Abstract: Understanding the acoustic emission signals from hydraulic turbines under flow cavitation and its variations with cavitation intensity is essential for monitoring cavitation. To overcome the difficulty in feature extraction from acoustic signals due to noise pollution, this paper develops a feature extraction method based on noise reduction through adaptive local iterative filtering and singular spectrum analysis, and on intrinsic time scale decomposition combined with feature parameters. First, an acoustic signal is initially denoised using the adaptive local iterative filtering combined with correlation coefficients to filter out evident noise components; the remaining components are reconstructed and further denoised via singular spectrum analysis. Adding the resulting signals to the trend component completes the whole noise reduction. Then, an intrinsic time scale decomposition algorithm is used to decompose the noise-reduced signal, screen out its effective components, and calculate their absolute energy and relative energy entropy. Finally, their variation trends with the cavitation coefficient are examined. The results show the variations in absolute energy and relative energy entropy with the cavitation coefficient manifest better regularity, offering an effective indicator of the developing status of hydraulic turbine cavitation.

Key words: hydraulic turbine, cavitation, acoustic emission, SSA algorithm, noise reduction, ITD algorithm