| Acoustic emission (AE) testing is a rising multidisciplinary nondestructive testing technology. In this dissertation, the pattern of the flaw in the wood-plastic composite (WPC) is confirmed. AE testing is applied to measure the flaw of the WPC. Wavelet transform is applied to denoise the AE signal and extract characteristic parameters from the AE signal. Artificial neural network (ANN) is applied to recognize the pattern of the flaw.Firstly, AE signals produced from WPC in the three-point flexural tests are collected. The pattern of the flaw is identified by metallographic examination of the WPC section. There are five kinds of flaws: fibre breaking, interfacial debonding, substrate cracking, void, interface friction.Secondly, three methods of wavelet denoising are compared, they are: the method based on wavelet transform modulus maximum, the method based on the interrelation of wavelet domain and the method based on the shrink of wavelet packet threshold. The result is the heuristic Stein's unbiased risk estimate method based on the shrink of wavelet packet hard-threshold is the fittest one to denoise the AE signals in this dissertation, and an ideal signal can be acquired after denoised.Thirdly, the frequency domain distribution of five kinds of AE signals are analysed by their time-frequency maps which are plotted based on wavelet packet decomposition. The energy ratios of each node to all are used as the characteristic information of the AE signal and the input of ANN.Finally, a 3 layer back-propagation network (BP) which both updates weight and bias values according to gradient descent with momentum is created. After trained 95437 epochs, the net constructs correct model by 36 sets of input-output training relationship, with 0.001 mean squared error performance. The trained net recognizes 33 from 35 sets successfully. Its accuracy rate is 94.3%. It shows the net has rational structure and can be used in practice.
|
PDF Full Text Request