eduzhai > Applied Sciences > Engineering >

Image-driven discriminative and generative machine learning algorithms for establishing microstructure-processing relationships

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: We investigate methods of microstructure representation for the purpose ofpredicting processing condition from microstructure image data. A binary alloy(uranium-molybdenum) that is currently under development as a nuclear fuel wasstudied for the purpose of developing an improved machine learning approach toimage recognition, characterization, and building predictive capabilitieslinking microstructure to processing conditions. Here, we test differentmicrostructure representations and evaluate model performance based on the F1score. A F1 score of 95.1 was achieved for distinguishing between micrographscorresponding to ten different thermo-mechanical material processingconditions. We find that our newly developed microstructure representationdescribes image data well, and the traditional approach of utilizing areafractions of different phases is insufficient for distinguishing betweenmultiple classes using a relatively small, imbalanced original data set of 272images. To explore the applicability of generative methods for supplementingsuch limited data sets, generative adversarial networks were trained togenerate artificial microstructure images. Two different generative networkswere trained and tested to assess performance. Challenges and best practicesassociated with applying machine learning to limited microstructure image datasets is also discussed. Our work has implications for quantitativemicrostructure analysis, and development of microstructure-processingrelationships in limited data sets typical of metallurgical process designstudies.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×