模式识别的斜拉桥损伤诊断动力指纹与识别算法

为有效并准确诊断出斜拉桥损伤,对基于模式识别的斜拉桥损伤诊断方法进行了研究。选取易于测试出的低阶模态频率和部分关键点竖向振型数据为动力指纹,无需模态扩展或模型缩聚。研究并采用全因子设计进行动力指纹库的创建,可精确评估设定的损伤因子及其交互作用对损伤识别结果的影响。设计并增加了带随机误差的动力指纹库样本集。编制了基于Matlab的模式识别的多种算法,重点研究了精确度高的多层感知器识别算法及其提高该算法预测准确率的装袋集成算法。最后给出一座单塔双跨双索面斜拉桥的多种识别算法的损伤诊断过程和结果,得到一种可包容测试随机误差的高精确度斜拉桥损伤诊断评估模型。

Dynamic fingerprint and identification algorithm for damage diagnosis of cable stayed bridge based on pattern recognition

In order to effectively and accurately diagnosis the damage of cable stayed bridge, the damage diagnosis method of cable stayed bridge based on pattern recognition was studied. The low order modal frequency and vertical vibrational mode of some key points were selected for dynamic fingerprints of no modal expansion or model condensation. The full factorial design was used to create the dynamic fingerprint database, which could accurately evaluate the damage factors and their interaction effects on the damage identification results. And the dynamic fingerprint database with random error was designed and added. The pattern recognition algorithms based on MATLAB were compiled. The high accuracy of the multilayer perceptron recognition algorithm and the algorithm to improve the prediction accuracy of the bagging ensemble algorithm were mainly studied. In the end, the damage diagnosis process and results of a single tower double span double cable planes cable stayed bridge were presented, and a high precision evaluation model covering random errors for damage diagnosis of cable stayed bridges was obtained.