The continuous indentation technique is increasingly being adopted as a
powerful tool for evaluating advanced mechanical properties of metallic
materials, such as tensile behavior and residual stress. With recent
advances in computational modeling, re...
The continuous indentation technique is increasingly being adopted as a
powerful tool for evaluating advanced mechanical properties of metallic
materials, such as tensile behavior and residual stress. With recent
advances in computational modeling, research has extended beyond
traditional analytical and empirical frameworks to incorporate machine
learning and deep learning methodologies. Most of these data-driven
approaches rely on synthetic datasets generated via finite element
analysis (FEA), offering an efficient means to simulate virtual materials
with a wide range of mechanical properties. However, current models
often neglect real-world complexities such as geometric imperfections of
the indenter and variations in specimen surface conditions. In this study,
we digitized the actual geometry of a spherical indenter using scanning
electron microscopy (SEM) imaging and incorporated it into an
FEA-based simulation framework to generate load–displacement curves.
A machine learning model was initially trained using simulation data
derived from an idealized spherical indenter, and its predictive
performance was subsequently evaluated through transfer learning using
load–displacement data reflecting the actual indenter geometry. This
approach underscores the feasibility and effectiveness of constructing
more realistic predictive models that account for practical indentation
conditions.