Posts by Collection

hobbies

Mathematically Animated Videos on Crystallography (Click for more)

Photos taken using Canon 700D 18-55 Kit Lens (Click for more)

Published in APL Machine Learning, 2023

A novel three-step machine learning methodology was developed using 26,000 experimental records to predict perovskite solar cell performance.
Random Forest model achieved the best performance (R² score of 0.70 for PCE) and was used to optimize non-toxic Sn-based perovskite devices.
The methodology led to significant PCE improvements in Sn-based devices, increasing by up to 28.35% through architecture and deposition optimizations.
Key features identified for optimization included device architecture, dimensionality, and deposition procedures for essential layers.
This data-driven approach offers an efficient alternative to traditional methods for designing high-efficiency, sustainable Sn-based perovskite solar cells.

Published in Computational Materials Science, 2024

Enhanced green channel thresholding for effective segmentation of graphene layers in optical images with variable lighting.
The thresholding can be fine-tuned and adjusted according to the lighting condition.
It can detect the subtle variation in contrast at the layer edges.
Image analysis shows a universal pattern in the green channel pixel median value.
Intersection over union metric (IoU) yielded 94 %, and 89 % accuracy for monolayer and bilayer respectively.

Published in AIP Advances, 2024

GLCM (Gray-level Co-occurrence Matrix) based texture features effectively classify CoCr-based magnetic thin films using EFTEM images.
LightGBM and ANN models achieved classification accuracies exceeding 85%, with ANN reaching 100% accuracy
Image segmentation strategy significantly impacts model performance, with 18 segments proving optimal for the ANN model.
SHAP analysis revealed feature importance, highlighting the ANN's ability to utilize a broader range of features for pattern identification.