Research And Optimization On Piezoelectric Drop-on-demand Three-dimensional Bioprinting

Bioprinting,based on rapid prototyping,is a widely used technology in biofabrication.It is used to build the 3D objects layer-by-layer according to the virtual design with reading data from computer-aided design drawings.Drop-on-demand(DOD)bioprinting is one of the most promising technologies currently due to the unique characteristics of high-throughput efficiency and cost-effectiveness.Compared with conventional cell seeding techniques,DOD bioprinting can establish 3D tissues constructs to provide a more physiological environment for the cells,and can print multiple cell types and biomaterials directly in specific spatial arrangements.To achieve the ultimate goal of organ fabrication,cells in the bio-ink have to be printed on the specific position stably with enough viability and proliferation to realize the complex biological structures and functions.However,they are still challenges for the development of DOD bioprinting technology.In this thesis,based on the piezoelectric DOD printing,the computational fluid dynamics(CFD)simulation,machine learning method and multi-objective optimization(MOO)method are selected to solve these challenges.Main contributions of this thesis are summarized as follows:(1)Shear stress analysis and its effects on cell viability and cell proliferation in DOD bioprinting based on CFD simulation:a simulation model of piezoelectric DOD print-head was developed and experiments were conducted to study the shear stress generated in the nozzle during printing as well as its action on cell viability and cell proliferation.The proposed simulation model,with the consideration of the piezoelectric effect,was used to analyze the distribution and variation of shear stress in the voltage pulse.The simulation model was validated by experiments.Parametric studies on shear stress were also carried out including the applied voltage,bio-ink viscosity and diameter of nozzle through simulation.Simulation results demonstrated that the shear stress is increased with the increase of voltage and bio-ink viscosity,as well as the decrease of nozzle diameter.With simulation results of shear stress,experiments showed that both cell viability and cell proliferation are decreased with the increase of shear stress,whereby shear stress has a larger influence on cell proliferation than on cell viability.Through proposed simulation model,computed shear stress during DOD bioprinting is able to link with engineering characteristics,such as printing parameters,and cell characteristics,such as cell viability and cell proliferation.The simulation model described here can be used to improve cell viability and cell proliferation through optimizing printing parameters to decrease shear stress in piezoelectric DOD bioprinting.(2)Learning-based cell injection control for precise DOD bioprinting:a novel machine learning technology based on Learning-based Cell Injection Control(LCIC)approach is demonstrated to replace the trial-and-error process for effective DOD printing control while eliminating satellite droplets automatically.The LCIC approach includes a specific CFD simulation model of piezoelectric DOD print-head considering inverse piezoelectric effect,which is used instead of repetitive experiments to collect data,and a fully connected neural network(FCNN)trained by simulation data based on artificial neural network algorithm,using the well-known classification performance of FCNN to optimize DOD printing parameters automatically.The test accuracy of the LCIC method was 90%.With the validation of LCIC method by experiments,satellite droplets from piezoelectric DOD printing are reduced significantly,improving the printing efficiency drastically to satisfy requirements of manufacturing precision for printing complex artificial tissues.The LCIC method can be further used to optimize the structure of DOD print-head and cell behaviors.(3)MOO design through machine learning for DOD bioprinting:an MOO design method of DOD printing parameters through the fully connected neural networks(FCNNs)is proposed to solve these challenges.The MOO problem comprises two objective functions.One is built with FCNNs to develop the satellites formation model.Another one is built for decrease of droplet diameter and increase of droplet speed.Hybrid multi subgradient descent bundle method with adaptive learning rate algorithm(HMSGDBA),combining multi subgradient descent bundle method(MSGDB)with Adam,is introduced to search the Pareto optimal set for the MOO problem.The superiority of HMSGDBA is proven through comparative studies with MSGDB,and experimental results showed that single droplet can be printed stably and droplet speed is increased from 0.88 m/s to 2.08 m/s after optimization with the proposed method.The proposed method can improve printing precision and stability,and is useful to realize precise cells arrays and complex biological functions.The proposed method can be used to obtain guidelines for setup of cell printing experimental platform.