Deep Learning for Microstructure Property Relationships - Results (MATLAB data)

Description

The data contains trained networks and interpretation data saved into ".MAT" files. The framework was orginally deployed on MATLAB R2022b. It has also been tested to be functional on MATLAB R2021a. The data used in this work has been collected from various literature which may contain copyrighted materials. Therefore we are unable to provide the data used for training. However, we have provided all the data sources in the file "Data_Sources.xlsx" located inside the "1_Alloy_Data" folder. MATLAB scripts for data preprocessing, optimization, training and interpretation of the deep learning framework are provided in a separate git repository: https://github.com/adityag23/Deep-Learning-Microstructure-Property Requirements: MATLAB Deep Learning Toolbox MATLAB Image Processing Toolbox

Steps to reproduce

To replicate this work and use the training and interpretation scripts: 1. Collect images from the article sequentially from each data source and save as the serial number mentioned in the "Data_Sources.xlsx" using .PNG format. Place all the raw images inside a folder named "image_data". 2. Simultaneously collect the respective compositions and Vickers hardness and save in spreadsheet named "compositions.xls" in the following column order:"Image no.", "Al", "Ti", "V", "Cr", "Mn", "Fe", "Co", "Ni", "Cu", "Zr", "Nb", "Mo", "Hf", "Ta", "W", "Re", "C", "B", "Si", "P", "S", "Hardness" 3. Use the "batchprocess_images_in.m" function with 'aug' parameter value 0 for non-augmented dataset and value 1 for augmented dataset. The dataset is ready for use. The ".MAT" files contain trained networks, the performance information of the networks, analysis results represented in plots.

Institutions

Categories

Funding

Licence