الفهرس 
Chapter 1Only 14 pages are availabe for public view
 |  | from | 115 |  |  |
| | | from | 115 | | |
Abstract
In the interest of improving the fault-tolerant control systems to better operate in the presence of faults, this thesis presents various methods of fault-tolerant control for discrete event systems subjected to single or multiple sensor faults.
The first methodology is active fault-tolerant control based on a timed automata model. The reconfiguration model block compares the observed signal and the time model system. If the time model system generates a new event and the discrete event systems are not able to reach this event in time. Hence a fault is detected and sent to the reconfiguration control law. Which is used to replace the normal behavior controller of faults with a timed behavior controller based on the time diagnosis results. The timed control law is then transmitted to the controller to adjust the faulty plant.
The second methodology is data-driven fault-tolerant control based once on a single recurrent neural network and again on a single Long Short-Term Memory that replaces diagnosis and control reconfiguration units. The diagnostic and process model are not required, which is a significant benefit of this technique over traditional model-based methods.
The third and final methodology is active fault-tolerant control based on long short-term memory networks. There are two long short-term memory networks are utilized; the first performs as a diagnosis for automated manufacturing systems and can identify faulty sensors, while the second acts as an inverse model of these systems and is used to determine the reconfigured control actions to take when sensors are not functioning as expected.
The Factory I/O simulator is interfaced with MATLAB based on the digital twin concept to simulate and verify the proposed approaches for automated material handling case study in faultless and multiple faulty sensors. The results of the simulation demonstrate the effectiveness of the proposed approaches. Furthermore, the results demonstrate the following characteristics of the proposed approaches:
a) LSTM is a useful network for modeling automated manufacturing systems and designing its controller.
b) The presented approach performs its proper control function in faultless and different faulty scenarios.
c) Four tolerate sensors out of six.
d) The presented fault diagnosis subunit depends on software redundancy, instead of hardware redundancy.
e) The fault diagnosis can detect and identify diagnosable and undiagnosable faulty sensors.