论文部分内容阅读
本文提出了结构健康监测中一种新颖的方法用于时变系统的损伤识别,选取FS-TARMA(函数基时变自回归移动平均)时间序列模型应用于一种随机振动信号中,这种振动信号在时变系统中用来估计TAR/TMA参数和创新方差。基于一种特征值分解技术,被估计的TAR/TMA参数和创新方差能够为损伤估计提供更多信息和数据,从而形成了一种新的理论LCs(潜在成分)。LCs被组合和分解成数值,接着输入概率神经网络中进行损伤识别。将这种新方法用于三自由度时变系统中进行评估,根据质量和弹簧刚度的降低来模拟不同级别的损伤。这种方法能够找出系统参数的时变性质和质量及刚度变化引起的损伤级别。结果表明:使用这种方法,与其他的非降维和普通的特征提取方法相比,识别的成功率有相当大的提高。
In this paper, we propose a novel method for structural damage detection in time-varying systems. The FS-TARMA time-series model is applied to a random vibration signal. The vibration The signal is used to estimate TAR / TMA parameters and the variance of innovation in time-varying systems. Based on an eigenvalue decomposition technique, the estimated TAR / TMA parameters and variance of innovation can provide more information and data for damage estimation, thus forming a new theoretical LCs (potential components). The LCs are combined and decomposed into numerical values, which are then input into a probabilistic neural network for damage identification. This new method was applied to a 3-DOF time-varying system to evaluate different levels of damage based on the reduction of mass and spring stiffness. This method can find out the time-varying properties of the system parameters and the damage level caused by the change of the quality and rigidity. The results show that using this method, compared with other non-dimensionality reduction and general feature extraction methods, the recognition success rate has been greatly improved.