Prediction of Sustainability Indicators of a Biomass Fueled Natural Convection Dryer Using Machine Learning

Efficient and sustainable drying technologies are crucial for reducing post-harvest losses and supporting agricultural communities in regions with challenging climates. This study presents a biomass-fueled natural convection dryer (BFNCD) and employs machine learning (ML) models to predict the exergetic sustainability indicators like waste exergy ratio (WER), sustainability index (SI), environmental destruction coefficient (EDC) and improvement potential (IP). The results show that SI ranges from 1.03-1.33 without heat storage materials (HSMs) and 1.03-1.47 with HSMs, while WER varies between 0.62-0.96, EDC from 3.3-28.67, and IP between 0.62–203.56 W depending on the case configuration. Using 112 experimental datasets with drying time, air temperature, moisture content, and biomass case as inputs, six multi-output ML algorithms were evaluated. The Support Vector Regression (SVR) multi-output model exhibited outstanding predictive capability, attaining a coefficient of determination of 0.9738 along with a mean absolute error of 0.0015 on the testing dataset. The observed versus predicted values through scatter plots further validated the precision of the model across all four outputs. SHAP (Shapley Additive Explanations) analysis revealed that moisture content and type of biomass case were the most influential parameters for SI and EDC, while drying temperature and case type dominated the prediction of WER and IP, respectively. The study emphasizes the effectiveness of ML models for rapid and accurate prediction of thermodynamic sustainability and facilitates the optimization of BFD systems for enhanced environmental and operational performance.