HEALTH MONITORING OF A MULTI-STOREY BUILDING USING ARTIFICIAL NEURAL NETWORK

Abstract

Multi-storey buildings are important infrastructural systems as they accomodate more people. They may have visible or invisible damages in their members due to ageing or environmental/tectonic forces such as wind, blast, earthquakes etc. Visible damages can be easily detected by visual supervising; but invisible damages can not be detected so as it is hidden in members. Vibration Based Damage Detection (VBDD) techniques are useful in detecting those invisible damages. VBDD exploits vibration signatures taken at different parts of the buildings and then use them in damage detection. In this thesis, VBDD technique is used with the powerful mapping tool called ANN to map damages with vibration characteristics of a selected multi-storey building. Vibration characteristics of the building, particularly modal parameters, were derived by conducting Ambient Vibration Testing (AVT). An equivalent analytical model was developed to simulate the study building . Modal parameters of the analytical model and that of the study building were compared. Since the analytical model had modeling error, a parametric study on model updating was carried out and the model was updated. Two types of model updating procedures , Eigen-sensitivity and ANN based methods, were considered for the parametric study. ANN based model updating procedure updated first frequency with higher accuracy while Eigen-sensitivity based model updating procedure updated the higher frequencies with maximum accuracy. Based on the results of the parametric study, a more accurate analytical model could be obtained for the next level of damage detection. Damage was assumed as any reduction of stiffness of floors/storeys, and floor/storey stiffness is equal to the sum of lateral stiffness of all the structural and non-structural members in that particular floor/storey. Three types of damage diagnostic parameters (relative percentage changes of natural frequencies, displacement mode shapes and curvature mode shapes) were used individually and combinedly along iii with Artificial Neural Network (ANN) in detecting damages at different storeys of the building. Quick Propagation algorithm in ANN was able to detect damages when all the three diagnostic parameters were supplied together as input. This result may be used to find which floors/storeys have developed damages rather than local members of the floors/storeys. Then, some devices shall be used to detect local member-level damages in that particular floors/storeys only. In this way, the findings of the thesis may reduce the effort of unnecessarily conducting tests to detect local level damage surveys at every floors/storeys of a multi-storey building by predicting which storey has damage and what the amount of damage is. iv