كاربرد روش كاهش تصادفي و معيار باتاچاريا در تشخيص خرابي سازه ها تحت شرايط محيطي و كاربري متغير

Application of random decrement technique and Bhattacharyya measure to damage detection under environmental and operational variability

كاربرد روش هاي تشخيص آماري الگو [1]در پايش سلامت سازه به منظور يافتن تغييرات در سازه توجه زيادي را به خود جلب كرده است. در اين زمينه، رويكردهاي متفاوتي توسط محققان به كار گرفته مي شود و موفقيت يك روش بستگي به سازه يا تغييرات بوجود آمده در سازه دارد. يكي از مسايلي كه در اعمال روش هاي تشخيص آماري الگو كمتر مورد توجه قرار گرفته است، متغير بودن شرايط محيطي و كاربري در هنگام ثبت داده ها در كاربردهاي عملي است. هدف از اين مقاله معرفي كاربرد جديدي از روش كاهش تصادفي [2] (RD) در تشخيص خرابي به منظور ارتقاي نتايج حاصل از تحليل سري زماني است. روش RD از طريق ميانگين گيري داده هاي سري زماني در بازه هاي زماني مشخص، سري زماني پاسخ سازه را به پاسخ ارتعاش آزاد كاهشي سازه [3] كه تنها شامل اطلاعاتي درباره ويژگي هاي ديناميكي سازه است، تبديل مي كند. از مدل اتورگرسيو- اتورگرسيو با ورودي خارجي [4] (AR-ARX) براي برازش يك مدل رياضي به داده هاي سري زماني پاسخ ارتعاش آزاد كاهشي سازه استفاده شده است. پس از محاسبه خطاي باقي مانده مدل، معيار جديدي به نام معيار باتاچاريا [5] براي تعيين وجود خرابي و مكان آن به كار گرفته شده است. . با استفاده از داده هاي حاصل از دو نمونه شاخص در حوزه پايش سلامت سازه به اعتبارسنجي روش ارايه شده پرداخته شده است. نتايج نشان مي دهد كه استفاده از روش RD به علت ميانگين گيري و نرمال سازي داده ها مي تواند تاثير تغييرات شرايط محيطي و كاربري را كاهش دهد و وضعيت سازه را به درستي تعيين كند. همچنين استفاده از معيار باتاچاريا مي تواند نتايج پايش سلامت سازه را در تشخيص وجود خرابي و تعيين مكان آن بهبود بخشد.

Damage detection is a necessary part for structural health monitoring (SHM), being beneficial to SHM and determining the severity of damage. Application of statistical pattern recognition methods for SHM has gained considerable attention to detect changes in a structure. One of the advantages of these methods is that only data from undamaged state is needed in training phase (unsupervised learning) as opposed to supervised learning where data from both undamaged and damaged conditions is required to train the model. There are different approaches used by researchers and the success of a certain one may depend on the type of structure or structural changes. Most of studies focused on the application of statistical pattern recognition methodologies for SHM utilize the time series analyses for extracting damage-sensitive features. These features are statistical properties of time series models that directly depend on damage. Extracting damage-sensitive features is a fundamental step in damage detection process because pattern recognition algorithms can identify the state of structure unless these features are just dependent on damage. The change in an environmental and operational condition during the data acquisition process is one of the problems that causes damage features to be depended on factors besides existence of damage. This can lead to incorrect structural state identification. On the other hand, after extracting damage-sensitive features, the application of a statistical novelty detection methodology for decision making on structural state is a significant topic in SHM. This paper proposes a new application of random decrement (RD) technique in order to choose appropriate and accurate damage features which are independent from environmental and operational conditions of structure. The RD technique transforms time series data of the structural response to free decay vibration form that only consist of dynamic properties information by averaging them at specific time. Moreover, a novel statistical method named as Bhattacharyya measure is applied as a robust method for damage diagnosis. The Bhattacharyya measure determines the discrepancy between damage features from different structural states through partitioning data and utilizing numerical information of each partition. Herein, before extracting damage features, time series data are averaged through RD technique. Then the Autoregressive-Autoregressive with exogenous output model (AR-ARX) is used to fit a mathematical model to the averaged time series data and the residuals are considered as damage features. The Bhattacharyya measure is utilized for damage identification and localization. The data obtained from an experimental study on a three-story frame structure model are exploited to validate the accuracy and reliability of the proposed algorithm. Random excitation is applied by varying amplitude level of the input force, simulating various environmental and operational conditions. Damage is induced at two different locations. The proposed algorithm is conducted on data from various environmental and operational conditions at two different locations. A comparative study is also carried out to demonstrate the superiority of the proposed algorithm over some exiting techniques. Results show that the application of random decrement technique reduces the influence of operational and environmental condition due to averaging and normalizing data and correctly determines the state of structure. In addition, using Bhattacharyya measure improves the structural health monitoring results in damage identification and localization.