R.J. Sheridan, L.C. Brinson
Duke University,
United States
Keywords: nanoindentation, dataset, viscoelastic, metrology
Summary:
We seek to develop accurate AFM nanoindentation metrology for soft material properties in the adhesive-viscoelastic contact (A-VE) regime, with the goal of extracting viscoelastic properties from features of the individual force curves. While adhesive-elastic contact models and non-adhesive-viscoelastic models are available, there is not yet a widely known or practically applied model for the combined A-VE case, let alone any kind of best practice. To move forward in this relatively unknown region of the adhesion map, we must first obtain reliable data upon which any potential A-VE contact models may be compared. While many papers have been published on materials exhibiting A-VE contact, most have opted to fit the retract or extend portions of the force curve without accounting for the hysteresis, inducing an upward (or downward, respectively) bias in any modulus estimates. Likewise, applying a non-A-VE indentation model to an adhesive case will result in biased modulus and relaxation estimates due to the additional work of adhesion. In prior work that required materials with very well characterized viscoelasticity and highly consistent adhesion characteristics, we developed relatively ideal "model" viscoelastic materials and indentation protocols. By collecting force-distance curves across a broad grid of rates, temperatures, and other indentation parameters, we have created a data set with many consistent examples of viscoelastic-adhesive indentations at each condition to help distinguish evidence of the underlying physics of the interaction from noise and nuisance artifacts. In this presentation we summarize our data sets, which reveal that force curves at varying rates and temperatures have a glass transition where indentation hysteresis is maximized. Then, the viscoelastic force curve dataset is combined with the known viscoelastic properties of the model material, allowing us to reduce the scope of the problem from developing a sufficient physical description of the A-VE contact to finding appropriate features. Further, we compare existing elastic and viscoelastic indentation models to the dataset and compare their (expected) mis-predictions in the presence of adhesion, and look beyond to potential new data-driven model forms. These forms include hand-crafted features and interpretable machine learning approaches, and use an information-theoretical approach to rate the performance of all these against a flexible deep learning model.