Abstract Scope |
The main consideration for using 9 to 12% Cr martensitic-ferritic steels is their relatively high microstructural stability at an operating temperature, with a design lifetime expectation of over 30 years. Due to high dimensionality of the problem, it requires very large datasets for the data-driven model development. The key experimental data collection, particularly on microstructural phases, is very challenging, which makes it particularly difficult to compile a high-quality database for unbiased machine learning (ML). Incorporation of the domain knowledge into ML graph structure, initialization and optimization processes, and informed cross-validation presents a viable mechanism for developing accurate models and reliable alloy design tools, with limited datasets. This presentation will describe the approach to digitize empirical domain knowledge, build the graph based on causality relationships, and use machine learning methodology to identify promising alloy compositions, rank factors affecting the alloys performance, and optimize the processing parameters for specific applications. |