Optimizing Sugarcane Detrashing: Design, Evaluation, and Predictive Modeling of a Mechanical Detrasher

Abstract

The evaluation of the developed whole-stalk sugarcane detrasher was conducted under various operational conditions to assess its effects on residual trash, detrashing efficiency, and overall performance. The detrasher was equipped with feeding and detrashing components powered by a 0.5 hp electric motor. Six distinct treatments (T1-T6) were examined, varying the speed ratio between the feeding and detrashing rollers and the speed of the feeding roller. One-way ANOVA revealed significant differences (p < 0.001) among the treatments for all three response variables. Trash left percentage ranged from 12.80 ± 2.74% (T2) to 30.20 ± 3.69% (T5), detrashing efficiency varied from 69.80 ± 3.19% (T5) to 87.45 ± 3.16% (T4), and output ranged from 0.82 ± 0.05 t h-1 (T6) to 1.42 ± 0.19 t h-1 (T1). Tukey's test identified distinct performance categories, with T4 demonstrating high efficiency, minimal residual trash, and a satisfactory output. Random Forest (RF) and Gradient Boosting (GB) regression models were employed to predict the response variables based on the operational parameters. GB outperformed RF in terms of efficiency (R² ≈ 0.89, RMSE ≈ 2.682%, MAE ≈ 1.543%) and residual trash (R² ≈ 0.90, RMSE ≈ 2.507%, MAE ≈ 1.391%), whereas RF slightly surpassed GB in predicting the output (R² ≈ 0.795, RMSE ≈ 0.101 t h-1, MAE ≈ 0.073 t h-1). The results suggest that ensemble tree-based methods provide an accurate framework for data-driven monitoring in agro-industrial detrashing applications and offer a foundation for advanced process control in the field.