Abstract:
Nanoindentation is often utilized to examine the mechanical properties of materials and the interactions between grain boundaries (GBs) and dislocations. During the nano-indentation, Load-Displacement (LD) curves usually display the load or displacement burst, which is known as the “Pop-in”. If the indentations are conducted in close proximity to a GB, in addition to the first pop-in, a second distinctive displacement event could be seen on the LD curve under higher loads, often referred to as the "GB pop-in". The transmission of dislocations to the adjacent grains is one of the causes of these secondary GB pop-in occurrences. The present study introduces a novel strategy that differs from traditional advanced characterization methods by using machine learning techniques to predict GB pop-in events. Genetic algorithm (GA) and particle swarm optimization (PSO)-based machine learning models, including Gaussian process regression (GPR), ensemble learning tree (ELT), support vector machine (SVM), and decision tree (DT), are developed. Model selection is based on coefficient of determination (R²) value. For GA the GPR, ELT, SVM, and DT, R2 values were found to be 0.9999, 0.9264, 0.9711 and 0.9811, respectively, whereas for PSO, GPR, ELT, SVM and DT were found to be 0.9999, 0.9976, 0.9611, and 0.9682, respectively. It is evident from the aforementioned R2 value that the GPR shows a value close to 1 as compared to the other three models, hence showing the best performance. Partial dependence plot (PDP) analysis underscores the significance of load and displacement parameters for precise prediction. Lastly, a user-friendly graphical interface (GUI) is meticulously designed based on the GA-GPR model. The integration of these novel methods enhances both their accuracy of predictions and the researchers' ability to detect the GB pop-in events, resulting in improving the fields of predictive modeling and materials science.
