ADJOINT-BASED OPTIMIZATION OF THE CONTROL BOUNDARIES OF A MICROFLUIDIC ACOUSTIC FLOW

Javier Lorente-Macias · 2025

Read ADJOINT-BASED OPTIMIZATION OF THE CONTROL BOUNDARIES OF A MICROFLUIDIC ACOUSTIC FLOW by Javier Lorente-Macias (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-08193.txt.

Introduction

Two engineers compute actuator settings that silence echoes inside inkjet printheads in a 2025 paper. Drop on demand heads fire droplets through a microchannel, and leftover acoustic waves spoil the next drop. The aim is an actuator motion that cancels the ringing at a chosen moment.

The cost is the total energy of the flow, counting channel sound plus droplet surface energy. An adjoint method yields the gradient of that cost against the actuator boundary, and a gradient search finds the best motion. Two head designs, thin film and bulk, both cut total energy by over a hundredfold. The winning motion first draws fluid back, then fires counter waves that erase the reverberation.

Catalog record

CREATORJavier Lorente-Macias
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-08193

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

ADJOINT-BASED OPTIMIZATION OF THE CONTROL BOUNDARIES OF A MICROFLUIDIC ACOUSTIC FLOW Javier Lorente-Macias Abstract \vskip-12.0pt In this study, we find the optimal control boundary (i.e., actuator velocity) that cancels the acoustic reverberations inside drop-on-demand inkjet printheads at a specific time. We formulate an optimization problem to minimize the total energy of the oscillating flow at a given time (i.e., the acoustic energy inside the microchannel and the surface energy of the droplet). We use the adjoint method to compute the gradient of the cost function with respect to the control boundary, and a gradient-based optimization algorithm to converge to the optimal solution. This methodology has been successfully applied to two generic inkjet printhead mechanisms: thin-film and bulk types. In both cases, the actuator first reduces the surface energy of the system by extracting fluid from the nozzle. In this process, acoustic waves also propagate through the channel and reverberate at the ends, which increases the acoustic energy of the system. The actuator then sends additional acoustic waves that cancel these reverberations. Both mechanisms have been able to reduce the total energy of the system by a factor of over 100 in comparison with the uncontrolled cases. Keywords: microfluidics, piezoacoustics, actuation design, inkjet printing 1. Introduction Drop-on-demand (DOD) inkjet printing is a technique in which ink droplets are generated on demand and jetted towards a moving substrate. An inkjet printhead typically contains several hundred microchannels and nozzles. Ink flows continuously through the channels at constant pressure difference (between one side and the other) and at constant pressure (relative to the atmosphere) just above the nozzle. An increase of the pressure must be generated inside the channel to expel droplets. In piezoelectric DOD printing, an actuator placed on one or more sides of the channel moves a few tens of nanometers, which generates acoustic pressure waves. These waves eventually reach the nozzle and a droplet is expelled. The piezoelectric actuators are made of ceramic materials that exhibit the inverse piezoelectric effect (i.e., they deform when a voltage is applied). Manufacturers seek to increase the jetting frequency while maintaining uniformity in droplet size and shape. The former objective speeds up the printing process, whereas the latter makes it more accurate, which is important in high precision applications (e.g. printing microcircuits [ 1 ] ; manufacturing of passive electronic components such as capacitors [ 2 ] , biomedicine [ 3 ] , or advanced manufacturing [ 4 ] ). As the time between droplet ejections decreases, the droplet shape and size becomes increasingly affected by the residual oscillations from the previous droplet ejection. Therefore, it is important to damp these reverberations within the desired time interval. In this study, we design the open-loop control for the actuator that damps the acoustic reverberations inside the microchannels within a given time interval. This active control strategy is easier to implement on already existing printheads than passive control strategies (e.g., geometric variations). We consider two simplified geometries of two inkjet printing strategies already employed in the industry. 2. Problem formulation 2.1 Governing equations We consider the flow inside the inkjet printhead microchannel. We derive the governing equations from a low Mach number asymptotic expansion [ 5 , 6 ] . We further consider that variations in temperature are negligible, and that the ratio of specific heats is \gamma_{th}=1 . Under these assumptions, we can express the nondimensionalized momentum and energy equations of the oscillating flow as: \displaystyle\frac{\partial\tilde{\boldsymbol{u}}}{\partial t}=-\nabla\tilde{p}+\frac{1}{\tilde{Re}}\nabla\cdot\tilde{\boldsymbol{\tau}} (1a) \displaystyle\frac{\partial\tilde{p}}{\partial t}+\nabla\cdot\tilde{\boldsymbol{u}}=0 (1b) where \tilde{\boldsymbol{u}} is the acoustic velocity, \tilde{p} is the acoustic pressure, \tilde{Re} is the Reynolds number of the acoustic flow, and \tilde{\boldsymbol{\tau}}=\nabla\tilde{\boldsymbol{u}}+\left(\nabla\tilde{\boldsymbol{u}}\right)^{T}-\frac{2}{3}\nabla\cdot\tilde{\boldsymbol{u}}\mathbb{I} is the viscous stress tensor. These equations have been nondimensionalized using the length of the channel, L_{c} , as the characteristic length, the speed of the sound, c_{0} , as the characteristic speed, and p_{c}=\rho c_{0}^{2} as the characteristic pressure. Therefore, the Reynolds number associated with the oscillating flow is \tilde{Re}=\frac{\rho c_{0}L_{c}}{\mu} , with \rho and \mu being, respectively, the density and the dynamic viscosity of the fluid. This allows us to define the time scale t_{c}=\frac{L_{c}}{c_{0}} , which is the time required for a wave to propagate through the entire length of the channel. We now consider the reduced-order model for the flow inside the droplet described in [ 6 ] . The meniscus is a spherical cap containing incompressible fluid. Thus the stress at the nozzle outlet is equal to the surface tension, \tilde{\boldsymbol{\sigma}}\cdot\boldsymbol{n}|_{\Gamma_{N}}=-\gamma_{s}\kappa\boldsymbol{n} , where \tilde{\boldsymbol{\sigma}}=-p\mathbb{I}+\frac{1}{\tilde{Re}}\tilde{\boldsymbol{\tau}} is the total stress tensor, \gamma_{s} is the nondimensional surface tension coefficient and \kappa is the nondimensional curvature. This approach decouples the flow inside the droplet from the flow inside the channel, so that we just need to simulate the flow inside the microchannel and consider the previous impedance boundary condition at the nozzle outlet boundary. The volume of the droplet, \Omega , varies according to the amount of mass flux through the nozzle outlet boundary: \frac{d\Omega}{dt}=-\epsilon\int_{\Gamma_{N}}\tilde{\boldsymbol{u}}\cdot\boldsymbol{n}d\Gamma (2) For a spherical cap, the volume of fluid inside the meniscus depends on the curvature through the following expression: \Omega\left(\kappa\right)=\frac{\pi}{3}R_{n}^{3}\frac{1}{\kappa^{3}}\left(2+\sqrt{1-\kappa^{2}}\right)\left(1-\sqrt{1-\kappa^{2}}\right)^{2} (3) This leads to an ordinary differential equation for the evolution of the curvature: \frac{d\Omega\left(\kappa\right)}{d\kappa}\frac{d\kappa}{dt}=-\epsilon\int_{\Gamma_{N}}\tilde{\boldsymbol{u}}\cdot\boldsymbol{n}d\Gamma (4) For the other boundaries, we consider homogeneous Dirichlet boundary conditions at the walls, \tilde{\boldsymbol{u}}_{\Gamma_{W}}=0 , nonhomogeneous Dirichlet boundary conditions at the actuators, \tilde{\boldsymbol{u}}_{\Gamma_{Act}}=\mathcal{U}_{Act}\left(t\right) , and stress free boundary conditions at the channel inlet/outlets \tilde{\boldsymbol{\sigma}}\cdot\boldsymbol{n}|_{\Gamma_{In/Out}}=0 . In this paper, the initial acoustic state of the flow is taken to be zero acoustic velocity, \tilde{\boldsymbol{u}}\left(t=0\right)=0 , and zero acoustic pressure, \tilde{p}\left(t=0\right)=0 . The initial meniscus curvature has a nonzero initial value, \kappa\left(t=0\right)=\kappa_{0} . We define the acoustic energy, \mathscr{E}_{ac}\left(t\right) , and the surface energy, \mathscr{E}_{s}\left(t\right) , as follows: \displaystyle\mathscr{E}_{ac}\left(t\right) \displaystyle=\frac{1}{2}\left(\tilde{u}^{2}+\tilde{p}^{2}\right) (5a) \displaystyle\mathscr{E}_{s}\left(t\right) \displaystyle=\frac{1}{2}\frac{\gamma_{s}}{\epsilon}R_{n}\left|\Gamma_{n}\right| (5b) where \gamma_{s} is the surface tension, R_{n} , is the nondimensional nozzle radius, and \Gamma_{n}=2\pi R_{n}^{2}\frac{1}{\kappa^{2}}\left(1-\sqrt{1-\kappa^{2}}\right) is the surface area of the spherical cap. In order to ensure droplet uniformity, the acoustic energy inside the channel and the surface energy of the spherical cap at the nozzle outlet must be the same before the next droplet is ejected. For simplicity, we set these energies to zero in this paper, but a non-zero curvature could be targeted if desired. Therefore, we define the cost function as the total energy (i.e., sum of acoustic energy plus surface energy) at a given time T : \mathscr{J}=\mathscr{E}_{ac}\left(T\right)+\mathscr{E}_{s}\left(T\right) (6) In order to minimize ( 6 ), we use the actuator velocity as the control. Therefore, we first propose a parametrization of the function \mathcal{U}_{Act} . The simplest way to parametrize this funtion in time is to use a parameter at every time step. However, this leads to a very large set of control parameters, and increases the dimension of the search space accordingly. This can cause there to be many local minima, and the optimization algorithm may not be able to find the global minimum. In order to regularize the problem, we reduce the dimensions of the search space by expressing U_{Act} through a truncated Fourier series expansion: \mathcal{U}_{Act}\left(t\right)=\sum_{n=0}^{N}a_{n}\cos{\left(\frac{n2\pi t}{T}\right)}+\sum_{i=0}^{N}b_{n}\sin{\left(\frac{n2\pi t}{T}\right)} (7) We then define our control vector \boldsymbol{c}\in\mathbb{R}^{(N+1)\times 1} as follows: \boldsymbol{c}=\left[\begin{array}[]{cccccccccc}a_{0}&a_{1}&\cdots&a_{N}&|&b_{0}&b_{1}&\cdots&b_{N}\end{array}\right]^{T} (8) We set the initial and final actuator velocity to zero, \mathcal{U}_{Act}\left(t=0\right)=\mathcal{U}_{Act}\left(t=T\right)=0 , which leads to the following constraint: \sum_{i=0}^{N}a_{n}=0 (9) Finally, we formulate the optimization problem as finding the

Read the complete text