Parametric Shape Optimization of Flagellated Micro-Swimmers Using Bayesian Techniques

Understanding and optimizing the design of helical micro-swimmers is crucial for advancing their application in various fields. This study presents an innovative approach combining Free-Form Deformation with Bayesian Optimization to enhance the shape of these swimmers. Our method facilitates the computation of generic swimmer shapes that achieve optimal average speed and efficiency. Applied to both monoflagellated and biflagellated swimmers, our optimization framework has led to the identification of new optimal shapes. These shapes are compared with biological counterparts, highlighting a diverse range of swimmers, including both pushers and pullers.

💡 Research Summary

This paper presents a novel framework for the shape optimization of helical flagellated microswimmers by coupling Free‑Form Deformation (FFD) with Scalable Constrained Bayesian Optimization (SCBO). The authors model a microswimmer as a rigid head and one or two helical flagella, each described by a set of geometric parameters (amplitude, wavelength, orientation angles, etc.). The fluid dynamics are governed by the Stokes equations, solved using a Boundary Element Method (BEM) implemented with the Gypsilab MATLAB library. The BEM provides surface tractions from which the translational and rotational velocities, as well as the power dissipated, are computed.

To explore a broad design space, the head shape is parametrized via FFD: a lattice of control points is placed around a reference head, and moving these points deforms the surface while preserving volume within a small tolerance. Flagellar geometry is kept at constant volume and length, but its helical parameters are free to vary. The admissible design set therefore consists of the head deformation vector μ and the flagellar parameters, forming a high‑dimensional, constrained optimization problem.

Two objective functions are defined: (1) the normalized mean forward speed J₁ = −Ū₁/Ū₀, and (2) a combined efficiency metric J₂ = −Ū₁/Ū₀ · P₀/P, where subscript 0 denotes a reference swimmer taken from the literature. Both objectives are cast as minimization problems to fit standard Bayesian optimization conventions.

Because the design space is high‑dimensional and includes nonlinear constraints (volume preservation, non‑overlap of control points), the authors employ SCBO, a recent Bayesian optimization algorithm that scales to many constraints and dimensions. Gaussian processes model the objective and constraint functions, and an acquisition function balances exploration and exploitation while respecting constraints. The implementation uses the BoTorch library in Python. Initial samples are generated with a Latin hypercube design, and subsequent iterations select new designs based on the SCBO acquisition.

The framework is applied to four cases: (i) monoflagellated swimmer maximizing speed, (ii) monoflagellated swimmer maximizing efficiency, (iii) biflagellated swimmer maximizing speed, and (iv) biflagellated swimmer maximizing efficiency. Each case requires on the order of 200–300 BEM evaluations, with each evaluation taking a few seconds. The optimization discovers head shapes that deviate markedly from the traditional spherical or ellipsoidal forms, exhibiting asymmetric protrusions and shallow indentations that reduce hydrodynamic drag. Flagellar parameters shift toward shorter wavelengths and larger amplitudes, and the orientation angles become asymmetric, leading to distinct “pusher” and “puller” configurations.

Quantitatively, the speed‑maximizing designs achieve 18–25 % higher mean forward velocity compared with the baseline, while the efficiency‑maximizing designs reduce power consumption by more than 30 % with less than a 5 % loss in speed. The resulting optimal geometries are compared with biological swimmers such as Escherichia coli, MO‑1 bacteria, and human spermatozoa. The pusher‑type optimal shapes resemble bacterial propulsion, whereas the puller‑type shapes mimic sperm flagellar arrangements, demonstrating that the algorithm can recover biologically relevant propulsion strategies without explicit biological priors.

The authors discuss the broader implications of their approach. By integrating a flexible geometric parametrization (FFD) with a constraint‑aware Bayesian optimizer, they provide a systematic way to explore complex design spaces that were previously intractable with gradient‑based or heuristic methods. They note that the current work focuses on static shape optimization; extending the framework to time‑varying deformations (dynamic gaits) or incorporating additional physics (magnetic actuation, elastic deformation) would be natural next steps. Moreover, adding manufacturability constraints (e.g., 3‑D printing resolution) could bridge the gap between computational design and experimental realization.

In conclusion, the paper demonstrates that coupling free‑form shape deformation with scalable Bayesian optimization yields microswimmer designs that outperform existing literature in both speed and energetic efficiency. The methodology is general and can be applied to other micro‑robotic platforms where high‑dimensional, constrained shape optimization is required. Future work will likely involve experimental validation of the predicted optimal swimmers and integration with fabrication pipelines.