Browsing by Author "Fourie, Bradley"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Development of a dynamic path planning system for autonomous mobile robots in a flexible manufacturing system
    (Stellenbosch : Stellenbosch University, 2023-02) Fourie, Bradley; Louw, Louis; Bitsch, Günter; Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering.
    ENGLISH ABSTRACT: Recent developments in Industry 4.0 have shifted consumer demand, resulting in a need for manufacturers to supply small batches of highly customised products. To enable profitable highly-customised production, Autonomous Mobile Robots (AMRs) have become the most important technology associated with flexible material handling. However, path planning for AMRs in dynamic environments is an unsolved problem and remains to be the largest barrier to practical implementation. Currently, most AMR implementations in practice require manual reprogramming of waypoints. However, for Flexible Manufacturing Systems (FMSs), manual reprogramming is not feasible due to the flexible nature of the layouts and the large variety of disturbances that can occur at both the production and consumer levels. As a result, FMS environments require systems that can dynamically adapt AMR paths to prevent unplanned downtime, extra expenses, and manual labour required for manual waypoint reprogramming. In this thesis, a path planning system for AMRs that is suitable for dynamic manufacturing environments was developed. A design science research methodology was used to develop the path planning system, where both aspects of intelligent optimisation methods and Multi-Agent Systems (MASs) were investigated and developed for the final design. The dynamic path planning system utilised a MAS design in software, where the multirobot conflict avoidance mechanism was implemented using the Iterative Exclusion Principle (IEP). Moreover, several Genetic Algorithm (GA) and Reinforcement Learning (RL) methods were developed for the intelligent path optimisation algorithm of the path planning system, integrating aspects from the informed heuristic search literature. The RL algorithms used a curriculum learning process, where training was completed on mediumlevel hardware to improve algorithm convergence. However, the generalisability required for FMSs could not be achieved when restricting the training time to what is allowable between shift changes. The GA had superior performance after evaluation on three separate environments and a total of 150 different transport order configurations and was, therefore, selected for the final design. The dynamic path planning system was further developed into a technology demonstrator for evaluation in an accurate simulation model of theWerk150 logistics learning factory at the ESB Business School for validation. Several disturbance scenarios prevalent in the Werk150 facility were identified to validate the design, and the associated experiments were created to investigate various flexibility parameters. The technology demonstrator of the dynamic path planning system collaboratively planned conflict-free paths in all disturbance scenarios and enabled the material handling flexibility required for the Werk150 facility.