Hongcai Zhang

A new clustering and training method for radial basis function networks

In view of the drawbacks of traditional learning algorithms for the radial basis function networks (RBFN), including improper selection of RBF centers, oversize problem of the network in the clustering stage and ill-condition in the training stage, a new clustering and training algorithm is proposed based on constructing an augmented vector consisting of both input and output, combined with the singular value decomposition (SVD) for selecting the significant basis function centers and for training the RBFN using an SVD-based recursive least squares (RLS) method so as to avoid the ill-conditioned problem. The new algorithm is superior to the RLS in convergence rate and mean square errors of training. The effectiveness and superiority of the proposed method are demonstrated via simulation examples.

Fault Detection and Identification of Dynamic Systems using Multiple Feedforward Neural Networks

Abstract Combining feedforward neural network (FNN) and multiple model adaptive estimator (MMAE), a new approach for fault detection and identification (FDI) of nonlinear systems as well as linear systems is proposed in this paper. Instead of Kalman filters, a bank of FNNs is used in the MMAE which are trained for the normal operation and possible fault situations. In order to overcome the drawbacks of the traditional BP training algorithm for FNN, singular value decomposition is used for the selection of hidden neurons, and then a new fast learning algorithm for training FNN by using a variable time-varying forgetting factor technique and U-D factorization baaed extended Kalman filter (EKF) is proposed. The new approach is then used for FDI of nonlinear systems as well as linear systems. The effectiveness of the method proposed is demonstrated by two simulation examples.

Essence of 2DPCA and Modification Method for Face Recognition

In this paper, the method of 2DPCA is analyzed and its nature is revealed, i.e., 2DPCA is equivalent to view rows of face images as training samples that constitute row training sets and then use PCA for feature extraction. We also have proved that principal component vectors extracted by 2DPCA contain redundancy in theory. Based on this result, this paper presents a new image feature extraction method. The proposed method provides a sequentially optimal image compression mechanism. Finally, the effectiveness of the proposed algorithm is verified using the ORL database

Model Predictive Control Research Based on Auto Landing System

Auto landing is a very complex stage of flight. Although percentage of auto landing to total flight is only 2~3%, about one third of flight accidents appear in this phase. Moreover half of flight accident of high speed jet-planes occur at this stage. In this paper, we discuss a kind of unmanned aerial vehicle which has a special figure and rigorous lateral requirements. Therefore this paper uses hierarchic control system which is made up of predictive controller plus traditional PID controller. The simulation results show MPC method can improve the landing precision efficiently; meanwhile control the roll angle rigorously.

On Robust Fault-Tolerant Control of Missile Control System

Abstract In order to avoid ballistic missile self-destruction because of sensor failures caused by element aging of long time storage, a reconstructed feedback robust fault-tolerant control scheme is presented. In our scheme, a fault detection method based on the adjacent coefficient of Mallat wavelet is proposed firstly so that the fault information, which is masked by several elastic oscillation signals with time-varying magnitudes and frequencies, can be extracted efficiently. As soon as a fault is detected, the switch control unit immediately isolates the failed sensor and switch from the original control model to the reconstructed one. To remain the stability of the system and control the tracking error in case of time-varying of missile parameters, a frequency domain robust stability control is designed, so that speed test feedback of engine swing angle and gain feedback of normal sensor are used to ensure the control stability and restrict the tracking error within a designed value. Finally, the effectiveness of our method is shown by an example of a Soviet missile flight simulation.

Signal Processing for Output Signals of the Inertia Sensors in the Missile Control System

To improve the reliability of missiles, the reconfiguring fault tolerant control system (RFTCS) should be built in the missile control system. The key problem for building RFTCS is to separate the rigid body angle signal and different order elastic oscillation caused angle signals (RSADS) from the output signals of inertia sensors (OSOIS). Firstly, the accurate location characteristics of FFT in frequency domain are used. OSOIS are transformed by FFT. The frequencies at extrema of frequency response obtained via FFT are the frequencies of RSADS. Secondly, an elliptical filter is designed to let the signals of the frequencies go through and the test signal is designed with combined sine signals whose initial phases are zero and whose frequencies are the frequencies at extrema. The test signal is filtered by the elliptical filter. The phase lag of each sine signal in the test signal is the initial phase of each filtered sine signal. OSOIS are also filtered by the elliptical filter. The filtered signal separated from inertia sensors are adjusted with the phase lags test by the test signal. Finally, RSADS are obtained in real-time. The simulation shows the effectiveness of the approach.

Performance analysis of interacting multiple model algorithm

Abstract A new non-simulation method for performance analysis of Interacting Multiple Model Algorithm is proposed. Firstly, how input-interaction effects model-conditional estimation is analyzed. Four conclusions are made qualitatively. Besides this, the compression ratio of model-conditional error is defined. So the parameters and modeling can be chosen quantitatively. Secondly, how input-interaction effects model probability is analyzed. Input-interaction is found not only to decide the upper and lower limits of model probability but also to lessen the difference among model probabilities, then in the sense of model probability, decaying-memory filtering, damping coefficient and regulating-time are defined. All these works may be useful to choose optimal parameters and design new adaptive filters.

Detection and identification of bias faults in nonlinear system

Although fault detection and identification (FDI) methods for linear systems have been developed extensively, FDI for nonlinear systems still deserves much attention. In order to detect bias type faults, a bias /spl chi//sup 2/ FDI method is proposed here on the basis of the pseudo separated-bias estimation (PSBE) algorithm. Estimates of biases obtained by PSBE are used to construct a statistical variable which obeys /spl chi//sup 2/ distribution in normal operational conditions. As a result, by testing if the constructed variable is /spl chi//sup 2/ distributed at every estimation step, one can detect input-output bias faults quickly. In order to identify where a bias fault occurs, a bias component /spl chi//sup 2/ detection scheme is proposed further. Simulation results of a paper machine illustrate the effectiveness of the method for real-time application.