Towards a Quantitative Analysis of the Mechanical Properties of Soft Materials at the Nanoscale—When AI Meets Materials!

Understanding the properties of materials at the nanoscale is fundamental to predict their macroscopic behavior, thus allowing the design of materials adapted to specific applications.<br/>This work explores the use of Machine Learning to improve the analysis of the mechanical properties of materials at the nanoscale, focusing on measurements made by the Atomic Force Microscope (AFM) in Peak Force Tapping and nano Dynamic Mechanical Analysis modes.<br/><br/>Both modes generate detailed sample maps, providing information on the topography, and the mechanical and viscoelastic properties. However, the quality of these measurements depends on the acquisition parameters, which must be adapted for each sample. To address this important issue, we have developed some Machine Learning-based tools to evaluate and score the quality of acquisitions, in particular via a force curve scoring module using supervised learning algorithms to predict three distinct classes (Unusable, Noisy, Excellent).<br/><br/>Another challenge is the determination of the rigidity modulus of materials, usually obtained by fitting a mathematical function to the force-separation curves. The nature of this function may vary (i.e. sometimes from one pixel to another pixel) depending on the local mechanical properties of the sample and the selected contact mechanics model. Most of the time, this crucial point is not considered by most of the SPM manufacturers. Therefore, our work proposes a novel method based on the Tabor coefficient to select the most suitable mechanical model for each pixel of the map, thus providing more accurate data.<br/><br/>To illustrate the power if this original approach, we have successfully applied it to different polymeric systems including hydrogels, multi (up to four) polymer blends, and nanocomposites. The obtained results show that the force curve scoring module has an accuracy of more than 90%, and that the rigidity module recalculation process offers a higher accuracy than the usual models.<br/><br/>In conclusion, our code, called PyCAROS (Python Code for Approach and Retract curve analysis of Organic and hybrid Soft materials), consists in three main modules: a module for reading the acquisition files, a module for scoring the quality of force curves, and a module for recalculating the mechanical properties aiming at helping the SPM users to be more confident in the data acquisition and analysis thanks to AI processes.