摘要
描述基于人工神经网络的结构损伤检测的基本步骤以及该方法在实际5层钢框架结构损伤检测上的应用.提出了一种改进的BP神经网络方法,它能够解决传统BP神经网络在实际应用中存在的两个问题:收敛速度慢并存在局部极小.其基本思想是引入动态自适应算子加速传统BP算法的梯度下降速度,从而提高运算速度,通过自调节保证学习过程中每一时刻具有较大的sigmoid函数值,从而可以避免局部极小.数值仿真结果表明基于该自适应神经网络的结构损伤检测方法具有强的鲁棒性,而且与传统的BP神经网络相比,不仅提高了计算速度,并且具有很高的精度.最后,实例的应用也证明了该方法的有效性.
A procedure based on the artificial neural networks in the damage detection of structures is developed and applied to an actual 5-story-steel frame. A new improved back-propagation neural network has been proposed to solve two practical problems encountered by the traditional back-propagation method, i.e., slow learning progress and convergence to a false local minimum. The basic idea is to accelerate steepest descent by adopting a dynamic adaptive algorithm, and to avoid local minima by keeping the sigmoid derivative relatively larger. Numerical simulation demonstrats that neural-networks-based method is a robust procedure and a practical tool for the detection of structural damage, and that the improved back-propagation algorithm may enhance computational efficiency as well as detection accuracy. The results of parameter identification of an actual 5-storey-steel-frame from measured data also validate the present approach.
出处
《力学学报》
EI
CSCD
北大核心
2003年第1期110-116,共7页
Chinese Journal of Theoretical and Applied Mechanics
基金
国家自然科学基金资助(59908003).
关键词
结构损伤
损伤检测
自适应BP神经网络
鲁棒性
neural network, improved back-propagation, self-adaptive algorithm, damage detection, robustness
