A machine learning-based temperature field reconstruction for food processing

This paper presents a reconstruction algorithm for estimating spatial temperature distribution using Optical Frequency Domain Reflectometry (OFDR) in food processing applications. OFDR provides high-resolution distributed sensing along an embedded fiber but is inherently limited to the sensor’s physical placement, leaving large regions of the matrix unsampled. To address this limitation, an interpolation-based algorithm that reconstructs thermal fields in areas beyond the direct reach of the fiber has been developed. Experimental validation was performed by embedding an OFDR sensor in mashed potatoes and subjecting the sample to controlled oven heating and cooling between 25 °C and 105 °C. Continuous temperature measurements were recorded and processed to generate reconstructed thermal maps across the tray. The results demonstrated good agreement with the expected heat transfer dynamics, capturing both spatial gradients and transient behaviors. This approach enables a more complete characterization of heat distribution in complex food systems using a minimal number of sensing fibers. The demonstrated feasibility provides a pathway toward advanced reconstruction techniques that may integrate physics-based modeling or data-driven approaches, ultimately supporting the development of intelligent thermal monitoring solutions for food safety, quality assurance, and process optimization.