Implementing Machine Learning Algorithms on Finite Element Analyses Data Sets for Selecting Proper Cellular Structure

Abstract

Numerous structure geometries are available for cellular structures, and selecting the suitable structure that reflects the intended characteristics is cumbersome. While testing many specimens for determining the mechanical properties of these materials could be time-consuming and expensive, finite element analysis (FEA) is considered an efficient alternative. In this study, we present a method to find the suitable geometry for the intended mechanical characteristics by implementing machine learning (ML) algorithms on FEA results of cellular structures. Different cellular structures of a given material are analyzed by FEA, and the results are validated with their corresponding analytical equations. The validated results are employed to create a data set used in the ML algorithms. Finally, by comparing the results with the correct answers, the most accurate algorithm is identified for the intended application. In our case study, the cellular structures are three widely used cellular structures as bone implants: Cube, Kelvin, and Rhombic dodecahedron, made of Ti–6Al–4V. The ML algorithms are simple Bayesian classification, K-nearest neighbor, XGBoost, random forest, and artificial neural network. By comparing the results of these algorithms, the best-performing algorithm is identified.

A multiphysics-based artificial neural networks model for atherosclerosis

“…However, more recently, to increase the computational speed of these models, machine learning techniques are being used to reduce the calculation time of the physics-based models (see e.g. [16] , [17] , [18] ). This combination has resulted in great achievements, and highly applicable tools are developed for medical applications (see e.g.…”