Validating the use of machine learning techniques in evaluating additive manufactured elastic materials

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT REQUEST OF AUTHOR.] As additive manufacturing continues to find widespread use in research and industry and the number of materials suitable for use in additive manufacturing, there is increasing desire to exploit the potential offered by these techniques and materials. Some materials exhibit auxetic mechanical responses, and some exotic designs can exhibit auxetic-like responses. Additive manufacturing allows the use of these designs and materials and will continue to support more of these designs and materials. Using traditional design analysis tools on exotic and rapidly evolving design can be expensive and time consuming, stifling innovation. Machine learning could allow for rapid and accurate predictions of the mechanical responses, but normally requires a very large number of learning samples which is uncommon for mechanical analysis. This project analyzed whether machine learning principles could be used for a low learning sample size. 143 unique structures were constructed from additively manufactured elastic samples and tested to find the buckling loads and effective moduli. A random forest regression model was applied to a learning set of the samples and tested on the remaining samples. It was found that the model used was accurate enough to predict responses despite the small sample sizes. Further work is encouraged on testing actual auxetic materials and structures as well as optimizing a model for limited sample size materials research.