Reinforcement learning-based optimal fault-tolerant control for offshore platforms

Abstract

This article investigates a novel reinforcement learning-based fault-tolerant control approach for steel-jacket offshore platforms. In the first step, the dynamic of the steal-jacket offshore platform with an active mass damper is considered, and the equivalent linear time-invariant model is obtained with the actuator fault. In fault-free conditions, an optimal controller is designed to keep the system stable under external wave force. Subsequently, in faulty conditions, the actuator fault is estimated by the fault observer. Next, by inserting the actuator fault estimation into the cost function, the fault-tolerant control problem transforms into the optimal control problem. The online policy iteration is used to minimize the new cost function. Finally, the final control law, which is a mixture of the nominal and the modified control law, stabilizes the offshore platform and improves its performance in the presence of the actuator fault without needing the complete knowledge of the offshore platform. The simulation results show the effectiveness of the proposed method.