other mechanical properties of materials, namely, Young’s modulus, E, yield strength, Y, and strain hardening exponent, n. The ability to extract these properties from load-indentation curves provides a practical platform for applications such as the characterization of a small volume of materials in thin films, thin surface layer, and other micro-electro-mechanical systems (MEMS).
The proposed material characterization process as illustrated in Figure 1 comprises two main components: (i) forward analysis to build up the database containing relationships between material properties and related characteristics of load-indentation curves; and (ii) reverse analysis via artificial neural network (ANN) models. In forward analysis, extensive 2D and 3D finite element analyses are carried out to investigate the response of indentation in the presence of friction. Simulated conical, spherical and three-sided pyramidal indentation tests covering a wide range of metallic material properties have been conducted and the database established. Both traditional neural network models and support vectors machines (SVM) have been adopted in the reverse analysis. Direct mapping via the proposed neural network models alleviates the use of iterative procedures in the reverse analysis. The tuned networks are able to accurately predict the mechanical properties of a new set of materials. They are shown
through sensitivity studies to be robust and with ability to provide accurate predicted values for mechanical properties of any elasto-plastic material, obeying power law strain hardening.
Though various combinations of results from conical, spherical and Berkovich indentation tests may be used in the reverse analysis, the combination should necessarily include results from spherical indentation tests. The latter always provides a distinct set of load-indentation curves when the material properties are different and consequently, when used with any other set always provides a unique set of practically accurate solutions. The proposed ANN and SVM algorithms are able to characterize any metallic material of small volume including those considered indistinguishable by other researchers, as shown in Figure 2. The proposed approach has great potential to be extended to applications for the characterization of a small volume of materials, including those in nano-electro-mechanical systems (NEMS), provided that the size effect is incorporated in the analysis to account for the well-known phenomenon of materials gaining in strength and hardness when the deformations are small at the nano-meter level.
