Chenglin Wen

An output delay approach to fault estimation for sampled-data systems

The problem of fault estimation for a class of non-uniformly sampled-data systems is investigated from the time delay point of view in this paper. Firstly, the output delay approach is employed to model the sampled-data system as a continuous-time one with time-varying delay output. Then, based on the analysis of the inapplicability of the adaptive fault diagnosis observer in such class of time-delay systems, a novel augmented fault estimation observer design method is proposed to guarantee the exponential convergence of the estimation errors. Furthermore, an extension to the case of time varying fault estimation for the noisy sampled-data systems is studied. Finally, simulation results of a flight control system are presented to demonstrate the effectiveness of the proposed method.

A Hybrid System Approach to Robust Fault Detection for a Class of Sampled-data Systems

Abstract In this paper, a novel direct design methodology of robust fault detection for a class of sampled-data systems with both continuous-time process noise and discrete-time measurement noise is presented. First, by using a linear system with finite discrete jumps as residual generator, the design of robust fault detection filter is formulated as a sampled-data filtering problem. Then, a bounded real lemma for the linear system with finite discrete jumps is developed in terms of linear matrix inequalities (LMIs). Based on this, a sufficient condition for the existence of the fault detection filter as well as the design parameters is derived. Furthermore, the case of sampled-data systems with model uncertainties is extended. The designed fault detection filter cannot only make the error between residual and weighted fault as small as possible but also exhibit robustness to all uncertainties including continuous-time process noise, discrete-time measurement noise, and model uncertainties. Finally, simulation results are provided to demonstrate the feasibility of the proposed method.

Multi-measurement Target Tracking by Using Random Sampling Approach

Abstract In multi-target tracking field, conventional algorithms supposed that target is a point source and produces at most one measurement. While with the development of modern sensor technology, a target may give multiple measurements. In this paper, we consider that targets have certain geometrical shapes and give multiple measurements and call these targets multi-measurement targets (MmTs). We first build rigid models for the targets in parameter space and then estimate their parameters using the Markov chain sampling approach. Next, we derive the moving state described by targets centroid with our proposed equivalent measurement. When the number of targets remains unknown, under the Poisson assumption, we use the ratios of Poisson intensities to estimate the number of targets. We also define the probabilistic vectors of type (PVoT) and propose a recursive process for the PVoT. To verify the proposed algorithm, the final experiment proposes three targets, with different shapes and distributions, moving in a 2-dimension plane with constant velocity (CV). The experimental results show that the estimation of target state has an excellent precision and the shape estimation can better and stably reflect the change of target shape. Besides, the target lost rate is around 1.4% in 500 Monte Carlo (MC) runs.

Optimal diagnostic observer for sampled-data systems

The direct design method of optimal diagnostic observer for sampled-data systems is investigated in this paper. Firstly, an appropriate diagnostic observer for sampled-data systems is constructed from the transfer function viewpoint and the dynamics of the discrete time residual with respect to continuous time unknown inputs and faults is derived. Then, the sampled-data fault detection problem is formulated as a class of ratio-type optimization one. By using the co-inner-outer factorization technique and algebra method, a general optimal solution is obtained to make the discrete-time residual robust to the continuous-time unknown inputs and sensitive to the faults. Finally, simulation results of a numerical example are provided to illustrate the efficiency of the proposed method.

A new fast rate fault detection approach for multirate sampled-data systems

An observer-based fault detection direct design method for multirate sampled-data systems (MSD) is investigated in this paper. Firstly, by use of the discrete lifting technique, the MSD fault detection problem is transformed into an equivalent slow rate pure discrete time problem based on the Hinfin optimal direct design method of single rate sampled-data systems. An Hinfin performance index is then optimized to make the obtained residuals sensitive to the faults and robust to the unknown inputs. An extend QR decomposition algorithm is put forward to ensure the residual generator satisfying the causality constraints. Finally the residual is inverse lifted to implement fast rate fault detection. Simulation results of a numerical example are provided to illustrate the efficiency of the proposed approach.

Decentralized Quantized Kalman Filter with Limited Bandwidth

Consider the decentralized estimation problem of dynamic stochastic process in a sensor network. Due to bandwidth constraints, only quantized messages of the original information from local sensor are available. For a class of vector state-vector observation model, an adaptive quantization strategy and sequential filter technique are introduced to design fusion algorithms in this paper. According to different forms of original information, two suboptimal Kalman filters are presented based on quantized measurements (KFQM) and quantized innovations (KFQI) respectively. The main advantages of these proposed filters include two aspects, the first is to adapt the general vector system, and another is that the data quantization and transmission strategies are both adaptive. In contrast, the latter has better estimation accuracy under the same bandwidth constraints because of the less information loss while quantizing innovations. Computer simulations show the effectiveness of two methods.

Kalman filter based on adaptive quantized information

When dealing with decentralized estimation problem of dynamic stochastic process in a sensor network, it is important to reduce the cost of communicating the local information due to bandwidth constraints. Thus, only quantized messages of the original information from local sensor are available. For a class of vector state-vector observation model, an adaptive quantization strategy and sequential filter technique are introduced to design fusion algorithms in this paper. According to different forms of original information, two suboptimal Kalman filters are presented based on quantized measurements (KFQM) and quantized innovations (KFQI) respectively. In contrast, the latter has better estimation accuracy under the same bandwidth constraints because of the less information loss while quantizing innovations. Computer simulations show the effectiveness of both methods.