Machine Learning-Based Recognition of Cavitation Regimes in a Venturi Tube

The Venturi tube is an important hydrodynamic cavitation generator. This study aims to use machine learning to automatically identify the cavitation regimes (σ = 0.7 and 0.3) in a Venturi tube without any prior knowledge. High-speed photography data were processed into a grayscale matrix. An unsupervised distance and density (DD) algorithm determined cavitation regime classification and quantity, while supervised Support Vector Machines (SVMs) identified cavitation regimes in Venturi tube flow. The results demonstrate that the combination of DD and SVM methods achieved the recognition accuracies over 95%. The unsupervised clustering and supervised identification methods can provide a new technical means for prediction and control of cavitation intensity in the Venturi.