Predicting elastic strain fields in defective microstructures using image colorization algorithms

In this work, an image colorization algorithm based on convolutional neural networks is explored as approach to predict tensile plane-strain field components of microstructures featuring porosity defects. For the same, various shapes, sizes, area fractions and number densities were sampled gage section ASTM-E8 sized numerical specimens whose deformation was simulated in plane strain mode using commercial finite element analysis package Abaqus. Subsequently, trained by treating microstructure defects gray scale image, its color layers, analogous red-green-blue traditional digital representations images. Towards CNN frameworks tested for optimization parameters, viz. filters each layer, stride, padding, activation function. An optimized framework presented that able fields randomly with high accuracy R2>0.91 at a fraction time would take. Various cross-validation tests performed test robustness learning features microstructures. Results indicated extremely robust can provide near-accurate generic scenarios.