Abstract
A novel approach for the regression on passive acoustic emission spectra is presented. Acoustic emission of the fluid flow through an orifice plate is partly characterized by frequency shifts in the spectra. Physical properties of the fluid affect the shifts and can therefore be predicted by exploiting the changes in peak positions. The positions and the corresponding amplitude of the most distinct peaks in the acoustic spectra were determined. The obtained peak positions and amplitude matrices were then used in partial least squares (PLS) regression. The proposed method was validated on two experimental datasets from passive acoustic measurements of liquid flow through an orifice plate. The liquids were a two‐component mixture of ethanol in water, and a three‐component mixture of ethanol and sucrose in water. The proposed method resulted in more parsimonious models and lower root mean squared errors of prediction (RMSEP), compared to the PLS regression on the whole acoustic spectra. The proposed method resulted in 43% lower RMSEP for prediction of ethanol in water, and 24 and 27% lower RMSEP for ethanol and water respectively in three‐component mixture. Only minor improvement in RMSEP was gained for the sucrose prediction. Copyright © 2007 John Wiley & Sons, Ltd.