A multimodal measurement and explainable AI framework for energy-efficient and quality-aware monitoring of food drying

Abstract

Simultaneously monitoring energy consumption, product quality, and process stability in food drying processes calls for robust measurement infrastructures and consistent decision-making mechanisms. To address these requirements, this study presents a smart measurement and decision-support framework for the drying process of Maras & cedil; red pepper (Capsicum annuum L.) that is measurement-based, data-driven, and interpretable. The proposed system integrates PLC-based sensor infrastructure, PV-supported energy monitoring, computer vision, multisource data fusion, and explainable artificial intelligence components. During the experiments, temperature, relative humidity, air velocity, product mass loss, and heater-fan energy consumption were continuously measured under different drying conditions. Meanwhile, geometric and color-based quality indicators, including shrinkage ratio, shrinkage rate, brightness, redness, and color difference, were extracted from product images. By integrating real-time sensor and image data, a multi-source dataset was created, enabling simultaneous evaluation of drying efficiency, energy consumption, and product quality. Moving to the data analysis stage, the Mask R-CNN-based instance segmentation model in the image processing layer achieved AP scores of 97.50% and 97.84% during validation and test, respectively. Subsequently, in the decision-making layer, decision tree rules were optimized using the Crested Porcupine Optimizer within the hybrid DT-CPO explainable AI model, which achieved a classification accuracy of 97%. These findings indicate that the proposed framework can support the selection of energy-efficient operating conditions and identify warning and risk states through interpretable rules based on visual quality indicators. Overall, the study provides a measurement-focused monitoring infrastructure that offers consistent and interpretable performance within the scope of the evaluated experimental conditions and rule-based constraints for smart drying applications.