Characterising paste extrusion behaviour by signal processing of pressure sensor data

Methods for control of the extrusion of concentrated solid–liquid pastes are limited by the lack of reliable process sensors. A previous study showed that quantitative indicators of the homogeneity of pastes and extrusion defect phenomena could be obtained from statistical analysis of die pressure data. This paper describes the application of more recent statistical techniques to the analysis of such data, including wavelet and Bayesian methods. Methods for calculating fractal properties (specifically, the Hurst parameter) were compared, and those which were not affected significantly by the presence of periodic components were applied to real data. Wavelet analysis proved very effective for de-noising pressure signal data, which were collected from ram extrusion tests using three materials: a starch-based snack food dough, a detergent and a ceramic paste, and analysed for measures of outliers, coefficients of variation, the Hurst parameter (a fractal property) and periodic behaviour linked to regular surface fracture. Multivariate cluster analysis methods were found to be more reliable in distinguishing between pastes and identifying suspect data sets. Different fracture behaviours were found to exhibit different pressure signal characteristics. The extension of these methods to continuously operating machines is discussed.