Skip to main content
  • Home
  • Happenings
  • Events
  • Materials informatics enabled quantification of structure property correlations
Materials informatics enabled quantification of structure property correlations

Materials informatics enabled quantification of structure property correlations

Date25th May 2022

Time03:00 PM

Venue Online (Google Meet :



Materials discovery, design, and deployment is a resource, and time-intensive process and roughly takes about 10-15 years. Analogous to Industry 4.0 paradigm, the Materials 4.0 paradigm uses the concepts like big data, machine learning, etc., for accelerated materials deployment. Among steels, Dual Phase (DP) steels are a heterogeneous mixture of soft ferrite and hard martensite phase. This phase incompatibility greatly influences its mechanical behavior. Ductile damage in DP steels is characterized by void nucleation, growth, and coalescence. However, it is crucial to understand and predict damage by accounting for the variability in the microstructures and corresponding phase behavior. The various DP steel microstructures exist in a large n-dimensional space, and evaluation of these microstructures for damage would require reliable structure-property correlations. Numerical models mimic this complex damage phenomenon but are usually computationally expensive and rely on heavy parameter calibration. In both coupled and uncoupled damage models, stress triaxiality remains at the heart of the damage evaluation process. The effect of stress triaxiality on void nucleation and growth is further influenced by the plastic behavior of the material.

In this work, we utilize the concepts of materials informatics to quantify these damage-based structure-property correlations and to develop a microstructure and hardening sensitive Reduced Order Model (ROM) for predicting damage initiation in DP steel. The regression-based model effectively predicts the damage initiation stress for various banded and non-banded DP steel morphologies. In addition to that, we have established a statistical-fitting based model to select the appropriate microstructure morphology for a service application. The regression-based and statistical fitting-based models can successfully quantify the damage initiation and group the various classes into three major clusters. To explore the large microstructure space, we use Generative Adversarial Networks (GAN) to reconstruct similar microstructures. The quantification of similarity assessment and physical awareness of these reconstructed DP steel microstructures is evaluated by various metrics and the above ROM. In this seminar, we elaborate on these concepts in detail.


Sanket Thakre (MM18D702)

Metallurgical and Materials Engineering