论文部分内容阅读
参数识别是结构健康监测、性能评估的关键问题之一。作为一种代表性的动力系统时域参数化模型方法,自回归滑动平均(Auto-regressive and moving average,ARMA)模型在机械和土木工程结构的参数识别中得到了广泛应用;另一方面,尽管一般而言神经网络模型的权重和阈值并不需要具备明确的物理意义,但由于神经网络具有描述复杂函数关系的能力,作为一种非参数化模型方法在结构动力系统的建模和控制领域发挥重要作用。该文首先通过结构运动平衡方程的离散时间解,证明了非参数化神经网络模型与ARMA模型在描述线性结构动力系统的响应时间序列上的等效性,在此基础上,提出了一种从结构的非参数化神经网络模型中抽取结构物理参数的新方法。通过一个多自由度系统的数值模拟结果和一个四层钢框架模型的动力试验实测数据验证了所提出的结构参数抽取方法的有效性。
Parameter identification is one of the key problems in structural health monitoring and performance evaluation. As a representative time-domain parametric model of dynamical systems, the auto-regressive and moving average (ARMA) model has been widely used in the parameter identification of mechanical and civil engineering structures. On the other hand, In general, the weights and thresholds of neural network model do not need explicit physical meaning. However, as neural networks have the ability to describe complex functional relationships, they can be used as a non-parametric modeling method in the field of structural dynamics modeling and control Important role. In this paper, firstly, the discrete time solution of the structural equation of motion equilibrium is used to prove the equivalence of the nonparametric neural network model and the ARMA model in describing the response time series of linear structural dynamical systems. Based on this, A New Method for Extracting Physical Parameters of Structure in Nonparametric Neural Network Model of Structure. The numerical simulation results of a multi-degree-of-freedom system and the dynamic test data of a four-story steel frame model validate the proposed structural parameter extraction method.