Internet of agriculture: Analyzing and predicting tractor ride comfort through supervised machine learning

Abstract

The aim of this study is to improve ride comfort among tractor drivers by utilizing Internet of Things (IoT) technology and Machine Learning (ML) techniques to analyze and predict vibration exposure. The study explores the association between tractor ride characteristics, including average speed, tool depth, and pulling force, and ride comfort during actual rotary soil tillage operations. Ride comfort was evaluated by calculating the Overall Vibration Value (OVV) at three measurement locations (i.e., the floor, seat pan, and seat backrest). The study employed spectral analysis i.e., Fast Fourier Transform (FFT) and Power Spectral Density (PSD) to determine the predominant resonant frequencies. In addition, ML approaches i.e., Linear Regression (LR), Decision Tree Regressor (DTR), Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Artificial Neural Network (ANN) were used to predict ride comfort response. Three hyperparameter optimization techniques, including Grid Search, Random Search, and Bayesian optimization were used. Tractor-tiller system was operated on a real agricultural field by five male tractor drivers. Vibration levels were measured along three translational axes at three measurement locations using a smart monitoring device based on the IoT concept. Average OVV magnitude was computed as 0.84 m/s2 i.e., exceeding the ISO 2631-1 (1997) threshold limits. FFT and PSD analysis revealed a variety of dominant frequencies, including resonance peaks at 2 Hz, 6–7 Hz, and 9–11 Hz. The analysis of variance (ANOVA) indicates that tractor speed and tool depth are significant variables (at the 5% level) in determining ride comfort. Regarding predictive accuracy, the ANN model exhibits the highest coefficient of determination of 0.90, followed by the SVM and GPR models with a coefficient of determination of 0.89. The DTR and LR models demonstrate slightly lower coefficient of determination values of 0.83 and 0.82, respectively. Furthermore, Bayesian optimization proved to be an effective approach in achieving more accurate predictions of tractor ride comfort. In general, the study outcomes can be utilized by tractor designers in numerous ways to optimize tractor design (such as suspension system, cab design, seat design, etc.) to improve tractor ride comfort in real field applications.