| Popis: |
The microstructure of a material is its inner morphological features. The microstructure of steel can be diverse and complex depending on the composition, heat treatment, and processing of the alloy, making it difficult to accurately predict the material’s property and composition without physically analyzing the microstructure. Since the microstructure of steel can determine its physical and chemical properties as well as its performance, cost, and efficiency, it is crucial to accurately classify the microstructure. Although microstructure characterization is widespread and well known, it is mostly conducted manually by human experts analyzing pictures taken by either a scanning electron microscope or a light optical microscope. This research aims to automate this processing using state-of-the-art Machine Learning architectures and models to train and learn to differentiate, classify, and interpret the microstructure pictures, employing a pixel-wise segmentation method via U-NET architecture built upon FCNN, Fully Convolutional Neural Network. The method employed several techniques ranging from data augmentation, Amazon computing service, to semantic segmentation. The system achieved a maximum classification accuracy of 98.689%, and predicted the mechanical property with 10% error, providing a robust, accurate approach for the difficult task of microstructure classification. |
