故障估计的快速梯度算法

加性故障的估计通常是依赖于状态空间模型，如滑动时域估计和未知的输入观测器。本文基于闭环系统的输入-输出观测数据，采用直接设计的方法来估计故障，并设计控制器。该直接方法无需建立系统和故障的状态空间模型，也无需采用子空间辨识来建模。对于故障的估计，利用闭环系统中输入-输出关系建立一个参数化的最小二乘问题。在施加故障的上下限约束条件下，采用快速梯度算法来估计故障。最后以直升机悬停状态为例，利用本文方法估计故障，验证方法的有效性。

Fast gradient algorithm of the fault estimation

Classical methods for estimating additive faults are based on state space models, such as moving estimation and unknown input observers. This paper contributes new direct design methods from closed loop input/output measured datas to estimate faults and design controllers, which do not require building state-space model of the system and fault. We use the input/output relationship of the closed loop system to parameterize least squares problems for estimating faults. When imposing the lower and upper bounds constrain conditions on the faults, we apply the fast gradient method to solve it. Finally by an example of helicopter, the efficiency of the proposed fault estimation can be easily realized.