Topological features (Euler Characteristic Surfaces) can automatically characterize complex physical phenomena and outperform traditional mechanistic models, even without labeled training data—showing that unsupervised topology-based methods can discover and correct scientific models.
Researchers developed the first mathematical definition of churn flow—a chaotic two-phase flow regime in vertical pipes—using topological analysis. They created an unsupervised learning system that combines topology-derived features with machine learning to automatically classify flow regimes and correct existing prediction models, achieving 95.6% accuracy without labeled training data.