Active Control of Structures Using Neural Networks and Fuzzy Logic

使用神经网络和模糊逻辑对结构进行主动控制

• 批准号: 9503209
• 负责人: Jamshid Ghaboussi
• 金额: $ 27万
• 依托单位: University of Illinois at Urbana-Champaign
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-05-01 至 1998-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9503209&HistoricalAwards=false
• 关键词: Active; Control; Structures; Using; Neural

9503209 Ghaboussi This project continues and extends the development of an active structural control method using neural networks by the investigators at the University of Illinois in 1992 under the NSF's structural control program. The main controller in the new method of structural control is a trained neural network, referred to as the neuro-controller. The training of the neuro-controller is accomplished with the aid of an emulator neural network, which is first trained to learn the structural behavior. Results from the numerical simulations showed that the new structural control method performs well compared to other control methods and is far more robust as well as being an inherently adaptive controller. In this project research on neural network based methods of structural control will be continued to consist of (1) experimental verification of the new structural control method; and (2) the extension into nonlinear control problems. The experimental work will be done in two stages. In the initial stage, extensive trial and error, adjustment and the online training of the emulator neural network and the neuro-controller will be done. Verification tests will be conducted on the shaking table at the University of Illinois Structural Engineering Laboratory, using a three story, one bay structural model, similar to those used in numerical simulations. The next stage of experimental work will be conducted at the NCEER facilities, on a larger structural model. Finally, the new method of structural control must be extended to address the problem of nonlinear control of structures sustaining unpredictable damage and develop a feasible and practical method of nonlinear structural control. ***

9503209 Ghaboussi该项目继续并扩展了1992年在美国国家科学基金会的结构控制计划下由伊利诺伊大学的研究人员使用神经网络的主动结构控制方法的发展。 在新的结构控制方法中，主控制器是一个经过训练的神经网络，称为神经控制器。 神经控制器的训练是在仿真器神经网络的帮助下完成的，该仿真器神经网络首先被训练以学习结构行为。 数值仿真结果表明，新的结构控制方法相比，其他控制方法表现良好，是更强大的，以及作为一个固有的自适应控制器。 在这个项目中，基于神经网络的结构控制方法的研究将继续包括（1）新的结构控制方法的实验验证;（2）扩展到非线性控制问题。 实验工作将分两个阶段进行。 在初始阶段，将进行大量的试验和错误，调整和在线训练的仿真神经网络和神经控制器。 验证试验将在伊利诺伊大学结构工程实验室的振动台上进行，使用三层，一个海湾结构模型，类似于数值模拟中使用的模型。 下一阶段的实验工作将在NCEER设施中进行，采用更大的结构模型。 最后，必须将新的结构控制方法扩展到解决承受不可预测损伤的结构的非线性控制问题，并开发出一种可行的和实用的非线性结构控制方法。 ***