Deep Learning-Based Identification of Cavitation Instabilities in Turbopump Inducers Using Raw Visualization Data

Abstract Conventional signal decomposition of raw flow visualizations with a limited field of view fails to accurately identify cavitation instabilities in turbopump inducers. Therefore, a new deep learning-based framework is proposed to directly analyze visualization data and identify such instabilities. The framework employs a hybrid convolutional-recurrent neural network to learn the spatio-temporal features of instability modes. The trained model successfully captures alternate blade cavitation in a two-bladed inducer and super-synchronous rotating cavitation in a three-bladed inducer. Notably, the approach requires neither pre-processing of the visualization data nor extensive measurement durations, as only a few shaft revolutions are sufficient for accurate identification. By eliminating the dependence on unsteady pressure transducers, the proposed framework provides a robust tool for identifying turbopump inducer cavitation instabilities and can be extended to other systems where low-order modal patterns are dominant.