Classification of microstructural defects in selective laser melted inconel 713C alloy using convolutional neural networks

Microstructural defects are commonly found in additively manufactured materials and can have significant effects on the material's bulk properties. This warrants defect detection and classification during microstructural evaluation, which is often laborious, costly, and can yield sub-optimal results if done manually. Previous attempts to facilitate automated classification in additively manufactured nickel-alloys have used supervised machine learning methods, such as kth-nearest neighbour classification and decision trees. This study proposes and evaluates the use of convolutional neural networks for microstructural defect classification in selective laser melted Inconel 713C samples. It outlines the process used to create, augment and expand the dataset, as well as hyperparameter selection for the neural network architecture to yield optimal classification performance.