Rare-Event Chance-Constrained Flight Control Optimization Using Surrogate-Based Subset Simulation

TL;DR

This paper presents a novel probabilistic control optimization method for flight systems that combines subset simulation with surrogate modeling to efficiently estimate rare-event probabilities and optimize control parameters under chance constraints.

Contribution

It introduces a surrogate-based subset simulation approach for accurate and efficient rare-event probability estimation in flight control system optimization.

Findings

Method reduces computational cost significantly.

Achieves accurate estimation of rare-event probabilities.

Demonstrated effectiveness on aircraft longitudinal model.

Abstract

A probabilistic performance-oriented control design optimization approach is introduced for flight systems. Aiming at estimating rare-event probabilities accurately and efficiently, subset simulation is combined with surrogate modeling techniques to improve efficiency. At each level of subset simulation, the samples that are close to the failure domain are employed to construct a surrogate model. The existing surrogate is then refined progressively. In return, seed and sample candidates are screened by the updated surrogate, thus saving a large number of calls to the true model and reducing the computational expense. Afterwards, control parameters are optimized under rare-event chance constraints to directly guarantee system performance. Simulations are conducted on an aircraft longitudinal model subject to parametric uncertainties to demonstrate the efficiency and accuracy of this method.