DAMAGE DETECTION IN FOUR STORIED STRUCTURE USING ANN WITH EXPERIMENTAL VALIDATION

Abstract

In this dissertation we proposed a novel ANN approach for damage detection in four storied steel structure and check its results using laboratory experiments. The four storied structure is fabricated in shake table laboratory and free vibration test is performed to extract modal properties using frequency domain decomposition FDD algorithm. The stick model is used for generation of training data, which is updated using ANN. For damage detection 1St four frequencies and associated mode shape vectors are used as input pattern to ANN. Damage is defined as reduction in storey. stiffness, which is used as output for ANN. New modified output layer MOP pattern is defined instead of traditional output layer TOP. Damage is quantified in four levels: safe, light, substantial and total. The new learning algorithm quick propagation is used for ANN training, which is found to _ be quicker and accurate than traditional back propagation. The analytical validation shows that ANN detected damage with accuracy more than 90%. For experimental validation damage is introduced artificially by loosening bracing bars. Experimental validation shows that, ANN predicted damage in l st floor and 2nd floor with higher accuracy, but 3rd and 4th floor with accuracy 60% and 40% respectively.