Application of Computational Intelligence to Determine the Effect of Different Shear Bar Positions on Chopping Length and Specific Cutting Energy Consumption in the Chopping of Silage Sorghum

Abstract

Flail forage harvesting machines are disadvantageous compared to other silage machines because of the uneven length of the chop obtained, the low distribution of the appropriate chopping length in the entire chopped material, and the high energy consumption. This issue is tackled by employing artificial neural networks (ANN), which serve as versatile mathematical instruments capable of capturing data. In this article, six different knife peripheral speeds (KPS) and three different shear bar (SB) positions were tested to determine the combination of KPS and SB positions that increased the proportion of suitable chopping length (CL) for silage and decreased the specific cutting energy consumption (SCEC) among all chopped material. The results have shown that, depending on the increase in KPS, the average CL varied between 165.28, 127.30, 100.24, 83.55, 77.06, and 65.09 mm. Depending on no shear bar (), shear bar positioned parallel to the feed unit (), and shear bar positioned at an angle of 45° to the feed unit () the average CL was determined as 114.73, 99.65, and 94.88 mm, respectively. Depending on the increase in KPS, the SCEC values vary between 0.97, 1.49, 1.89, 2.20, 3.05, and 4.12 kWh t−1. Depending on , , and the SCEC values were determined as 1.64, 3.05, and 2.17 kWh t−1, respectively. The effects of KPS and SB positions on CL and SCEC were statistically very significant (p < 0.001). An ANN model with a 3‐(5‐5)‐1 architecture, utilizing a backpropagation learning algorithm, was developed to forecast SCEC. This model outperformed traditional statistical models and was constructed using data on KPS, SB, and CL. The ANN model exhibited the highest efficiency, outperforming the polynomial models. In this particular ANN model, key metrics such as coefficient of determination (R2), root mean square error (RMSE), and mean error (ε) (0.9970%, 0.0159%, and 3.86% respectively) are significantly positive. This research convincingly demonstrated the efficacy of ANN in accurately predicting SCEC, leveraging data on KPS, SB, and average CL.