In conventional procedures, the air flow around the object being tested is modeled using a complex system of equations. To conclude the calculations takes many hours. An Austrian and a Japanese researcher have now accelerated the process so that the flow lines and pressure field are available in real time, reports the Institute of Science and Technology (IST) Austria.
The breakthrough came with an idea for using machine learning in this area of application. Usually the algorithms require structured input and output data. In the case of two-dimensional images, that works well. But 3D objects are represented by small units such as a network of triangles, which, to put it simply, can confuse the computers. Nobuyuki Umetani and Bernd Bickel used so-called polycubes to solve the problem. The procedure is subject to strict rules and ensures that objects with similar shapes are given similar data structures.
A lot more is still to come from machine learning in automobile production. According to a study by McKinsey , it could create up to $215 billion of additional value worldwide over the next seven years. Increases in efficiency in production alone will make savings of up to $61 billion possible.