Multiobjective optimization of a food process based on expert knowledge Example of 0.1 µm skim milk microfiltration

International audienceOptimizing food filtration processes is complex due to the lack of knowledge about process performances mechanisms, the complexity of the food product itself and the heterogeneity of the variables involved in prediction models. This is the case for skimmed milk crossflow microfiltration with 0.1 µm pore size (MF). This operation is commonly used to separate the two major milk proteins and produce fractions with high interest in cheese making and ingredient preparation. In this process, the choices of membrane configurations (Uniform transmembrane Pressure system, membranes with gradient of permeability or spiral wound membranes), processing designs and operating conditions are mainly based on the know-how of equipment manufacturers and the available expert knowledge. However, despite its high interest, this process has never been optimized regarding stakeholder's conflicting objectives (high quality, low costs). This work aims to optimize skimmed milk MF 0.1 µm by considering conflicting objectives defined by stakeholders and integrating expert knowledge into the formulation of a multiobjective problem. The methodology used was based on “Knowledge integration” and composed of four steps: 1) the acquisition of scientific and expert knowledge about the objectives to optimize; 2) the modeling of objectives as computable functions; 3) the solving of the multiobjective problem by using an adapted metaheuristic multiobjective optimization algorithm; 4) the decision support. The multiobjective MF problem was formulated by considering the product, operating variables, design and economic costs. 1000 pareto-optimal solutions were found, including solutions close to those used in industry and at lower cost. The multiobjective optimization of the MF has opened up new thinking about installation designs and combinations of operating conditions by improving product characteristics while keeping investment and production costs constant. The result of the optimization is a proof of concept whose feasibility and viability at an industrial scale must be validated. This work opens new perspectives for multiobjective optimization of food membrane processes and generally to food processe