High-fidelity surrogate models for parametric shape design in microfluidics

• Autores: Luca Borchini
• Directores de la Tesis: Antonio Huerta Cerezuela (dir. tes.) , Matteo Giacomini (codir. tes.) , Rubén Sevilla Cárdenas (codir. tes.)
• Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2020
• Idioma: español
• Programa de doctorado: Programa de Doctorado Erasmus Mundus en Simulación en Ingeniería y Desarrollo de la Iniciativa Emprendedora / Simulation in Engineering and Entrepreneurship Development por la Universidad Politécnica de Catalunya; Instituto Superior Técnico de Lisboa(Portugal); Swansea University (Reino Unido); Technische Universiteit Eindhoven-Eindhoven University of Technology(Países Bajos); Technische Universität München(Alemania); Università degli Studi di Pavía(Italia); Université Libre de Bruxelles(Bélgic
• Calificación: Sobresaliente Cum Laude
• Texto completo no disponible (Saber más ...)
• Resumen

  • Nowadays, the main computational bottleneck in computer-assisted industrial design procedures is the necessity of testing multiple parameter settings for the same problem. Material properties, boundary conditions or geometry may have a relevant influence on the solution of those problems. Consequently, the effects of changes in these quantities on the numerical solution need to be accurately estimated. That leads to significantly time-consuming multi-query procedures during decision-making processes. Microfluidics is one of the many fields affected by this issue, especially in the context of the design of robotic devices inspired by natural microswimmers. Reduced-order modelling procedures are commonly employed to reduce the computational burden of such parametric studies with multiple parameters. Moreover, highfidelity simulation techniques play a crucial role in the accurate approximation of the flow features appearing in complex geometries. This thesis proposes a coupled methodology based on the high-order hybridisable discontinuous Galerkin (HDG) method and the proper generalized decomposition (PGD) technique. Geometrically parametrised Stokes equations are solved exploiting the innovative HDG-PGD framework. On the one hand, the parameters describing the geometry of the domain act as extra-coordinates and PGD is employed to construct a separated approximation of the solution. On the other hand, HDG mixed formulation allows separating exactly the terms introduced by the parametric mapping into products of functions depending either on the spatial or on the parametric unknowns. Convergence results validate the methodology and more realistic test cases, inspired by microswimmer devices involving variable geometries, show the potential of the proposed HDG-PGD framework in parametric shape design. The PGD-based surrogate models are also utilised to construct separated response surfaces for the drag force. A comparison between response surfaces obtained through the apriori and the a posteriori PGD is exposed. A critical analysis of the two techniques is presented reporting advantages and drawbacks of both in terms of computational costs and accuracy.