Two-way design of alloys for advanced ultra supercritical plants based on machine learning

Materials for advanced ultra-supercritical (A-USC) power plants with steam temperatures of 700 degrees C and above are extremely needed in order to achieve high efficiency and low CO2 emissions. Alloy design based on machine learning is of great importance to explore the space to decide connections between compositions and performances. In this work, we employed an artificial neural network (ANN) in the machine learning framework to compete a two-way design which is defined by predicting the target properties and designing alloys over the dataset consisted of experimental data. Combined with Genetic Algorithm (GA), the ANN model was optimized to improve the accuracy over 98% by training and testing the full dataset. Meanwhile, the model can find the global optimization values of two performances: yield strength and creep rupture life eventually. With a true accuracy of over 90%, we designed a group of compositions of Ni based superalloy to meet the requirements of microstructures and properties for A-USC plants. Further experimental validation was also conducted, which proved that our ANN model optimized by GA can be used to predict and design superalloys for A-USC.