A digital twin integrating multi-objective optimization to support fryer operators in managing potato crisps production

The production of potato crisps currently relies on the expertise of human operators, known as fryers, whose training is long and demanding. With the decline in fryer vocations and the increase in consumer quality expectations, it is now essential to develop decision-support tools to make fryer work easier and better control product quality. This study proposes a digital twin (DT) approach that incorporates multi-objective optimization to assist fryers in managing their crisp production line. Data from crisp production is collected and used to model key crisp physicochemical indicators (Fat content, Moisture content, and Lightness) using a Machine Learning approach, specifically the Random Forest method. Then, a multi-objective optimization is carried out using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) algorithm, which identifies optimal adjustments to enhance the three crisp physicochemical indicators. The optimization is tested on 2 batches of potatoes as inputs. The algorithm generates a set of optimal solutions, from which a final solution is selected using the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) multi-criteria decision-making method. This solution provides practical recommendations for adjusting production parameters according to a given potato batch quality. The results show that physicochemical parameters of the crisps are similar after optimization, regardless of the quality the potato batch. Variation in potato batch quality is compensated by appropriate adjustments of the crisp manufacturing process parameters ensuring consistent and optimal production. In conclusion, this digital twin, integrating multi-objective optimization, proves to be a valuable tool for improving fryer decision-making and optimizing production line management.