Dimensionality reduction using principal component analysis to classify different grain boundary slip transfer regimes in BCC-Titanium Ti21S: A feature engineering-based approach - Université de Lorraine Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2021

Dimensionality reduction using principal component analysis to classify different grain boundary slip transfer regimes in BCC-Titanium Ti21S: A feature engineering-based approach

Résumé

In this study, grain boundary slip transfer events have been analyzed in β-Ti21S by using a semi-automated procedure for detecting slip traces on the specimen surface and using a distance-based metric to detect the traces closest to and across grain boundaries. The crystallography of the detected slip traces was used to extract several geometric parameters associated with slip transfer. The parameters were grouped to form a multi-dimensional dataset with the aim of performing dimensionality reduction and obtaining a lower-dimensional and perceivable dataset using principal component analysis (PCA). The preference for specific values of the angular variables related to the alignments between the slip planes and directions for the different slip transfer regimes were not perceivable without the PCA. Our analysis reveals the structure in the multi-dimensional dataset which is clearer in terms of the clustering between different slip transfer regimes. Statistical distributions obtained from the clustered datasets clearly show a correlation between the different classifiers used for clustering, such as the slip transmission factors and the geometric parameters (m’ and Nfactor values) computed for the different slip transfer events. The analysis procedure is capable of defining domains associated with certain slip-transfer regimes, expressed as a linear combination of measurable input parameters. Thus, a feature extraction approach has been presented in this study and though clusters associated with different slip transfer regimes become evident by considering just four geometric parameters, the methodology is versatile enough to be extended to data obtained from other experimental approaches such as localized strain-fields obtained from digital image correlation or topographic information obtained from atomic force microscopy.
Fichier non déposé

Dates et versions

hal-03372274 , version 1 (10-10-2021)

Identifiants

  • HAL Id : hal-03372274 , version 1

Citer

Kaustubh Venkatraman, Vincent Taupin, Antoine Guitton. Dimensionality reduction using principal component analysis to classify different grain boundary slip transfer regimes in BCC-Titanium Ti21S: A feature engineering-based approach. 2021. ⟨hal-03372274⟩
150 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More