Prediction of Tensile Properties from Micro-structure in Metallic Alloys using Machine Learning

Design for 'fail-safe' of a structural component includes early prediction of its remnant life under known service loading conditions, preventing catastrophic failure of the component during service. Currently, numerical techniques to determine the remnant life are popular among structural engineers but are time consuming and need substantial experimentation to validate predictions. Off late, researchers are exploring use of appropriate Machine Learning Algorithms (MLAs) for life prediction and seems an alternative approach as it would be quick and relies on data available in the literature. Predicting the tensile strength properties from microstructural information of a structural material is envisaged in the present work. Eventually, the tensile data information would be used in predicting the remnant life of the component. Exhaustive data is available in literature on the microstructural information of most structural alloys. The present work uses the convolutional neural network (CNN) technique to predict the tensile properties in metals and alloys using microstructural images. For this purpose, 20% of the data was used for testing while the rest was used for training. The CNN takes the material microstructure as input to detect the patterns and map necessary features to predict the tensile properties viz., yield stress and ultimate strength. The prediction shows good correlation (< 20% error in few cases) with the experimental results from the literature and exhibits a 'proof-of-concept' for future work.