Performance evaluation and prediction of optimal operational conditions for a compact date seeds milling unit using feedforward neural networks

Abstract

Listen

Translate article

Date seed grinding remains a significant challenge limiting the utilization of this valuable agricultural by-product." In this study, a compact date seeds grinding unit was designed, tested, and evaluated. The machine has two primary: a pair of toothed cylinders and a hammer mill. The machine’s performance was assessed in terms of throughput, specific energy consumption, and mean particle size of the product. First, the cylindrical section was tested under various conditions, including cylinder rotational speed (150, 250, 350, and 450 rpm), feed gate opening size (30, 37.5, and 45 cm2), and the clearance between cylinders (0, 1, and 2 mm). The feedforward neural network (FNN) framework predicated the optimal operating conditions for this part, which were recorded as 150 rpm cylinder rotational speed, 45 cm2 feed gate opening, and 2 mm cylinder clearance. This optimal operational condition was utilized as the starting conditions for subsequent testing of the hammer mill section. Then, the hammer mill was tested with different hammer rotational speeds (1250, 1500, and 1750 rpm) and screen hole diameters (2, 4, and 6 mm) underneath the hammers. The FNN model was again employed to predicate the most suitable operating parameters for the grinding unit. The key results included the optimal operational parameters at 150 rpm cylinder rotational speed, 2 mm clearance, 45 cm2 feeding area, 1750 rpm hammer speed, and 6 mm screen hole diameter. That operational condition resulted in 30 kg/h for machine’s throughput, 49 kW h/ton specific energy consumption, and 2.14 mm mean product size. With FNN model accuracy R2 of 0.99974, demonstrating high prediction reliability. Meanwhile, the operating cost was 0.027 $/kg, suitable for small to medium-scale operations. The significance of these findings lies in the development of an efficient, versatile milling solution for date seeds and similar agricultural materials. This research pioneers the application of machine learning in optimizing date seed processing, potentially revolutionizing agricultural waste valorization and opening new avenues for sustainable resource utilization.