应用神经网络评估叠合结构的耐久性

来源期刊：中南大学学报(自然科学版)2002年第2期

论文作者：侯小美 黄赛超

文章页码：125 - 128

关键词：神经网络;叠合结构;耐久性;二次受力

Key words：neural network; composite construct; durability; two-time loading

摘    要：由于叠合结构在预制和现浇两阶段受力截面高度不同,在荷载的作用下叠合结构产生的裂缝宽度、挠度、曲率总是大于钢筋混凝土和预应力混凝土在相同条件下产生的相应数值.而且在实际操作中,只是按照结构规范简单地将荷载作用下的两阶段结果叠加,导致误差很大,很难准确地进行叠合结构的耐久性评估.为此,作者提出了应用神经网络技术对叠合结构进行耐久性评估新方法.该方法以叠合结构的裂缝宽度、挠度作为结构耐久性评估的网络输入,对网络采用动量法和自适应学习率调整的方法,从而较好地解决了数据空间分布不均对网络收敛速度及精度的影响问题.计算结果与试验结果较好吻合,真实地反映了叠合结构二次受力的实际情况,能准确地预测出叠合结构的裂缝宽度、挠度,且避免了实际操作中人为因素、环境因素的影响,是一种有效的评估方法.

Abstract: Considering properties of composite constructs under two-stage loading, its amounts of curvature, crack and deflection are more excessive than reinforced concrete and stressed concrete′s at the same parameters. On the other hand,during the estimation of durability of composite constructs, the amounts of curvature, crack and deflection under two-stage are only added simply, so the results aren′t accurate. The authors applied neural network to estimate durability of com- posite construct. The crock deflection of composite constructs is taken as inputs to the neural network for parameter identification, and the methods of dynamics, selfadaptation adopted, thus the neural network can learn the training patterns efficiently as well as accurately. The results indicate that this method can be converged fast bymeans of dynamic learning algorithm.