报告人：Jin-Song Pei（Assistant Professor，School of Civil Engineering and Environmental Science，the University of Oklahoma，USA）

时间：2008年7月4日（周五）下午14：30

地点：航空宇航学院学术报告厅（A18-529）

主办单位：国际合作交流处、科协、航空宇航学院

报告人简介：

Jin-Song Pei received her B.Eng. and M.Eng. degrees in Structural Engineering from Xi’an Jiaotong University, Xi’an, China and Nanyang Technological University, Singapore, respectively, and a Ph.D.degree in Civil Engineering and Engineering Mechanics from Columbia University, New York, NY. She is an Assistant Professor at School of Civil Engineering and Environmental Science at the University of Oklahoma. She was an assistant engineer at the Real Estate Division of Construction & Development Corporation, Xiaman, China, worked as an engineer at Indeco Consultants, Singapore and also practiced at Weidlinger Associates, Inc., Cambridge, MA before joining the faculty at the University of Oklahoma in 2002.

报告内容简介：

Rapid advances in sensor technologies and sensor networks are providing researchers in different fields of science and engineering with new opportunities for understanding complex large-scale dynamic systems based on valuable first-hand information - data measurements collected from the real world. This is true in the discipline of structural engineering as in many others, where multidisciplinary expertise is in demand to handle complicated data sets and extract models to fit specific applications, such as structural control and health monitoring.

This seminar will first present the development of a novel neural network initialization methodology in modeling complex nonlinear systems, where the insight to the underlying mathematical and/or physical characteristics of the systems is injected into this popular data-driven technique. Here, data-driven techniques refer to a wide range of techniques that are not directly/explicitly built upon the physics of the problem. Highly dependent upon data sets, data-driven techniques are inherently adaptive and thus effective in producing a good fit. This is a strong merit, however, meaningful interpretation of non-unique results (i.e., non-unique non-parametric parameters), especially the quantification of damages by relating

these parameters to physics-based parameters, is very difficult. Various design issues associated with data-driven techniques can be subjective, lacking analytic or even empirical guidance. These challenging practical needs prompt the speaker to explore a possible fusion of data-driven techniques and physicsbased modeling of nonlinear dynamic systems to develop system identification tools that are powerful as well as rational. This fundamental study has a wide range of applications including designing active control systems, validating and updating structural models (e.g., for earthquake simulation), and performing structural health monitoring and damage detection.

A proof-of-concept investigation of Field-Programmable Gate Arrays (FPGAs) as a complement and/or an alternative to microprocessors to interrogate local data at sensors will also be introduced in this seminar. This smart sensing technology together with wireless communication have several advantages including increased power efficiency, reduced installation and maintenance time and costs, and more importantly, improved performance reliability compared to transmitting raw time histories out of wireless sensors. A Field-Programmable-Gate-Array (An FPGA) is in effect an integrated circuit consisting of an array of programmable logic cells, which spatially composes primitive operations rather than temporally composing them as in a traditional microprocessor. This fundamental difference can be viewed as parallel computing architecture in FPGAs versus serial computing in microprocessors. The growth of smart structures technologies demands a high level of integration of sensing, wireless communication, data processing, embedded systems, control and simulation processes to guard structures from aging problems and mitigate natural and man-made hazards, it is envisioned that FPGAs will greatly enhance onboard data processing (i.e., DSP) data interpretation (i.e., system identification) capabilities, facilitate various  task-specific application needs for optimized efficiency, flexibility and performance by working with microprocessors, and many more.