Title : Development and industrial validation of an ai-based decision support system for dairy processing
Abstract:
Artificial intelligence (AI) is emerging as a key technology for the digital transformation of food manufacturing. However, most industrial dairy processes are still operated using fixed control strategies and operator experience, limiting their ability to anticipate process deviations and maintain optimal operating conditions under continuously changing process dynamics. This study presents the development and industrial validation of an AI-based Decision Support System (AI-DSS) designed to provide predictive supervision and operational decision support for complex dairy processing. The proposed AI-DSS integrates four complementary layers, including real-time industrial data acquisition, a Digital Twin for dynamic process state estimation, machine learning models for predictive analytics, and a multi-objective Genetic Algorithm for process optimization. Unlike conventional monitoring systems, the proposed framework continuously evaluates process conditions, predicts future process behavior, estimates non-measurable process variables, and recommends corrective operational actions before significant process instability or product quality deterioration occurs. The system was experimentally evaluated using industrial-scale datasets obtained from two representative dairy unit operations. Ceramic microfiltration experiments consisted of twenty independent industrial production runs performed over eight-hour operating cycles. Process variables including feed temperature, viscosity, feed pressure,
permeate pressure, transmembrane pressure, crossflow velocity and pH were continuously monitored, while membrane hydraulic resistance, mass transfer coefficient, permeate flux, critical flux, limiting flux, fouling percentage and fouling mechanisms were estimated and incorporated into the decision engine. A second evaluation was conducted using twenty industrial spray-drying production runs under eight-hour operating conditions. Real-time measurements included feed viscosity, inlet and outlet air temperatures, drying air flow rate, chamber vacuum, cyclone temperature, fluidized-bed operating conditions and powder outlet temperature, whereas powder quality attributes including bulk density, solubility, angle of repose, moisture content, compressibility, flowability and particle-size distribution were incorporated into the optimization framework. Experimental evaluation demonstrated that the proposed AI-DSS successfully integrated Digital Twin modeling, predictive machine learning and Genetic Algorithm optimization into a unified supervisory platform capable of monitoring, evaluating and supporting operational decisions for both membrane filtration and spray-drying processes. The proposed framework provides a scalable foundation for intelligent dairy manufacturing and represents a practical step toward future smart dairy factories.
