The soft computing approaches in optimising multi-objective mechanical design of a weeding robot

Abstract

• A mechanical multi-objective design of a weeding robot was performed to remove weeds under trees and on the ground surface. • Artificial neural network models were used to predict the maximum weight and load acting on the components of the weeding robot. • The maximum weight reduction for the suspension system was 24.6 % (from 1.75 kg to 1.32 kg). • The increase in the safety factor for the wheels was 36.0 % (from 1.50 to 2.51). • The maximum absolute difference between the ANN and FEM models was <6.6 %. Weed control is one of the biggest challenges in agriculture, given the considerable variations in the shape, size, speed, and type of weed growth. Weeding robots present a promising solution, as they do not require human labor and can operate under various conditions. In this study, a first of its kind, we propose a multipurpose weeding robot that is designed to remove weed from under trees and on the soil surface. Mechanical weeding techniques were considered for the development of the desired robot. Initially, we developed the kinematics of the robot's arm, which has five degrees of freedom, to facilitate weeding under trees and along narrow paths. A rotating blade, capable of adjusting its height, was subsequently designed to effectively remove weeds from the ground. Given that weight is a critical factor in evaluating robots, this study aims to minimize the weight of the robot. To achieve this, we optimized the design of the robot's components while considering design constraints to minimize mass and load. To this end, we determined the dimensions, weights, and loads acting on the components of 60 existing weeding robots. Artificial neural network (ANN) models were then trained based on the dataset from these 60 specimens. The optimized dimensions and masses were derived using a multi-objective genetic algorithm (MGA), and a finite element method (FEM) analysis of the resulting models was performed in Ansys R18.2. The results indicated that the maximum weight reduction for the suspension system and the increase in the safety factor for the wheels achieved were 24.6 % (from 1.75 kg to 1.32 kg) and 36.0 % (from 1.50 to 2.51), respectively. Furthermore, the maximum absolute difference between the ANN and FEM models was <6.6 %.