Date of Award

12-2017

Degree Type

Thesis

Degree Name

Master of Science in Industrial Engineering (MSIE)

Department

Industrial Engineering

Committee Chair

Shimon Y. Nof

Committee Member 1

Vaneet Aggarwal

Committee Member 2

Charles Robert Kenley

Abstract

In recent years, more attempts on mechanization, intensification, and automation have been implemented to increase agricultural productivity. However, only few of the concepts are financially feasible for most agricultural fields, mostly due to the high cost of stationary on-ground sensors that is required in such concepts. In response to this issue, a novel approach of using sensor-mounted mobile robots to perform daily inspection has been proposed. By moving the sensors towards different object location, the investment cost is lowered and the framework become more affordable for most agricultural fields. In response to the background mentioned above, this research aims to develop a fault-tolerance learning algorithm to process the data of the moving sensors. The scope and application of this research is limited to a controlled environment within the agricultural robotic system. The sensor data and actual state of the object are generated computationally as a function of error and conflict rate. Two learning algorithms, Adaptive Learning Algorithm (ALA) and Cumulative Learning Algorithm (CLA) are proposed, developed, illustrated, and validated in this thesis. The performance of algorithms is measured in terms of Mean Square Error (MSE) between the predicted and the actual data. A ratio between the MSE of the algorithm and the MSE of baseline scenario is calculated as Conflict and Error Prevention Ratio (CEPR). In terms of mean CEPR, the ALA can reduce the potential error and conflict by 66.4% compared to the baseline scenario. Meanwhile, CLA manages to reduce potential fault to 86.91% on average. Between the two proposed algorithms, CLA’s performance is 30.88% higher compared to ALA.

Download

Share

COinS