Predictive modelling of hexagonal boron nitride nanosheets yield through machine and deep learning: An ultrasonic exfoliation parametric evaluation

A systematic and comprehensive approach for efficiently exfoliating bulk hBN into hexagonal boron nitride nanosheets (hBNNs) via ultrasonication-assisted liquid-phase exfoliation (UALPE) is presented in this investigation. The optimal condition for UALPE was identified as 330 W ultrasonication power for 90 min using a 1:1 (v/v) isopropanol/deionized water mixture, achieving a 12.68 % yield of hBNNs. Successful exfoliation to hBNNs was confirmed by characterization techniques such as UV-Visible spectroscopy, X-ray diffraction, FTIR spectroscopy, Raman spectroscopy, and electron microscopic analysis with preserved crystallinity, improved interplanar spacing, and reduced crystallite size. The study extends beyond optimizing exfoliation parameters by employing machine learning (ML) and deep learning (DL) techniques to forecast the yield of hBNNs. A suite of machine learning regression models including Adaptive Boosting (AdaBoost) Regressor, Random Forest (RF) Regressor, Linear Regressor (LR), and Classification and Regression Tree (CART) Regressor, was employed alongside a deep neural network (DNN) architecture optimized using various algorithms such as Adaptive Moment Estimation (Adam), Root Mean Square Propagation (RMS Prop), Stochastic Gradient Descent (SGD), and Limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS). The DNN with the Adam optimizer achieved the highest accuracy (R2 = 0.98423), demonstrating superior predictive capability compared to other ML and DL models. This novel research optimizes hBN exfoliation and establishes a new framework for yield prediction using machine and deep learning, empowering researchers for targeted hBNNs production.