论文部分内容阅读
为通过声发射技术识别铝合金蜂窝板超高速撞击(HVI)的损伤状态,提出一种基于神经网络的损伤模式识别方法。通过超高速撞击实验获取声发射信号,结合精确源定位技术、时频分析技术、小波分析技术及模态声发射技术,提出了10个与损伤相关的特征参数,通过非参数检验分析其与损伤的关系,设计了一种基于贝叶斯正则化BP神经网络的超高速撞击损伤模式识别方法。建立最优网络模型,通过不同参数组合识别能力分析,优选出2种特征参数组合,通过非同源样本对其损伤模式识别能力进行验证。结果表明:传播距离与损伤模式无关,却是识别损伤模式的重要参数;125~250kHz频域的自动加窗小波能量比会降低损伤模式的识别能力;采用贝叶斯正则化的BP神经网络可以较好地识别蜂窝板超高速撞击损伤模式,参数组合为传播距离、上升时间、持续时间、截止频率、4个自动加窗小波能量比及小波能量熵,共9个参数,对任意选取非同源样本识别错分率仅为9.38%。
In order to identify the damage state of hypersonic impact (HVI) of aluminum alloy honeycomb panels by acoustic emission technique, a damage pattern recognition method based on neural network is proposed. The acoustic emission signals were obtained by hypervelocity impact experiments. Ten precise damage-related parameters were proposed by using precise source localization, time-frequency analysis, wavelet analysis and modal acoustic emission techniques. , A hyper-speed impact damage pattern recognition method based on Bayesian regularized BP neural network was designed. The optimal network model is established. Through the analysis of the recognition ability of different parameters, two kinds of characteristic parameters are optimized, and their damage pattern recognition ability is verified by non-homologous samples. The results show that the propagation distance is not related to the damage mode, but it is an important parameter to identify the damage mode. The auto-windowed wavelet energy ratio in the 125 ~ 250kHz frequency domain will reduce the recognition ability of the damage mode. The Bayesian regularization BP neural network The damage modes of hypervelocity impact of honeycomb panels are well identified. The parameters are composed of nine parameters, including propagation distance, rise time, duration, cutoff frequency, four auto-windowed wavelet energy ratios and wavelet energy entropy. Source sample identification misclassification rate was only 9.38%.