System Identification of a Building with Mid-story Seismic Isolation using Artificial Neural Networks

近年來，拜各種抗震技術的發明，天然災害所導致的土木結構危害得以減緩，其中又以隔震技術在各大行政、商業以及住宅區等皆廣為使用，若能將其技術以簡化的方式明瞭地顯示其特性，必定對隔震建築的相關設計研究有很大的幫助，故隔震建築物識別模型隨之而生。 本文旨在建置一套中間層隔震建築物之系統識別模型，期望以一簡化的模型識別出具有意義之物理參數。利用合理假設將中間層隔震建築物簡化成三個集中質量節點，分別為上部結構、隔震結構與下部結構，而每個節點有長向、短向以及扭轉向，共9個自由度。在上部結構採用等效線性模態質量法，將隔震層上方多自由度簡化成一單自由度系統，並且考慮其偏心所造成之扭轉耦合效應；而裝設在隔震層之鉛心橡膠支承墊其力學反應為雙線性行為，將以Masing Rule 之背骨曲線去做描述。 本研究試將上述簡化理論與類神經網路進行整合，分析不同地震作用下隔震建築物之所發生樓層反應，並利用類神經網路中倒傳遞網路之概念來開發系統識別模型，且在開發模型過程中，依序建置線性模型與非線性模型，主以線性模型來建立類神經網路在系統識別下架構，並以此架構來確立隔震層非線性模型的完整性。並在各階段模型建置完成下皆進行數值驗證，將以適當之參數代入模型中，以 El Centro 地震歷時藉由 Newmark β線性加速度法做動力分析，以此求得各自由度之歷時反應，將其作為量測資料進行系統識別，驗證模型正確性與識別流程可行性。 在實例識別方面，本研究視台大土研大樓為分析對象，藉由臺大土研大樓結構物強震監測系統裝設計畫，所收集到歷時資料配合適當之基線修正作為量測資料進行分析，將其進行系統識別求取各自由度之勁度、阻尼等參數，最後計算量測資料之歷時反應與識別之歷時反應做比較， 並且量化成誤差指標看其差異以及探討其誤差來源之原因，以確認整合類神經網路與系統識別之模型於實際應用的可行性。In the recent years, the techniques against the earthquake are used to mitigate the effects of nature hazards on civil infrastructure, especially in isolation technology, which widely used in a lot of administrative, commercial and residential buildings. It will help a lot if we can clearly get the characteristics by simplifying isolation technology. So an identification model of isolation building can be established. The objective of the research is to establish a system identification model of building with mid-story isolation, expecting to identify meaningful and physical parameters of building with a simplified model. We use lump mass method, simplifying a building with mid-story isolation into a three lump mass system. Respectively, these are superstructure, isolated-structure and substructure. And we consider there are three degrees of freedom (longitudinal, transverse and torsional) in every lump mass, total of nine degrees of freedom. For the superstructure, we use effective modal mass method to simplify multiple degrees of freedom system, and consider torsional coupling effect with it. Bilinear mechanical behavior assumed in isolation story with LRB, following the Skeleton curve in Masing Rule. In the research, integrating following theory and Artificial Neural Network (ANN), we analyze the relationship between the structure vibration and ground-motion and use Back Propagation Network (BPN) to build the model. We can divide it into two stages, linear and nonlinear model, in the process of development. Linear model is major to build the framework with ANN. By the framework, we can set up bilinear mechanical behavior into nonlinear model. After developing, the model will be verified by numerical analysis after developing at each stage of model development. We design the structure parameter with nine degrees of freedom of mid-story isolation building model. Next, we input El Centro earthquake and calculated dynamics response of all degrees of freedom by Newmark β linear acceleration method. Subsequently, regarding the behavior of each degree of freedom as measurement data, the identification model is executed to verify the suitability of identification theory and modified network framework. In identification of real case, the data gathered by strong motion instrumentation program in the New Research Building of Civil Engineering department of National Taiwan University is adopted as measurement data to implement practical analysis, both the single and multi-section identification are accomplished and discussed in the research. At final, the reliability of the frame of network is verified by calculating the error index between the result of identification model and measurement data.口試委員審定書 I 誌謝 II 摘要 III ABSTRACT IV 目錄 VI 表目錄 VIII 圖目錄 IX 第一章 緒論 1 1.1. 研究動機與目的 1 1.2. 文獻回顧 3 1.3. 論文章節內容 5 第二章 中間層隔震建築物模型識別理論 6 2.1. 結構物隔震原理 6 2.1.1. 鉛心橡膠支承墊 7 2.2. 中間層隔震建築物簡化模型 8 2.3. 類神經網路 10 2.3.1. 生物神經元與人工神經元 10 2.3.2. 類神經網路模式 12 2.3.3. 類神經網路運作過程 13 2.3.4. 類神經網路之特性 13 2.3.5. 倒傳遞類神經網路與演算法 14 第三章 中間層隔震建築物模型建置過程 18 3.1. 中間層隔震建築物線性模型 18 3.1.1. 線性模型數值驗證 21 3.1.2. 線性模型數值驗證結果 22 3.2. 中間層隔震建築物非線性模型 23 3.2.1. 遲滯–雙線性模型 25 3.2.2. 非線性模型數值驗證 27 3.2.3. 非線性模型數值驗證結果 28 3.2.4. 小結 29 第四章 實例之系統識別 30 4.1. 台大土木系新建研究大樓結構物強震監測計畫 30 4.2. 地震資料處理 31 4.3. 非識別參數推估 32 4.4. 實例識別範例 34 4.4.1. 實例識別範例（Case A） 35 4.4.2. 實例識別範例（Case B） 36 4.5. 小結 37 第五章 結論與展望 38 參考文獻 40 附表 42 附圖 52 附錄A 運動方程式矩陣表示式 10