Two complementary methods for the computational modeling of cleaning processes in food industry. (6th April 2020)

• Record Type: Journal Article
• Authors: Deponte, Hannes; Tonda, Alberto; Gottschalk, Nathalie; Bouvier, Laurent; Delaplace, Guillaume; Augustin, Wolfgang; Scholl, Stephan
• Is Part Of: Computers & chemical engineering. Volume 135(2020)
• Publication Date: 2020-04-06
• Subjects: Statistical analysis -- Symbolic regression -- Dimensional analysis -- Machine learning -- Cleaning process -- Food industry; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2020.106733
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: Two approaches for the comprehensive modeling of cleaning processes in food industry are introduced. A computational method for the dimensional analysis is applied. Three dimensionless numbers for the description of cleaning processes are derived. The results of statistical analysis and machine learning are compared and discussed. Abstract: Insufficient cleaning in the food industry can create serious hygienic risks. However, when attempting to avoid these risks, food-processing plants frequently tend to clean for too long, at extremely high temperatures, or with too many chemicals, resulting in high cleaning costs and severe environmental impacts. Therefore, the optimization of cleaning processes in the food industry has significant economic and ecological potential. Unfortunately, in-situ assessments of cleaning processes are difficult, and the multitude of different cleaning situations complicates the definition of a comprehensive approach. In this study, two methodological approaches for the comprehensive modeling of cleaning processes are introduced. The resulting models facilitate comparisons of different cleaning processes and they can be scaled up for processes with similar conditions, using cleaning time as a response. A dimensional analysis is performed to obtain general results and to allow transfer of the approaches to other cleaning situations. The models are established according to the statistical rules for the deduction of multiple regression equations for the prediction of the response based on the input parameters. The terms of the model equation are confirmed with a significance analysis. A machine learning approach is also used to create model equations with symbolic regression. Both methods and the obtained model equations are validated. The two applied approaches reveal similar significant terms and models. Significant dimensionless numbers are the Reynolds number, the density number that describes the ratio of the density of the soil to the density of the cleaning agent, and the soil number, which is a new dimensionless number that characterizes the properties of food soils. The methodology of both approaches is transparent; therefore, the resulting equations can be compared and similarities are found. Both methods are deemed applicable for the computational modeling of cleaning processes in food industry.