Machine learning-based estimation of agricultural tyre sinkage: A streamlit web application

Abstract

This study investigates the impact of wheel slip, drawbar pull, and soil strength on agricultural tyre sinkage under varying normal loads and inflation pressures. A controlled experiment was conducted with a 13.6–28 bias ply tyre using single wheel tester in a soil bin, measuring tyre sinkage, drawbar pull, and wheel slip across different conditions. Machine learning models, including Artificial Neural Network (ANN) and Support Vector Regression (SVR), were developed to predict tyre sinkage based on key variables, with hyperparameter tuning to optimize model performance. The SVR model outperformed the ANN model, with Coefficient of determination (R²) and Mean Squared Errors (MSE) as 0.997 and 0.8 for training; 0.981 and 4.3 mm for testing, respectively. The Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) were also significantly lower for SVR, with MAPE values of 2.58 % (training) and 6.94 % (testing). The optimized SVR model was integrated into a Streamlit web application, offering a user-friendly platform for real-time predictions of tyre sinkage. This application had significant potential for enhancing tractive efficiency and minimizing soil degradation in agricultural practices. The study highlighted the efficacy of machine learning techniques in modelling tyre sinkage.