Identification of gas–liquid two‐phase cavitation states in Venturi reactors based on acoustics and deep learning

Abstract Simultaneous implementation of ozone oxidation and cavitation oxidation within Venturi reactors enables an efficient synergistic oxidation process, whose performance optimization hinges on precise identification and regulation of cavitation states. However, non‐condensable gases generated by ozone within the reactor create a complex gas–liquid flow environment, posing challenges for traditional monitoring methods. Thus, this study proposes a framework based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), continuous wavelet transform (CWT), and deep residual shrinkage network (DRSN). This framework utilizes CEEMDAN to decompose non‐stationary acoustic signals and correlation analysis for noise reduction. After CWT conversion into time–frequency maps, the signals are input into DRSN for state identification. Results show that this framework effectively captures the essential characteristics of gas–liquid cavitation. CEEMDAN‐CWT markedly enhances the distinguishability of time–frequency features, while DRSN achieves an accuracy of 98.75% on an independent test set, providing a reliable solution for cavitation monitoring in complex environments.