This paper proposes a real-time fault-tolerant estimation approach for combined sensor fault diagnosis and air data reconstruction. Due to simultaneous influence of winds and latent faults on monitored sensors, it is challenging to address the tradeoff between robustness to wind disturbances and sensitivity to sensor faults. As opposed to conventional fault-tolerant estimators that do not consider any constraints, we propose a constrained fault-tolerant estimator using moving horizon estimation (MHE). By exploiting wind bounds according to the weather or flight conditions, this approach improves fault sensitivity without sacrificing disturbance robustness. This improvement is attributed to active inequality constraints caused by faults, as shown in sensitivity analysis of the formulated MHE problem. The challenge of real-time nonlinear MHE is addressed by adopting an efficient structure-exploiting algorithm within a real-time iteration scheme. In order to facilitate the industrial validation and verification, the algorithm is implemented using an Airbus graphical symbol library to be compliant with the actual flight control computer, and its feasibility of real-time computation has been validated. The simulation results on the RECONFIGURE benchmark, which is a high-fidelity Airbus simulator, over a wide range of the flight envelop show the efficacy of the proposed approach.

In order to improve fault detection (FD) performance, integrated design of residual generation and evaluation is investigated in this paper for trade-offs between fault detection rate and false alarm rate (FAR). A set-membership approach is proposed in residual evaluation by adopting a threshold ellipsoid, which enables more design freedom than a conventional threshold value. With the set-based definitions of fault detection rate and FAR, the integrated design problem is formulated by maximizing the FD performance under a predefined FAR. The joint optimal selection of a residual generator and a threshold ellipsoid is equivalently transformed into a simplified optimization problem of determining an optimal threshold ellipsoid for any given residual generator. A suboptimal solution for the set-membership-based integrated FD system design is obtained based on approximated computation of the FD performance. Monte Carlo simulations show the performance improvement of the proposed method compared with an existing integrated design method.

This paper presents a real-time nonlinear moving horizon observer (MHO) with pre-estimation and its application to aircraft sensor fault detection and estimation. An MHO determines the state estimates by minimizing the output estimation errors online, considering a finite sequence of current and past measured data and the available system model. To achieve the real-time implementability of such an online optimization-based observer, 2 particular strategies are adopted. First, a pre-estimating observer is embedded to compensate for model uncertainties so that the calculation of disturbance estimates in a standard MHO can be avoided without losing much estimation performance. This strategy significantly reduces the online computational complexity. Second, a real-time iteration scheme is proposed by performing only 1 iteration of sequential quadratic programming with local Gauss-Newton approximation to the nonlinear optimization problem. Since existing stability analyses of real-time moving horizon observers cannot address the incorporation of the pre-estimating observer, a new stability analysis is performed in the presence of bounded disturbances and noises. Using a nonlinear passenger aircraft benchmark simulator, the simulation results show that the proposed approach achieves a good compromise between estimation performance and computational complexity compared with the extended Kalman filtering and 2 other moving horizon observers.

An adaptive decentralized strategy for active queue management of TCP flows over communication networks is presented. The proposed strategy solves locally, at each link, an optimal control problem, minimizing a cost composed of residual capacity and buffer queue size. The solution of the optimal control problem exploits an adaptive optimization algorithm aiming at adaptively minimizing a suitable approximation of the Hamilton-Jacobi-Bellman equation associated with the optimal control problem. Simulations results, obtained by using a fluid flow based model of the communication network and a common network topology, show improvement with respect to the Random Early Detection strategy. Besides, it is shown that the performance of the proposed decentralized solution is comparable with the performance obtained with a centralized strategy, which solves the optimal control problem via a central unit that maintains the flow states of the entire network.