency vibrations, and the high control costs [
11,
12]. Therefore, integrated active and passive control can combine the advantages of the first two control methods, and the disadvantages compensate for each other [
13,
14], as proposed by Baz [
15] with the active constrained layer damping (ACLD) model. The difference from the PCLD structure is the replacement of the uppermost constraint layer with a piezoelectric layer [
16,
17]. Piezoelectric ceramics are usually used as the piezoelectric layer. Piezoelectric ceramics are suitable for use in structures such as plates, shells, and beams because of the small excitation required and the fast response time [
18].
Figure 2 shows a typical ACLD structure.
The establishment of a mathematical model of the structure is a prerequisite for vibration control, and, currently, the standard methods are analytical, numerical, and experimental [
19,
20]. For a system with a relatively simple structure and relatively regular shape, the analytical method can be used. For complex structures, numerical methods should be used, in which the finite element method (FEM) is a common modeling method. The experimental method can be used to verify the accuracy of the models established by the analytical method and numerical method, and can also be used to describe the model characteristics of complex structures with the principle of system identification [
21,
22]. The model characteristics of the viscoelastic layer (VEM) should also be considered when modeling ACLD structures, where standard models include: complex constant shear modulus models, modal strain energy models, anelastic displacement field (ADF) models, and Golla–Hughes–McTavish (GHM) models. The complex constant shear modulus model is simple in structure, but it cannot reflect the phenomenon that the physical parameters of viscoelastic materials change with frequency. The modal strain energy model will produce large errors when the structure frequency is dense. Both the GHM model and ADF model describe the mechanical properties of viscoelastic materials in the frequency domain through a series of micro-oscillator terms, which can be effectively combined with FEM modeling methods to form a second-order differential equation [
23,
24]. However, the two models will introduce dissipative degrees of freedom in the use process, so that the dimension of degrees of freedom of the system is too large and must be used together with the order reduction method. Common model reduction methods include the GUYAN condensation method, the high-precision dynamic condensation method, the modal reduction method, the balanced reduction method, the robust reduction method, etc. The models can also be divided into physical order reduction and state-space order reduction. Physical order reduction can greatly reduce the degrees of freedom of the system, but it cannot guarantee the observability and controllability of the reduced order system; state-space order reduction can ensure the observability and controllability of the reduced system, but it is not suitable for systems with too high degrees of freedom [
25,
26]. Therefore, two order reduction methods are usually used together. Cao YQ [
21] used the ADF model to characterize the relationship between the shear modulus of viscoelastic materials and temperature and used the finite element method to model ACLD beam and plate structures. Lu [
23] introduced the GHM model to model the smart constrained layer damping (SCLD) thin plate structure and then obtained the decentralized subsystem control model through balanced order reduction and complex mode truncation. Li [
27] established an experimental system for the ACLD beam structure and established the dynamic equation of the cantilever beam by using the finite element method, combining the piezoelectric equation, and introducing the complex constant modulus damping characteristics. Ray [
28], based on the constitutive equations of piezoelectric materials and viscoelastic materials, deduced the dynamic equations of the partially covered ACLD cylindrical shell structure using the energy method. Huang and Mao [
29] used the modal method and Hamilton principle to establish the dynamic equations of piezoelectric sandwich plates and solved them using the Rayleigh–Ritz method.
Common active control methods include proportional–integral differential (PID) control, linear quadratic optimal control (LQR), boundary control, robust control, etc. PID control is simple and easy to implement, but it is difficult to obtain the optimal gain. LQR control requires a higher accuracy for the model, all states can be fed back, resulting in a theoretically optimal feedback gain matrix, and robust control increases the stability of the system and resistance to disturbances. Li [
30] proposed a hybrid PID–FXLMS algorithm combining the feedback FXLMS algorithm with a conventional PID controller and applied it to improve the efficiency of piezoelectric cantilever beam vibration control. Lu [
31] designed a modal controller and modal state estimator in order to solve the full-state feedback problem, which made the system more suitable for real-time control. Zheng [
32] designed a closed-loop control system with proportional (PD) feedback and derivative feedback based on sensor voltage and discussed the effect of control gain on the vibration suppression of ACLD cylindrical shell and plate structures. Zhang [
33] designed a feedback-based LQG and feedforward FXLMS adaptive control composite controller to simulate and analyze the vibration problems of ACLD plate structures and compare the performance of the composite controller. Liu and Shi [
34] designed a robust controller for active control of the first four orders of plate vibration using a robust control method. Liao [
35] further applied the LQR control method to investigate the control force requirements for different configurations of ACLD structures and the response to the vibration effects. Lam [
36] designed LQR controllers to optimize the controller parameters according to the characteristics of the actively constrained layer-damping structure. Baz [
37] designed boundary controllers for ACLD beams, enabling the system to be kept globally stable in all vibration modes in boundary control.
Through the literature review, it is found that many scholars have conducted a lot of research in this field, but there are still some deficiencies, such as the ACLD sandwich board accurate modeling problem. For the active control of the ACLD structure, parameter selection in the LQR controller is always a hot issue, and the parameter optimization of the piezoelectric plate needs further study. Therefore, we conducted the following work, as described in this paper. In
Section 2, modeling was performed for the ACLD sandwich plate based on the FEM method, and dynamic equations under different boundary conditions were obtained. In
Section 3, the observability and controllability systems were obtained by the high-precision dynamic condensation and internal balance method. In
Section 4, the accuracy of the models established by the method in this paper was verified, respectively, before and after order reduction. In
Section 5, the control effect of the system under different boundary conditions and different incentives was explored through simulation analysis. The parameter optimization of the controller and the piezoelectric plate is discussed from the performance and cost points of view.
4. Finite Element Model Verification
In this section, the accuracy of the original model and the accuracy of the model after joint reduction were verified through the analysis of numerical examples. The optimal number of discrete elements of the finite element model was obtained through the convergence test. The geometric and material parameters of each layer are as follows.
Piezoelectric layer: .
Damping layer: ,
Base layer: ;
Base size: The total length is 0.2 m and the width is 0.1 m. Piezoelectric constant:
GHM model parameters were taken according to the literature [
20]:
= 5.542 × 10
5,
= [3.96, 65.69, 1.447], = [8.962 × 105, 9.278 × 105, 7.613 × 105], = [148, 12.16, 810.4], where N = 3, there are 3N + 1 = 10 parameters.
4.1. Finite Element Model Element Convergence Test
The convergence of the elements is an important property of the finite element method, which is directly related to whether the convergence solution can be obtained and whether the appropriate number of elements can be used to obtain a satisfactory solution. The purpose of the convergence test is to test the influence of the number of finite element elements on the calculation results. When the number of elements increases and the result does not change significantly (i.e., tends toward a stable value), the calculation result is considered to be convergent. If the number of finite element elements increases, the calculation results always fluctuate greatly, which means that the element does not converge and a reasonable finite element solution cannot be obtained.
To investigate the convergence of the finite element model in this paper, that is, to discuss the discrete effect of the number of units on the result is to discuss how much is needed, it was appropriate to discretize the unit. Because too few units may lead to non-convergence in results, too many elements will lead to an increase in the workload of element assembly and equation solving. Therefore, a simple numerical example was designed to study this problem. The ACLD plate structure was discretized by the four-node, 28-degree-of-freedom plate element derived in this paper, and the number of elements was two, four, six, eight, and ten. The calculated natural frequency changes with the number of elements are shown in
Figure 5.
It can be seen from
Figure 5 that this element has good convergence. When the number of elements is six, the calculation results of the natural frequency of the system begin to have an obvious convergence trend. When the number of elements is eight, the curve is basically unchanged and the convergence requirements can be considered.
4.2. Model Validation before Order Reduction
Under different boundary conditions, the modal parameters of the base plate and ACLD sandwich plate were calculated by using Ansys2021 finite element analysis software and Matlab2016 finite element program, respectively, and the correctness of the model was further verified by comparing with the modal experiments results from the literature [
23,
31,
33]. The boundary conditions of the four sides of the plate are represented by letters, and their meanings are: C—fixed support and F—free. CFFF means that one side is fixed and the other three sides are free; CFCF means that two opposite edges are fixed and the other two edges are free; CCCC stands for four-sided fixed support.
Figure 6 shows the finite element model of the ALCD thin plate structure. According to the convergence analysis results, the bottom plate was divided into eight elements and 15 nodes. The element type in the finite element model is SHELL63, an elastic plate and shell element, and a mesh contact was used between the plate and shell elements. The linear contact mode was adopted between the layers, and the materials of each layer were firmly pasted without relative sliding between layers. The ACLD unit was partially covered. The coverage position is represented by element 1 in the figure.
It can be seen from
Table 1 that the difference between the numerical solution obtained based on Matlab programming and the results of Ansys2021 software, references, and modal experiments is very small. It can be seen from
Table 2 that the difference between the numerical solution of Matlab and the results of Ansys and references is also very small. Compared with the results in
Table 1 and
Table 2, the coverage of the piezoelectric layer and viscoelastic layer leads to a reduction in natural frequency, which is consistent with the actual results. The maximum error of the first four natural frequencies of the system is not more than 5%, and the unit of natural frequencies is Hz.
Figure 7 shows the first four modal shapes of the cantilever plate obtained via Ansys2021 software.
Figure 8 shows the first four modal shapes of the cantilever plate obtained with a Matlab numerical solution.
Figure 7 and
Figure 8 show that the mode of shape deformation of the cantilever plate is similar. Therefore, the above results show that the finite element modeling method in this paper is accurate.
4.3. Model Verification after Order Reduction
The full coverage ACLD structure model under one side fixed support was selected as the research object, and the natural frequencies before and after the model reduction were calculated through Matlab, the Bode diagram in the frequency domain and the pulse response diagram in the time domain were drawn, and the controllability and observability of the system were judged.
As shown in
Table 3, the error of the first- to fourth-order natural frequencies of the model is less than 3% after two-order reduction methods are adopted, which has little impact on the natural characteristics of the system. As shown in
Table 4, before the reduction, the system dimension was 210, the WC and WO matrices were not full of rank, and the system was uncontrollable and unobservable. After dynamic condensation, the system dimension was greatly reduced, but the observability matrix and controllability matrix were not full of rank. Next, we balanced the reduced order model, arranged the state variable
from small to large, and retained a value greater than 0.002. Finally, the system model only had six degrees of freedom, and the observability and controllability matrices had full rank. The vibration characteristics of the system were mainly determined by the first- to fourth-order modal characteristics, and the subsequent active control was mainly carried out around the low-frequency modes.
Figure 9 shows the Bode diagram of the system in the frequency domain. After the model was reduced twice, it could still accurately and reliably describe the dynamic characteristics of the original system in the low-frequency region.
Figure 10 shows the time–domain response diagram of the original system under impulse excitation after two order reductions. From the degree of fitting of the time–domain response curve, the time–domain effect of the two order reductions on the system is small. To sum up, the order reduction method proposed in this paper is effective, which not only greatly reduces the degree of freedom of the system but also, ultimately, makes the system observable and controllable, providing favorable help for the design of an active controller.