Uncertainty-aware constrained optimization for air convective drying of thin apple slices using machine-learning-based response surface methodology

Air convective drying is an important food processing technology contributing to moisture reduction and food product preservation. Optimization of air convective drying is crucial to achieve high food quality and process efficiency. However, existing drying optimization methods have two critical limitations. First, conventional response surface methodology cannot adequately account for the intricate relationships between process variables and responses, and fails in optimization of multiple drying objectives including drying quality, drying time, and energy consumption. Second, process uncertainties are ubiquitous in industrial food drying, but existing modeling approaches often neglect these uncertainties. To address these limitations, this paper develops an uncertainty-aware constrained optimization framework for air convective drying of thin apple slices. Specifically, we employ machine learning techniques to establish variable-response relationships. The Monte Carlo simulation-based approach is utilized for uncertainty quantification. A constrained optimization method is then used to identify feasible design spaces and find the optimal process parameters. To validate our framework, we conduct drying experiments simulating real-world settings featured by thin apple slices and process uncertainties (e.g., sample thickness). Further, multiple key quality characteristics including color, texture, and water activity are measured and considered within the proposed framework. The developed response surface model demonstrates excellent prediction accuracy with an average mean absolute percentage error of 5.2%. The constrained optimization method leads to 17.9% energy savings and 19.4% reduction in drying time.