betaFeedback

WorksComputer Science

Optimal design of unimorph-type cantilevered piezoelectric energy harvesters using level set-based topology optimization by considering manufacturability

Abstract

In this study, we propose a design methodology for a piezoelectric energy-harvesting device optimized for maximal power generation at a designated frequency using topology optimization. The proposed methodology is adapted to the design of a unimorph-type piezoelectric energy harvester, wherein a piezoelectric film is affixed to a singular side of a silicon cantilever beam. Both the substrate and the piezoelectric film components undergo concurrent optimization. Constraints are imposed to ensure that the resultant design is amenable to microfabrication, with specific emphasis on the etchability of piezoelectric energy harvesters. Several numerical examples are provided to validate the efficacy of the proposed method. The results show that the proposed method yields optimized substrate and piezoelectric designs with an enhanced electromechanical coupling coefficient, while allowing the eigenfrequency of the device and the minimum output voltage to be set to the desired values. Furthermore, the proposed method can provide solutions that satisfy the cross-sectional shape, substrate-dependent, and minimum output voltage constraints. The solutions obtained by the proposed method are manufacturable in the field of microfabrication.

Introduction

Piezoelectric energy harvesters convert mechanical energy into electrical energy through the piezoelectric effect. These devices have gained significant attention in recent years owing to their potential to scavenge energy from various sources such as vibrations [1, 2] and fluid flow [3]. The design of a piezoelectric energy harvester plays a critical role in determining its efficiency and performance [4]. Various piezoelectric energy harvester designs, including cantilever beams, multilayered structures, and cylindrical structures, have been proposed and studied [5, 6]. The unimorph cantilevered piezoelectric energy harvester has several advantages over other piezoelectric energy harvester designs. A unimorph cantilevered piezoelectric energy harvester consists of a cantilever beam with one end fixed and the other end free to move. The beam is typically composed of two primary domains: a substrate domain and a piezoelectric material domain. Additionally, it includes electrode domains for electrical energy extraction. When the device is subjected to external vibrations or mechanical loads, the beam bends, causing mechanical strain in the piezoelectric domain. This strain generates electrical energy through the piezoelectric effect. The simple and compact design of this harvester makes it suitable for applications where size and weight are critical factors, such as in portable or wearable devices. Furthermore, the unimorph cantilevered design allows easy integration with other systems, such as sensors or wireless transmitters [7]. However, the performance of a unimorph cantilevered piezoelectric energy harvester is limited by the amount of mechanical energy harvested from external vibrations. To overcome these limitations, researchers have explored the use of advanced materials and optimization techniques to improve their performance. Some studies have investigated piezoelectric materials with improved electromechanical coupling factors to enhance the piezoelectric effect and improve the efficiency of energy conversion [8, 9]. Other researchers have focused on designing the geometry of the cantilever beam and piezoelectric element for maximum energy output. For instance, rectangular, triangular, and trapezoidal shapes have been proposed for unimorph piezoelectric vibration energy harvesters [10]. However, as these structures have simple geometries, significant improvements can be achieved using topological optimization techniques [11].

Among the early forays in the topology optimization of piezoelectric materials, the study by Silva et al. [12] stood out as a pioneering study. This study explored the optimal design of periodic piezocomposites and underscored the profound impact of the unit cell topology on the performance of piezoelectric materials. Their insights provided a foundational framework for subsequent advancements in the field. The optimization of a piezoelectric energy harvester requires meticulous attention to numerous parameters, including the geometry and material properties of the piezoelectric element, the resonant frequency of the device, and the electrical load impedance. Zeng et al. [13] introduced a topology optimization technique to maximize the harvesting performance of piezoelectric energy harvesters. Rupp et al. [14] conceived a design for a piezoelectric energy harvester integrated with an external electrical circuit with due regard to the electrical load impedance. Kögl et al. [15] advanced the “PEAMAP-P” analysis model, specifically tailored for the analytical modeling of piezoelectric materials. Their design paradigm singularly targeted the piezoelectric material domain and deliberately omitted the substrate domain. Kang et al. [16] considered both the piezoelectric material and electrode domains in their design. Adding a unique perspective, Zhang et al. [17] focused on the dynamic responses in piezoelectric structures, emphasizing transient load optimization. Their approach highlights the need for a precisely tailored design in active-control smart structures. Chen et al. [18] and Luo et al. [19] designed tubular energy harvesting devices made of multiple materials. Notably, they employed level set-based topology optimization to eliminate grayscale discrepancies between the material and void domains. He et al. [20] and de Almeida et al. [21] focused on designing the thickness profile of a piezoelectric energy harvester by considering the piezoelectric material, substrate, and electrode domains. In addition, Kim et al. [22] optimized both the substrate and piezoelectric material domains. They delineated distinct objective functions for each domain and sequentially optimized the substrate and piezoelectric film structures. Similarly, in the field of acoustic energy focusing, Yoon et al. [23] demonstrated the application of topology optimization for piezoelectric acoustic focusers. By considering the intricate interactions between the electric, mechanical, and acoustic phenomena, their approach offers a more comprehensive understanding and optimization of piezoelectric devices in acoustically challenging environments. In a notable advancement, Salas et al. [24] extended topology optimization of piezoelectric actuators beyond traditional metal substrates and piezoelectric films to include the optimization of fibers and polarization in laminated piezocomposite multi-entry actuators (LAPAs). This holistic approach presents a significant challenge in the design of sophisticated actuators with enhanced performance capabilities. To further develop their methods, Salas et al. [25] introduced the hybrid interpolation model for fiber orientation (HYIMFO) method, which harmonizes continuous and discrete fiber orientation optimization for the LAPA. This hybrid approach represents a progressive stride in the optimization of complex multi-material actuators.

However, these methodologies often yield designs that face inherent manufacturing challenges. Consequently, various strategies have been proposed to realize manufacturable designs through topology optimization. Vatanabe et al. [26], Michailidis et al. [27], and Sato et al. [28] developed advanced methodologies to assess manufacturability. Vatanabe et al. [26] proposed a unified projection-based approach to incorporate manufacturing constraints into topology optimization, enabling the generation of manufacturable designs with minimum member size, minimum hole size, symmetry, and other geometric constraints. This method ensures that the optimized structures are not only optimal in terms of performance but also feasible for manufacturing processes like casting, milling, turning, forging, and rolling. Michailidis et al. [27] focused on multi-phase shape and topology optimization via a level-set method, incorporating manufacturing constraints to ensure feasible designs for practical applications. Sato et al. [28] developed a method for evaluating manufacturability in molding processes using fictitious physical models to ensure that the geometrical features of molded parts are amenable to manufacturing. Building upon the advancements in manufacturability assessments, various studies have proposed constraints related to manufacturability specifically in the context of additive manufacturing. Liu et al. [29] reviewed the current trends and future directions in topology optimization for additive manufacturing, highlighting the need for support structure optimization and addressing issues related to overhangs and material anisotropy. Yamada et al. [30] focused on additive manufacturing and proposed a topology optimization method that addressed closed cavity exclusion constraints and enhanced the design process of additive manufacturing. Tajima et al. [31] introduced a coupled fictitious physics model aimed at improving the convergence in topology optimization by integrating the objectives of both the fictitious physical and mechanical models, particularly for additive manufacturing with geometric constraints. These methodologies collectively underscore the importance of incorporating manufacturability considerations into topology optimization, which is essential for practical applications in various manufacturing processes. The level set-based topology optimization championed by Yamada et al. [32] affords precision in modulating the geo-平特一肖 metrical intricacy of the resultant optimal configurations. Moreover, earlier studies developed various methodologies for multi-material topology optimization, such as those by Yulin and Xiaoming [33] and Wang and Wang [34], which laid the groundwork for subsequent advancements in this field. Building on this foundation, Noda et al. [35] augmented the level set method to address the specific challenges of multi-material topology optimization. This innovation ensures a meticulous orchestration of the complexity intrinsic to each deployed material, culminating in a multi-material assembly that is conducive to straightforward manufacturing. However, the previously underscored methodologies are predominantly conceptualized for traditional, macroscopic machining techniques such as milling. Therefore, their direct application to piezoelectric energy harvesters, typically realized through microfabrication techniques such as etching [9, 36], may not be feasible.

In this study, we introduce a methodology for the optimal design of piezoelectric energy harvesters by employing topology optimization based on the level set method. This approach emphasizes manufacturability in the field of microfabrication, with etching being the predominant process under consideration. The design criteria for piezoelectric energy harvesters are delineated based on the operational frequency and a threshold for the excitation voltage output. Furthermore, the proposed method uniquely facilitates the concurrent optimization of three-dimensional structures within both the substrate and piezoelectric material domains. The resulting designs not only satisfy the stipulated specifications but are also readily adaptable to microfabrication procedures.

The remainder of this paper is organized as follows: In § 2, the architectural framework of the designated device is described. After this initial exposition, in that section we further elaborate on the optimal design approach, articulating the pivotal design criteria, including the eigenfrequency specifications and constraints regarding the minimum output voltage. In § 3, we explain the formulation of constraints pertinent to manufacturability. Specifically, we introduce two constraints to enhance manufacturability: ensuring a consistent cross-sectional shape across each domain and stipulation that no piezoelectric material is positioned without an accompanying substrate. In § 4, the implementation of the proposed method is described. In § 5, numerical illustrations employing a benchmark model to test the efficacy of the introduced method are detailed. Finally, in § 6, the conclusions of this study are summarized. Additionally, the methodologies tailored for computing the output voltage are detailed in Appendix A, and the sensitivity analyses are discussed in Appendix B.

Formulation for the optimal design of a piezoelectric energy harvester

Unimorph cantilevered piezoelectric energy harvester

This study predominantly focuses on the unimorph cantilevered energy harvester. A unimorph cantilevered piezoelectric energy harvester comprises three main components: substrate, piezoelectric material, and weight. These elements are depicted in Figure 1, which shows the theoretical structure of a unimorph cantilevered piezoelectric vibration energy harvester. A unimorph cantilevered piezoelectric energy harvester consists of a cantilever beam with one end fixed and the other end free to move. The beam has a structure typically composed of three primary domains: a substrate domain, a piezoelectric material domain, and a weight domain at the free end. Additionally, it includes electrode domains for electrical energy extraction. The substrate domain is nonpiezoelectric and serves as the base layer to which the piezoelectric material is bonded. In this study, the substrate is composed of silicon, and the piezoelectric material utilized is lead zirconate titanate (PZT). To ensure electrical insulation between the electrodes and the substrate domain, an insulating layer is formed on both the upper and lower surfaces of the substrate domain. When the device is subjected to external vibrations or mechanical loads, the weight causes the beam to bend more significantly, inducing mechanical strain in the piezoelectric domain. This strain generates electrical energy through the piezoelectric effect. The presence of the weight enhances the bending of the beam, resulting in greater deformation of the piezoelectric layer and thus a higher electrical output.

Figure 1. Concept of a piezoelectric energy harvester

The piezoelectric effect is a phenomenon where certain materials, such as quartz and PZT, generate an electric charge in response to applied mechanical stress. In a piezoelectric material, when mechanical strain is induced, an electric potential is developed across the material due to the displacement of charge carriers. This property allows piezoelectric materials to be used in energy harvesting devices, converting mechanical energy from vibrations into electrical energy.

The unique advantage of the unimorph design is its capability to produce a superior electrical output for a given mechanical input compared to a conventional cantilevered piezoelectric energy harvester. This superiority is attributed to the mechanical amplification effect induced by the substrate. domain, which results in an enhanced deformation of the piezoelectric layer for a given input vibration. From a design viewpoint, the unimorph-type harvester features a simpler structure with a single piezoelectric layer, allowing for easier fabrication and reduced manufacturing costs. The design flexibility is also increased as both the substrate and the piezoelectric layer can be independently etched, allowing for more complex geometries and optimized vibration characteristics.

Formulation of piezoelectric phenomena

The governing equations for a piezoelectric energy harvester are articulated in the following set of simultaneous piezoelectric equations, as elaborated in [4]:

σij=CijklEskl+ekijφ,kdi=eiklskl−εikSφ,k(1)\begin{aligned} \sigma_{ij} &= C^{E}_{ijkl}s_{kl} + e_{kij}\varphi_{,k} \\ d_i &= e_{ikl}s_{kl} - \varepsilon^{S}_{ik}\varphi_{,k} \tag*{(1)} \end{aligned}

Here, σ\sigma is the stress tensor; dd is the electric displacement tensor; ss represents the strain tensor; CEC^{E} is the elasticity tensor at a constant electric field, and we assume that CEC^{E} represents an isotropic linear elastic material; ee represents the piezoelectric stress constant tensor; φ\varphi is the electric potential; φ,k\varphi_{,k} is a derivative with respect to the kk th variable; and εS\varepsilon^{S} is the dielectric constant tensor at a constant strain.

The piezoelectric coefficient tensor ee plays a crucial role in these equations, as it describes the coupling between the mechanical and electrical states of the material. In the second equation, ekije_{kij} represents the piezoelectric effect, where mechanical strain skls_{kl} induces an electric displacement did_i. Conversely, in the first equation, the application of an electric field φ,k\varphi_{,k} results in mechanical stress σij\sigma_{ij}. This bidirectional coupling is the hallmark of piezoelectric materials, allowing them to convert mechanical energy into electrical energy and vice versa. Therefore, the piezoelectric effect creates a strong interaction between the electric field and the elastic field within the material, enabling it to function efficiently as an energy harvester, sensor, or actuator.

Matrix and vector formulation of piezoelectric phenomena

To facilitate numerical computations and implementation, it is often convenient to convert these tensor equations into matrix and vector forms. This involves representing the strain tensor ss in its vector form, commonly using Voigt notation, and expressing the piezoelectric stress constant tensor ee and other relevant tensors in matrix forms.

Using Voigt notation, the strain tensor ss can be represented as a column vector s(u)s(u), and the elasticity tensor CEC^{E} and piezoelectric stress constant tensor ee are transformed accordingly. The electric field EkE_k is related to the electric potential φ\varphi by Ek=−φ,kE_k=-\varphi_{,k}. Consequently, the piezoelectric equations in matrix-vector forms are given by:

{σ=CEs(u)+eT∇φd=es(u)−εS∇φ(2)\begin{cases} \boldsymbol{\sigma}=C^{E}s(u)+e^{T}\nabla\varphi\\ d=es(u)-\varepsilon^{S}\nabla\varphi \tag*{(2)} \end{cases}

In these equations, σ\boldsymbol{\sigma} represents the stress vector (6×1)(6\times1); dd is the electric displacement vector (3×1)(3\times1); s(u)s(u) is the strain vector (6×1)(6\times1); CEC^{E} is the elasticity matrix (6×6)(6\times6); ee is the piezoelectric coefficient matrix (3×6)(3\times6); εS\varepsilon^{S} is the dielectric constant matrix (3×3)(3\times3); and ∇φ\nabla\varphi is the electric field vector (3×1)(3\times1) derived from the electric potential φ\varphi.

In this study, we primarily focus on the steady-state vibration of a piezoelectric energy harvester under harmonic excitation at a frequency ω\omega. Within this context, the displacement vector uu and the electric potential φ\varphi are designated as state variables.

Assuming a time-harmonic form for the displacement and electric potential, such that u(t)=ueiωtu(t)=u e^{i\omega t} and φ(t)=φeiωt\varphi(t)=\varphi e^{i\omega t}, the time-dependent piezoelectric equations with external forcing can be expressed as follows:

{CEs(u)+eT∇φ=ρ∂2u∂t2+f^es(u)−εS∇φ=q^(3)\begin{cases} C^{E}s(u)+e^{T}\nabla\varphi=\rho\frac{\partial^{2}u}{\partial t^{2}}+\hat{f} \\ es(u)-\varepsilon^{S}\nabla\varphi=\hat{q} \tag*{(3)} \end{cases}

Here, ρ\rho denotes the material’s density, f^\hat{f} represents the external forcing function, and q^\hat{q} indicates the externally provided electric charge. By substituting u(t)=ueiωt\mathbf{u}(t) = \mathbf{u}e^{i\omega t} and φ(t)=φeiωt\varphi(t) = \varphi e^{i\omega t} into these equations, we obtain the following:

{CEs(u)−ω2ρu+eT∇φ=f^es(u)−εS∇φ=q^.(4)\left\{ \begin{aligned} C^{E}s(\mathbf{u})-\omega^{2}\rho\mathbf{u}+e^{T}\nabla\varphi&= \hat{f}\\ e s(\mathbf{u})-\varepsilon^{S}\nabla\varphi&= \hat{q}. \end{aligned} \right. \tag*{(4)}

Note that the term −ω2ρu-\omega^{2}\rho\mathbf{u} arises from the second derivative of the harmonic displacement ueiωt\mathbf{u}e^{i\omega t} with respect to time, representing the inertia force.

To determine the eigenfrequencies of the system, we consider the free vibration case with no external forces or charges. This leads to the eigenvalue problem of the piezoelectric field under open-circuit and short-circuit conditions. In the open-circuit condition, the top and bottom electrodes are not connected, implying the electrical resistance between the electrodes is assumed to be infinite. This configuration inhibits any charge transfer between the electrodes, thereby preserving the voltage difference induced by the polarization of the piezoelectric material. Conversely, in the short-circuit condition, the top and bottom electrodes are connected with zero electrical resistance. This allows any charge accumulation due to the polarization of the piezoelectric material to be immediately neutralized, resulting in no electric field within the piezoelectric material. As indicated by the equations (4), the change in the electric field within the piezoelectric material directly affects its stiffness. In practical applications, the device typically operates with a finite electrical resistance connected to electronic circuits, resulting in a state between these two conditions.

Under open-circuit conditions (where q^=0\hat{q}=0), the eigenvalue problem is expressed as follows:

{CEs(uoci)−ωoci2ρuoci+eT∇φoci=0es(uoci)−εS∇φoci=0.(5)\left\{ \begin{aligned} C^{E}s(\mathbf{u}_{oci})-\omega_{oci}^{2}\rho\mathbf{u}_{oci}+e^{T}\nabla\varphi_{oci} &= 0\\ e s(\mathbf{u}_{oci})-\varepsilon^{S}\nabla\varphi_{oci} &= 0. \end{aligned} \right. \tag*{(5)}

Here, ωoci\omega_{oci} is the ii-th mode eigenfrequency under open-circuit conditions, uoci\mathbf{u}_{oci} is the eigenvector corresponding to the eigenvalue ωoci\omega_{oci}, and φoci\varphi_{oci} denotes the electric potential for the ii-th vibration mode under open-circuit conditions. Under open-circuit conditions, the piezoelectric effect increases the stiffness, leading to a higher eigenfrequency ωoci\omega_{oci}. Conversely, under short-circuit conditions (where ∇φ=0\nabla\varphi=0), the eigenvalue problem reduces to the following equation, unaffected by the influence of the piezoelectric material.

CEs(usci)−ωsci2ρusci=0.(6)C^{E}s(\boldsymbol{u}_{sci})-\omega_{sci}^{2}\rho\boldsymbol{u}_{sci}=0. \tag*{(6)}

Here, ωsci\omega_{sci} is the ii-th mode eigenfrequency under short-circuit conditions, and usci\boldsymbol{u}_{sci} is the eigenvector corresponding to the eigenvalue ωsci\omega_{sci}. Under short-circuit conditions, the stiffness is reduced due to the neutralization of the piezoelectric effect, resulting in a lower eigenfrequency ωsci\omega_{sci}.

Comparing the eigenfrequencies under these two conditions provides insight into the impact of the piezoelectric effect on the device’s behavior. The difference in stiffness between open-circuit and short-circuit conditions results in a difference between the eigenfrequencies ωoci\omega_{oci} and ωsci\omega_{sci}. By comparing these eigenfrequencies, the electromechanical coupling coefficient ki2k_i^2 can be calculated, serving as a measure of the effectiveness of the piezoelectric material in converting mechanical energy into electrical energy. The electromechanical coupling coefficient for each vibration mode is given by the following equation [37]:

ki2=ωoci2−ωsci2ωoci2.(7)k_i^2=\frac{\omega_{oci}^{2}-\omega_{sci}^{2}}{\omega_{oci}^{2}}. \tag*{(7)}

In the context of a unimorph cantilevered piezoelectric energy harvester design, the eigenfrequency of the device is established as a crucial design parameter for the following reasons. First, these devices generate maximum output when they are resonating at their eigenfrequency, as the energy conversion efficiency is highest in this state. Second, by tuning the eigenfrequency to match the frequency of the ambient vibration source, the harvester can achieve optimal performance at its installation location by effectively capturing and converting the environmental vibration energy into electrical energy [4]. Consequently, in this study, we considered strategies for concurrently aligning the eigenfrequency with the desired value. Adopting the objective function proposed by Yamada et al. [38], we normalized the eigenfrequency specification problem with respect to the ii-th mode desired frequency ωˉi\bar{\omega}_i and formulated the objective function FωF_\omega as follows:

Fω=∑i=1n∣ωoci−ωˉi∣2ωˉi2,(8)F_\omega=\sum_{i=1}^{n}\frac{\lvert\omega_{oci}-\bar{\omega}_i\rvert^2}{\bar{\omega}_i^2}, \tag*{(8)}

where nn donates the number of modes to evaluation. In this study, for the design of a 3-dimensional beam, we set n=4n=4. This is because we need to consider not only the first mode of vibration but also the second mode and bending vibrations in the direction perpendicular to the primary vibration direction, as well as torsional vibrations of the beam, which collectively require evaluating up to the fourth mode.

In this study, we aim to design a high-performance piezoelectric energy harvester by maximizing the electromechanical coupling coefficient ki2k_i^2. To transform the maximization problem into a minimization problem, we formulated the objective function FkF_k using the electromechanical coupling coefficient kik_i as follows:

Fk=∑i=1n1ki2ωˉi2=∑i=1nωoci2(ωoci2−ωsci2)ωˉi2.(9)F_k = \sum_{i=1}^{n} \frac{1}{k_i^2\bar{\omega}_i^2} = \sum_{i=1}^{n} \frac{\omega_{oci}^2}{(\omega_{oci}^2-\omega_{sci}^2)\bar{\omega}_i^2}. \tag*{(9)}

By weighting the performance kik_i by the modes ωˉi\bar{\omega}_i, we emphasized the lower-order modes.

For a unimorph cantilevered piezoelectric energy harvester design, we perform the simultaneous optimization of the two objective functions FkF_k and FωF_\omega using a weighting factor α\alpha as follows:

inf⁡ΩF=αFk+(1−α)Fω.(10)\inf_{\Omega} \quad F = \alpha F_k + (1-\alpha)F_\omega. \tag*{(10)}

This approach allows us to obtain an optimized design that achieves enhanced power generation efficiency among solutions with the desired eigenfrequency. Furthermore, to optimize both the substrate and piezoelectric material regions, we employ the following method.

We formulate an optimization problem that simultaneously minimizes distinct objective functions for different material domains of the energy harvester, specifically, FsbF_{sb} for the substrate material domain Ωsb\Omega_{sb} and FpeF_{pe} for the piezoelectric material domain Ωpe\Omega_{pe}. The substrate material domain Ωsb\Omega_{sb} consists of silicon, while the piezoelectric material domain Ωpe\Omega_{pe} is composed of PZT. The eigenvectors uoci\mathbf{u}_{oci} and usci\mathbf{u}_{sci}, as well as the electric potential φoci\varphi_{oci}, are continuous across the boundary between Ωsb\Omega_{sb} and Ωpe\Omega_{pe}.

The governing equations for each domain under open-circuit and short-circuit conditions are as follows. In the piezoelectric material domain Ωpe\Omega_{pe}, the governing equations under open-circuit conditions are:

CpeEs(uoci)−ωoci2ρpeuoci+eT∇φoci=0in Ωpe(11)C_{pe}^{E}s(\mathbf{u}_{oci})-\omega_{oci}^{2}\rho_{pe}\mathbf{u}_{oci}+e^{T}\nabla\varphi_{oci}=0 \quad\text{in }\Omega_{pe} \tag*{(11)}
es(uoci)−εS∇φoci=0in Ωpe(12)es(\mathbf{u}_{oci})-\varepsilon^{S}\nabla\varphi_{oci}=0 \quad\text{in }\Omega_{pe} \tag*{(12)}

and under short-circuit conditions it is:

CpeEs(usci)−ωsci2ρpeusci=0in Ωpe.(13)C^{E}_{pe}s(\mathbf{u}_{sci})-\omega_{sci}^{2}\rho_{pe}\mathbf{u}_{sci}=0 \quad\text{in }\Omega_{pe}. \tag*{(13)}

Similarly, in the substrate material domain Ωsb\Omega_{sb}, the governing equations under open-circuit conditions is:

CsbEs(uoci)−ωoci2ρsbuoci=0in Ωsb.(14)C^{E}_{sb}s(\mathbf{u}_{oci})-\omega_{oci}^{2}\rho_{sb}\mathbf{u}_{oci}=0 \quad\text{in }\Omega_{sb}. \tag*{(14)}

and under short-circuit conditions is:

CsbEs(usci)−ωsci2ρsbusci=0in Ωsb.(15)C^{E}_{sb}s(\mathbf{u}_{sci})-\omega_{sci}^{2}\rho_{sb}\mathbf{u}_{sci}=0 \quad\text{in }\Omega_{sb}. \tag*{(15)}

Here, CpeEC^{E}_{pe} and CsbEC^{E}_{sb} denote the elastic matrices of the piezoelectric and substrate materials, respectively, while ρpe\rho_{pe} and ρsb\rho_{sb} represent the densities of the piezoelectric and substrate materials. The eigenvector fields uoci\mathbf{u}_{oci} and usci\mathbf{u}_{sci} are shared between the piezoelectric and substrate domains under open-circuit and short-circuit conditions, respectively.

This optimization incorporates weighting factors αsb\alpha_{sb} and αpe\alpha_{pe} to balance the contributions of each domain to the overall objective. The formulation of the objective functions FsbF_{sb} and FpeF_{pe}, and the incorporation of the weighting factors αsb\alpha_{sb} and αpe\alpha_{pe}, are demonstrated below, following the methodology proposed by Kim and Lee [22]:

inf⁡ΩsbFsb=αsbFk+(1−αsb)Fωinf⁡ΩpeFpe=αpeFk+(1−αpe)Fωsubject toCpeEs(uoci)−ωoci2ρpeuoci+eT∇φoci=0in Ωpees(uoci)−εS∇φoci=0in ΩpeCpeEs(usci)−ωsci2ρpeusci=0in ΩpeCsbEs(uoci)−ωoci2ρsbuoci=0in ΩsbCsbEs(usci)−ωsci2ρsbusci=0in Ωsb(16)\begin{aligned} \inf_{\Omega_{sb}}\quad F_{sb} &= \alpha_{sb}F_k+(1-\alpha_{sb})F_{\omega}\\ \inf_{\Omega_{pe}}\quad F_{pe} &= \alpha_{pe}F_k+(1-\alpha_{pe})F_{\omega}\\ \text{subject to}\quad C^{E}_{pe}s(\mathbf{u}_{oci})-\omega_{oci}^{2}\rho_{pe}\mathbf{u}_{oci}+e^{T}\nabla\varphi_{oci} &= 0 \quad\text{in }\Omega_{pe}\\ e s(\mathbf{u}_{oci})-\varepsilon^{S}\nabla\varphi_{oci} &= 0 \quad\text{in }\Omega_{pe}\\ C^{E}_{pe}s(\mathbf{u}_{sci})-\omega_{sci}^{2}\rho_{pe}\mathbf{u}_{sci} &= 0 \quad\text{in }\Omega_{pe}\\ C^{E}_{sb}s(\mathbf{u}_{oci})-\omega_{oci}^{2}\rho_{sb}\mathbf{u}_{oci} &= 0 \quad\text{in }\Omega_{sb}\\ C^{E}_{sb}s(\mathbf{u}_{sci})-\omega_{sci}^{2}\rho_{sb}\mathbf{u}_{sci} &= 0 \quad\text{in }\Omega_{sb} \tag*{(16)} \end{aligned}

Minimum output voltage constraint

Concomitant with the eigenfrequency, the output voltage is a pivotal performance metric in the design of piezoelectric energy harvesters. This attribute is significant because devices harnessing the generated energy often require a minimum viable output voltage. Owing to the physical constitution of the piezoelectric energy harvester, wherein the piezoelectric material is encapsulated between electrodes, the resultant output voltage from the vibratory motion is closely correlated not only with the charge induced on the electrodes but also with the capacitance between the piezoelectric material and the electrodes. A larger piezoelectric material area results in a greater charge, albeit at the cost of increased capacitance. Hence, the capacitance must be meticulously regulated to match the required output voltage. Macroscopically, the output voltage VEV_E is expressed as follows:

VE=QCp=∫Ωpenz⋅∇φn dΩ∫ΩpedΩ/(εzLz2)(17)V_E = \frac{Q}{C_p} = \frac{\int_{\Omega_{pe}} n_z \cdot\nabla\varphi_n \,\mathrm{d}\Omega}{\int_{\Omega_{pe}} \mathrm{d}\Omega/(\varepsilon_z L_z^2)} \tag*{(17)}

where CpC_p signifies the capacitance of the piezoelectric energy harvester, QQ denotes the electrical charge elicited by the piezoelectric material, nzn_z represents a unit vector in the polarization direction, εz\varepsilon_z is the component of the piezoelectric material’s dielectric constant tensor in the polarization direction, and LzL_z is the thickness of the piezoelectric domain. φn\varphi_n denotes the electric potential calculated using the modal superposition method. Details of the modal superposition method are described in Appendix A.

The constraint on the minimum output voltage is formulated as follows:

VE=∫Ωpenz⋅∇φn dΩ∫ΩpedΩ/(εzLz2)≥Vˉmin⁡V,(18)V_E = \frac{\int_{\Omega_{pe}} n_z \cdot\nabla\varphi_n \,\mathrm{d}\Omega}{\int_{\Omega_{pe}} \mathrm{d}\Omega/(\varepsilon_z L_z^2)} \geq\bar{V}_{\min V}, \tag*{(18)}
∫ΩpedΩ≤εzLz2∫Ωpenz⋅∇φn dΩVˉmin⁡V,(19)\int_{\Omega_{pe}} \mathrm{d}\Omega\leq\frac{\varepsilon_z L_z^2 \int_{\Omega_{pe}} n_z \cdot\nabla\varphi_n \,\mathrm{d}\Omega}{\bar{V}_{\min V}}, \tag*{(19)}

where Vˉmin⁡V\bar{V}_{\min V} denotes the minimum output voltage. This formulation allows the minimum output voltage constraint to be treated similarly to the volume constraint.

Concept of topology optimization

Topology optimization is a type of structural optimization method [39, 40, 41]. Structural optimization is used to obtain a structure Ω\Omega that minimizes or maximizes an objective function. The objective function often includes physical properties, such as stiffness [11, 39], as well as thermal [42, 43, 44, 45, 46], fluid [47, 48], electromagnetic [49, 50], and acoustic properties [51, 52, 53, 54].

Formulation of topology optimization

Typically in topology optimization methods, the optimal structure is obtained under the assumption that the objective function satisfies the governing equations that describe the physical phenomena. The governing equations are considered as constraints in the optimization problem, and the basic structural optimization problem can be formulated as follows:

inf⁡ΩF(u,Ω)=∫Ωf(u) dΩsubject tothe governing equations of the physical domain,additional constraint equations,(20)\begin{aligned} \inf_{\Omega}\quad&F(u,\Omega)=\int_{\Omega} f(u)\,\mathrm{d}\Omega\\ \text{subject to}\quad&\text{the governing equations of the physical domain,}\\ &\text{additional constraint equations,} \tag*{(20)} \end{aligned}

where uu is the state variable obtained as the solution of the governing equation and f(u)f(u) is the objective function.

Next, we consider the application of topology optimization to the structural optimization problem (20). We introduce a domain D⊂RnD \subset\mathbb{R}^{n} (n=2n=2 or 33) where the structure can be placed. Here, the domain DD is called a fixed design domain because it does not change during the optimization process. The fixed design domain is a domain filled with the structure (hereafter referred to as the material domain) and a domain not filled with the structure (hereafter referred to as the void domain), and these domains are expressed by the characteristic function χ\chi, which is defined as follows:

χ(x):={1for x∈Ω,0for x∈D∖Ω,(21)\chi(x) := \begin{cases} 1 & \text{for } x \in\Omega,\\ 0 & \text{for } x \in D\setminus\Omega, \end{cases} \tag*{(21)}

where the boundary between the material and void domains is included in the material domain. Using the characteristic function χ\chi, the topology optimization problem can be formulated as follows:

inf⁡χF(u,χ)=∫Df(u)χ dΩsubject tothe governing equation system,constraint equations.(22)\begin{aligned} \inf_{\chi}\quad&F(u,\chi)=\int_{D} f(u)\chi\,\mathrm{d}\Omega\\ \text{subject to}\quad&\text{the governing equation system,}\\ &\text{constraint equations.} \tag*{(22)} \end{aligned}

In topology optimization, the problem described in (20) is reformulated as a material distribution problem. This allows for topological changes, such as an increase or decrease in the number of holes, during the optimization procedure.

However, topology optimization problems are commonly ill-posed [39]; therefore, the space of admissible design should incorporate relaxation or regularization techniques to render the problem well-posed. A typical method based on the relaxation in space of the admissible designs is the homogenization method [11]. In this study, we use a level set-based topology optimization method [32] to transform an ill-posed problem into a well-posed problem. In this method, the boundary surface of the material domain is represented by the isosurface of a scalar function called the level set function, and changes in the level set function represent changes in the shape of the material domain. The topology optimization problem is regularized by ensuring the proper smoothness of the level set function. This method is described in subsection 2.3.2.

Level set-based topology optimization

In the level set-based method, the scalar function ϕ(x)∈H1(D)\phi(x) \in H^{1}(D), also called the level set function, illustrated in the following equation is introduced to represent the shape [55]:

{−1≤ϕ(x)<0for x∈D∖Ωϕ(x)=0for x∈∂Ω1≥ϕ(x)>0for x∈Ω,(23)\begin{cases} -1 \le\phi(x) < 0 & \text{for } x \in D \setminus\Omega\\ \phi(x) = 0 & \text{for } x \in\partial\Omega\\ 1 \ge\phi(x) > 0 & \text{for } x \in\Omega, \tag*{(23)} \end{cases}

where ∂Ω\partial\Omega denotes the boundaries between the material and void domains. We redefined the characteristic function using the level set function as follows:

χϕ(ϕ(x))≔{1for ϕ(x)≥00for ϕ(x)<0.(24)\chi_{\phi}(\phi(x)) \coloneqq \begin{cases} 1 & \text{for } \phi(x) \ge0 \\ 0 & \text{for } \phi(x) < 0. \end{cases} \tag*{(24)}

In the level set-based method, the topology optimization problem is formulated using the characteristic function defined in (24). However, as previously discussed, optimization that employs characteristic functions as design variables constitutes an ill-posed problem. Consequently, the level set-based method offers regularization to this optimization challenge, which is formulated as follows [56]:

inf⁡ϕFR(u,ϕ)=∫Df(u)χϕ(ϕ) dΩ+∫D12∣∇ϕ∣2 dΩsubject tothe governing equations of the physical domain,additional constraint equations,(25)\begin{aligned} \inf_{\phi}\quad& F_R(u,\phi) = \int_D f(u)\chi_{\phi}(\phi)\,\mathrm{d}\Omega+ \int_D \frac{1}{2} \lvert\nabla\phi\rvert^2\,\mathrm{d}\Omega\\ \text{subject to}\quad& \text{the governing equations of the physical domain,} \\ & \text{additional constraint equations,} \tag*{(25)} \end{aligned}

where τ∈R+\tau\in\mathbb{R}_{+} is the regularization coefficient. The optimization problem represented by Eq. (25), is reformulated using Lagrange’s method of undetermined multipliers. Let the Lagrangian be LL and the Lagrange multiplier of the governing equations be vv, and that of the constraint equations be λ\lambda. The optimization problem is transformed into an unconstrained problem as follows:

inf⁡ϕ L=∫Df(u)χϕ(ϕ) dΩ+∫D12τ∣∇ϕ∣2 dΩ+v J(u)+λ G(u).(26)\inf_{\phi}\ L = \int_D f(u)\chi_{\phi}(\phi)\,\mathrm{d}\Omega+ \int_D \frac{1}{2}\tau|\nabla\phi|^2\,\mathrm{d}\Omega+ v\,J(u) + \lambda\,G(u). \tag*{(26)}

Next, we describe a method for updating the level set function. Assuming that the level set function is a function of fictitious time tt, it is updated by the reaction-diffusion equation as follows [32]:

∂ϕ(x,t)∂t=−K(ϕ){−c~F′−τ∇2ϕ(x,t)},c~:=c∫DdΩ∫D∣F′∣ dΩ,F′:=dLdϕ,(27)\begin{aligned} \frac{\partial\phi(x,t)}{\partial t} &= -K(\phi)\left\{-\tilde{c}F' - \tau\nabla^2\phi(x,t)\right\},\\ \tilde{c} &:= \frac{c\int_D \mathrm{d}\Omega}{\int_D |F'|\,\mathrm{d}\Omega},\\ F' &:= \frac{\mathrm{d}L}{\mathrm{d}\phi}, \tag*{(27)} \end{aligned}

where K(ϕ)∈R+K(\phi) \in\mathbb{R}_{+} is the proportionality coefficient, c∈R+c \in\mathbb{R}_{+} is the normalization coefficient, LL is the Lagrangian function, and F′F' is the design sensitivity defined as gradient of the Lagrangian function. K(ϕ)K(\phi) behaves as a coefficient to adjust the magnitude of sensitivity. In this study, we set K=1.0K = 1.0 and c=2.0c = 2.0.

Formulation of constraints for manufacturability

In this section, we describe the formulation of prerequisites related to manufacturing processes. This study sets forth specific manufacturing criteria for piezoelectric energy harvesters, emphasizing configurations amenable to etching methodologies, predominantly utilizing the deep reactive ion etching (DRIE) apparatus [9, 57]. Etching is a crucial step in the fabrication of microstructures, allowing for the precise removal of material. There are various etching techniques, primarily categorized into wet etching and dry etching. Wet etching involves chemical solutions to remove material, while dry etching uses plasma or gases. For the purposes of this study, we utilize the deep reactive ion etching (DRIE) method. Figure 2 outlines the DRIE process. DRIE is a highly anisotropic etching process that allows for the creation of complex structures in the planar direction using photolithography. This anisotropy means that etching can be precisely controlled in one direction, typically perpendicular to the substrate, while controlling etching in other directions is significantly more challenging. As a result, ensuring structural uniformity in the thickness direction is critical when designing structures using DRIE.

Figure 2. Steps of the DRIE process creating precise microstructures. (a) Application of resist: A layer of photoresist is applied to the surface of the material. (b) Photolithography: (1) A photomask is aligned over the photoresist-coated material. (2) UV exposure through the photomask transfers the pattern onto the photoresist. (3) The photoresist is developed, revealing the pattern. (c) Etching: (1) Etching begins in the exposed areas by exposing the surface to reactive ions. (2) Etching progresses to achieve the desired depth and pattern. (3) Etching is completed with precise control in the vertical direction. (d) Removal of resist: The remaining photoresist is removed, leaving behind the etched pattern.

To this end, constraints are introduced to ensure structural uniformity in the thickness direction for both the piezoelectric and substrate domains. Such constraints prevent the emergence of configurations with apertures oriented in oblique or orthogonal directions relative to the processing trajectory.

When etching is performed from only one side, the manufacturing constraints can be satisfied by imposing a constraint that the entire device has a uniform cross-section using a single level set function. This approach is mainly used for comparison in § 5.1. On the other hand, when etching is performed from both sides, the structure does not need to have a uniform cross-section collectively. It is permissible for the piezoelectric material and the substrate to each have uniform cross-sections individually while maintaining overall manufacturability. By using the extended level set method [35], a type of multi-material topology optimization method, we can satisfy the constraints for etching from both sides while ensuring design flexibility.

In addition to ensuring structural uniformity, it is crucial to impose a substrate-dependent constraint on the piezoelectric domain. This is to preclude the occurrence of piezoelectric materials in areas devoid of a substrate, which can result in mechanical instability and reduced device performance. Without this constraint, the piezoelectric material may be deposited in areas where it lacks adequate support from the substrate, leading to potential delamination or cracking during the fabrication process. Figure 3 illustrates the difference between constrained and not constrained structures. Ensuring that the piezoelectric materials are only present where there is a supporting substrate helps maintain the mechanical integrity and reliability of the device.

Figure 3. Illustration of the substrate-dependent constraint on the piezoelectric domain. The left side shows a ”Constrained structure,” where the piezoelectric material is properly supported by the substrate, ensuring mechanical stability and integrity. The right side shows a ”Not constrained structure,” where piezoelectric material is deposited without substrate support, leading to potential delamination or cracking.

In this section, we present different methods to model and address the etching fabrication constraint. § 3.1 introduces methods to ensure that the cross-sectional shape of the device is uniform. Specifically, § 3.1.1 describes the formulation of constraints for etching from a single side using the single level set function method, which is mainly used for comparative purposes. § 3.1.2 presents the formulation of constraints for etching from both sides using the extended level set function method. § 3.2 details the formulation of substrate-dependent constraints. Finally, § 3.3 formulates the optimization problem incorporating both substrate-dependent constraints and the extended level set method.

Constraint: the same cross-sectional shape

To ensure manufacturability via etching, it is crucial to impose constraints that maintain structural uniformity in the thickness direction. This section describes methods to achieve a uniform cross-sectional shape for the piezoelectric and substrate domains. The concept involves increasing the value of the regularization coefficient τ\tau in the thickness direction to achieve uniform cross-sections. In this study, as shown in Figure 4, we have two different design domains: DsbD_{sb} for the substrate material and DpeD_{pe} for the piezoelectric material. The entire design domain is defined as D=Dpe∪DsbD = D_{pe} \cup D_{sb}. We address two approaches to enforce this constraint: one using a single level set function for etching from a single side, and the other using the extended level set method for etching from both sides. Figure 5 illustrates the conceptual structures that satisfy each constraint. Structures as shown in Step (b) can be achieved by single-side etching, while structures as shown in Step (c) can be obtained through double-side etching.

Figure 4. The design domains used in this study. 3D view: Visualization of the 3D structure showing the division into the piezoelectric design domain (DpeD_{pe}) and the substrate design domain (DsbD_{sb}). Side view: Cross-sectional side view illustrating the components contained within each domain. DpeD_{pe} is the design domain for the piezoelectric material, while DsbD_{sb} is design domain for the substrate material.

Figure 5. Schematic illustration of the etching process for a unimorph cantilevered piezoelectric energy harvester. Initial configuration: Before any etching process begins. Step (a) Weight formation: The structure after etching from the backside to form the weight. Step (b) Beam and PZT formation: The structure after etching from the backside to form the silicon beam and PZT shapes. This step represents single-side etching. Step (c) Final PZT formation: The structure after etching from the front side to form the PZT shape differing from the beam. This step represents double-side etching.

Constraint: the same cross-sectional shape throughout the design domain

Within the framework of topology optimization obtained via the level set method, qualitative control over complexity can be regulated by the regularization coefficient τ\tau, as defined in Eq. (25) [32, 56]. To leverage this control over complexity, we adopt the approach used by Yamada et al. [32] to assign regularization coefficients that curtail the complexity in the thickness direction of the structure, thereby facilitating manufacturability via etching. In this study, we extend the regularization coefficient to a tensor τ\boldsymbol{\tau} and modify Eqs. (25) and (27) as follows:

inf⁡ϕFR(u,ϕ)=∫Df(u)χϕ(ϕ) dΩ+∫D12∣(∇ϕ)Tτ∇ϕ∣ dΩ(28)\inf_{\phi}\quad F_{\mathrm{R}}(u,\phi)=\int_D f(u)\chi_{\phi}(\phi)\,\mathrm{d}\Omega+\int_D \frac{1}{2}\left\lvert(\nabla\phi)^{T}\boldsymbol{\tau}\nabla\phi\right\rvert\,\mathrm{d}\Omega \tag*{(28)}

subject to the governing equations of the physical domain, additional constraint equations,

∂ϕ(x,t)∂t=−K(ϕ){−cF~′−∇Tτ∇ϕ(x,t)},(29)\frac{\partial\phi(x,t)}{\partial t}=-K(\phi)\left\{-c\tilde{F}^{\prime}-\nabla^{T}\boldsymbol{\tau}\nabla\phi(x,t)\right\}, \tag*{(29)}

where

τ=[τx000τy000τz].(30)\boldsymbol{\tau}= \begin{bmatrix} \tau_x & 0 & 0\\ 0 & \tau_y & 0\\ 0 & 0 & \tau_z \end{bmatrix}. \tag*{(30)}

In this context, augmenting the component of the regularization coefficient tensor associated with the gradient in a specific direction of the level set function simplifies the complexity along that direction. By defining the thickness direction as zz and assigning τx,τy≪τz\tau_x,\tau_y\ll\tau_z, we can obtain a structure exhibiting minimal complexity in the thickness direction, thus enhancing manufacturability via etching.

To maintain a consistent cross-sectional shape throughout the design domain, the optimization problem is formulated using a single level set function ϕ\phi and a single weight factor α\alpha. In this study, since the physical properties of the object domain differ between the design domain of the piezoelectric material DpeD_{pe} and the design domain of the substrate DsbD_{sb}, a constant function ϕps\phi_{ps} is used to define the governing equations as follows:

{ϕps(x)=1if x∈Dpeϕps(x)=−1if x∈Dsb,(31)\begin{cases} \phi_{ps}(x)=1 & \text{if }x\in D_{pe}\\ \phi_{ps}(x)=-1 & \text{if }x\in D_{sb}, \tag*{(31)} \end{cases}
inf⁡ϕF=αFk+(1−α)Fω,\inf_{\phi}\quad F=\alpha F_k+(1-\alpha)F_{\omega},

subject to

(CpeEχϕ(ϕps)+CsbEχϕ(1−ϕps))χϕ(ϕ)s(uoci)−ωoci2(ρpeχϕ(ϕps)+ρsb(1−χϕ(ϕps)))χϕ(ϕ)uoci+eTχϕ(ϕps)∇φoci=0in D,eχϕ(ϕps)χϕ(ϕ)s(uoci)−ϵSχϕ(ϕ)∇φoci=0in D,(CpeEχϕ(ϕps)+CsbEχϕ(1−ϕps))χϕ(ϕ)s(usci)−ωsci2(ρpeχϕ(ϕps)+ρsb(1−χϕ(ϕps)))χϕ(ϕ)usci=0in D.(32)\begin{aligned} \left(C_{pe}^{E}\chi_{\phi}(\phi_{ps})+C_{sb}^{E}\chi_{\phi}(1-\phi_{ps})\right)\chi_{\phi}(\phi)\boldsymbol{s}(u_{oci})\\ -\omega_{oci}^{2}\left(\rho_{pe}\chi_{\phi}(\phi_{ps})+\rho_{sb}\left(1-\chi_{\phi}(\phi_{ps})\right)\right)\chi_{\phi}(\phi)u_{oci}\\ +e^{T}\chi_{\phi}(\phi_{ps})\nabla\varphi_{oci}=0 \qquad\text{in }D,\\ e\chi_{\phi}(\phi_{ps})\chi_{\phi}(\phi)\boldsymbol{s}(u_{oci})-\epsilon^{S}\chi_{\phi}(\phi)\nabla\varphi_{oci}=0 \qquad\text{in }D,\\ \left(C_{pe}^{E}\chi_{\phi}(\phi_{ps})+C_{sb}^{E}\chi_{\phi}(1-\phi_{ps})\right)\chi_{\phi}(\phi)\boldsymbol{s}(u_{sci})\\ -\omega_{sci}^{2}\left(\rho_{pe}\chi_{\phi}(\phi_{ps})+\rho_{sb}\left(1-\chi_{\phi}(\phi_{ps})\right)\right)\chi_{\phi}(\phi)u_{sci}=0 \qquad\text{in }D. \tag*{(32)} \end{aligned}

Constraint: the same cross-sectional shape in each domain

Although the method described above controls the complexity of the structure throughout the design domain, the substrate and piezoelectric components can be readily processed via etching, even if their structures differ. It is desirable to control the complexity in the thickness direction within the substrate and piezoelectric domains without affecting the structural differences at the interface of each domain. To this end, in this study, we defined separate level set functions for the substrate and piezoelectric structures, invoking the extended level set method [35], which is capable of handling multiple materials and updating each level set function concurrently. In the extended level set method, when dealing with a void domain and two types of object domains, making three types of domains in total, the level set function matrix is defined as follows:

ϕ=[ϕ00ϕ01ϕ02ϕ10ϕ11ϕ12ϕ20ϕ21ϕ22],(33)\phi= \begin{bmatrix} \phi_{00} & \phi_{01} & \phi_{02} \\ \phi_{10} & \phi_{11} & \phi_{12} \\ \phi_{20} & \phi_{21} & \phi_{22} \end{bmatrix}, \tag*{(33)}
χa=∏b∈[0,1,2]∖aχϕ(ϕba),(34)\chi_a = \prod_{b\in[0,1,2]\setminus a}\chi_{\phi}(\phi_{ba}), \tag*{(34)}
ϕaa=0(a∈[0,1,2])(35)\phi_{aa}=0 \qquad(a\in[0,1,2]) \tag*{(35)}
ϕab=−ϕba(a≠b)(36)\phi_{ab}=-\phi_{ba} \qquad(a\ne b) \tag*{(36)}

In this framework, ϕpψ\phi_{p\psi} is defined as ϕ10\phi_{10}, representing the boundary between the piezoelectric and void domains, and ϕsψ\phi_{s\psi} is defined as ϕ20\phi_{20}, representing the boundary between the substrate and void domains. Therefore, the level set function matrix can be redefined as follows:

ϕ=[ϕψψϕψpϕψsϕpψϕppϕpsϕsψϕspϕss],(37)\phi= \begin{bmatrix} \phi_{\psi\psi} & \phi_{\psi p} & \phi_{\psi s} \\ \phi_{p\psi} & \phi_{pp} & \phi_{ps} \\ \phi_{s\psi} & \phi_{sp} & \phi_{ss} \end{bmatrix}, \tag*{(37)}
ϕψp=−ϕpψ,(38)\phi_{\psi p}=-\phi_{p\psi}, \tag*{(38)}
ϕψs=−ϕsψ,(39)\phi_{\psi s}=-\phi_{s\psi}, \tag*{(39)}
ϕsp=−ϕps,(40)\phi_{sp}=-\phi_{ps}, \tag*{(40)}
ϕψψ=ϕpp=ϕss=0,(41)\phi_{\psi\psi}=\phi_{pp}=\phi_{ss}=0, \tag*{(41)}
χp(x)={1if x∈Ωpe0if x∈D∖Ωpe,(42)\chi_p(\mathbf{x}) = \begin{cases} 1 & \text{if } \mathbf{x} \in\Omega_{pe} \\ 0 & \text{if } \mathbf{x} \in D \setminus\Omega_{pe}, \end{cases} \tag*{(42)}
χs(x)={1if x∈Ωsb0if x∈D∖Ωsb,(43)\chi_s(\mathbf{x}) = \begin{cases} 1 & \text{if } \mathbf{x} \in\Omega_{sb} \\ 0 & \text{if } \mathbf{x} \in D \setminus\Omega_{sb}, \end{cases} \tag*{(43)}
χψ(x)={1if x∈D∖(Ωpe∪Ωsb)0if x∈Ωpe∪Ωsb,(44)\chi_\psi(\mathbf{x}) = \begin{cases} 1 & \text{if } \mathbf{x} \in D \setminus(\Omega_{pe} \cup\Omega_{sb}) \\ 0 & \text{if } \mathbf{x} \in\Omega_{pe} \cup\Omega_{sb}, \end{cases} \tag*{(44)}

where χp\chi_p denotes the piezoelectric material domain, χs\chi_s denotes the substrate material domain, and χψ\chi_\psi denotes void domain. The fixed design domains for the substrate and piezoelectric material, denoted as DsbD_{sb} and DpeD_{pe}, respectively, are assumed to remain invariant throughout the optimization procedure. The entire fixed design domain, denoted as DD, encompasses both DpeD_{pe} and DsbD_{sb}. The level set function ϕps\phi_{ps}, representing the boundary between the domains ωpe\omega_{pe} and ωsb\omega_{sb}, is defined as a constant value during the optimization process, as shown in the following equation:

{ϕps(x)=1if x∈Dpeϕps(x)=−1if x∈Dsb,(45)\begin{cases} \phi_{ps}(\mathbf{x}) = 1 & \text{if } \mathbf{x} \in D_{pe} \\ \phi_{ps}(\mathbf{x}) = -1 & \text{if } \mathbf{x} \in D_{sb}, \tag*{(45)} \end{cases}

The level set functions ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi} are updated as follows:

∂ϕpψ∂t=−K(−c~Fpψ′−∇(τpψ∇ϕpψ)),(46)\frac{\partial\phi_{p\psi}}{\partial t} = -K\left(-\tilde{c}F'_{p\psi} - \nabla\left(\boldsymbol{\tau}_{p\psi}\nabla\phi_{p\psi}\right)\right), \tag*{(46)}
∂ϕsψ∂t=−K(−c~Fsψ′−∇(τsψ∇ϕsψ)),(47)\frac{\partial\phi_{s\psi}}{\partial t} = -K\left(-\tilde{c}F'_{s\psi} - \nabla\left(\boldsymbol{\tau}_{s\psi}\nabla\phi_{s\psi}\right)\right), \tag*{(47)}

where Fpψ′F'_{p\psi} and Fsψ′F'_{s\psi} are the gradients of the objective function with respect to ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}, respectively, and τpψ\boldsymbol{\tau}_{p\psi} and τsψ\boldsymbol{\tau}_{s\psi} are the regularization coefficient tensors for ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}, respectively. By individually defining the regularization coefficient tensors for the piezoelectric and substrate domains, the complexity in the thickness direction for each domain can be controlled. This approach allows for an increase in design flexibility without compromising manufacturability, compared to using a single level set function for the entire fixed design domain DD.

Constraint: no piezoelectric material is placed without a substrate

Herein, we elucidate an approach that allows the imposition of geometric constraints by using a fictitious physical function field denoted by ξ\xi. The concept of fictitious physical models described by a steady-state anisotropic advection-diffusion equation is initially introduced by Sato et al. [28]. A further elaboration of the fictitious physical model using the anisotropic diffusion equation is provided by Yamada et al. [30], which is described below.

−div⁡(κξ∇ξ)=0in D,ξ=ξˉsin Ωsb,ξ=−ξˉ0on Γξ,∇ξ⋅n=0on ∂D∖Γξ.(48)\begin{aligned} -\operatorname{div}(\boldsymbol{\kappa}_{\xi}\nabla\xi) &= 0 && \text{in } D,\\ \xi&= \bar{\xi}_{s} && \text{in } \Omega_{sb},\\ \xi&= -\bar{\xi}_{0} && \text{on } \Gamma_{\xi},\\ \nabla\xi\cdot\boldsymbol{n} &= 0 && \text{on } \partial D \setminus\Gamma_{\xi}. \tag*{(48)} \end{aligned}
κξ=(κx000κy000κz),(49)\boldsymbol{\kappa}_{\xi} = \begin{pmatrix} \kappa_x & 0 & 0\\ 0 & \kappa_y & 0\\ 0 & 0 & \kappa_z \end{pmatrix}, \tag*{(49)}

where ∂D\partial D denotes the outer boundary of the entire fixed design domain DD; κξ\boldsymbol{\kappa}_{\xi} represents the conductivity tensor of ξ\xi; κx\kappa_x, κy\kappa_y, and κz\kappa_z represent the anisotropic conductivities along the xx, yy, and zz directions, respectively; and ξˉs∈R+\bar{\xi}_{s} \in\mathbb{R}_{+} and ξˉ0∈R+\bar{\xi}_{0} \in\mathbb{R}_{+} are the constants for the domain and boundary conditions of ξ\xi. In this study, the thickness direction is represented by the zz-axis, which is employed as a constraint to determine the presence or absence of the substrate. The anisotropic conductivities κx\kappa_x and κy\kappa_y are set to values substantially lower than κz\kappa_z. The boundary Γξ\Gamma_{\xi}, subjected to the Dirichlet boundary condition, is identified as the boundary opposite to domain DpeD_{pe} within the domain DsbD_{sb}. The second equation of (48) ensures that ξ=ξˉs\xi= \bar{\xi}_{s} within the substrate material domain Ωsb\Omega_{sb}. The physical field ξ\xi behaves as a fictitious heat within the fixed design domain DD. In the domain Ωsb\Omega_{sb}, it acts as a heat source with a temperature ξ=ξˉs\xi= \bar{\xi}_{s}, and a cold heat bath with ξ=−ξˉ0\xi= -\bar{\xi}_{0} is applied at the boundary Γξ\Gamma_{\xi}. Within the substrate material domain Ωsb\Omega_{sb}, the fictitious heat ξ\xi is applied and propagates in the zz-direction towards the domain DpeD_{pe} when a material is present within Ωsb\Omega_{sb}. Conversely, in the absence of a material within DsbD_{sb}, the boundary condition ξ=−ξˉ0\xi= -\bar{\xi}_{0} is conveyed to the domain DpeD_{pe} owing to the lack of ξ\xi propagation in Ωsb\Omega_{sb}. Consequently, within the piezoelectric design domain DpeD_{pe}, ξ\xi assumes positive values above the substrate and negative values elsewhere. By defining Ωpe\Omega_{pe} as regions where ϕpψ>0\phi_{p\psi} > 0 and ξ>0\xi> 0, we ensure that the structure adheres to the substrate-dependent constraint:

χp(x)={1for x∈Ωpe0for x∈D\Ωpe,(50)\chi_p(\mathbf{x}) = \begin{cases} 1 & \text{for } \mathbf{x} \in\Omega_{pe} \\ 0 & \text{for } \mathbf{x} \in D \backslash\Omega_{pe}, \end{cases} \tag*{(50)}
χp(x):=χϕ(ϕpψ(x))χϕ(ξ(x))(51)\chi_p(\mathbf{x}) := \chi_{\phi}(\phi_{p\psi}(\mathbf{x}))\chi_{\phi}(\xi(\mathbf{x})) \tag*{(51)}

Figure 6 visually elucidates the behavior of the fictitious heat field ξ\xi and its role in defining the piezoelectric material domain. The figure depicts the initial structure without constraints, followed by the propagation of the fictitious heat field within both the substrate and piezoelectric domains. The fictitious heat field ensures that no piezoelectric material is placed without the presence of a substrate.

Figure 6. Schematic illustration of the fictitious heat field used to enforce the substrate-dependent constraint. Non-compliant Structure: Initial configuration without constraints. Fictitious Heat Bath at Γξ\Gamma_{\xi}: Cold heat bath with ξ=−ξˉ0\xi= -\bar{\xi}_0 is applied at Γξ\Gamma_{\xi}, the boundary opposite to domain DpeD_{pe} within domain DsbD_{sb}. Fictitious Heat Field in DsbD_{sb}: Substrate material domain Ωsb\Omega_{sb} as a heat source with ξ=ξˉs\xi= \bar{\xi}_s. Fictitious Heat Propagation: Heat propagation in the zz-direction towards DpeD_{pe}. In the absence of material in Ωsb\Omega_{sb}, ξ=−ξˉ0\xi= -\bar{\xi}_0 is conveyed to DpeD_{pe}. Piezoelectric Domain Represented by Fictitious Heat Field: Ωpe\Omega_{pe} expressed as χp(x):=χϕ(ϕpψ(x))χϕ(ξ(x))\chi_p(\mathbf{x}) := \chi_{\phi}(\phi_{p\psi}(\mathbf{x}))\chi_{\phi}(\xi(\mathbf{x})) to satisfy the substrate-dependent constraint. Compliant Structure: Structure satisfying the substrate-dependent constraint.

Formulation of the optimal design with proposed constraints

By incorporating the aforementioned methodologies, the optimal design problem is formulated as follows:

inf⁡ϕpψFpeinf⁡ϕsψFsbsubject tofor i=1,…,naoci(ϕpψ,ϕsψ,ωoci,uoci,voci,ϕoci,ξ)=0,boci(ϕpψ,ϕsψ,uoci,ϕoci,vϕoci,ξ)=0,asci(ϕpψ,ϕsψ,ωsci,usci,vsci,ξ)=0,aξ(ϕpψ,ϕsψ,ξ,vξ)=0,GV(ϕpψ,ϕn)≤0.(52)\begin{aligned} \inf_{\phi_{p\psi}} \quad& F_{pe} \\ \inf_{\phi_{s\psi}} \quad& F_{sb} \\ \text{subject to} \quad& \text{for } i=1,\ldots,n \\ & a_{oci}(\phi_{p\psi},\phi_{s\psi},\omega_{oci},\mathbf{u}_{oci},v_{oci},\phi_{oci},\xi)=0, \\ & b_{oci}(\phi_{p\psi},\phi_{s\psi},\mathbf{u}_{oci},\phi_{oci},v_{\phi_{oci}},\xi)=0, \\ & a_{sci}(\phi_{p\psi},\phi_{s\psi},\omega_{sci},\mathbf{u}_{sci},v_{sci},\xi)=0, \\ & a_{\xi}(\phi_{p\psi},\phi_{s\psi},\xi,v_{\xi})=0, \\ & G_V(\phi_{p\psi},\phi_n)\leq0. \tag*{(52)} \end{aligned}
aoci(ϕpψ,ϕsψ,ωoci,uoci,voci,φoci)=(∫Ωpes(voci)TCpeEs(uoci) dΩ−ωoc2∫ΩpeρpevociTuoci dΩ+∫Ωpes(voci)TeT∇φoci dΩ)+(∫Ωsbs(voci)TCsbEs(uoci) dΩ−ωoc2∫ΩsbρsbvociTuoci dΩ)=0,(53)\begin{aligned} a_{oci}(\phi_{p\psi},\phi_{s\psi},\omega_{oci},\mathbf{u}_{oci},\mathbf{v}_{oci},\varphi_{oci}) ={}&\left(\int_{\Omega_{pe}}\mathbf{s}(\mathbf{v}_{oci})^{T}\mathbf{C}_{pe}^{E}\mathbf{s}(\mathbf{u}_{oci})\,\mathrm{d}\Omega-\omega_{oc}^{2}\int_{\Omega_{pe}}\rho_{pe}\mathbf{v}_{oci}^{T}\mathbf{u}_{oci}\,\mathrm{d}\Omega\right.\\ &\left.+\int_{\Omega_{pe}}\mathbf{s}(\mathbf{v}_{oci})^{T}\mathbf{e}^{T}\nabla\varphi_{oci}\,\mathrm{d}\Omega\right)\\ &+\left(\int_{\Omega_{sb}}\mathbf{s}(\mathbf{v}_{oci})^{T}\mathbf{C}_{sb}^{E}\mathbf{s}(\mathbf{u}_{oci})\,\mathrm{d}\Omega-\omega_{oc}^{2}\int_{\Omega_{sb}}\rho_{sb}\mathbf{v}_{oci}^{T}\mathbf{u}_{oci}\,\mathrm{d}\Omega\right)=0, \tag*{(53)} \end{aligned}
boci(ϕpψ,ϕsψ,uoci,φoci,vφoci)=∫Ωpe∇vφociTes(uoci) dΩ−∫Ωpe∇vφociTεS∇φoci dΩ=0,(54)b_{oci}(\phi_{p\psi},\phi_{s\psi},\mathbf{u}_{oci},\varphi_{oci},v_{\varphi_{oci}}) =\int_{\Omega_{pe}}\nabla v_{\varphi_{oci}}^{T}\mathbf{e}\mathbf{s}(\mathbf{u}_{oci})\,\mathrm{d}\Omega-\int_{\Omega_{pe}}\nabla v_{\varphi_{oci}}^{T}\varepsilon^{S}\nabla\varphi_{oci}\,\mathrm{d}\Omega=0, \tag*{(54)}
asci(ϕpψ,ϕsψ,ωsci,usci,vsci)=(∫Ωpes(vsci)TCpeEs(usci) dΩ−ωsci2∫ΩpeρpevsciTusci dΩ+∫Ωsbs(vsci)TCsbEs(usci) dΩ−ωsci2∫ΩsbρsbvsciTusci dΩ)=0,(55)\begin{aligned} a_{sci}(\phi_{p\psi},\phi_{s\psi},\omega_{sci},\mathbf{u}_{sci},\mathbf{v}_{sci}) ={}&\left(\int_{\Omega_{pe}}\mathbf{s}(\mathbf{v}_{sci})^{T}\mathbf{C}_{pe}^{E}\mathbf{s}(\mathbf{u}_{sci})\,\mathrm{d}\Omega-\omega_{sci}^{2}\int_{\Omega_{pe}}\rho_{pe}\mathbf{v}_{sci}^{T}\mathbf{u}_{sci}\,\mathrm{d}\Omega\right.\\ &\left.+\int_{\Omega_{sb}}\mathbf{s}(\mathbf{v}_{sci})^{T}\mathbf{C}_{sb}^{E}\mathbf{s}(\mathbf{u}_{sci})\,\mathrm{d}\Omega-\omega_{sci}^{2}\int_{\Omega_{sb}}\rho_{sb}\mathbf{v}_{sci}^{T}\mathbf{u}_{sci}\,\mathrm{d}\Omega\right)=0, \tag*{(55)} \end{aligned}
aξ(ϕpψ,ϕsψ,ξ,vξ)=∫D∇vξκξ∇ξ dΩ+∫Dh(ϕsp)(ξ−ξˉs(2χ(ϕsψ)−1))vξ dΩ=0,(56)a_{\xi}(\phi_{p\psi},\phi_{s\psi},\xi,v_{\xi})=\int_{D}\nabla v_{\xi}\kappa_{\xi}\nabla\xi\,\mathrm{d}\Omega+\int_{D}h(\phi_{sp})(\xi-\bar{\xi}_{s}(2\chi(\phi_{s\psi})-1))v_{\xi}\,\mathrm{d}\Omega=0, \tag*{(56)}
GV(ϕpψ,ϕn)=∫ΩpedΩ−εzLz2∫Ωpenz⋅∇φn dΩV‾min⁡V≤0,(57)G_{V}(\phi_{p\psi},\phi_{n})=\int_{\Omega_{pe}}\mathrm{d}\Omega-\frac{\varepsilon_{z}L_{z}^{2}\int_{\Omega_{pe}}\mathbf{n}_{z}\cdot\nabla\varphi_{n}\,\mathrm{d}\Omega}{\overline{V}_{\min}V}\leq0, \tag*{(57)}

where ρpe\rho_{pe} and ρsb\rho_{sb} represent the densities of the piezoelectric and substrate materials, respectively, whereas voci\mathbf{v}_{oci}, vsci\mathbf{v}_{sci}, vφociv_{\varphi_{oci}}, and vξv_{\xi} serve as test functions. The test functions are detailed as follows: voci\mathbf{v}_{oci} and vsci\mathbf{v}_{sci} are test functions for the eigenvectors uoci\mathbf{u}_{oci} and usci\mathbf{u}_{sci}, respectively. vφociv_{\varphi_{oci}} is a test function for φoci\varphi_{oci}. vξv_{\xi} is a test function for the fictitious physical function ξ\xi. The potential, symbolized by φn\varphi_{n}, is obtained from the displacement u\mathbf{u} derived via the modal superposition method. This is distinct from φoci\varphi_{oci}, which is determined based on the eigenvector uoci\mathbf{u}_{oci}. A comprehensive exposition on the derivation of the potential φn\varphi_{n} through the modal superposition method can be found in Appendix A.

Numerical implementation

Optimization algorithm

The optimization algorithm is implemented as follows:

Step 1: The initial level set function is set.

Step 2: The fictitious physical function field ξ\xi based on Eq. (56) is solved using the finite element method (FEM).

Step 3: The eigenvector fields uoci\boldsymbol{u}_{oci} and usci\boldsymbol{u}_{sci} and eigenvalues ωoci\omega_{oci}, ωsci\omega_{sci}, and φoci\varphi_{oci} derived from Eqs. (53)–(55) are solved using the FEM.

Step 4: The displacement u\boldsymbol{u}, the electric potential φn\varphi_n, the objective functions FpeF_{pe} and FsbF_{sb} derived from Eq. (16) together with the output voltage VEV_E at a frequency ωˉi\bar{\omega}_i and the constraint function GVG_V, which is based on Eq. (57), are evaluated.

Step 5: If the objective functions converge, the optimization procedure is terminated; otherwise, the sensitivity with respect to the objective functions is computed.

Step 6: The level set function is updated using the time evolution equation given by Eq. (27), and the optimization procedure returns to step 2.

All computations, including those involving the FEM solver, are carried out using FreeFEM++ [58]. In this study, convergence is considered achieved when the ratio of the objective function value from the previous iteration to that of the current iteration remains below 1.0×10−61.0 \times10^{-6} for 10 consecutive iterations. Additionally, the computation is terminated if the number of iterations reaches 1000. In the following subsections, we explain the approximations of the displacement and electrical potential fields and the sensitivity analysis.

Approximate solution of the displacement and electrical potential fields based on the Eulerian coordinate system

Within the Eulerian coordinate system, the fixed design domain necessitates the generation of finite elements during each iteration of the optimization process. To mitigate the computational expenditure, an ersatz material approach, as delineated in [56], is employed. In particular, the void domain is assumed to be a structural material with a relatively small Young’s modulus, and the material properties are assumed to be smoothly distributed in the vicinity of the interface between the weak and strong phases. The governing equations, which dictate the displacement field within the FEM, originally pertain to the material domain Ω\Omega. Utilizing the level set, characteristic, and fictitious physical functions delineated in Section 3, coupled with an approximated Heaviside function h(ϕ)h(\phi) to be presented subsequently, we extend these equations to the entire fixed design domain DD as follows:

χϕ(ϕ)≈h(ϕ),\chi_{\phi}(\phi) \approx h(\phi),
h(ϕ):={dfor ϕ<−w,{12+ϕw[1516−ϕ2w2(58−3ϕ216w2)]}(1−d)+dfor −w≤ϕ≤w,1for w<ϕ.(58)h(\phi) := \begin{cases} d & \text{for } \phi< -w, \\ \left\{\frac{1}{2}+\frac{\phi}{w}\left[\frac{15}{16}-\frac{\phi^2}{w^2}\left(\frac{5}{8}-\frac{3\phi^2}{16w^2}\right)\right]\right\}(1-d)+d & \text{for } -w \le\phi\le w, \\ 1 & \text{for } w < \phi. \end{cases} \tag*{(58)}
CϕpsE:=CpeEh(ϕps)h(ϕpψ)h(ξ)+CsbEh(ϕsp)h(ϕsψ),(59)C_{\phi ps}^{E} := C_{pe}^{E}h(\phi_{ps})h(\phi_{p\psi})h(\xi)+C_{sb}^{E}h(\phi_{sp})h(\phi_{s\psi}), \tag*{(59)}
eϕps:=eh(ϕps)h(ϕpψ)h(ξ),(60)e_{\phi ps} := eh(\phi_{ps})h(\phi_{p\psi})h(\xi), \tag*{(60)}
εϕpsS:=ε0I(1−(h(ϕps)h(ϕpψ)h(ξ)+h(ϕsp)h(ϕsψ)))+εSh(ϕps)h(ϕpψ)h(ξ),(61)\varepsilon_{\phi ps}^{S} := \varepsilon_{0}I\left(1-\left(h(\phi_{ps})h(\phi_{p\psi})h(\xi)+h(\phi_{sp})h(\phi_{s\psi})\right)\right)+\varepsilon^{S}h(\phi_{ps})h(\phi_{p\psi})h(\xi), \tag*{(61)}
ρϕps:=ρpeh(ϕps)h(ϕpψ)h(ξ)+ρsbh(ϕsp)h(ϕsψ),(62)\rho_{\phi ps} := \rho_{pe}h(\phi_{ps})h(\phi_{p\psi})h(\xi)+\rho_{sb}h(\phi_{sp})h(\phi_{s\psi}), \tag*{(62)}

where CϕpsEC_{\phi ps}^{E} is the extended elastic tensor, eϕpse_{\phi ps} is the extended piezoelectric constant tensor, εϕpsS\varepsilon_{\phi ps}^{S} is the extended permittivity tensor, ρϕps\rho_{\phi ps} is the extended density, ww is the transition width of the Heaviside function, and dd is a sufficiently small positive number. In this study, we set w=0.9w=0.9 and d=0.01d=0.01.

Sensitivity analysis

Here, we describe a procedure for determining the sensitivity to update level set functions. In this study, based on Eqs. (53)–(57), we define the Lagrangian functions LpeL_{pe} and LsbL_{sb} of the level set functions with respect to ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}, respectively, as follows:

Lpe=Fpe+∑i=1n(aoci(ϕpψ,ϕsψ,ωoci,uoci,vocpe,φoci)+boci(ϕpψ,ϕsψ,uoci,φoci,vφocpe)+asci(ϕpψ,ϕsψ,ωsci,usci,vscpe))+aξ(ϕpψ,ϕsψ,ξ,vξ)+λGV(ϕpψ,ϕn),(63)L_{pe}=F_{pe}+\sum_{i=1}^{n}\left(a_{oci}(\phi_{p\psi},\phi_{s\psi},\omega_{oci},\mathbf{u}_{oci},\mathbf{v}_{ocpe},\varphi_{oci})+b_{oci}(\phi_{p\psi},\phi_{s\psi},\mathbf{u}_{oci},\varphi_{oci},v_{\varphi ocpe})+a_{sci}(\phi_{p\psi},\phi_{s\psi},\omega_{sci},\mathbf{u}_{sci},v_{scpe})\right)+a_{\xi}(\phi_{p\psi},\phi_{s\psi},\xi,v_{\xi})+\lambda GV(\phi_{p\psi},\phi_n), \tag*{(63)}
Lsb=Fsb+∑i=1n(aoci(ϕpψ,ϕsψ,ωoci,uoci,vocsb,φoci)+boci(ϕpψ,ϕsψ,uoci,φoci,vφocsb)+asci(ϕpψ,ϕsψ,ωsci,usci,vscsb))+aξ(ϕpψ,ϕsψ,ξ,vξ),(64)L_{sb}=F_{sb}+\sum_{i=1}^{n}\left(a_{oci}(\phi_{p\psi},\phi_{s\psi},\omega_{oci},\mathbf{u}_{oci},v_{ocsb},\varphi_{oci})+b_{oci}(\phi_{p\psi},\phi_{s\psi},\mathbf{u}_{oci},\varphi_{oci},v_{\varphi ocsb})+a_{sci}(\phi_{p\psi},\phi_{s\psi},\omega_{sci},\mathbf{u}_{sci},v_{scsb})\right)+a_{\xi}(\phi_{p\psi},\phi_{s\psi},\xi,v_{\xi}), \tag*{(64)}

where vocpev_{ocpe}, vocsbv_{ocsb}, vscpev_{scpe}, vscsbv_{scsb}, vφocpev_{\varphi ocpe}, vφocsbv_{\varphi ocsb}, and λ\lambda are the Lagrange multipliers. Using the adjoint method, sensitivity is determined as follows:

Fpe′=∑i=1n(cocpe{aoci(1,ϕsψ,ωoci,uoci,uoci,φoci)+boci(1,uoci,φoci,φoci)}+cscpeasci(1,ϕsψ,ωsci,usci,usci))+λ,(65)F'_{pe}=\sum_{i=1}^{n}\left(c_{ocpe}\left\{a_{oci}(1,\phi_{s\psi},\omega_{oci},\mathbf{u}_{oci},\mathbf{u}_{oci},\varphi_{oci})+b_{oci}(1,\mathbf{u}_{oci},\varphi_{oci},\varphi_{oci})\right\}+c_{scpe}a_{sci}(1,\phi_{s\psi},\omega_{sci},\mathbf{u}_{sci},\mathbf{u}_{sci})\right)+\lambda, \tag*{(65)}
Fsb′=∑i=1n(cocsb{aoci(ϕpψ,1,ωoci,uoci,uoci,φoci)}+cscsb{asci(ϕpψ,1,ωsci,usci,usci)}).(66)F'_{sb}=\sum_{i=1}^{n}\left(c_{ocsb}\left\{a_{oci}(\phi_{p\psi},1,\omega_{oci},\mathbf{u}_{oci},\mathbf{u}_{oci},\varphi_{oci})\right\}+c_{scsb}\left\{a_{sci}(\phi_{p\psi},1,\omega_{sci},\mathbf{u}_{sci},\mathbf{u}_{sci})\right\}\right). \tag*{(66)}

The sensitivity analysis is explained in detail in Appendix § 8.

Numerical examples

In this section, several numerical examples are presented to demonstrate the utility and validity of the proposed method.

The design domain and associated boundary conditions are illustrated in Figure 7. Let DpeD_{pe} denote a square, colored in blue in the figure, with a side length of 500 mm and a thickness of 4 mm. The material in this domain is PZT, which is a piezoelectric material. Similarly, let DsbD_{sb} symbolize a square, indicated in grey in the figure, with a side length of 500 mm and a thickness of 36 mm. The material within this domain is Silicon, which is a nonpiezoelectric material. The fixed design domain DD consists of the combined regions of DpeD_{pe} and DsbD_{sb}. The domain measuring 500 mm in length, 20 mm in width, and 36 mm in thickness, depicted in dark purple in the figure, is a nondesign domain composed of Silicon. This domain will be referred to as the ”Nondesign domain (Silicon)”.

Correspondingly, the domain that is 500 mm in length, 20 mm in breadth, and 4 mm in thickness, illustrated in light purple in the figure, represents a nondesign domain filled with PZT. This domain will be referred to as the ”Nondesign domain (PZT)”.

A fixed boundary condition is imposed on the surface of the Nondesign domain (Silicon) that is opposite to the surface adjacent to the domain DsbD_{sb}. The domain colored red on the right side of the figure, delineating a 20-mm square with a thickness of 40 mm, is designated as the weight domain, which is a nondesign domain filled with objects devoid of piezoelectric properties. To reduce computational load, the weight part is represented as a nondesign domain with a high density. The density is assumed to be 100 times that of the substrate material within this weight domain. This domain will be referred to as the ”Nondesign domain (Weight)”. The mesh configuration for different domains within this example is illustrated in Figure 8. In the nondesign domain (PZT), the mesh size is uniformly set to 10×10×210 \times10 \times2. The design domain DpeD_{pe} (PZT) also has a uniform mesh size of 20×10×220 \times10 \times2. For the nondesign domain (Silicon), the mesh size is 20×10×8.520 \times10 \times8.5 for zz coordinates less than 34, and 20×10×220 \times10 \times2 for zz coordinates 34 and above, near the interface with DpeD_{pe}. The nondesign domain (Weight) follows the same pattern, with a mesh size of 20×10×8.520 \times10 \times8.5 for zz coordinates less than 34, and 20×10×220 \times10 \times2 for zz coordinates 34 and above. Finally, the design domain DsbD_{sb} (Silicon) has a mesh size of 20×10×8.520 \times10 \times8.5 for zz coordinates less than 34, except at the interface with DpeD_{pe}, where the mesh size is 20×10×220 \times10 \times2 for zz coordinates 34 and above. For each condition, the computation time took approximately 100 hours. The number of modes to be contemplated is fixed at n=4n = 4, and the specified natural frequencies are [ωˉ0,ωˉ1,ωˉ2,ωˉ3]=[70 Hz,435 Hz,450 Hz,500 Hz][\bar{\omega}_0, \bar{\omega}_1, \bar{\omega}_2, \bar{\omega}_3] = [70\ \mathrm{Hz}, 435\ \mathrm{Hz}, 450\ \mathrm{Hz}, 500\ \mathrm{Hz}]. The initial structure is conjectured to be such that ϕpψ=1\phi_{p\psi} = 1 in DpeD_{pe} and ϕsψ=1\phi_{s\psi} = 1 in DsbD_{sb}, indicating that the entire fixed design domain is occupied by the material domain.

Figure 8. Mesh configuration in different domains. In the nondesign domain (PZT), the mesh size is uniformly set to 10×10×210 \times10 \times2. The design domain DpeD_{pe} (PZT) has a uniform mesh size of 20×10×220 \times10 \times2. For the nondesign domain (Silicon), the mesh size is 20×10×8.520 \times10 \times8.5 for zz coordinates less than 34, and 20×10×220 \times10 \times2 for zz coordinates 34 and above, near the interface with DpeD_{pe}. The nondesign domain (Weight) has a mesh size of 20×10×8.520 \times10 \times8.5 for zz coordinates less than 34, and 20×10×220 \times10 \times2 for zz coordinates 34 and above. The design domain DsbD_{sb} (Silicon) has a mesh size of 20×10×8.520 \times10 \times8.5 for zz coordinates less than 34, except at the interface with DpeD_{pe}, where the mesh size is 20×10×220 \times10 \times2 for zz coordinates 34 and above.

Figure 7. Design domain and boundary conditions.

Effectiveness of the proposed constraints for manufacturability

In this subsection, we present numerical examples to confirm the effectiveness of the constraints for manufacturability proposed in § 3. In order to verify the effect of ”Constraint: the same cross-sectional shape throughout the design domain” mentioned in subsection 3.1.1, we test the regularization factor τ\tau. Similarly, to verify the effect of ”Constraint: the same cross-sectional shape in each domain” discussed in subsection 3.1.2, we examine the representation of level set functions. Lastly, to verify the effect of ”Constraint: no piezoelectric material is placed without a substrate” described in subsection 3.2, we evaluate the status of constraint implementation via a fictitious field ξ\xi.

We conduct numerical experiments under nine conditions, as listed in Table 1. Each condition is characterized by a distinct setting of the regularization factor τz\tau_z in the thickness direction, the representation of level set functions either a singular function ϕ\phi or a pair of functions ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}, and the status of constraint implementation via a fictitious field ξ\xi. Specifically, for each condition, the regularization factor τz\tau_z is set to 1.0×10−61.0 \times10^{-6}, 1.0×10−41.0 \times10^{-4}, or 1.0×10−21.0 \times10^{-2}. Furthermore, the level set functions are either represented solely by ϕ\phi or by a combination of ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}, and the constraints

conditionτz\tau_zlevel set functionξ\xi constraintFkF_kFωF_\omega
(a)1.0×10−61.0 \times10^{-6}ϕ\phinot constrained4.47×10−54.47 \times10^{-5}1.52×10−51.52 \times10^{-5}
(b)1.0×10−41.0 \times10^{-4}ϕ\phinot constrained7.07×10−57.07 \times10^{-5}6.70×10−66.70 \times10^{-6}
(c)1.0×10−21.0 \times10^{-2}ϕ\phinot constrained8.43×10−58.43 \times10^{-5}4.98×10−64.98 \times10^{-6}
(d)1.0×10−61.0 \times10^{-6}ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}not constrained4.50×10−54.50 \times10^{-5}1.51×10−51.51 \times10^{-5}
(e)1.0×10−41.0 \times10^{-4}ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}not constrained5.80×10−55.80 \times10^{-5}1.33×10−51.33 \times10^{-5}
(f)1.0×10−21.0 \times10^{-2}ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}not constrained7.20×10−57.20 \times10^{-5}1.08×10−51.08 \times10^{-5}
(g)1.0×10−61.0 \times10^{-6}ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}ξ\xi constrained4.70×10−54.70 \times10^{-5}1.25×10−51.25 \times10^{-5}
(h)1.0×10−41.0 \times10^{-4}ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}ξ\xi constrained4.53×10−54.53 \times10^{-5}1.28×10−51.28 \times10^{-5}
(i)1.0×10−21.0 \times10^{-2}ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}ξ\xi constrained8.14×10−58.14 \times10^{-5}4.79×10−64.79 \times10^{-6}

Table 1. Summary of parameter settings and objective functions: regularization factor τz\tau_z in the thickness direction, the representation of level set functions, and the status of constraint implementation via a fictitious field ξ\xi and the corresponding objective functions FkF_k and FωF_\omega for each condition.

are applied using the fictitious field ξ\xi for conditions (g)–(i). Each of these nine configurations is evaluated through 1000 iterations of computation. The parameters corresponding to each condition, coupled with the average values of the objective function FkF_k and FωF_\omega taken over the last 100 iterations, are also presented in Table 1. Under all conditions, the weighting factors αpe\alpha_{pe} and αsb\alpha_{sb} are set to 0.95 to achieve a balanced contribution between FkF_k and FωF_\omega in both the piezoelectric and the substrate domains. The resulting structures for each condition are shown in Figure 9.

The tradeoff relationship between the objective functions FkF_k and FωF_\omega for conditions (a)–(i) delineated in Table 1 is depicted in Figure 10. From the figure, a clear tradeoff relationship can be discerned between FωF_\omega and FkF_k, the balance of which is influenced by changes in the parameter τz\tau_z. Moreover, conditions (d)–(f) yield smaller FkF_k values than conditions (a)–(c), given the same τz\tau_z value. This phenomenon is attributed to the fact that, in conditions (d)–(f), the architecture of the piezoelectric film remains uninfluenced by the structure of the substrate, and, apart from regularization with respect to ϕpψ\phi_{p\psi}, it does not adhere to any constraints. This allows for a wider variety of potential solutions, thereby enhancing the structural degrees of freedom of piezoelectric films. In conditions (g)–(i), although the structure of the piezoelectric film is influenced by the structure of the substrate, the converse does not hold true, resulting in FωF_\omega values that are lower than those under conditions (d)–(f).

Next, the following paragraph explains the effectiveness of the proposed constraints. We compute two normalized quantities: Nϕ1N_{\phi1} and Nϕ2N_{\phi2}. Nϕ1N_{\phi1} is calculated to confirm the extent to which boundaries in the zz direction exist within the entire design domain DD. This Nϕ1N_{\phi1} verifies whether the constant cross-sectional constraint is effective throughout the fixed design domain DD. Nϕ1N_{\phi1} is the proportion of nodes where the level set function changes sign with adjacent nodes in the negative zz-direction, indicating a boundary. Conversely, Nϕ2N_{\phi2} is the proportion of nodes, among those extracted by Nϕ1N_{\phi1}, that are not located at the boundaries of domains DsbD_{sb} and DpeD_{pe}. By calculating Nϕ2N_{\phi2}, we verify whether the constraint of having a constant cross-section in each of DpeD_{pe} and DsbD_{sb} is effective. These values are normalized by the total number of nodes. The computed values for Nϕ1N_{\phi1} and Nϕ2N_{\phi2} for each condition are provided in Table 2.

As delineated in Table 2, irrespective of the level set function configuration and the existence of a ξ\xi constraint, an increase in the τz\tau_z value corresponds to a reduction in the zz-directional complexity. When two level set

Figure 9. Optimal configuration of a unimorph cantilevered energy harvester for conditions (a)–(i) listed in Table 1. The red region denotes the piezoelectric material domain Ωpe\Omega_{pe}, and the blue region represents the substrate material domain Ωsb\Omega_{sb}.

Figure 10. Tradeoff relationship between the objective functions FkF_k and FωF_\omega under conditions (a)–(i) listed in Table 1.

functions are utilized, as in conditions (d)–(i), the Nϕ1N_{\phi1} value surpasses that in conditions (a)–(c), where a single level set function embodied the entirety of the design domain. Contrastingly, Nϕ2N_{\phi2}, excluding the interface between DsbD_{sb} and DpeD_{pe}, does not exhibit a significant difference from Nϕ1N_{\phi1} under conditions (a)–(c) but is less than Nϕ1N_{\phi1} in conditions (d)–(i). This phenomenon can be attributed to the fact that, when two level set functions are employed, the structural complexity is not restrained at the boundary between domains DsbD_{sb} and DpeD_{pe}, even upon altering the τz\tau_z value. Hence, the Nϕ1N_{\phi1} value is considered higher than that in conditions (a)–(c).

conditionτz\tau_zNϕ1N_{\phi1}Nϕ2N_{\phi2}
(a)1.0×10−61.0 \times10^{-6}0.088180.07629
(b)1.0×10−41.0 \times10^{-4}0.019650.01819
(c)1.0×10−21.0 \times10^{-2}0.000490.00042
(d)1.0×10−61.0 \times10^{-6}0.088900.07138
(e)1.0×10−41.0 \times10^{-4}0.050580.03523
(f)1.0×10−21.0 \times10^{-2}0.018980.00079
(g)1.0×10−61.0 \times10^{-6}0.141240.10782
(h)1.0×10−41.0 \times10^{-4}0.086600.06932
(i)1.0×10−21.0 \times10^{-2}0.003400.00200

Table 2. Number of nodes where the product of their level set function values with nodes adjacent to the negative direction of the z-axis is negative for each condition

Optimal configuration subjected to a minimum output voltage constraint

Herein, the efficacy of the introduced constraint with respect to output voltage is critically examined. Consistently across all scenarios, the substrate-dependent constraint is enforced utilizing the fictitious physical function ξ\xi. Moreover, the material domain is characterized by two level set functions, ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi}, with the parameter τz\tau_z assigned a value of 1.0×10−21.0 \times10^{-2}. Table 3 enumerates the parameters for the minimum output voltage constraint under various conditions, alongside their respective objective functions FkF_k, FωF_\omega, and output voltage VEV_E. The optimal configurations for each scenario are graphically represented in Figure 11.

Figure 11. Optimal configuration of the unimorph cantilevered energy harvester for conditions (j)–(q), as detailed in Table 3. The red region denotes the piezoelectric material domain Ωpe\Omega_{pe}, and the blue region represents the substrate material domain Ωsb\Omega_{sb}.

As shown in Table 3, the output voltage VEV_E obtained under condition (j) exceeds the stipulated minimum voltage constraint. However, under the other conditions, the output voltages are aligned exactly with their respective constraints, highlighting the effectiveness and precision of the minimum

Table 3: Summary of constraint parameters and objective functions: minimum output voltage constraint VˉminV\bar{V}_{minV} and the corresponding objective functions FkF_k and FωF_\omega, the output voltage VEV_E, and the volume of piezoelectric material domain Ωpe\Omega_{pe} for each condition.

conditionConstraint VˉminV\bar{V}_{minV}Objective function FkF_kObjective function FωF_\omegaOutput voltage VEV_EVolume of Ωpe\Omega_{pe} VpeV_{pe}
(j)9.0×10−39.0 \times10^{-3}8.44×10−58.44 \times10^{-5}4.56×10−64.56 \times10^{-6}9.73×10−39.73 \times10^{-3}4.47×1054.47 \times10^{5}
(k)9.5×10−39.5 \times10^{-3}1.82×10−41.82 \times10^{-4}4.87×10−64.87 \times10^{-6}9.50×10−39.50 \times10^{-3}2.31×1052.31 \times10^{5}
(l)1.0×10−21.0 \times10^{-2}2.03×10−42.03 \times10^{-4}5.60×10−65.60 \times10^{-6}1.00×10−21.00 \times10^{-2}2.07×1052.07 \times10^{5}
(m)1.05×10−21.05 \times10^{-2}2.21×10−42.21 \times10^{-4}6.23×10−66.23 \times10^{-6}1.05×10−21.05 \times10^{-2}1.90×1051.90 \times10^{5}
(n)1.1×10−21.1 \times10^{-2}2.34×10−42.34 \times10^{-4}6.72×10−66.72 \times10^{-6}1.10×10−21.10 \times10^{-2}1.80×1051.80 \times10^{5}
(o)1.15×10−21.15 \times10^{-2}2.45×10−42.45 \times10^{-4}7.11×10−67.11 \times10^{-6}1.15×10−21.15 \times10^{-2}1.71×1051.71 \times10^{5}
(p)1.2×10−21.2 \times10^{-2}2.39×10−42.39 \times10^{-4}1.02×10−51.02 \times10^{-5}1.20×10−21.20 \times10^{-2}1.66×1051.66 \times10^{5}
(q)1.3×10−21.3 \times10^{-2}2.61×10−42.61 \times10^{-4}1.11×10−51.11 \times10^{-5}1.30×10−21.30 \times10^{-2}1.49×1051.49 \times10^{5}

Table 3.

output voltage constraint. Furthermore, as the output voltage constraint values increase, there is a concomitant reduction in the volume of the piezoelectric material domain, represented by Ωpe\Omega_{pe}. As elucidated by Eq. (19), this reduction in the volume of the piezoelectric material domain results in a diminished capacitance, which, in turn, amplifies the production voltage. Concomitantly, an increase in the V‾minV\overline{V}_{minV} value invariably leads to a corresponding increase in both objective functions FkF_k and FωF_\omega. It is cogent to infer that imposing a minimum output voltage constraint invariably curtails the potential for optimizing the objective functions FkF_k and FωF_\omega.

As shown in Figure 11, under conditions (k)–(q) with an active minimum output voltage constraint, the piezoelectric material domain is predominantly located near the fixed boundary and at the midpoint of the beam. Close to the fixed boundary of the beam, the aggregation of the piezoelectric material domain can be attributed to the stress concentration observed in the first-order eigenmode. The region encompassing the piezoelectric material domain located around the center of the beam is aligned with the zone of maximal displacement in the second-order eigenmodes.

Conclusion

In this study, we propose a method for designing piezoelectric energy harvesters while improving their manufacturability. The proposed method derives both the substrate and piezoelectric designs that maximize the electromechanical coupling coefficient. This method allows the eigenfrequency of the device and minimum output voltage to be set to the desired values. Furthermore, the proposed method yields a design that can be manufactured using a microfabrication process. The results of this study are summarized as follows:

  1. A topology optimization method is formulated to achieve the desired values for the eigenfrequency and minimum output voltage while concurrently maximizing the electromechanical coupling coefficient. An optimization algorithm is constructed for the numerical analysis.

  1. Cross-sectional shape constraints and substrate-dependent constraints are formulated to improve manufacturability with microfabrication.

  1. Several numerical examples are provided to demonstrate the utility and validity of the proposed method. The proposed method can provide

solutions that satisfy the cross-sectional shape, substrate-dependent, and minimum output voltage constraints.

Limitations and future work

Despite the promising results, this study has some limitations. As the constraints become more stringent, the value of the objective function deteriorates. This indicates that the proposed method may struggle to provide solutions with significant electrical-mechanical coupling coefficients under highly restrictive conditions.

In future work, we aim to actually manufacture devices based on the designs obtained through the proposed method. This will allow us to validate the practicality of the solutions obtained and further demonstrate the utility of the proposed method.

Final remarks

The proposed method for designing piezoelectric energy harvesters has shown significant potential for improving their manufacturability. We believe that this research contributes to the field of energy harvesting and opens up new possibilities for the design of efficient and manufacturable energy harvesters.

Computation of the output voltage using the modal superposition method

In this study, the output voltage is computed using the modal superposition method. In Appendix A.1, we explain the method used to solve the eigenvalue problem using the finite element method (FEM). In Appendix A.2, we explain the method used to yield the output voltage using the modal superposition method based on the results obtained in Appendix A.1.

Eigenvalue and eigenvector determination

In this subsection, we elucidate an approach to solve the eigenvalue problem for piezoelectric materials using the FEM. Under open-circuit conditions, the relationships can be expressed in the form of finite element equations, as described below [4, 24].

(K−ωoc2M)u+Pφ=0,(67)(K - \omega_{oc}^{2} M)u + P\varphi= 0, \tag*{(67)}
PTu−Gφ=0,(68)P^{T}u - G\varphi= 0, \tag*{(68)}

wherein u\mathbf{u} and φ\varphi correspond to the displacement and electric potential under open-circuit conditions, respectively, and P\mathbf{P} and G\mathbf{G} denote the piezoelectric and dielectric matrices, respectively. In the absence of an external charge, denoted by q\mathbf{q}, and an external force, symbolized by f\mathbf{f}, the potential φ\varphi can be eliminated. This leads to:

PTu=Gφ,φ=G−1PTu,(69)\begin{aligned} \mathbf{P}^{T}\mathbf{u} &= \mathbf{G}\varphi,\\ \varphi&= \mathbf{G}^{-1}\mathbf{P}^{T}\mathbf{u}, \tag*{(69)} \end{aligned}
(K−ω2M+PG−1PT)u=0.(70)\left(\mathbf{K}-\omega^{2}\mathbf{M}+\mathbf{P}\mathbf{G}^{-1}\mathbf{P}^{T}\right)\mathbf{u}=0. \tag*{(70)}

Consequently, the eigenvalue equation pertinent to the open-circuit conditions can be expressed as follows:

M−1(K+PG−1PT)uoc=ωoc2uoc,(71)\mathbf{M}^{-1}\left(\mathbf{K}+\mathbf{P}\mathbf{G}^{-1}\mathbf{P}^{T}\right)\mathbf{u}_{oc}=\omega_{oc}^{2}\mathbf{u}_{oc}, \tag*{(71)}

where uoc\mathbf{u}_{oc} represents the eigenvector associated with the eigenvalue ωoc\omega_{oc}. Conversely, for a short-circuited electrode, the potential difference between the electrodes is zero. By postulating that both the external charge q\mathbf{q} and the external force f\mathbf{f} are zero, analogous to the open-circuit condition, and by grounding the electrode to establish a zero potential, Eq. (6) yields the following equation:

(K−ω2M)u=0.(72)\left(\mathbf{K}-\omega^{2}\mathbf{M}\right)\mathbf{u}=0. \tag*{(72)}

The eigenvalue equation for the short-circuit condition is given as follows:

M−1Kusc=ωsc2usc,(73)\mathbf{M}^{-1}\mathbf{K}\mathbf{u}_{sc}=\omega_{sc}^{2}\mathbf{u}_{sc}, \tag*{(73)}

where usc\mathbf{u}_{sc} represents the eigenvector corresponding to the eigenvalue ωsc\omega_{sc}.

Evaluation of displacement and output voltage using the modal superposition method

In this subsection, we describe the process of obtaining the displacement field using the modal superposition method, utilizing the eigenvectors obtained in the previous subsection [4]. To evaluate the output voltage, it is necessary to determine the displacement field. Since the eigenvectors are normalized, we need to consider the magnitude of the response for each mode to accurately determine the overall displacement. Therefore, we employ the modal superposition method to determine the displacement field necessary for this evaluation.

To obtain the displacement field under forced vibration conditions, we initially express the equation of motion as:

Mu¨+Cu˙+Ku=f.(74)M\ddot{u}+C\dot{u}+Ku=f. \tag*{(74)}

In the previous subsection, we discussed the general eigenvalue problem and denoted the solutions as uocu_{oc}, ωoc\omega_{oc}, uscu_{sc}, and ωsc\omega_{sc}. Using the modal transformation matrix Λ\Lambda, we perform a coordinate transformation such that u=Λqu=\Lambda q. The modal transformation matrix Λ\Lambda and the modal amplitude vector qq are given by:

Λ=[uoc1,uoc2,uoc3,⋯ ,uocn],(75)\Lambda=[u_{oc1},u_{oc2},u_{oc3},\cdots,u_{ocn}], \tag*{(75)}
q=[q1,q2,q3,⋯ ,qn].(76)q=[q_1,q_2,q_3,\cdots,q_n]. \tag*{(76)}

Multiplying Eq. (A.8) on the left with ΛT\Lambda^T, we obtain:

ΛTMΛq¨+ΛTCΛq˙+ΛTKΛq=ΛTf.(77)\Lambda^T M\Lambda\ddot{q}+\Lambda^T C\Lambda\dot{q}+\Lambda^T K\Lambda q=\Lambda^T f. \tag*{(77)}

Through normalization, we transform each coefficient term as follows:

ΛTMΛ=I,(78)\Lambda^T M\Lambda=I, \tag*{(78)}
ΛTCΛ=D=[2ζ1ω10⋯002ζ2ω2⋯0⋮⋮⋱⋮00⋯2ζnωn],(79)\Lambda^T C\Lambda=D= \begin{bmatrix} 2\zeta_1\omega_1 & 0 & \cdots& 0 \\ 0 & 2\zeta_2\omega_2 & \cdots& 0 \\ \vdots& \vdots& \ddots& \vdots\\ 0 & 0 & \cdots& 2\zeta_n\omega_n \end{bmatrix}, \tag*{(79)}
ΛTKΛ=[ω120⋯00ω22⋯0⋮⋮⋱⋮00⋯ωn2],(80)\Lambda^T K\Lambda= \begin{bmatrix} \omega_1^2 & 0 & \cdots& 0 \\ 0 & \omega_2^2 & \cdots& 0 \\ \vdots& \vdots& \ddots& \vdots\\ 0 & 0 & \cdots& \omega_n^2 \end{bmatrix}, \tag*{(80)}
ΛTf=[F1(t)0⋯00F2(t)⋯0⋮⋮⋱⋮00⋯Fn(t)].(81)\Lambda^T f= \begin{bmatrix} F_1(t) & 0 & \cdots& 0 \\ 0 & F_2(t) & \cdots& 0 \\ \vdots& \vdots& \ddots& \vdots\\ 0 & 0 & \cdots& F_n(t) \end{bmatrix}. \tag*{(81)}

where ζi\zeta_i is the modal damping ratio for the iith mode. In this transformation, we assume that the damping is small and the off-diagonal terms of the damping matrix are negligible compared with the diagonal terms. This transformation allows us to express the motion Eq. (A.8) as nn independent equations, one for each eigenmode as follows:

q¨i(t)+2ζiωiq˙i+ωi2qi(t)=Fi(t)(i=1,2,⋯ ,n),(82)\ddot{q}_i(t) + 2\zeta_i\omega_i\dot{q}_i + \omega_i^2q_i(t) = F_i(t) \qquad(i=1,2,\cdots,n), \tag*{(82)}
qi(t)=Fi(t)ωi(1−(ωˉi/ωi)2)2+4ζi2(ωˉi/ωi)2.(83)q_i(t) = \frac{F_i(t)}{\omega_i\sqrt{\left(1-(\bar{\omega}_i/\omega_i)^2\right)^2+4\zeta_i^2(\bar{\omega}_i/\omega_i)^2}}. \tag*{(83)}

From the above, the displacement is obtained as:

u(t)=∑i=1nqi(t)uoci.(84)u(t)=\sum_{i=1}^{n}q_i(t)u_{oci}. \tag*{(84)}

Incorporating this u\mathbf{u} into Eq. (A.3) yields the potential φn\varphi_n. Notably, ζi=0.01\zeta_i=0.01 for i=1,2,⋯ni=1,2,\cdots n. Finally, incorporating φn\varphi_n into Eq. (17) yields the output voltage.

Sensitivity analysis

In this section, we provide a detailed description of the sensitivity analysis using mathematical formulations.

At the stationary point of the Lagrangian, the following optimality conditions hold:

⟨∂Lpe∂u^oc,δuoc⟩∣opt=⟨∂(Fpe+aoc+boc)∂u^oc,δuoc⟩∣opt=0(85)\left\langle\frac{\partial L_{pe}}{\partial\hat{u}_{oc}},\delta u_{oc}\right\rangle\bigg|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{oc}+b_{oc})}{\partial\hat{u}_{oc}},\delta u_{oc}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(85)}
⟨∂Lpe∂vocpe,δvocpe⟩∣opt=⟨∂(Fpe+aoc)∂vocpe,δvocpe⟩∣opt=0(86)\left\langle\frac{\partial L_{pe}}{\partial v_{ocpe}},\delta v_{ocpe}\right\rangle\bigg|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{oc})}{\partial v_{ocpe}},\delta v_{ocpe}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(86)}
⟨∂Lpe∂φ^oc,δφoc⟩∣opt=⟨∂(Fpe+aoc+boc+λdo)∂φ^oc,δφoc⟩∣opt=0(87)\left\langle\frac{\partial L_{pe}}{\partial\hat{\varphi}_{oc}},\delta\varphi_{oc}\right\rangle\bigg|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{oc}+b_{oc}+\lambda d_{o})}{\partial\hat{\varphi}_{oc}},\delta\varphi_{oc}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(87)}
⟨∂Lpe∂vφocpe,δvφocpe⟩∣opt=⟨∂(Fpe+boc)∂vφocpe,δvφocpe⟩∣opt=0(88)\left\langle\frac{\partial L_{pe}}{\partial v_{\varphi ocpe}},\delta v_{\varphi ocpe}\right\rangle\bigg|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+b_{oc})}{\partial v_{\varphi ocpe}},\delta v_{\varphi ocpe}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(88)}
∂Lpe∂ωoci∣opt=⟨∂(Fpe+aoc(ωoci))∂ωoci,δωoci⟩∣opt=0(89)\left.\frac{\partial L_{pe}}{\partial\omega_{oci}}\right|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{oc}(\omega_{oci}))}{\partial\omega_{oci}},\delta\omega_{oci}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(89)}
⟨∂Lpe∂u^sc,δusc⟩∣opt=⟨∂(Fpe+asc)∂u^sc,δusc⟩∣opt=0(90)\left\langle\frac{\partial L_{pe}}{\partial\hat{u}_{sc}},\delta u_{sc}\right\rangle\bigg|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{sc})}{\partial\hat{u}_{sc}},\delta u_{sc}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(90)}
⟨∂Lpe∂vscpe,δvscpe⟩∣opt=⟨∂(Fpe+asc)∂vscpe,δvscpe⟩∣opt=0(91)\left\langle\frac{\partial L_{pe}}{\partial v_{scpe}},\delta v_{scpe}\right\rangle\bigg|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{sc})}{\partial v_{scpe}},\delta v_{scpe}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(91)}
∂Lpe∂ωsci∣opt=⟨∂(Fpe+asc(ωsci))∂ωsci,δωsci⟩∣opt=0(92)\left.\frac{\partial L_{pe}}{\partial\omega_{sci}}\right|_{\mathrm{opt}}=\left\langle\frac{\partial(F_{pe}+a_{sc}(\omega_{sci}))}{\partial\omega_{sci}},\delta\omega_{sci}\right\rangle\bigg|_{\mathrm{opt}}=0 \tag*{(92)}
⟨∂Lsb∂uoc,δuoc⟩∣opt=⟨∂(Fsb+aoc+boc)∂uoc,δuoc⟩∣opt=0(93)\left\langle\frac{\partial L_{sb}}{\partial u_{oc}},\delta u_{oc}\right\rangle\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{oc}+b_{oc})}{\partial u_{oc}},\delta u_{oc}\right\rangle\Big|_{opt}=0 \tag*{(93)}
⟨∂Lsb∂vocsb,δvocsb⟩∣opt=⟨∂(Fsb+aoc)∂vocsb,δvocsb⟩∣opt=0(94)\left\langle\frac{\partial L_{sb}}{\partial v_{ocsb}},\delta v_{ocsb}\right\rangle\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{oc})}{\partial v_{ocsb}},\delta v_{ocsb}\right\rangle\Big|_{opt}=0 \tag*{(94)}
⟨∂Lsb∂φoc,δφoc⟩∣opt=⟨∂(Fsb+aoc+boc+λdo)∂φoc,δφoc⟩∣opt=0(95)\left\langle\frac{\partial L_{sb}}{\partial\varphi_{oc}},\delta\varphi_{oc}\right\rangle\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{oc}+b_{oc}+\lambda d_{o})}{\partial\varphi_{oc}},\delta\varphi_{oc}\right\rangle\Big|_{opt}=0 \tag*{(95)}
\left\langle\frac{\partial L_{sb}}{\partial v_{\varphiocsb}},\delta v_{\varphiocsb}\right\rangle\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+b_{oc})}{\partial v_{\varphiocsb}},\delta v_{\varphiocsb}\right\rangle\Big|_{opt}=0 \tag*{(96)}
∂Lsb∂ω^oci∣opt=⟨∂(Fsb+aoc(ωoci))∂ω^oci,δωoci⟩∣opt=0(97)\frac{\partial L_{sb}}{\partial\hat{\omega}_{oci}}\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{oc}(\omega_{oci}))}{\partial\hat{\omega}_{oci}},\delta\omega_{oci}\right\rangle\Big|_{opt}=0 \tag*{(97)}
⟨∂Lsb∂usc,δusc⟩∣opt=⟨∂(Fsb+asc)∂usc,δusc⟩∣opt=0(98)\left\langle\frac{\partial L_{sb}}{\partial u_{sc}},\delta u_{sc}\right\rangle\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{sc})}{\partial u_{sc}},\delta u_{sc}\right\rangle\Big|_{opt}=0 \tag*{(98)}
⟨∂Lsb∂vscsb,δvscsb⟩∣opt=⟨∂(Fsb+asc)∂vscsb,δvscsb⟩∣opt=0(99)\left\langle\frac{\partial L_{sb}}{\partial v_{scsb}},\delta v_{scsb}\right\rangle\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{sc})}{\partial v_{scsb}},\delta v_{scsb}\right\rangle\Big|_{opt}=0 \tag*{(99)}
∂Lsb∂ω^sci∣opt=⟨∂(Fsb+asc(ωsci))∂ω^sci,δωsci⟩∣opt=0(100)\frac{\partial L_{sb}}{\partial\hat{\omega}_{sci}}\Big|_{opt}=\left\langle\frac{\partial(F_{sb}+a_{sc}(\omega_{sci}))}{\partial\hat{\omega}_{sci}},\delta\omega_{sci}\right\rangle\Big|_{opt}=0 \tag*{(100)}

where the expressions within the brackets represent the directional derivatives of the functional. The variables denoted with a hat, such as u^\hat{u}, represent the variables with respect to which the variational analysis is performed and should not be confused with the state variables, which are the solutions of the state equations. This distinction is made to clearly differentiate between the variables used in the sensitivity analysis and the state variables.

Eqs. (B.1), (B.3), and (B.5) can be used to derive vocpev_{ocpe}, vocsbv_{ocsb}, vscpev_{scpe}, vscsbv_{scsb}, v_{\varphiocpe}, and v_{\varphiocsb} as follows:

v_{ocpe}=c_{ocpe}u_{oc},\qquad v_{scpe}=c_{scpe}u_{sc},\qquad v_{\varphiocpe}=c_{ocpe}\varphi_{oc} \tag*{(101)}
cocpe=−αpeωsc2(ωoc2−ωsc2)2+(1−αpe)∑i=1n(ωoci−ωˉi)ωociωˉi2(102)c_{ocpe}=\frac{-\alpha_{pe}\omega_{sc}^{2}}{(\omega_{oc}^{2}-\omega_{sc}^{2})^{2}}+(1-\alpha_{pe})\sum_{i=1}^{n}\frac{(\omega_{oci}-\bar{\omega}_{i})}{\omega_{oci}\bar{\omega}_{i}^{2}} \tag*{(102)}
cscpe=αpeωoc2(ωoc2−ωsc2)2(103)c_{scpe}=\frac{\alpha_{pe}\omega_{oc}^{2}}{(\omega_{oc}^{2}-\omega_{sc}^{2})^{2}} \tag*{(103)}
vocsb=cocsbuoc,vscsb=cscsbusc,vϕocsb=cocsbφoc,(104)v_{ocsb}=c_{ocsb}u_{oc},\quad v_{scsb}=c_{scsb}u_{sc},\quad v_{\phi oc sb}=c_{ocsb}\varphi_{oc}, \tag*{(104)}
cocsb=−αsbωoc2(ωoc2−ωsc2)2+(1−αsb)∑i=1n(ωoci−ωˉi)ωociωˉi2,(105)c_{ocsb}=\frac{-\alpha_{sb}\omega_{oc}^{2}}{\left(\omega_{oc}^{2}-\omega_{sc}^{2}\right)^{2}}+(1-\alpha_{sb})\sum_{i=1}^{n}\frac{\left(\omega_{oci}-\bar{\omega}_{i}\right)}{\omega_{oci}\bar{\omega}_{i}^{2}}, \tag*{(105)}
cscsb=αsbωoc2(ωoc2−ωsc2)2.(106)c_{scsb}=\frac{\alpha_{sb}\omega_{oc}^{2}}{\left(\omega_{oc}^{2}-\omega_{sc}^{2}\right)^{2}}. \tag*{(106)}

From the above equations, the sensitivities for updating ϕpψ\phi_{p\psi} and ϕsψ\phi_{s\psi} can be obtained as follows:

Fpe′=∂Lpe∂ϕpψ=cocpe{∂aoc(ϕpψ,ϕsψ,ωoc,uoc,uoc,φoc)∂ϕpψ+∂boc(ϕpψ,uoc,φoc,φoc)∂ϕpψ}+ cscpe{∂asc(ϕpψ,ϕsψ,ωsc,usc,usc)∂ϕpψ}+λ,=cocpe{aoc(1,ϕsψ,ωoc,uoc,uoc,φoc)+boc(1,uoc,φoc,φoc)}+ cscpeasc(1,ϕsψ,ωsc,usc,usc)+λ,(107)\begin{aligned} F'_{pe}=\frac{\partial L_{pe}}{\partial\phi_{p\psi}}=c_{ocpe}\left\{\frac{\partial a_{oc}(\phi_{p\psi},\phi_{s\psi},\omega_{oc},\mathbf{u}_{oc},\mathbf{u}_{oc},\varphi_{oc})}{\partial\phi_{p\psi}}+\frac{\partial b_{oc}(\phi_{p\psi},\mathbf{u}_{oc},\varphi_{oc},\varphi_{oc})}{\partial\phi_{p\psi}}\right\} +\ c_{scpe}\left\{\frac{\partial a_{sc}(\phi_{p\psi},\phi_{s\psi},\omega_{sc},\mathbf{u}_{sc},\mathbf{u}_{sc})}{\partial\phi_{p\psi}}\right\}+\lambda, \\ &=c_{ocpe}\left\{a_{oc}(1,\phi_{s\psi},\omega_{oc},\mathbf{u}_{oc},\mathbf{u}_{oc},\varphi_{oc})+b_{oc}(1,\mathbf{u}_{oc},\varphi_{oc},\varphi_{oc})\right\} +\ c_{scpe}a_{sc}(1,\phi_{s\psi},\omega_{sc},\mathbf{u}_{sc},\mathbf{u}_{sc})+\lambda, \tag*{(107)} \end{aligned}
Fsb′=∂Lsb∂ϕsψ=cocsb{∂aoc(ϕpψ,ϕsψ,ωoc,uoc,uoc,φoc)∂ϕsψ+∂boc(ϕpψ,uoc,φoc,φoc)∂ϕsψ}+ cscsb{∂asc(ϕpψ,ϕsψ,ωsc,usc,usc)∂ϕsψ},=cocsb{aoc(ϕpψ,1,ωoc,uoc,uoc,φoc)}+ cscsb{asc(ϕpψ,1,ωsc,usc,usc)}.(108)\begin{aligned} F'_{sb}=\frac{\partial L_{sb}}{\partial\phi_{s\psi}}=c_{oc sb}\left\{\frac{\partial a_{oc}(\phi_{p\psi},\phi_{s\psi},\omega_{oc},\mathbf{u}_{oc},\mathbf{u}_{oc},\varphi_{oc})}{\partial\phi_{s\psi}}+\frac{\partial b_{oc}(\phi_{p\psi},\mathbf{u}_{oc},\varphi_{oc},\varphi_{oc})}{\partial\phi_{s\psi}}\right\} +\ c_{scsb}\left\{\frac{\partial a_{sc}(\phi_{p\psi},\phi_{s\psi},\omega_{sc},\mathbf{u}_{sc},\mathbf{u}_{sc})}{\partial\phi_{s\psi}}\right\}, \\ &=c_{oc sb}\left\{a_{oc}(\phi_{p\psi},1,\omega_{oc},\mathbf{u}_{oc},\mathbf{u}_{oc},\varphi_{oc})\right\} +\ c_{scsb}\left\{a_{sc}(\phi_{p\psi},1,\omega_{sc},\mathbf{u}_{sc},\mathbf{u}_{sc})\right\}. \tag*{(108)} \end{aligned}

References

  1. [1]Y. B. Jeon, R. Sood, J.-h. Jeong, S.-G. Kim, Mems power generator with transverse mode thin film pzt, Sensors and Actuators A: Physical 122 (2005) 16–22.DOI
  2. [2]S. Roundy, P. K. Wright, J. Rabaey, A study of low level vibrations as a power source for wireless sensor nodes, Computer communications 26 (2003) 1131–1144.DOI
  3. [3]S. M. Yun, C. Kim, The vibrating piezoelectric cantilevered generator under vortex shedding excitation and voltage tests, International Journal of Precision Engineering and Manufacturing 17 (2016) 1615–1622.DOI
  4. [4]A. Erturk, D. J. Inman, Piezoelectric energy harvesting, John Wiley & Sons, 2011.
  5. [5]K. A. Cook-Chennault, N. Thambi, A. M. Sastry, Powering mems portable devices – a review of non-regenerative and regenerative power supply systems with special emphasis on piezoelectric energy harvesting systems, Smart materials and structures 17 (2008) 043001.DOI
  6. [6]S. P. Beeby, M. J. Tudor, N. White, Energy harvesting vibration sources for microsystems applications, Measurement science and technology 17 (2006) R175.
  7. [7]S. Saxena, R. Sharma, B. Pant, Design and development of MEMS based guided beam type piezoelectric energy harvester, Springer, 2021.DOI
  8. [8]T. Yoshimura, S. Murakami, K. Wakazono, K. Kariya, N. Fujimura, Piezoelectric vibrational energy harvester using lead-free ferroelectric bifeo3 films, Applied Physics Express 6 (2013) 051501.
  9. [9]S. Murakami, T. Yoshimura, K. Satoh, K. Wakazono, K. Kariya, N. Fujimura, Development of piezoelectric mems vibration energy harvester using (100) oriented bifeo3 ferroelectric film, Journal of Physics: Conference Series 476 (2013) 012007.
  10. [10]P. Glynn e-Jones, S. P. Beeby, N. M. White, Towards a piezoelectric vibration-powered microgenerator, IEE Proceedings-Science, measurement and technology 148 (2001) 68–72.
  11. [11]M. P. Bendsøe, N. Kikuchi, Generating optimal topologies in structural design using a homogenization method, Computer Methods in Applied Mechanics and Engineering 71 (1988) 197–224.
  12. [12]E. C. N. Silva, J. S. Ono Fonseca, N. Kikuchi, Optimal design of periodic piezocomposites, Computer Methods in Applied Mechanics and Engineering 159 (1998) 49–77.
  13. [13]B. Zheng, C. J. Chang, H. C. Gea, Topology optimization of energy harvesting devices using piezoelectric materials, Structural and Multidisciplinary Optimization 38 (2009) 17–23.
  14. [14]C. J. Rupp, A. Evgrafov, K. Maute, M. L. Dunn, Design of piezoelectric energy harvesting systems: a topology optimization approach based on multilayer plates and shells, Journal of Intelligent Material Systems and Structures 20 (2009) 1923–1939.
  15. [15]M. Kögl, E. C. N. Silva, Topology optimization of smart structures: design of piezoelectric plate and shell actuators, Smart materials and Structures 14 (2005) 387.
  16. [16]Z. Kang, L. Tong, Integrated optimization of material layout and control voltage for piezoelectric laminated plates, Journal of Intelligent Material Systems and Structures 19 (2008) 889–904.
  17. [17]X. Zhang, Z. Kang, Dynamic topology optimization of piezoelectric structures with active control for reducing transient response, Computer Methods in Applied Mechanics and Engineering 281 (2014) 200–219.
  18. [18]S. Chen, S. Gonella, W. Chen, W. K. Liu, A level set approach for optimal design of smart energy harvesters, Computer Methods in Applied Mechanics and Engineering 199 (2010) 2532–2543.
  19. [19]Z. Luo, L. Tong, J. Luo, P. Wei, M. Y. Wang, Design of piezoelectric actuators using a multiphase level set method of piecewise constants, Journal of Computational Physics 228 (2009) 2643–2659.
  20. [20]M. He, M. He, X. Zhang, L. Xia, Topology optimization of piezoelectric energy harvesters for enhanced open-circuit voltage subjected to harmonic excitations, Materials 15 (2022) 4423.
  21. [21]B. V. de Almeida, D. C. Cunha, R. Pavanello, Topology optimization of bimorph piezoelectric energy harvesters considering variable electrode location, Smart Materials and Structures 28 (2019) 085030.
  22. [22]C. Kim, J. Lee, Topology optimum design of unimorph piezoelectric cantilevered mindlin plates as a vibrating electric harvester, Journal of Mechanical Science and Technology 28 (2014) 4131–4138.
  23. [23]G. H. Yoon, H. Choi, S. Hur, Multiphysics topology optimization for piezoelectric acoustic focuser, Computer Methods in Applied Mechanics and Engineering 332 (2018) 600–623.
  24. [24]R. A. Salas, F. J. Ramírez-Gil, W. Montealegre-Rubio, E. C. N. Silva, J. Reddy, Optimized dynamic design of laminated piezocomposite multi-entry actuators considering fiber orientation, Computer Methods in Applied Mechanics and Engineering 335 (2018) 223–254.
  25. [25]R. A. Salas, A. L. F. da Silva, E. C. N. Silva, HYIMFO: Hybrid method for optimizing fiber orientation angles in laminated piezocomposite actuators, Computer Methods in Applied Mechanics and Engineering 385 (2021) 114010.
  26. [26]S. L. Vatanabe, T. N. Lippi, C. R. de Lima, G. H. Paulino, E. C. Silva, Topology optimization with manufacturing constraints: A unified projection-based approach, Advances in Engineering Software 100 (2016) 97–112.
  27. [27]G. Michailidis, Manufacturing constraints and multi-phase shape and topology optimization via a level-set method, Ph.D. thesis, Ecole Polytechnique X, 2014.
  28. [28]Y. Sato, T. Yamada, K. Izui, S. Nishiwaki, Manufacturability evaluation for molded parts using fictitious physical models, and its application in topology optimization, The International Journal of Advanced Manufacturing Technology 92 (2017) 1391–1409.
  29. [29]J. Liu, A. T. Gaynor, S. Chen, Z. Kang, K. Suresh, A. Takezawa, L. Li, J. Kato, J. Tang, C. C. Wang, et al., Current and future trends in topology optimization for additive manufacturing, Structural and multidisciplinary optimization 57 (2018) 2457–2483.
  30. [30]T. Yamada, Y. Noguchi, Topology optimization with a closed cavity exclusion constraint for additive manufacturing based on the fictitious physical model approach, Additive Manufacturing 52 (2022) 102630.
  31. [31]M. Tajima, T. Yamada, Topology optimization with geometric constraints for additive manufacturing based on coupled fictitious physical model, Computer Methods in Applied Mechanics and Engineering 417 (2023) 116415.
  32. [32]T. Yamada, K. Izui, S. Nishiwaki, A. Takezawa, A topology optimization method based on the level set method incorporating a fictitious interface energy, Computer Methods in Applied Mechanics and Engineering 199 (2010) 2876–2891.
  33. [33]M. Yulin, W. Xiaoming, A level set method for structural topology optimization with multi-constraints and multi-materials, Acta Mechanica Sinica 20 (2004) 507–518.
  34. [34]M. Y. Wang, X. Wang, “color” level sets: a multi-phase method for structural topology optimization with multiple materials, Computer Methods in Applied Mechanics and Engineering 193 (2004) 469–496.
  35. [35]M. Noda, Y. Noguchi, T. Yamada, Extended level set method: A multiphase representation with perfect symmetric property, and its application to multi-material topology optimization, Computer Methods in Applied Mechanics and Engineering 393 (2022) 114742.
  36. [36]M. Aramaki, T. Yoshimura, S. Murakami, K. Satoh, N. Fujimura, Demonstration of high-performance piezoelectric mems vibration energy harvester using bifeo3 film with improved electromechanical coupling factor, Sensors and Actuators A: Physical 291 (2019) 167–173.
  37. [37]N. W. Hagood, A. von Flotow, Damping of structural vibrations with piezoelectric materials and passive electrical networks, Journal of sound and vibration 146 (1991) 243–268.
  38. [38]T. Yamada, T. Matsumoto, S. Nishiwaki, Generating structural configurations having a number of specified eigenfrequencies using a level set-based topology optimization method, Civil-Comp Proceedings 99 (2012) 1–11.
  39. [39]G. Allaire, Shape optimization by the homogenization method, volume 146, Springer Science & Business Media, 2001.
  40. [40]M. P. Bendsøe, O. Sigmund, Topology optimization: theory, methods, and applications, Springer Science & Business Media, 2003.
  41. [41]H. Azegami, et al., Shape optimization problems, Springer, 2020.
  42. [42]M. Zhou, Y. Liu, Z. Lin, Topology optimization of thermal conductive support structures for laser additive manufacturing, Computer Methods in Applied Mechanics and Engineering 353 (2019) 24–43.
  43. [43]T. Yamada, K. Izui, S. Nishiwaki, A level set-based topology optimization method for maximizing thermal diffusivity in problems including design-dependent effects, Journal of Mechanical Design 133 (2011). 031011.
  44. [44]G. Jing, H. Isakari, T. Matsumoto, T. Yamada, T. Takahashi, Level set-based topology optimization for 2D heat conduction problems using BEM with objective function defined on design-dependent boundary with heat transfer boundary condition, Engineering Analysis with Boundary Elements 61 (2015) 61–70.
  45. [45]C. Wu, J. Fang, Q. Li, Multi-material topology optimization for thermal buckling criteria, Computer Methods in Applied Mechanics and Engineering 346 (2019) 1136–1155.
  46. [46]T. Miki, T. Yamada, Topology optimization considering the distortion in additive manufacturing, Finite Elements in Analysis and Design 193 (2021) 103558.
  47. [47]F. Feppon, G. Allaire, C. Dapogny, P. Jolivet, Topology optimization of thermal fluid-structure systems using body-fitted meshes and parallel computing, Journal of Computational Physics 417 (2020) 109574.
  48. [48]K. Guan, K. Matsushima, Y. Noguchi, T. Yamada, Topology optimization for rarefied gas flow problems using density method and adjoint IP-DSMC, Journal of Computational Physics 474 (2023) 111788.
  49. [49]J. S. Choi, J. Yoo, Simultaneous structural topology optimization of electromagnetic sources and ferromagnetic materials, Computer Methods in Applied Mechanics and Engineering 198 (2009) 2111–2121.
  50. [50]T. Yamada, H. Watanabe, G. Fujii, T. Matsumoto, Topology optimization for a dielectric optical cloak based on an exact level set approach, IEEE Transactions on Magnetics 49 (2013) 2073–2076.
  51. [51]E. Wadbro, M. Berggren, Topology optimization of an acoustic horn, Computer Methods in Applied Mechanics and Engineering 196 (2006) 420–436.
  52. [52]O. Sigmund, J. S. Jensen, Systematic design of phononic band–gap materials and structures by topology optimization, Philosophical Transactions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences 361 (2003) 1001–1019.
  53. [53]Y. Noguchi, T. Yamada, Topology optimization for acoustic structures considering viscous and thermal boundary layers using a sequential linearized navier-stokes model, Computer Methods in Applied Mechanics and Engineering 394 (2022) 114863.
  54. [54]Y. Noguchi, T. Yamada, Topology optimization of acoustic metasurfaces by using a two-scale homogenization method, Applied Mathematical Modelling 98 (2021) 465–497.
  55. [55]M. Y. Wang, X. Wang, D. Guo, A level set method for structural topology optimization, Computer methods in applied mechanics and engineering 192 (2003) 227–246.
  56. [56]G. Allaire, F. Jouve, A. M. Toader, Structural optimization using sensitivity analysis and a level-set method, Journal of computational physics 194 (2004) 363–393.
  57. [57]F. Laermer, Method of anisotropically etching silicon, United States Patent 5,501,893 (1996).
  58. [58]F. Hecht, New development in freefem++, J. Numer. Math. 20 (2012) 251–265. URL: https://freefem.org/.

Paper details

vlnc:W_n9mat74x
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents