Optimization of Thermal Flow Dynamics in Tobacco Leaf Curing via IoT-Integrated Deep-Neural-Network-Assisted CFD

In bulk tobacco curing barns, thermo-aerodynamic interactions critically determine product quality through coupled heat-mass transfer processes. To address persistent discrepancies between computational fluid dynamics (CFD) simulations and experimental results caused by insufficient sensor validation, we develop an Internet of Things-enabled CFD-deep neural network-integrated platform (IoT-CFD-DNN IP) integrated platform based on data-physics fusion and edge-cloud adaptive learning. This framework synthesizes IoT sensor data with high-fidelity CFD simulations to overcome spatial sensing limitations, constructs dual-source inputs for DNN modeling, and autonomously optimizes curing conditions by learning airflow-heat-leaf couplings through a two-stage adaptive strategy: offline pre-training establishes thermo-aerodynamic mappings while online fine-tuning dynamically updates weights via real-time sensor data. The cybernetic loop is completed by executing control commands on barn controllers via IoT. CFD simulations were conducted for key tobacco curing stages: yellowing (38°C), color-fixing (48°C), and stem-drying (62°C). A multi-parameter monitoring platform, coupled with DNN, validated and refined these simulations. The results demonstrated that optimization using the IoT-CFD-DNN IP significantly improved simulation accuracy, with temperature and wind speed field fitting accuracy increasing by 86.7% and 55.9% during the yellowing phase, 65.3% and 55.4% during the color-fixing phase, and 75.3% and 61.6% during the stem-drying phase. Beyond accuracy gains, the platform enhances tobacco quality consistency and reduces energy consumption versus conventional methods. This study establishes a cyber-agronomic convergence framework, integrating multi-sensor cyber-physical systems with adaptive DNNs to enable self-optimizing curing through closed-loop fan/temperature control. The paradigm advances precision agriculture by autonomously generating high-fidelity digital twins, providing an intelligent foundation for smart curing operations.