Study On Fault Diagnosis For Analog Circuit Based On Wavelet And Neural Network

In this dissertation, we focus on wavelet and neural network theory and their applications to fault diagnosis for analog circuit. What's more, Hilbert-Huang transform and its application to speech enhancement is studied. The main research work and results are as follows:1. There are many noises in analog circuit, which involves certain difficulties for fault diagnosis. Under the circumstances, denoising algorithms are studied in this paper and a novel thresholding function is presented based on the wavelet shrinkage put forward by D.L.Donoho and I.M.Johnstone. This new thresholding function has many advantages versus DJ's soft- and hard-thresholding function. It is simple in expression and continuous as the soft-thresholding function, and has high order derivative which make it convenient for some kinds of mathematics disposals. It also overcomes the shortcoming that there is a invariable dispersion between the estimator wavelet coefficients and the decomposed wavelet coefficients in soft-thresholding. At the same time, the new thresholding function is more elastic than the soft- and hard-thresholding function. Simulation experiments indicate that the de-noising method adopting the new thresholding function is effective and gives better performance.2. A new method of energy-fault diagnosis based on the wavelet packet transform is presented, which overcomes the disadvantages that the traditional diagnosis approaches need topology structure of analog circuit. Simulation results show that this new method is available for fault diagnosis for analog circuit.3. Two improved BP neural network algorithms of fault diagnosis for analog circuit are presented through using optimal wavelet packet transform (OWPT) or incomplete wavelet packet transform (IWPT) as preprocessor. At first, the response signal of an analog circuit is preprocessed by OWPT or IWPT, and the normalization energy of each frequency band is worked out. The normalization energy is then used to train a BP neural network to diagnose faulty components in the analog circuit. These two algorithms need small network size, while have faster learning and convergence speed. Finally, simulation results illustrate the two algorithms are effective for fault diagnosis.4. Binding wavelet packet, neural network and fuzzy theory together, we present a new neural network algorithm of fault diagnosis for analog circuit base on wavelet packet and fuzzy rules. Simulation experiments were given and show that this algorithm is effective, needs small neural network size, and has faster learning and...