Xu Lijia

Studying on state prediction for radar transmitter using GM(1,1)

Through analyzing the importance of component modules of a kind of radar transmitter, both the cathode current of multiple-beam klystron and the ripple voltage of high-voltage power supply are selected as the prediction signals. By studying on their characteristics, a improved GM (1,1) is proposed to predict the state of transmitter, where the idea of metabolism is applied to make the parameters of GM (1,1) on-line change. Particle Swarm Optimization(PSO) is used to select the best prediction dimension. The performance of the proposed method is tested, and the experiment results show that the proposed method has better prediction capability than the popularly used ARMA model, and then it provides technical support for preventive maintenance.

Study on Fault Diagnosis for radar's transmitter based on Bayesian network

Fault Diagnosis for radars transmitter is very crucial, yet very difficult. Transmitter is complex electrical equipment whose units are full of uncertain factors and information. The paper puts forward a new fault diagnosis method based on Bayesian network for the transmitter. The cause-result graph of transmitter is gained by analyzing its structure firstly, and then the fault model based on Bayesian network is built. The next step is to study its parameters by using certain algorithm. Finally the fault model is tested, and the experiment results show that the proposed method is very effective for diagnose the faults of transmitter.

A novel method for monitoring state and evaluating health of analog circuit

Aiming at monitoring health degradation of analog circuit, the paper proposes a new method about state monitoring and health evaluation. Firstly original voltage features are extracted from analog circuit with some components changing gradually; Then LDA is used to reduce the dimension of the original features to obtain effective features; In the following with the effective features of normal state, a DHMM trained by an improved training algorithm is applied to calculate KL distance to monitor analog circuits state and evaluate its health. Applying this new method to an analog circuit, the experiment results show that the proposed method can successfully convert the unconspicuous change of early fault process into the obvious change of KL distance and owns excellent capability of state monitoring.

Image fusion method based on non-sampled lifting wavelet transforms

A new algorithm for multi-sensor image fusion is presented. Based on lifting wavelet transform, the non-sampled lifting wavelet transform with the trait of shift invariance can be obtained by equivalently shifting to the pumping values and interpolation of the lifting wavelet transform. The adaptive weighted average method based on neighborhood gradient is applied to fuse the low-frequency coefficients and the weighted method based on edge information is chosen in the course of fusing the high-frequency coefficients. Then the fusion image can be obtained through inverse non-sampled lifting wavelet transform. Applying the proposed method to multi-sensor image fusion, the experimental results show that the novel algorithm is better than previous ones, both in visual effect and objective evaluation criteria.