Using Unsupervised Machine Learning to Examine Microstructure of 17-4 Stainless Steel
Open Access
Author:
Bina, Thomas F
Area of Honors:
Materials Science and Engineering
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Robert Allen Kimel, Thesis Supervisor Robert Allen Kimel, Thesis Honors Advisor Joan Marie Redwing, Faculty Reader
Keywords:
machine learning material characterization microstructure
Abstract:
This work provides an unsupervised machine learning method to examine the microstructure of 17-4 stainless steel. Using a variant of a k-means algorithm, features of steel samples imaged via transmission electron microscopy were analyzed and clustered into unique regions. Each of these regions may correspond to an individual phase in the material. This technique did not require a priori description or labeling of the target material system. The described method may be used in an automated manner which has the potential to be effective in rapid identification of phases across a large data set. The work presented here is the first step in developing a larger automated method to identify and characterize the microstructure of 17-4 stainless steel.