Threshold-Feedforward Neural Networks with Application to Tractor Automated Ground Leveling

Abstract

This dissertation details an improved supervised machine learning method, referred to as the Threshold-Feedforward Neural Network (TFNN). The TFNN operates on continuous inputs, generates discrete outputs, and effectively produces superior classifications of outputs in a noisy environment. The TFNN is successfully applied to a tractor ground leveling system, increasing operator comfort and improving ground leveling with consistent quality. Training of the TFNN and Tractor Automated Ground Leveling (TAGL) simulations was formulated, applied, and verified using a virtual profile that detailed the terrain. A tractor was equipped with a GPS receiver to verify the simulation result partially. Data were collected to show the receiver's ability to locate the tractor's position, altitude, and pitch angle. The rear implement arm angle was detected using motor position feedback sensing. The data gathered showed all necessary inputs and output information to feed into the simulation model to realize the theoretical TFNN results. Further work will equip the tractor with a display to show the tractor operator real-time leveling error input, enabling the completion of TAGL training data gathering and implementation