B. Zhang, M. Porro, A. Parmar, Y. Shin
Purdue University,
United States
Keywords: microstructure-property relationship, additive manufacturing, data-driven modeling, stainless steel
Summary:
The high upfront cost of determining the resultant mechanical properties of metal parts built by additive manufacturing significantly hinders widespread adoption in the industry. The objective of this study is to establish a procedure to reduce this cost by using data-driven modeling to predict the mechanical properties of AM-built various steel parts. A data-driven model can provide a quicker way of predicting mechanical properties based on given microstructure information. To build a data-driven model the widely available but scattered data in the literature are collected and utilized, and additional tensile specimens were built by binderjet and laser powderbed fusion processes. Microstructural details such as grain size distributions, phase percentages and porosity are extracted from the microstructural images using machine learning techniques such as convolution neural networks. The collected data from the literature and corresponding mechanical property data are used to build data-driven models of structure-property relationships for different stainless steels. The established data-driven models exhibited good prediction capabilities for yield strength, ductility and ultimate tensile strength for a number of validation cases.