Developing versatile alloys of aluminum can now take half the time, with the use of computation science and neural networks.
As a versatile metal, Aluminum has various applications because it is lighter than iron and easier to mold and craft. However, it is usually used as alloy containing other metals because pure aluminum is soft.
Conventional aluminum alloys lose tensile strength at 100 degrees Celsius or higher. Therefore, developing alloys that can withstand higher operating temperatures is an industry goal.
However, developing high-performance aluminum alloys usually takes time because designing aluminum alloys requires deep metallurgy knowledge and repetition of trial and error.
Now, with the use of AI, a Japanese firm has developed neural network models to predict mechanical properties of 2000-series aluminum alloys from their design with high accuracy.
In collaboration with the National Institute for Material Science (NIMS) and The University of Tokyo (UTokyo), Showa Denko K.K. (SDK) developed computational models that accelerate the process of exploring optimal metallurgic compositions and heat-treatment methods to create alloys that can maintain tensile strength at high temperatures. The development time can be shortened by about half to one-third of that with conventional methods.
Utilizing the design data of 410 types of aluminum alloys listed in public databases, SDK developed neural network models that accurately predict the strength of aluminum alloys at various temperatures ranging from room temperature to high temperature. This neural network model can estimate the strengths of aluminum alloys under 10,000 different conditions within two seconds, which allows for greater experimentation compared to traditional methods that take hours.
Furthermore, the AI model allows users to define a set of desired alloy properties for an arbitrary working temperature for the neural engine to offer possible metallurgic compositions that satisfy the desired parameters.