A Data-Driven Approach to Forecast Engine Torque of an Agricultural Tractor Across Varied Operational Range Using Machine Learning

Machine learning has repeatedly found its way into the realm of agricultural applications due to its advantages over traditional frameworks. Their capacity for prediction and forecasting through parallel reasoning stands as their primary strength. This research endeavour aimed to establish a predictive model for engine torque for M-575 tractor across diverse operating conditions. A total of 6 machine-learning models were developed for tractor engine torque prediction, and the Extreme Gradient Boosting Regressor was selected as the emerging optimal choice for machine-learning-based predictive model development. The performance of the developed model was calculated based on the evaluation metrics like the coefficient of determination (R ² ) and mean squared error. Notably, an R2 value of 0.997 for training and 0.971 for testing datasets was achieved, accompanied by a mean squared error of 0.0009 for training and 0.0074 for testing datasets. The findings indicated a direct relationship between engine torque and engine speed under different load and throttle settings. The developed ML algorithm exhibited the ability to accurately forecast engine torque within the given power range, accounting for varying load and throttle configurations (100% and 50%), with a significant level of precision (MAPE = 3.45%). Model validation involved comparing the predicted and actual engine torque data of a specific tractor model, M-575, subjected to testing at FMTTI India.