Abstract
Efficient mechanized dehulling and cleaning of acha (Digitaria exilis) remain technically challenging due to the grain’s extremely small size and the complex interaction between mechanical impact and aerodynamic separation processes. This study developed and comparatively evaluated predictive and optimization models for an acha dehulling and cleaning machine using Response Surface Methodology (RSM) and Support Vector Regression (SVR). A four factor experimental design was employed to examine the effects of feed rate, cylinder speed, fan speed, and tray inclination angle on dehulling efficiency, cleaning efficiency, and grain loss. Sequential model testing and ANOVA indicated that quadratic RSM models adequately represented the system within the experimental domain, with cylinder speed emerging as the dominant factor influencing all responses (p ≤ 0.0004). Significant interaction terms involving cylinder speed (BC and BD) revealed strong coupling between impact intensity, airflow conditions, and grain residence dynamics. Multi response optimization predicted optimal operating conditions at a feed rate of 30 g min⁻¹, cylinder speed of 2000 rpm, fan speed of 186.5 rpm, and tray inclination angle of 45°, corresponding to dehulling efficiency of 90.24%, cleaning efficiency of 90.26%, and grain loss of 9.94%. The RSM models demonstrated moderate explanatory capability (R² = 0.716–0.800; RMSE = 3.19–3.87), whereas SVR achieved superior predictive performance (R² = 0.80–0.85; RMSE = 2.30–2.95) under cross validation. The results indicate that while RSM provides structured experimental insight and reliable optimization within the design space, SVR offers enhanced nonlinear generalization capacity for predictive modeling in complex small grain processing systems.

Key words: Acha dehulling, Response Surface Methodology, Support Vector Regression, process optimization, small-grain mechanization