Unsupervised learning-aided extrapolation for accelerated design of superalloys

Machine learning has been widely used to guide the search for new materials by learning the patterns underlying available data. However, the pure prediction-oriented search is often biased to interpolation due to the limited data in a large unexplored space. Here we present a sampling framework towards extrapolation, that integrates unsupervised clustering, interpretable analysis, and similarity evaluation to sample target candidates with improved properties from a vast search space. Using the design of superalloys with improved \({\gamma }^{{\prime} }\)-phase solvus temperature (\({T}_{{\gamma }^{{\prime} }}\)) as a model case, we start with sparse data, and by a few experiments, we find nine new superalloys with chemistries distinct to those in the training data. Three of them show improved \({T}_{{\gamma }^{{\prime} }}\) by about 50 °C, a large enhancement for superalloys. Moreover, we find two features characterizing mismatch in atomic size and mixing enthalpy linearly effect. This work demonstrates the capability of unsupervised learning to search for new materials when limited data is available.