2018-01-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/672415摘要：本年度研究將朝向自主式的結構健康監測與損傷評估平台進行研發。在第一年度研究中，主要是以集中式結構健康監測與損傷評估平台之研發為主，主要研究課題是透過時頻域分析的技巧進行訊號淨化前處理、工作模態分析法及結構損傷快速評估，且開發、設計及實驗結構健康監測與損傷評估平台，其中包括標竿鋼構架、監測與損傷方法於資料擷取及運算系統上運行且快速呈現評估結果，探求研發方法的準確性及可實用性，以利未來於實際結構物中使用；第二年主要是將第一年的時頻域結構健康監測方法與工作模態分析法，改變成離散式結構損傷方法，其中時頻的方法適用即時與震後的快速損傷評估，而工作模態分析法則可以佐證時頻域的分析方法，另外也加入卡氏濾波器組的結構健康監測方法，以利於多重感測網的應用，並可即時探測結構損傷程度、損傷時間、損傷位置，將損傷程度利用機率方式呈現，利於後續決策之參考。第三年主要著重在自主式的結構健康監測與損傷評估平台之研發，由於過去兩年的研究成果，可把單一感測網改變成多重感測網，因此首先會著重在最佳配置的離散感測網，於感測往中配置多樣性感測儀器，如加速規、應變計等，增加損傷評估之能力；工作模態分析法則著重在高頻模態之識別，並改善識別之成效，降低識別之誤差，確保萃取之動力特徵正確性；引進人工智慧之方法，利用工作模態分析法及具輸入、輸出之系統識別方法，取得結構之簡化模型，基於該簡化模型建置各種破壞模式，並將其損傷之簡化模型進行動力特徵分析，將模態振形轉化成多重損傷指標(如模態應變能、模態驗證指標、模態小波轉換、模態曲率等)，利用多重損傷指標作為訓練標準，爾後建置卷積式神經網路，通過該卷積式神經網路判別損傷程度，並估計殘餘結構之性能；最後利用實驗的手段，驗證整體方法之可行性與效能，並將該方法應用在中科管理局之建築與高科技廠房中。<br> Abstract: The last-year project will mainly emphasize the development of an autonomous structural health monitoring and damage diagnosis system based on the vibrational measurements. In the first year, the research was focused on the development of signal preprocessing, operational modal analysis, and rapid damage assessment of structures using the time-frequency analysis. Additionally, a platform consisting of a benchmark multi-bay steel frame, a monitoring and damage assessment system, and embedded software was established for testing and verifying of all proposed methods. In second year, the focus was redirected to the development of a decentralized monitoring and damage detection system, which includes the real-time and after-event damage detection method based on time-frequency distributions and the frequency-domain, two-stage operational modal analysis that can verify the results obtained from the time-frequency damage detection method. Additionally, the damage detection method using multiple banks of Kalman estimators was developed that allowed damage occurrence, locations and severity to be determined in terms of probability. In the third year, the probject will be aimed at the development of an autonomous structural health monitoring and damage diagnosis system by integrating the previous developments with an aritificial intelligent model for more detailed damage diagnosis. In this autonomous system, the optimal sensor placement will be established by combining various types of sensors (e.g., accelerometers and strain gauges). This method will also offer an opportunity of constructing multiple structural health monitoring systems in a structure. The developed operational modal analysis will be modified to be capable of extracting the dynamic characteristics at higher modes. The modal parameters will be improved in terms of accuracy and will be exploited to better understand the changes in dynamics of a structure. Subsequently, an aritificial intelligent model will be built based on multiple types of damage indices and will be used to accurately diagnose damage severity as well as to successfully estimate the remaining perofmrnace of a structure. To construct the model, the input-output system identification will be utilized to establish a simplified model. Based on this model, several damage patterns will be examined and turned into the modal parameters. A convolutional neural network model will be trained by mode-shape damage indices (e.g. the modal strain energy, coordinate modal assurance criterion, continuous wavelet transform of mode shapes, and mode shape curvature) as input and the artificial created damage as output. The remaining perofmrance of the structure will be eventually determined by this neural network model. Finally, an experiment will be conducted to verify the proposed autonomous structural health monitoring and damage diagnosis system. This system will be also implement in the office building in the Central Sicence Park as well as the high-tech fabrication plants.結構健康監測多重感測網人工智慧損傷診斷Structural Health MonitoringMulti-Sensing MonitoringArtificial Intelligent Damage Detection