Penalty Function Approaches for Proactive Fault-Tolerant Model Predictive Control
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2352551Utgivelsesdato
2015Metadata
Vis full innførselSamlinger
Sammendrag
In the event of actuator faults in systems, standard control algorithms might notbe sufficient for stabilizing the system and keeping the performance at an accept-able level. Because of this, fault-tolerant control methods have been an active areaduring the last decade, and several significant contributions to the reliability ofsafety-critical systems have been made.
Model predictive control (MPC) has shown to be a powerful control scheme formulti-variable control problems, and provide a natural framework for integratingreceding-horizon optimization, while also achieving system reconfiguration in theevent of faults. However, almost all the efforts on incorporating fault-tolerance inMPC are focused on reactive fault-tolerance, which aim to handle a fault after ithas occurred. In contrast, proactive fault-tolerant control seeks to utilize an es-timated, conservative time window between the warning of an incipient fault andthe time at which the faulty component is rendered useless to steer the state insidea recoverable region before the fault occurs. As such, a proactive approach circum-vents the issues of possible infeasibilities and destabilization often encountered inreactive approaches, while allowing the system to continue operation during thesubsequent system repair.
Furthermore, economic MPC (EMPC) has received increasing attention in the re-cent years. Rather than separating real-time optimization and control, an economicMPC scheme merges dynamic economic operations with the feedback properties ofconventional MPC. However, there has not been paid much attention to includingfault-tolerance and economic optimization in a unified framework.
This thesis proposes a proactive EMPC algorithm for handling incipient faults,that also takes economic profits in to account. The scheme utilizes an exact penaltyfunction to steer the system inside an invariant set ensuring stability during theloss of actuation in the system. Additionally, the scheme is extended to be robustin terms of handling unknown disturbances to the system, while still achieving thedesired fault-tolerance. Stability for the proposed scheme is proven, both for sys-tems with and without disturbances. The merits and shortcomings of the proposedscheme is demonstrated through several numerical examples