Damage detection in steel plates using feed-forward neural network coupled with hybrid particle swarm optimization and gravitational search algorithm

Abstract

Over recent decades, the artificial neural networks (ANNs) have been applied as an effective approach for detecting damage in construction materials. However, to achieve a superior result of defect identification, they have to overcome some shortcomings, for instance slow convergence or stagnancy in local minima. Therefore, optimization algorithms with a global search ability are used to enhance ANNs, i.e. to increase the rate of convergence and to reach a global minimum. This paper introduces a two-stage approach for failure identification in a steel beam. In the first step, the presence of defects and their positions are identified by modal indices. In the second step, a feedforward neural network, improved by a hybrid particle swarm optimization and gravitational search algorithm, namely FNN-PSOGSA, is used to quantify the severity of damage. Finite element (FE) models of the beam for two damage scenarios are used to certify the accuracy and reliability of the proposed method. For comparison, a traditional ANN is also used to estimate the severity of the damage. The obtained results prove that the proposed approach can be used effectively for damage detection and quantification.

Abstract

目的

使用模态损伤指数建立一个简单, 有效的结构健康监测评估工具, 并对钢板进行数值研究, 以确认该方法的可行性。

创新点

为使研究可应用于实际结构, 本文放弃了目前的大量研究中的刚度折减假设, 并在有限元模型中模拟了钢板的切割, 以代表实际结构的失效。

方法

1. 一个有名的混合优化算法, 即粒子群优化-引力搜索算法(PSOGSA), 被用于优化前馈神经网络(FNN)的连接权重和偏差, 以增强其训练性能。2. 模型的输入变量为由柔度矩阵变化推导出的两个损伤指数, 而输出变量则是损伤严重程度。3. 预测值和目标值之间的均方误差(MSE)是优化过程的适应度函数。

结论

1. 随机的FNN-PSOGSA方法获得了比传统人工神经网络(ANN)更好的损伤量化结果; 其在两种破坏场景下目标和估计之间的严重性差异分别为−0.06%和0.89%, 而在ANN中为−1.91%和1.01%。2. 所提出的方法可以在损伤指数和相应的严重程度之间建立联系, 而如果仅使用损伤指数则无法观察到该联系。3. FNN-PSOGSA方法的准确性和易实施性说明它具有作为真实结构损伤评估工具的潜力。