Abstract

Welded hollow spherical joints (WHSJs) are commonly used in reticulated shell structures. Corrosion on the surface of WHSJs can remarkably reduce their compression capacity. Pitting corrosion is a typical corrosion type on steel structures. Artificial neural network (ANN) is utilized to predict the compression capacity of WHSJs with random corrosion. Corrosion occurring at different positions can affect the compression capacity in different degrees. The spherical body of WHSJs is divided into several parts, and the mass loss ratio χ is utilized as the representation of corrosion severity. The influences of the number of divided corrosion locations and the probabilistic distribution of T_c/T on prediction accuracy is investigated in this study. The applicability of trained ANN for WHSJs with different geometric sizes is also validated. Results indicated that ANNs can be utilized for predicting compression capacity with high accuracy, and the mass loss ratio can be used as the input variable.

摘要

焊接空心球形接头 (WHSJ) 通常用于网壳结构。WHSJ 表面的腐蚀会显着降低其压缩能力。点蚀是钢结构上典型的腐蚀类型。人工神经网络 (ANN) 用于预测具有随机腐蚀的 WHSJ 的压缩能力。不同部位的腐蚀会不同程度地影响压缩能力。WHSJs的球体被分成几个部分，用质量损失比χ作为腐蚀严重程度的表示。划分腐蚀位置数量和Tc / T_{概率分布的影响}本研究对预测精度进行了研究。训练的人工神经网络对不同几何尺寸的 WHSJ 的适用性也得到了验证。结果表明，人工神经网络可用于高精度预测压缩能力，质量损失率可作为输入变量。