Conclusions

Development of a universal machine learning system for predicting material properties for a wide range of diversified materials is a very large undertaking, which eventually results in the development of hundreds of specific ML models, integrated into one solution. Although novel, the system has already demonstrated capability to effectively model mechanical and physical properties of tens of thousands stainless steels, aluminums, coppers, refractory alloys and polymers, at the accuracy level of property data collections that are typically used in engineering and CAE simulations.

The original methodology used for the development of the system [], provides multiple possibilities for further research and enhancements. One direction is certainly to further enlarge the number and types of materials that can be modeled, for instance to structural and microalloyed steels, as well as some additional types of polymers and composites. The second is to increase the number of properties that are being modeled by involving nonlinear properties such as plasticity, fatigue, creep and crack growth, as well as environmental impacts such as corrosion. Finally, the system can be further enhanced by adding additional ML architectures, such as Boosted Decision Tree and Decision Forest Regression, and putting diversified ML models to work together in committee mode, which can provide even more robust and reliable results.