Whey protein fouling prediction in plate heat exchanger by combining dynamic modelling, dimensional analysis, and symbolic regression

Abstract

Heat treatment of whey protein solution is a common industrial practice to texturise dairy derived products and meet shelf-life requirements. Thermal treatment is frequently interrupted for cleaning which consumes a large amount of water at different pH to remove deposits from the heating surface. Although it has been a research topic for decades, fouling growth models are still poorly predicted beyond the model training dataset. Here, parameters in a dynamic 2D plate heat exchanger (PHE) model were fitted to capture deposit mass when three variables are manipulated. These are whey protein concentration (0.25–2.5% w/w), calcium concentration (100 and 120 ppm) in the feed and PHE configuration, represented by the number of heating channels (5 and 10 channels). The PHE model consists of thermal, reaction, and fouling sub-models to account for the key events behind deposit formation. The PHE fouling model has a single parameter that needs re-estimation if the processed whey protein solution and process conditions are slightly changed. In the past, this case specific re-estimation has hindered the prediction capability of the model. In this regard, dimensional analysis of the PHE and symbolic regression were used to create a mathematical relationship for the fouling model adjustable parameter, enabling estimation of deposit mass for a wider range of whey derivatives and process conditions. The modelling approach was validated for three different scenarios representing different thermal profiles and whey powder. The proposed methodology increases the ability to predict fouling for different operating conditions and whey protein solutions.