Thumbnail Image

Surrogate-assisted multi-objective Bayesian optimization of bio-printing process parameters

Rane, Aditya
3D bioprinting is an additive manufacturing method that allows bio-mimeticfabrication of living cells through layer deposition of bio-ink. Tissue constructs resulting from such additive manufacturing process have the potential to emulate functional characteristics of native human tissue microenvironment. This emerging technology has various biological research applications mainly in regenerative medicine, in-vitro disease modelling and drug screening. Due to this fact, it could be possible to transplant damaged human organs and study drug responses to diseases. Extrusion-based bio-printing (EBB) is widely used for the process of biofabrication, provides a practical solution for the fabrication of tissue constructs. Its principle of operation is based on exerting pressure on bioink through the orifice of nozzle to form filaments or droplets. As a consequence of this operation, cells are damaged either due to mechanical forces or shear stress. Furthermore, the interplay of bioink composition with nozzle design parameters has a detrimental effect on living cells. Unlike non-biological printing, 3D bioprinting has additional complexities of cell viability and post-printing functionality.
In bioprinting, design optimization of printer parameters, to minimize the exerted shear stress on wall of nozzle and maximizing the survival as well as functional characteristic of cells is crucial. An experimental design approach is time consuming and cost intensive. This thesis introduces data-driven Gaussian Process Regression Multi-Objective Bayesian Optimization (GPR-MOBO) framework. The approach optimizes five design parameters including nozzle and bio-ink rheology: outlet diameter (d), maximum and minimum viscosity (η), power law (n) and flow consistency index (K). Initial experiments are conducted with two-level factorial design by varying ranges of design inputs. Using CFD simulations of 18G and 25G nozzle in Ansys FLUENT three fluid shear responses are extracted at the X-Y-Z location of the internal fluid shear profile. The proposed GPR-MOBO framework combines the principle of black-box modeling (a function without closed-form relationship) using the Gaussian process and integrates sampling and optimization using Bayesian optimization. On account of integration, the experimental approach becomes sequentially adaptive and determines the Pareto optimal sampling point in 21 iterations following preliminary 24 runs. The proposed framework implements adaptive sampling method for creating an iterative framework and optimally selecting process inputs to minimize shear stress and maximize cell functionality.