.png){kind=icon}
.png){kind=icon}
Return
Model-free data-driven identification algorithm enhanced by local manifold learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s00466-022-02255-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Reliable and consistent material data identification is essential to the data-driven computational mechanics paradigm. This paper presents a generalized data-driven identification (DDI) approach to constructing material databases of high quality. We integrate the locally convex reconstruction method into DDI to formulate the local-convexity DDI (LCDDI) method. The LCDDI method can learn the local structures of material data and thus produce structure-informed optimal material data points for solving stresses with given strains. The effectiveness of the LCDDI method at addressing large acquisitions of material data with a complex heterogeneous strain field is demonstrated through two numerical experiments: a perforated elastic plate and a center-holed elasto-plastic plate. Convergence studies show that results using the LCDDI method are dramatically improved. We further explain how the LCDDI method manages to accurately identify mechanical stress fields and a high-fidelity material database under the condition of imbalanced distribution of elastic and plastic deformation. Discussion of the LCDDI method in regards to the oversampling issue, the capability of full-domain analysis, and importance sampling is given. Finally, we conclude that the LCDDI method can extract a vast amount of material data points with improved quality from full-field strain measurements, and can serve as a more reliable technique for material data acquisition.
Keywords:
Data-driven computing
Material data acquisition
Manifold learning
Material data structure
Local-convexity data-driven identification
Material database quality
Journal
IF:
3.8
Papers:
3.2K
Citations:
9.0K