Development of an intelligent structural testing system utilizing neural networks

在线或伪动力试验已被认为是建筑物地震模拟的有力工具。该方法基于动力分析计算结构系统的位移，并根据该响应驱动执行器直接从试件中测量恢复力。本标准完全适用于混凝土刚性板或剪力墙板的中高层建筑。在这种情况下，系统的质量集中在底层，元件的位移被认为是与质量相关的侧向位移的函数。但对于由柔性梁柱组成的钢框架，在非质量相关自由度中出现不平衡力，并且位移作为横向位移的函数表示是无效的。本文提出了一种在线混合结构模拟的一般方案，以获得试验钢梁柱的变形位置。该方案使用了对试件的增量力的预测器。提出了弹性梁单元预测器、双分量双线性梁单元预测器、人工神经网络预测器和多弹簧非弹性梁柱单元预测器四种预测器。简要说明了该方案在混合分析中的适用性。此外，还显示了达到一般变形构型所需的转换。提出了三个系列试验来评估不同的预测指标及其方案的适用性，并描述了受控旋转和变轴向载荷或恒定轴向载荷下的试验。结果表明：神经网络预测器成功地应用于不平衡矩的去除方案中，使不平衡矩降至可忽略的水平。

The on-line or pseudo dynamic test has been considered as a powerful tool for the earthquake simulation of building. The method computes the displacements of the structural system based on the dynamic analysis and drives the actuators with that response to measure the restoring force directly from the test specimen. This criteria is perfectly applied to middle or high rise buildings with concrete stiff slabs or shear wall panels. In such a case the masses of the system are concentrated at the floor levels and the displacement of the elements are considered as a function of the mass associated lateral displacement. But in the case of steel frames composed of flexible beam-columns unbalanced forces appear in the non mass assosiated degree of freedom, and the representation of displacements as function of the lateral ones is not valid. In the present report a general scheme for on-line hybrid substructuring simulation is presented, to achieve the deformed position of testing steel beam-column. This scheme uses a predictor of the incremental forces on the specimen. Four predictors are presented, the elastic beam element predictor, the two component bilinear beam element predictor, the artificial neural network predictor, and the multi-spring inelastic beam-column element predictor. A brief description of the applicability of the scheme in the hybrid analysis is presented. Also the transformations required to reach the general deformed configuration are shown. Three series tests are presented to evaluate the different predictors and the applicability of the scheme, tests under controlled rotation and varying or constant axial load are described. From the report is conclude : the neural netwark predictor is applied successfully in the removal scheme of the unbalance moment, which is reduced to a negligible level.