According to foreign media reports, scientists from Japan's National Institute for materials science have developed a machine learning process to obtain aluminum alloys with specific and required mechanical properties. This method may promote the discovery of new materials.

    Aluminum alloy is a light and energy-saving material, which is mainly made of aluminum, but also contains magnesium, manganese, silicon, zinc and copper and other elements. The combination of various elements and manufacturing process determines the elasticity of the alloy in the face of various stresses. For example, 5000 series aluminum alloys contain magnesium and several other elements, which can be used as welding materials for buildings, automobiles and pressure vessels; while 7000 series aluminum alloys contain zinc, usually magnesium and copper, which are commonly used in bicycle frames.

    To overcome this problem, Ryo Tamura of Japan's National Institute of materials science, along with colleagues and Toyota Motor, is a time-consuming and expensive way to produce aluminum alloys Corporation) developed a material information technology, which can input the known database data of aluminum alloy into the machine learning model, so as to train the model to understand the relationship between different mechanical properties and different components of alloy, and the relationship between different types of heat treatment applied in the production process. Once the model is provided with sufficient data, it can predict what elements and production processes are required to produce new alloys with specific mechanical properties, and all of the above work does not require manual input or supervision.

    For example, the model found that 5000 Series Aluminum Alloys with high stress and deformation resistance can be produced by increasing the content of manganese and magnesium and reducing the content of aluminum. "This kind of information can help develop new materials such as alloys to meet industrial needs," Tamura said

    The model uses a statistical method called Markov chain Monte Carlo, which can obtain information by using the algorithm, and then display the results with graphics, so as to visualize the association between different variables. By inputting larger data sets in the training process, the machine learning method becomes more and more reliable.

Source: Gasch