Development of Drying-Based Hybrid CFD–MLPNN Application for Predicting the Coccinia grandis Characteristics

Abstract

Objective: Develop a drying based application platform to predict the initial characteristics of Coccinia grandis, which vary with geographic location, climate, harvesting method, harvesting time, and biological variability. Methodology: A computational fluid dynamic (CFD) model was developed based on different partial differential equations that consider heat, mass, and momentum transfer and validated against an experimental drying process (hot air oven at 60°C and an air velocity of 1 m/s). The model was further used to check the sensitivity of various thermal and structural input properties. Based on the highly sensitive characteristics (within the specified ranges and their possible combinations), bulk data (including surface temperature, centre temperature, and average moisture ratio) were extracted over a 20-minute duration. An inverse methodology, multilayer perceptron neural network (MLPNN), was employed to predict the sensitive characteristics based on bulk transport data at different time-steps. After validating the trained model, it was deployed as an application for ease of use using Python libraries. Results and Conclusions: The results indicate that porosity, water saturation, and solid density are most sensitive parameters affecting drying. After extracting bulk data through CFD and further training with MLPNN, the model results closely matched experimental data. Validation metrics yielded R² values of 0.984, 0.986, and 0.975, and RMSE values of 0.010, 1.119, and 2.365 for moisture ratio, surface temperature, and centre temperature, respectively. However, the deployed web-based application reliably provides a robust solution for addressing variations caused by geographic location, climate, harvesting method, harvesting time, and biological variability.