Traffic management and control of automated guided vehicles using artificial neural networks

Preview

PDF WRAP_THESIS_Barbosa_1997.pdf - Requires a PDF viewer. Download (21MB)

Request Changes to record.

Abstract

An industrial traffic management and control system based on Automated Guided Vehicles faces
several combined problems. Decisions must be made concerning which vehicles will respond, or are
allocated to each of the transport orders. Once a vehicle is allocated a transport order, a route has to
be selected that allows it to reach its target location. In order for the vehicle to move efficiently along
the selected route it must be provided with the means to recognise and adapt to the changing
characteristics of the path it must follow. When several vehicles are involved these decisions are
interrelated and must take into account the coordination of the movements of the vehicles in order to
avoid collisions and maximise the performance of the transport system. This research concentrates on
the problem of routing the vehicles that have already been assigned destinations associated with
transport orders.
In nearly all existing AGV systems this problem is simplified by considering there to be a fixed route
between source and destination workstations. However if the system is to be used more efficiently,
and particularly if it must support the requirements of modern manufacturing strategies, such as Justin-
Time and Flexible Manufacturing Systems, of moving very small batches more frequently, then
there is a need for a system capable of dealing with the increased complexity of the routing problem.
The consideration of alternative paths between any two workstations together with the possibility of
other vehicles blocking routes while waiting at a particular location, increases enormously the number
of alternatives that must be considered in order to identify the routes for each vehicle leading to an
optimum solution. Current methods used to solve this type of problem do not provide satisfactory
solutions for all cases, which leaves scope for improvement. The approach proposed in this work
takes advantage of the use of Backpropagation Artificial Neural Networks to develop a solution for the
routing problem. A novel aspect of the approach implemented is the use of a solution derived for
routing a single vehicle in a physical layout when some pieces of track are set as unavailable, as the
basis for the solution when several vehicles are involved. Another original aspect is the method
developed to deal with the problem of selecting a route between two locations based on an analysis of
the conditions of the traffic system, when each movement decision has to be made. This lead to the
implementation of a step-by-step search of the available routes for each vehicle.
Two distinct phases can be identified in the approach proposed. First the design of a solution based on
an ANN to solve the single vehicle case, and subsequently the development and testing of a solution
for a multi-vehicle case. To test and implement these phases a specific layout was selected, and an
algorithm was implemented to generate the data required for the design of the ANN solution.
During the development of alternative solutions it was found that the addition of simple rules provided
a useful means to overcome some of the limitations of the ANN solution, and a "hybrid" solution was
originated. Numerous computer simulations were performed to test the solutions developed against
alternatives based on the best published heuristic rules. The results showed that while it was not
possible to generate a globally optimal solution, near optimal solutions could be obtained and the best
hybrid solution was marginally better than the best of the currently available heuristic rules.

Request changes or add full text files to a record

Repository staff actions (login required)

View Item