Machine learning-based draft prediction for mouldboard ploughing in sandy clay loam soil

Abstract

Machine learning (ML) models are developed to predict draft for mouldboard ploughs operating in sandy-clay-loam soil. The draft of tillage tools is influenced by soil cone-index, tillage-depth, and operating-speed. We used a three-point hitch dynamometer to measure draft force, a cone penetrometer for soil cone-index, rotary potentiometers for tillage-depth, and proximity sensors for operating-speed. Draft requirements were experimentally measured for a two-bottom mouldboard plough at three different tillage-depths and various operating-speeds. We developed prediction models using recent ML algorithms, including Linear-Regression, Ridge-Regression, Support-Vector-Machines, Decision-Trees, k-Nearest-Neighbours, Random-Forests, Adaptive-Boosting, Gradient-Boosting-Regression, Light-Gradient-Boosting-Machine, and Categorical-Boosting. These models were trained and tested using a dataset of field measurements including soil cone-index, tillage-depth, operating-speed, and corresponding draft values. We compared the measured draft with the commonly used ASABE model, which resulted in an R² of 0.62. Our ML models outperformed the ASABE model with significantly better performance. The test data set achieved R² values ranging from 0.906 to 0.983. These results demonstrate that the developed ML models effectively capture the complex nonlinear relationship between input parameters and draft of mouldboard plough.