Estudo do algoritmo de clusterização K-means no processo de seleção de materiais

Abstract

The use of machine learning algorithms has become common in everyday life, bringing to realize experiments in several fields of material engineering. This study complements the Ashby methodology with the clustering machine learning technic Kmeans, by the application of an algorithm in the material selection process for a screwdriver. For that, two materials processes were performed, where the firsts consisted in use integrally the Ashby’s methodologies. In the second one, materials resulting from the screening step of first process were used and applied the clustering algorithm K-means several times in order to reduce the number of materials. After that, comparisons are made with the two obtained results by each process, making it possible to observe the advantages and disadvantages present in the use of K-means. Four identical materials were obtained in both results. The main advantage observed is the possibility of working with a large number of variables with K-means, whereas in Ashby's methodology it is only possible to use two materials per property map. The main disadvantage is the loss of the algorithm's capacity to differentiate the materials after each run, due to the similarity between them.