Fuzzy Decisions on Fluid Instabilities: Autoencoder-Based Reconstruction meets Rule-Based Anomaly Classification

TL;DR

This paper introduces a hybrid autoencoder and fuzzy inference system for classifying shockwave anomalies in fluid flow images, enabling interpretable, unsupervised, and real-time diagnostics despite limited labeled data.

Contribution

It presents a novel hybrid framework combining autoencoders with fuzzy rules for fluid instability analysis, improving interpretability and effectiveness over existing methods.

Findings

The $eta$-VAE autoencoder with fuzzy rules best captured shock features.

The approach enables unsupervised, interpretable anomaly classification.

It lays the foundation for real-time, physics-informed diagnostics.

Abstract

Shockwave classification in shadowgraph imaging is challenging due to limited labeled data and complex flow structures. This study presents a hybrid framework that combines unsupervised autoencoder models with a fuzzy inference system to generate and interpret anomaly maps. Among the evaluated methods, the hybrid $\beta$-VAE autoencoder with a fuzzy rule-based system most effectively captured coherent shock features, integrating spatial context to enhance anomaly classification. The resulting approach enables interpretable, unsupervised classification of flow disruptions and lays the groundwork for real-time, physics-informed diagnostics in experimental and industrial fluid applications.