Digital Twin-Enabled Predictive Maintenance for Climate-Smart Food Dehydration in Tropical Climates

Abstract Post-harvest losses and the high energy intensity of conventional drying remain stubborn bottlenecks for decentralised food processing in tropical economies. This study develops and evaluates a digital twin (DT) that couples a physics-based drying-kinetics engine with an autoencoder-driven predictive-maintenance (PdM) engine for a hybrid solar-electric dehydrator operated under real tropical conditions in Jos, Nigeria. Streams of chamber temperature, relative humidity, airflow and power draw were synchronised with a virtual model at one-minute resolution across 42 drying batches. The twin predicted moisture ratio with a root-mean-square error of 0.031 and tracked humidity disturbances that a static recipe model missed entirely. The PdM engine detected incipient fan-bearing degradation 5.3 h before functional failure, achieving 94.0% precision and 91.2% recall while correctly rejecting weather-driven anomalies. Relative to static recipe operation, twin-optimised drying cut specific energy use by 31%, batch-level CO2e emissions by 36% and spoilage by roughly two-thirds. The results show that modestly instrumented, climate-smart dehydrators can deliver industrial-grade reliability alongside material decarbonisation gains, offering a replicable template for smallholder-linked food enterprises in emerging economies. Keywords: digital twin; predictive maintenance; food dehydration; drying kinetics; climate-smart agriculture