Research On Acoustic Emission Characteristics Of Plywood Strength

Plywood is a widely used engineering material. Its mechanical strength and failure mechanism are important indicators of safety assessment. However, the traditional mechanical method is slightly inadequate to characteristic the damage progress of plywood. Meanwhile, acoustic emission technique is one of important methods of nondestructive testing and its parameter characteristics can be used as indicators for the damage modes of materials. In this dissertation, the relation of plywood mechanical strength and its failure mechanism is studied. The failure process of plywood in loading is monitored by acoustic emission technique. It is expected that acoustic emission technique is a useful supplement for the traditional mechanical method.Firstly, plywood in loading is tested and monitored by AE instrument. Then different types of damage characteristics of specimen are distinguished by AE parameter analysis. The results reveal that different specimen show different stress-strain relationship during loading. Meanwhile, the activities and characteristics of AE signal are also distinct in the detecting process. Furthermore, the process of material damage can be divided into several stages according to different characteristics of AE parameter.Secondly, seven AE signals de-noising methods based on the shrink of wavelet threshold are compared to find the heuristic Stein's unbiased risk estimate method based on the shrink of wavelet hard-threshold is regarded as the fittest one for in this dissertation. An ideal signal can be acquired after de-noising to meet smoothness and similarity.Thirdly, the AE signals are decomposed in five layers ones by wavelet packet, and frequency spectrum, time-frequency map and wavelet packet energy spectrum of five different damage modes are also compared. The frequency domain distribution of five kinds of AE signals are analyzed by above three methods. Wavelet packet energy spectrum reveals that the energy ratios of each node to all are different from each other. So the energy ratios can be used as the characteristic information of the AE signals and can be used for the input of ANN.Finally, a three layers back-propagation network (BP) is created. Three different training functions are adopted to train the network, and Levenberg-Marquardt algorism is used as study algorism by comparing three different algorism convergence speeds. After trained 81 epochs, the net constructs correct model by 26 sets of training samples, to acquired minimum mean squared error performance. The trained net recognizes 25 from 27 sets successfully. It shows the net has rational structure and can be used in practice.