Discrete event system tools for fault diagnosis and collision prevention
- Autores: Xu Wang
- Directores de la Tesis: Cristian Mahulea (dir. tes.), Manuel Silva Suárez (dir. tes.)
- Lectura: En la Universidad de Zaragoza ( España ) en 2017
- Idioma: español
- Tribunal Calificador de la Tesis: Francisco Javier Campos Laclaustra (presid.), Miquel Àngel Piera i Eroles (secret.), Francesco Basile (voc.)
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Discrete Event Systems (DES) are widely used tools in practical systems including manufacturing systems, robot motion, logistics systems and so on. Petri Nets (PN) are proposed to model and analyze DES in a compact and efficient way. In this thesis, we focus on two topics in DES and deal with them using PN.
One problem is fault diagnosis. We propose an on-line approach for fault diagnosis of timed discrete event systems modeled by Time Petri Net (TPN). The set of transitions is partitioned into two subsets containing observable and unobservable transitions, respectively. Faults correspond to a subset of unobservable transitions. In accordance with most of the literature on discrete event systems, we define three diagnosis states, namely normal, faulty and uncertain states, respectively. The proposed approach uses fault diagnosis graph, which is incrementally computed using the state class graph of the unobservable TPN. After each observation, if the part of FDG necessary to compute the diagnosis states is not available, the state class graph of the unobservable TPN is computed starting from the consistent states. This graph is then optimized and added to the partial FDG keeping only the necessary information for computation of the diagnosis states.We provide algorithms to compute the FDG and the diagnosis states. The method is implemented as a software package and simulation results are included.
The other problem is collision avoidance in robot planning and we deal with two problems: deadlock prevention with and without real time control. In the case of deadlock prevention with real time control, we consider the problem of design a deadlock prevention control policy for a team of mobile robots that should follow some trajectories in order to accomplish a given task. We assume that some regions have limited capacity (i.e., the numbers of robots that can be simultaneously in that regions are limited) that can be seen as limited available resources in a resource allocation system (RAS). We propose a method, based on inhibitor arcs that can be applied in a decentralized way. This is an alternative to the deadlock prevention strategy based on monitor places that could be used in several applications since the implementation cost could be smaller. In the second case, we address a collision avoidance problem in a multi-robot system, where real time control is not applicable. Each robot has a set of possible trajectories, each trajectory fulfilling its individual task. The trajectories consist of sequences of regions to be followed, and the time for moving inside each region is known. The problem is to avoid inter-robot collisions by imposing initial time delays for each trajectory. Two solutions are developed, depending on the possibility of imposing a certain trajectory from the available set of paths for each robot. The solutions have the form of mixed integer linear programming optimizations that return the initial time delays and, when necessary, the chosen trajectory for each robot. Finally, we perform a statistical study on the proposed solutions and we conclude that one formulation is preferable to the others.
