PCA-based weighted similarity calculation algorithm for steel materials matching

In order to improve the reliability and feasibility of steel materials matching methods, a new scheme based on weighted similarity calculation is proposed, which assimilates both metallurgic knowledge and data-driven method to maximize the useful information about steels for similarity calculation. While metallurgic knowledge is used to locate steels in different steel classes and identify dominant chemical elements in certain steel class, as data-driven method Principal Component Analysis (PCA) is introduced to discover latent patterns of steel properties influenced by chemical elements and to assign weighting factors to all these chemical elements. Subsequently weighted similarity calculation is realized to quantify likeness between steel materials and to add new similar steels to existing interchangeable steel clusters. This algorithm is applied to free-cutting steel class in Gang.so Online Multinational Steels Database, and the matching results demonstrate its superiorities of supplementing current empirical interchangeable steels and providing quantitative similarity values among these alternative materials to facilitate clients´ selection.