Application of Artificial Neural Networks for the Replenishment of Nickel-Based Superalloys Catalogues

Abstract

Nickel-based superalloys are compositions with complex doping. To date, a lot of data on the chemical composition and service properties of the alloys has been accumulated. However, a significant amount of information is still missing, which is a problem for practitioners. Novel information technologies make it possible to substantiate a new approach to analyzing the service properties of metallic materials and filling information gaps. The approach is based on artificial neural networks. The previously obtained real test data formed the basis for estimating the missing values. Special conversion of input data greatly improves the accuracy of computations. As a result, a database for 210 types of nickel-based superalloys and their tensile strength under various conditions was obtained. Using the database of chemical composition and tensile strength of alloys as input parameters of the ANN the model allowed us to obtain a series of curves. These curves represent the dependences of tensile strength on temperature-time conditions, which are converted into the Larson-Miller parameter. Analysis of dependences allows one to select the parameters of the initial tensile strength and thermal stability of the alloys. The computation results coincide with the experimental data.