Interpretable Machine Learning for Acoustic Classification of Incipient Boiling Regimes

Acoustic emissions provide a non-intrusive window into the mechanisms and regimes of incipient boiling triggered by localized heat leaks—a phenomenon relevant to Cryogenic Fuel Management (CFM). This Technical Memorandum (TM) is an explicit machine-learning (ML) companion to NASA/TM–20250009668, which establishes the experimental foundation and physics-based interpretation of accelerometer signals synchronized with high-speed video in a surrogate benchtop setup. Here we focus on operationalizing that regime understanding into an interpretable classification workflow using 441 short-duration boiling runs labeled by human annotators. We extract time- and frequency-domain features designed to capture event rate, rhythmic structure, and spectral content, and we use (i) unsupervised clustering for regime discovery and (ii) a decision-tree classifier selected for transparency and auditability. We also provide a web-based application that reproduces the same feature pipeline for interactive exploration and consistent classification of new runs. The resulting framework supports scientifically grounded, interpretable mapping from acoustic signatures to boiling regimes, complementing the physics-first narrative and enabling future physics-informed ML diagnostics.