| Composite materials consist of two or more components , it both has the advantages of each material and overcome the defects of single material, so it expands the application scope of materials。The damage forms of composite material are complicated, and its corresponding acoustic emission signal is transient non-stationary signal. The traditional analysis method of acoustic emission signal has certain limitations in the analysis for this kind of signal, so how to adopt effective and reasonable signal analysis methods, which from a lot of AE signals to identify characterization of different types of damage signal, become the key technique for researching FRP composite material acoustic emission signal recognition of damage.This paper use wavelet analysis relevant theoretical knowledge in the signal de-noising and feature extraction, and adopting threshold de-noising method to get rid of noise signal which combining with actual acoustic emission signal characteristics. To decompose four layers wavelet packet for processed signals, and extraction energy of each node as neural network's input. The structure is "tight" type wavelet neural network . Morlet wavelet function as hidden excitation function. E-learning training process based on the reverse transmission of error, and according to the gradient descent to adjust the network parameters, meanwhile, the introduction of genetic algorithm to optimize network initial parameters optimization so as to avoid network into local optimal solution, then it can improve network convergence and stable performance. The constructed wavelet neural network is applied to the analysis of experimental data, which results show that the network not only satisfy higher error precision requirement but also has good recognition and generalization ability. In order to achieve the pattern recognition of FRP composite material damage procedural, this paper finished interface design by computer language.
|
PDF Full Text Request