Data-Driven Phase Analysis of Immiscible Polymer Blends via Multi-parameter AFM Imaging

In recent years, the combination of atomic force microscopy (AFM) and machine learning has gained increasing attention for analyzing the phase structure of polymer blends. However, the potential advantages of using multi-parameter images for analyzing complex and practical polymer blends have rarely been explored. In this study, we applied k-means clustering to multi-parameter AFM images acquired using two measurement modes–nanomechanical and scanning thermal microscopy–to identify the phase structure of a ternary polymer blend composed of polypropylene, polyolefin elastomer, and high-density polyethylene, a system that serves as a practical model for high-impact polypropylene. It was revealed that appropriately combined parameters exhibited a higher phase identification ability than single-parameter images. We also proposed a strategy to select effective parameter combinations based on the silhouette score and information comprehensiveness. The proposed method, based on multi-parameter AFM imaging, offers a generalizable framework for the phase structure analysis of various polymer blend systems.