材料期刊网

以复合材料为对象，以宽频带传感器及线阵列方式对各类模式试样采集了波形及信号参数，比较波形、信号参数、频谱及小波谱的特征，筛选出六类1300个样本，采用多分辨小波变换提取了5个特征向量，实现了特征空间的降维处理，采用B—P型反向传播神经网络构成了智能化模式分类器，研究了网络模型的学习效果和对与复合材料主要损伤机制有关的六类声发射信号的识别能力。试验结果表明，神经网络对六类信号的平均正确识别率达到90.4%。最佳识别率为97.2%。该方法成功用于90°、0°光滑和0°缺口三种试样的破坏过程分析，获得了满意的效果。

The comprehensive messages of six types of pattern signals such as broadband waveforms,AE parameters,analog parameters were collected with linear location simultaneously.Features of signals were analyzed for wave form,AE parameters,FFT spectrum and wavelet transform,in which 1300 samples were selected.The dimensional reduction of feature space was brought out by algorithm of multi-resolution wavelet transform,in which 5 features were extracted from each sample.An intelligent pattern classifier with B-P neural network was used in studying the learnt effect of network and the recognition ability for unknown flaw.Experiments showed that the results of recognition were satisfactory when the wavelet spectrum was taken as a sample feature vector.The average recognition accuracy of the six types of flaws was about 90.4%,and the best recognition accuracy amounted to 97.2%.The method has been applied to describing the entire three-point flexural failure process successfully for 90° and 0° specimens with no notch and 0° notched specimens.