Deformable microcapsules are widely used in industries and also serve as a mechanical model of living biological cells.In this study,we develop a novel method,by integrating a deep convolutional neural network(DCNN)wi...Deformable microcapsules are widely used in industries and also serve as a mechanical model of living biological cells.In this study,we develop a novel method,by integrating a deep convolutional neural network(DCNN)with high-fidelity mechanistic capsule modelling,to identify the membrane constitutive law and estimate associated parameters of a microcapsule from its steady deformed profile in a capillary tube.Compared with conventional inverse methods,the present approach is more accurate and can increase the prediction throughput rate by a few orders of magnitude.It can process capsules with large deformation in inertial flows.Furthermore,the method can predict the capsule membrane shear elasticity,area dilatation modulus and initial inflation from a single steady capsule profile.We explore the mechanism that the DCNN makes decisions by considering its feature maps,and discuss their potential implication on the development of inverse methods.The present method provides a promising tool which may enable high-throughput mechanical characterisation of microcapsules and biological cells in microfluidic flows.展开更多
基金supported by the UK Engineering and Physical Science Research Council(EP/K000128/1)and the China Scholarship Council.
文摘Deformable microcapsules are widely used in industries and also serve as a mechanical model of living biological cells.In this study,we develop a novel method,by integrating a deep convolutional neural network(DCNN)with high-fidelity mechanistic capsule modelling,to identify the membrane constitutive law and estimate associated parameters of a microcapsule from its steady deformed profile in a capillary tube.Compared with conventional inverse methods,the present approach is more accurate and can increase the prediction throughput rate by a few orders of magnitude.It can process capsules with large deformation in inertial flows.Furthermore,the method can predict the capsule membrane shear elasticity,area dilatation modulus and initial inflation from a single steady capsule profile.We explore the mechanism that the DCNN makes decisions by considering its feature maps,and discuss their potential implication on the development of inverse methods.The present method provides a promising tool which may enable high-throughput mechanical characterisation of microcapsules and biological cells in microfluidic flows.
