自适应小波概率神经网络损伤识别方法  

Damage Identification Approach of Adaptive Wavelet Probabilistic Neural Network

作　　者：姜绍飞[1] 张帅[1] 杨晓楠[2] 

机构地区：[1]沈阳建筑大学土木工程学院,辽宁沈阳110168 [2]同济大学结构工程与防灾研究所,上海200092

出　　处：《沈阳建筑大学学报（自然科学版）》2006年第1期45-48,共4页Journal of Shenyang Jianzhu University：Natural Science

基　　金：国家自然科学基金(50408033);辽宁省自然科学基金(20022136)

摘　　要：目的为了提高大型结构健康监测系统的监测能力与损伤诊断率,降低误报率.方法以小波变换作为动力信号处理工具,利用其可以降低噪声以及在时域-频域表征信号特征的强大能力,提取小波能量作为特征参数;以贝叶斯推理作为模式识别原理的概率神经网络(PNN)为损伤识别分类器,利用遗传算法来优化PNN模型中的圆滑参数σ,提出自适应小波概率神经网络(AWPNN)损伤识别方法.并对AS-CE的基准结构模型进行损伤识别研究以验证该方法的有效性.结果研究结果表明,在噪声程度达40%时,AWPNN的识别正确率高达98%.结论AWPNN具有较强的抗噪声能力和较高的损伤识别率,在结构健康监测与损伤识别领域具有很大的潜力.The main factors of the assessment on the structural healthy conditions are noise, feature extraction from the dynamic responses and effective damage identification approach. In order to improve the abilities of structural health monitoring and damage diagnosis ratio for a large structural health monitoring system, wavelet energy feature was extracted from dynamic response based on wavelet transform, which could not only reduce the noise from signals but also describe signal characteristics in both time and frequency domains. Probabilistic neural network （PNN）, whose fundamental was Bayesian inference, was classifier. A new damage identification approach was proposed, i. e., adaptive wavelet probabilistic neural network （AWPNN）, in which the wavelet energy feature was the input of PNN, damage classifier was adaptive PNN, whose scaling parameter a was optimized by genetic algorithms. To validate the efficiency of the proposed approach, the damage identification of a Benchmark structure presented by ASCE was investigated. The results show that damage identification accuracy amounts up to 98 %, whilst the noise level is 40 %. It is confirmed that the proposed approach, i. e., AWPNN, possesses high identification accuracy and good noise-resistant, which has tremendous potential in the structural health monitoring and damage detection.

关 键 词：损伤识别 小波变换 能量特征 概率神经网络 噪声 

分 类 号：TU318[建筑科学—结构工程] TU393