Modelling near-infrared signals for on-line monitoring in cheese manufacture

This paper considers the analysis of a continuously monitored near‐infrared reflectance signal at a single wavelength for the calibration of a process parameter in an application to food engineering for quality control. We describe how we may summarize the information in the observed signals by postulating an explicit statistical model on the signal. An exploratory data analysis may then be carried out on the profile summaries to evaluate whether and how the functional data provide information on the parameter which we would like to calibrate. From a conceptual point of view, such an approach is not dissimilar to principal component regression methods which use an intermediate decomposition through which information is summarised for calibration of a response. The paper demonstrates how the approach leads to important insights into the manner in which the functional data provide the required information on the desired outcome variable, in the context of the practical application in quality control which is discussed and by using a designed experiment. Calculations are implemented through the Gibbs sampler. Calibration of the prediction equation takes place through meta‐analysis of the summarized profile data in order to take the uncertainty inherent in the summaries into account. Copyright © 2002 John Wiley & Sons, Ltd.