基于模糊神经网络和证据理论的结构损伤识别新方法

A new structural damage identification method based on fuzzy neural network and evidential theory

为了有效利用结构健康监测系统中的多源不确定数据,提高损伤识别的正确率,通过构造模糊神经网络(FNN)分类器,提出了一种新的概率赋值函数构造方法和数据融合损伤识别新方法.该损伤识别方法先对数据预处理,提取有效的特征参数,接着将它作为FNN的输入,构造FNN分类器,最后运用数据融合中的D-S证据理论计算出融合决策结果.为了验证所提方法的有效性,通过一个七层剪切型框架结构的数值模型,分别用单一FNN分类器和数据融合损伤识别方法进行了损伤识别和比较.研究结果表明,本文所提方法比单一决策结果更准确,具有更高的可靠度。

In order to make full use of multi-resource and uncertain data and to improve the damage identification accuracy of the objective from a large structural health monitoring system, a construction method of basic belief assignment function and a new data-fusion damage identification method were proposed in this paper via constructing an FNN model. In this proposed damage identification method, the original data is preprocessed and the feature parameters are extracted. Thus these parameters are regarded as input vectors of an FNN, this FNN classifier is constructed and then decision results were drawn by this model. Finally, the fusion decision-making results are computed and drawn by the data fusion algorithm of D-S evidential theory. A 7-story shear frame numerical model was utilized to validate this proposed method, and a comparison was made between this method and single FNN classifier. The results show that the proposed damage identification method is more exact and reliable than that of single FNN classifier.