ABSTRACT

A novel assessment method for rapid structural safety state assessment is developed based on state-of-the-art machine learning technology. In this paper, a deep neural network (DNN), which originated in computer science, is first introduced, including its concept, structure and related training algorithms. Then the evaluation procedure to determine the structural safety state based on the DNN is developed. In the procedure, the pseudo-acceleration spectra and the structural safety states are selected as the input and output of the DNN model, respectively. Finally, the procedure of the novel assessment method is illustrated using a five-story reinforced concrete (RC) frame as an example. The effectiveness and accuracy are validated, and the influence of the number of hidden layers and size of the training data on the performance of the DDN is also investigated. The results demonstrate that the proposed method can evaluate the structural safety state rapidly based on the DNN. The most appropriate number of hidden layers is two considering both training accuracy and test accuracy. The training accuracy and test accuracy were 93.45% and 93.14%, respectively, using the two-hidden-layer DNN model trained on a training dataset of size 66,612.

摘要

基于最先进的机器学习技术，开发了一种用于快速结构安全状态评估的新型评估方法。本文首先介绍了一种起源于计算机科学的深度神经网络（DNN），包括它的概念、结构和相关的训练算法。然后开发了基于 DNN 确定结构安全状态的评估程序。在该过程中，伪加速度谱和结构安全状态分别被选为 DNN 模型的输入和输出。最后，以五层钢筋混凝土（RC）框架为例说明了新评估方法的过程。有效性和准确性得到验证，并研究了隐藏层数和训练数据大小对 DDN 性能的影响。结果表明，所提出的方法可以基于 DNN 快速评估结构安全状态。考虑到训练精度和测试精度，最合适的隐藏层数是两个。使用在大小为 66,612 的训练数据集上训练的两层 DNN 模型，训练准确率和测试准确率分别为 93.45% 和 93.14%。