عليرضا اطلاعي

In this research, a new approach based on reinforcement learning has been considered to design a fault tolerant control structure. In this approach, an attempt has been made to deal with different types of defects and disturbances by properly defining the state-space of the reinforcement learning agent. These state variables include controlled outputs, residue, tracking error, and fault type. In addition, the design of the controller signal has been done in such a way that this signal is limited in terms of amplitude and variation. In order to enable agent learning in interaction with continuous state space and continuous action space, the soft actor critic algorithm is used, which is one of the new algorithms in the field of deep reinforcement learning. A chemical reactor model has been used to eva‎luate the performance of the proposed structure. The dynamics of this reactor is modeled as a nonlinear four-order differential equation with variable coefficients, and one of its operating points has challenging features, including changing the steady-state gain sign and changing zero dynamic stability characteristics. The results of the implementation of the proposed fault tolerant control structure indicate the desirable performance of this controller in the face of various types of faults and disturbances that are applied to the system suddenly or gradually.

كنترل تحمل‌پذير عيب , يادگيري تقويتي , كنترل فرآيندهاي صنعتي

Fault tolerance control , Reinforcement learning , Industrial processes control

Alireza Ettelaei

Dr. Seyyed Majid Esmaeilzadeh