Deprem Sonrasında Algılayıcılar Kullanılarak Geçiş Alanlarında Hasara Bağlı Blokaj Tespiti

Abstract

Yalnızca İstanbul’da gerçekleşebilecek depremlerin senaryolaştırılmış sonuçları incelendiğinde, can kayıplarının tüm İstanbul nüfusunun %0,2’sine ve yapısal hasar kaynaklı mali kaybın 26 Milyar TL’ye varabileceği görülmüştür. Ekonomiyi olumsuz etkileyen ve ülkemizde gerçekleşen afetlerin büyük bir kısmını oluşturan depremler sonucunda meydana gelen can kayıplarını en aza indirmek için bina sakinleri ve acil kurtarma ekipleri için eş zamanlı olarak yapılarda oluşan hasar bilgisine ihtiyaç duyulmaktadır. Geçiş alanlarında oluşan blokaj bilgisinin eş-zamanlı olarak sağlanması ile gerekli noktalarda kullanılabilir hale gelecektir. Bu tez çalışmasında yapı içerisinde bulunan gerek yapısal, gerek yapısal olmayan elemanların durumları hakkında veri sağlayabilecek şekilde kapalı devre kabloları, ultrasonik mesafe ölçerler ve kameralar yapı içerisine yerleştirilmiş ve deprem ile eş zamanlı olarak veri aktarımları sağlanmıştır. Farklı tür algılayıcılardan elde edilen, farklı özellikteki veriler ortak bir biçimde ifade edilerek(1,0) anlamlandırılmıştır. Manuel olarak yapılacak değerlendirmelerde kullanılmak üzere, kameralardan elde edilen görüntü verileri için derinlik algılayabilen bir yöntem ve derinlik algılamadan geçiş alanının tamamını değerlendiren iki yöntem tanımlanmıştır. Verilerin anlamlandırılmasından ve kamera görüntü işleme yöntemlerinin belirtilmesinden sonra, çalışma grubu içerisinde yer aldığım TÜBİTAK projesi kapsamında seçilen pilot binanın geniş ve dar koridor modeli üzerinde deprem senaryoları gerçekleştirilmiştir. Kapalı devre kabloları ve ultrasonik mesafe ölçerler dolap, duvar ve asma tavan gibi yapısal olmayan elemanlar üzerine yerleştirilmiş ve toplamda 168 geniş koridor, 21 dar koridor deneyi gerçekleştirilmiştir. Deneyler sonucunda algılayıcılardan elde edilen veriler ile ilk olarak her bir algılayıcının denetlemesi gereken elemanın durumunu tek başına tespit edebilme başarısı manuel olarak irdelenmiştir. Sonrasında tek tip algılayıcı türünün ve kameranın bir geçiş alanını değerlendirme başarısı ayrı ayrı tespit edilmiştir. Algılayıcıların elemanları değerlendirmedeki başarı sonuçları göz önünde bulundurularak elde edilen verilerin birleştirilmesi ile algılayıcıların beraber kullanılması sonucunda blokajı tespit edebilme başarısına sağlanabilecek katkılar araştırılmıştır. Gerçekleştirilen çalışmalarda kullanılan veriler C4.5 ve REPTree algoritmaları ile karar ağacı oluşturmada kullanılmış, karar ağacı sonuçları ile manuel olarak gerçekleştirilen sonuçların birbirini ne oranda destekledikleri değerlendirilmiştir. Karar ağacı dışında 6 farklı algoritma (GainRatioAttributeEval, ChiSquaredAttributeEval, FilteredAttributeEval, OneRAttributeEval, ReliefFAttributeEval, SymmetricalUncertAttributeEval) ile geçiş alanının durumunu değerlendirmede kullanılan algılayıcı verilerinin önem katsayıları belirlenmiş ve manuel olarak yapılan değerlendirmelerin doğruluğu desteklenmiştir. Son olarak, farklı sayıda ve türde algılayıcı içeren sistemlerin farklı yapılarda kullanılmasına olanak sağlamak amacıyla pilot binanın hasar tespit doğruluk oranlarını ve maliyet analizlerini içeren bir çizelge oluşturulmuş ve ideal olan sistemler belirtilmiştir. Genel anlamda, geçiş alanlarının blokaj seviyelerini belirlemede kullanılacak algılayıcıların ne şekilde anlamlandırılması ve birleştirilmesi gerektiği konusunda çalışma gerçekleştirilmiştir. Tüm elemanların durumunu belirlemede KDK’lerin UMÖ algılayıcılarından daha başarılı sonuç verdikleri belirlenmiştir. Kamera görüntü işleme yöntemleri karşılaştırıldığında, derinlik algılayabilen yöntemin geçiş alanının blokaj seviyesinin tespitinde en başarılı yöntem olduğu ve UMÖ algılayıcısı ile bu tespitin daha da güçlendirilebileceği gözlemlenmiştir. Maliyet analizi sonucunda ise en ideal sistemin %96 doğruluk oranına sahip yalnızca KDK’lerin kullanıldığı sistem olduğu sonucuna varılmıştır. %92 doğruluk seviyesi ile KDK miktarlarında azaltma gerçekleştirmek ve maliyeti azaltmak da mümkün olabilmektedir. Bu çalışma yardımıyla geliştirilebilecek bilgisayar sistemi, acil durum ekipleri ve bina sakinleri için ihtiyaç duyulan bilgilere zamanında ulaşılabilmesine olanak sağlayacaktır. Acil durum ekipleri yapının aldığı hasar bilgisi ile daha fazla hasar ve can kaybı meydana gelmeden öncelikli müdahale gerektiren noktalara yönelebileceklerdir. Bina sakinleri ise, deprem ile eş zamanlı olarak çalışır durumda olan bir yönlendirme modeli ile aşina olmadıkları bir yerde dahi en elverişli çıkış yolunu bulabileceklerdir.

The economic damage from natural disasters has been doubled by 10 in the past 50 years. Earthquakes cause unpreventable direct impacts like economic damages and live lose as they can collapse buildings but in most cases, most of the structures remain in the safety limit like in Gölcük earthquake on 17 August 1999 in Turkey. To lower the damage from earthquakes, which have the biggest percentage of disasters that occur in our country, there is a need for the information about the blockage related to damage in transition areas in buildings. This information can be used both by the occupants inside the structure and by the emergency rescue teams for the determination of the shortest path to safe zone. Although some other kinds of information are available at about the buildings structural system, these kinds of real-time information were not available since now for both parts. According to this deficiency, in this thesis ultrasonic distance measurement tools, closed circuit cables and cameras were used to obtain the information about the situation of both structural and non-structural elements in buildings. By the help of this information, blockage level caused by elements in transition areas of structures was aimed to be determined. A three-story pilot university structure was selected in Nevşehir and 1/3 scaled prototype models of wide and narrow corridors were created in 109M263 numbered TÜBİTAK project that I have been involved. In wide corridor prototype model suspended ceilings, wardrobes and walls were placed where in narrow corridor prototype model only wardrobes and walls were used for the blockage tests. Ultrasonic distance measurement tools were used on suspended ceilings and walls one on each to detect their final situations after disaster. One ultrasonic distance measurement tool was placed bottom of the upper floor, above the suspended ceiling to watch that specific suspended ceiling and one ultrasonic distance measurement tool was placed opposite of a wall to watch that specific wall’s situation after disaster. As the suspended ceiling’ final situation was thought to be in seven different situations after disaster, four closed circuit cables were used on suspended ceilings. Only two closed circuit cables were used on walls and, two closed circuit cables were used on wardrobes to detect their final situations. The camera was placed to detect whole transition areas blockage level instead of a single non-structural element’s final situation. Related with the structure’s story area, 2 to 3 accelerometers were determined in TÜBİTAK project to be used as triggers for the sensors to start sensing with the earthquake. Gyroscopes were decided to be used on several columns in story to determine the structural final situation of the building. Although the transition area in real buildings with a damaged column planned to be considered as blocked, as the structural system of the 1/3 scaled prototypes were not damaged with real earthquakes, any of the accelerometers or gyroscopes were used on tests. After that, 168 wide corridor tests and 21 narrow corridor tests were performed according to the previously specified earthquake scenarios in TÜBİTAK project. The data were collected with the placed sensors in all tests. The data collected from ultrasonic distance measurement tool was in meters where closed circuit cable was in 1 and 0 and, camera was in image format. To strengthen the information’s reliability about the structural and non-structural elements’ final situation, data fusion was needed. In addition, to perform the fusion of the data collected from different kind of sensors, all data were organized in 1, 0 format. For ultrasonic distance measurement tool, threshold values were determined for elements individually, considering collected values from tests in TÜBİTAK project. If the ultrasonic distance measurement tools sensed value exceeds the threshold value, the output was 1 and the detected element considered as damaged. Otherwise, elements situation detected with ultrasonic distance measurement tool remained as 0. For the camera, one perspective detectable and two non-detectable in total three different image processing systems were introduced and used to determine the transitions areas’ blockage level. All three systems general idea was similar as they all calculate the first accessible part of the transition area and compare it with the final situation. From A to E, with 20% increment five class intervals were determined for the camera results and a table was created according to this consideration of the ratios. To start with, all collected sensor data were compared with the real element situations viewed from external camera records and all false negative, false positive situations were detected. The values detected were analyzed to understand the causes and noted for further experiments to prevent collecting that kind of unexpected values. In this thesis, all false negative and false positive values were used for calculations as they were considered that they may also appear in real life and also it was one of the main aims to determine every sensors performance on detecting the elements final situation. Data obtained from different sensors on different non-structural elements were evaluated separately and manually to specify every sensor’ success on determining every element’s final situation individually. All sensors were compared on their performance detecting different kind of non-structural elements. Then, a manual study was performed for every kind of sensor placed on different non-structural elements in a transition area to determine the success of sensors on detecting transition areas final situation. To enhance the success level of sensors on detecting the transition areas blockage level, considering the success levels obtained from previous study, the data were fused. First, fusion was performed on single elements, with same kind of sensors like two closed circuit cables on wardrobes or on walls. Then, both ultrasonic distance measurement tools and closed circuit cables sensors were fused on elements and enhance to the damage detection levels was observed. To observe the ultrasonic distance measurement tools contribution to the camera image processing results, both data were fused with three different processing systems. Next, manually obtained data were supported by decision trees for validation. C4.5 algorithm was selected as it was widely used for that kind decision tree systems and ranker method was used for attribute ranking as it was defined as the processor of the C4.5 algorithm. In addition, a decision tree with REPTree algorithm was created, as it was known to be faster than C4.5 algorithm. 24 input and 1 output attributes were entered as data to Weka program to obtain trees. Confusion matrixes’ recall and precision values were evaluated to compare the effectiveness of decision tree algorithms and C4.5 performed better in both recall and precision values. Then attributes’ rankings were obtained with 6 different attribute evaluator algorithms (GainRatioAttributeEval, ChiSquaredAttributeEval, FilteredAttributeEval, OneRAttributeEval, ReliefFAttributeEval, SymmetricalUncertAttributeEval). GainRatioAttributeEval was used to rank all 24 attributes in tests where the rest of the algorithms were just used to rank and compare closed circuit cables’ and ultrasonic distance measurement tools’ success on detecting suspended ceilings and, walls and transition areas’ final situations. As previously found from manually performed calculations, closed circuit cables were performed better performance on detecting transition areas’, suspended ceilings’ and walls’ final situation compared to ultrasonic distance measurement tools. For last, the cost analysis of different kinds of sensor-involved systems had been made in previously selected pilot model. The sensors were considered as they were placed in the same way as they have been used in prototype during the tests. Nine different selections of sensors were listed in a table with their success in detection of the transition areas final situation. The one that only includes closed cable circuit sensors were determined as one of the cheapest and the most successful with 96% success percentage. The second successful one was determined with some reduction in the numbers of closed circuit cable sensors on walls and on wardrobes with a success percentage of 92%. Any of the camera’ image processing systems seem as it was contributing to the results as their success were clearly low. The maximum camera success percentage was 52% and obtained from the perspective detectable processing system. It was concluded as closed circuit cables performed better on detecting all elements and transition areas situations compared to ultrasonic distance meters and cameras. However, ultrasonic distance meters strengths the information obtained from closed circuit cables in element level. As with the selected processing systems, cameras’ success levels were low, they were not useful for the fusion. Besides that, the processing systems can be upgraded and they can be useful for detecting transition areas’ blockage level without any visual pollution in the area as the processing systems can be set in security cameras. From the results obtained from this study, the information needed for the emergency rescue teams and occupants can be generated during the earthquakes. According to the damage information about the building, emergency rescue teams will be able to response through less blocked transition areas and avoid more live loses. In addition, occupants can be directed with a visual system, even if they were not familiar with the building. It is proposed that, more and different earthquake scenarios for different pilot models should be created for this work. Although the system mentioned in this thesis was validated, as the more scenarios will be performed, the more system will reflect the real earthquake situation.