Nanoindentation is a widely used technique to characterize the mechanical properties of polymeric materials at the nanoscale. Extreme surface stiffening has been reported for soft polymers such as poly(dimethylsiloxane) (PDMS) rubber. Our recent work [J. Polym. Sci. Part B Polym. Phys. 2017, 55, 30-38] provided a quantitative model which demonstrates such extreme stiffening can be associated with experimental artifacts, for example, error in surface detection. In this work, we have further investigated the effect of surface detection error on the determination of mechanical properties by varying the sample modulus, instrument surface detection criterion, and probe geometry. We have examined materials having Young's moduli from 2 MPa (PDMS) to 3 GPa (polystyrene) using two different nanoindentation instruments (G200 and TI 950) which implement different surface detection methods. The results show that surface detection error can lead to apparent large stiffening. The errors are lower for the stiffer materials, but can still be significant if care is not taken to establish the range of the surface detection error in a particular experimental situation. We have also examined the effect of pressure beneath the probe on the nanoindentation-determined modulus of polystyrene with different probe geometries.

No abstract

High throughput nanoindentation techniques can provide rapid materials screening and property mapping and can span millimeter length scales and up to 106 data points. To facilitate rapid sorting of these data into similar groups, a necessary task for establishing structure–property relationships, use of an unsupervised machine learning analysis called clustering has grown in popularity. Here, a method is proposed and tested that evaluates the uncertainty associated with various clustering algorithms for an example high entropy alloy data set and explores the effect of the number of data points in a second Damascus steel data set. The proposed method utilizes the bootstrapping method of Efron to resample a modeled probability distribution function based upon the original data, which allows the uncertainty related to the clustering to be evaluated in contrast to the classical standard error on the mean calculations. For the Damascus, it was found that results data from a 104 point subsample are comparable to those from the full 106 set while representing a significant reduction in data acquisition.