Analysis of Wavelet Neural Network in Fault Diagnosis in Analog Circuits

Paper Keywords: analog circuits, fault diagnosis, wavelet neural network
Abstract: This paper analyzes the importance of fault diagnosis of analog circuits and the current difficulties, the small transition theory based on neural network theory and analog circuit fault diagnosis methods are reviewed. That the wavelet neural network to analog fault diagnosis problems and future prospects.


Analog circuit fault diagnosis, in theory, can be summarized as follows: the known network topology, the input excitation and fault response or a known pArt of the device parameters may be the case, find the parameters and location of faulty components.

Although ģ�� circuit fault diagnosis theory and method have achieved a lot, made a lot of fault diagnosis method Ru fault dictionary France, fault parameter identification method, fault Yanzheng law. But the analog circuits and diagnosis �� own difficulties, Progress has been slow. The main difficulties are: Analog circuit fault model is more complex and difficult to make a simple quantitative; analog circuits with tolerance in the component parameters, increasing the difficulty of diagnosis; in the simulation of nonlinear circuits there were extensive problem for fault diagnosis of more difficult; in a practical analog circuits, almost without exception, there is a feedback loop, the simulation requires a lot of complex calculations; actual analog circuit voltage can be measured in a very limited number of nodes . lead can be used for fault diagnosis of inadequate information, resulting in uncertainty of fault location and fuzzy.

Therefore, in the past for analog circuit fault diagnosis of the main stay in the small and medium-scale non-linear tolerance tolerance or small the situation, some methods have been successfully applied in engineering practice. But how to effectively address the analog circuits and non-tolerance linear problem, how to solve the fault diagnosis of ambiguity and uncertainty is an urgent need to address the problem of the future. wavelet neural network is beneficial because of its simulation of human problem solving process, easy to take into account the human experience and has some learning ability characteristics, so in this area has been widely used.

A wavelet analysis theory in analog circuit fault diagnosis Analysis
Simply put, wavelet is a beginning and an end of the small "wave" signal from the wavelet analysis, from the function of the dilation and translation, is the Fourier analysis, Gabor analysis and short-time Fourier analysis of the development of the direct results. Wavelet analysis-based wood principle is the wavelet mother function, the expansion in scale and time-domain analysis of signals on the pan to properly select the generating function. can make the expansion function has a better locality, wavelet analysis is effective in the low frequency signal of the layer by layer decomposition, the wavelet packet analysis is an improvement on the wavelet analysis, it provides a more precise signal analysis method, the signal layer in the whole frequency band for the effective decomposition of the characteristics more conducive to extract the signal. Therefore, It is a time-frequency analysis. in good time and frequency domain localization performance and features with multi-resolution analysis is very suitable for non-stationary signal analysis of the singularity. such as: use of continuous wavelet transform can detect signal singularity distinguish between signal and noise mutations, the use of discrete wavelet transform can detect the structure of random mutation frequency.

Transform fault diagnosis mechanisms include: the use of signal singularity observer for fault diagnosis and signal frequencies using the observed changes in the structure for fault diagnosis. Wavelet transform has no mathematical model of the system, sensitive and accurate fault detection, computational load of noise reduction capability and requirements of the advantages of low input signal. However, on large scales due to the filter in time domain width larger test will produce delay time and the selection of different wavelets also have an impact on the diagnosis. In simulation circuit fault diagnosis, wavelet transform has been effectively used to extract fault features that wavelet pre-processor, the fault information and then give it away these fault classification processor for fault diagnosis. the Application of wavelet analysis are generally limited to small-scale range, mainly because of large-scale Application of wavelet construction and the cost of larger storage needs.

2 Neural networks in analog circuit fault diagnosis of
Artificial neural network (ANN) is the basis of modern scientific Research on the neural put forward, is an abstract mathematical model is the simulation of human brain function. After ten years of development, Artificial neural network has been formed a few ten kinds of networks, including multilayer perceptron Kohomen self-organizing maps, Hopfield networks, Adaptive Resonance Theory, Art network, RBF network, probabilistic neural networks. These networks because of different structures, different scope of Application. As Artificial neural network itself is not only non-linear, adaptive, parallel, fault tolerance, etc., and to distinguish cause of the malfunction, failure type skills, and trained neural network can store the knowledge of the process, directly from quantitative history fault information to learn. so in the 20th century, late 80s, it has been applied to fault diagnosis of analog circuits. With the growing maturity of artificial neural networks and the large number of Applications, the neural network is widely used in analog circuit fault diagnosis has been a trend . BY neural network pattern classification due to good capacity, especially for analog circuit fault diagnosis and therefore in the analog circuit fault diagnosis system has wide Application, is currently used analog circuit fault diagnosis have more and more effective a neural network. reposted elsewhere in the Research Papers Download http://www.hi138.com 3 Application of wavelet neural network analysis
3,1 wavelet analysis and neural network theory with the need for
In neural network theory is applied to analog circuit fault diagnosis process, the neural network hidden layer neurons for the determination of the number of nodes, the initialization of various parameters and the structure of neural network structure such as a lack of theoretical guidance for more effective methods of which will directly affect the practical application of neural network results. wavelet analysis in time domain and frequency domain also has good localization properties, while the neural network is a self-learning, parallel processing, adaptive, fault tolerance and generalization ability of two so wavelet analysis and neural networks to combine the advantages of both applied to the fault diagnosis is the need for objective reality.

Wavelet analysis and neural networks present a combination of two forms, one is to use wavelet transform to preprocess the signal to extract the signal feature vectors as the neural network input, the other is used to form the wavelet function and scaling function neurons, to wavelet analysis and neural networks combined with the direct integration of the first approach is loosely based on wavelet neural network combined with the second approach is based on wavelet neural network combined with the compact.

3.2 Wavelet analysis and neural network theory of the combined form
Wavelet and Neural Networks combination of loose, namely: using wavelet analysis or wavelet packet analysis as a means of pre-processing neural network for neural network input features to the fish, specifically the use of wavelet analysis or wavelet packet analysis, the signal decomposed into separate frequency bands within the energy of the band the value of the formation of a child to feel that different to the child corresponding to different values of failure, which can be used as neural network input feature vector Once the neural network input features to children, used according to experiences determine which neural network and hidden layers and the number of hidden units, etc. to the test sample can be used to train the neural network to adjust weights in order to establish the necessary neural network model.

Wavelet and Neural Networks combination of compact type, namely: the wavelet function and scaling function with the formation of neurons and neural networks to wavelet analysis of direct integration, called the narrow sense of the wavelet neural network, which is often said that the wavelet neural network. It is based on the wavelet function or the scaling function as the activation function, its mechanism and the use of MLP Sigmoid function is basically the same. fault diagnosis is the essence of failure to achieve symptom space to space mapping, this mapping function approximation can also be used to represent. wavelet neural network formation can be from the perspective of function approximation description. common neural networks: the use of scaling functions as a neural network activation function of neurons in the orthogonal basis of wavelet network, adaptive wavelet neural network, multi- resolution wavelet network, the Interval Wavelet networks.

3.3 The wavelet analysis combined with the advantages of neural network theory
Wavelet neural network has the following advantages: First, avoid the M LY and other neural network design purpose of education; second approach is a strong, fast learning convergence, there are theoretical guidance for the selection of parameters, effectively avoid local minimum problems advantages.

In the field of analog circuit fault diagnosis, wavelet neural network is a new, promising application of Research. With the wavelet analysis theory and the continuous development of neural network theory, wavelet neural network used in the field of analog circuit fault diagnosis will become increasingly mature.

4 Conclusion
Wavelet analysis and neural network theory in the field of analog circuit fault diagnosis has broad application prospects. Wavelet neural application of the theory will further promote the analog circuit fault diagnosis theory and method of development, to make it more complete and more extensive applicability for complex fault diagnosis for large analog circuit to provide a more efficient, more practical approach is to fault diagnosis of analog circuits in the future direction of development. Links http://www.hi138.com Research Papers Download