Study Of The Method For Damage Detecting Based On Neural Network

Structure damage because of its potential to cause catastrophic failures are a grave threat to an uninterrupted operation and performance of the modern day machines. Thus, damage detection has become the most important research topics. In the present, many methods for detecting damage have been studied and implemented. The vibration based approach is the most widely used on the basis of which a new method for detecting damage based on neural network is presented.The inner energy distribution always changes when the damage occured. By using the Short Time Fourier Transformation (STFT) of collected signals, the energy distribution of the structure is manifested in the obtained two dimensional time-frequency spectra. From this point, the method for extracting features of signals by combining the STFT and pulse coupled neural network (PCNN) is proposed. The extracted features are then used to train self-organizing map (SOM) neural network, and thus the classification and identification of structure damage is achieved.The basic principles of STFT, PCNN, wavelet threshold denoising and SOM neural network are discussed, and the application of PCNN model for extracting image features is mainly introduced. The example of extracting entropy sequences of three images by means of PCNN is given which indicates the PCNN model has identical property in extracting entropy sequence feature and has certain noise resistivity which can be used in identifying images. Thus, the fact that the usage of PCNN in extracting features of spectra and identifying damage has certain theoretical foundation is convinced.The experiment of cantilever beams have three different damage condition stimulated by stochastic white noise is given. The collected acceleration signals are short time fourier transformed and their features are extracted to train the SOM neural network which is then used to classify and detect the cantilever beam damages. The feasibility and validity is verified by the experimental results.Overall, by means of theoretical analysis and experimental verification, the PCNN model which was used in extracting image features is successfully applied in damage detection for cantilever beams so as to lay the foundation for its further usage in practical application.