The Study Of The Application Of Neural Network To Reinforced Concrete Framed Structural Damage Detection

Reinforced concrete framed structure has been extensively used in the world, because it has many merits such as its materials can easily be got, its cost is low, its endurance is good etc. However, a large amount of existing constructions have approached their designed standard time, and some have been damaged because of overload and accidental load and so on, so there exist some hidden quality troubles. These damages and hidden troubles may cause the constructions badly destroyed even collapse, it will seriously threaten nation and people's life and property's safety. Therefore, it is necessary to intensify the research for the damage detection of reinforced concrete framed structure. Recently, many interior and oversea scholars have been all along endeavoring to find the methods that can detect the structural damage accurately. Because artificial neural network's nonlinear mapped capacity is strong, calculating speed is rapid, and it has highly robustness and fault-tolerance, it is appropriate to be used in structural damage detection. So many scholars have been carrying out exploring and research in this area and got much achievement, but there still exist some problems, for example, how to choose the type of the network and the network's input parameter, how to determine the training sample's number and the actual testing data are incomplete. In this paper, based on the analysis of preceding research, a parameter, called mode shape curvature ratio, is put forward for damage identification of reinforced concrete framed structure. Using this parameter, only the first mode shapes of the intact and damaged structure need to be got to calculate its mode shape .curvature ratio and then the structural damage locations and degree can be identified. ANSYS is used to make the modal analysis of a plane reinforced concrete framed structure model, then APDL is used to compile procedures to calculate mode shape curvature ratio, based on above work, a BP neural network is set up, the mode shape curvature ratio is its input parameter, and the former numerical case's data is taken as its training and testing samples. When the object is a large-scale and complex structure, neural network will need large amount of training samples, aiming at this shortcoming, from numerical simulation, a method to decrease the training sample's quantity is obtained and verified. Meantime, the simulation forecasting results indicate that using mode shape curvature ratio as the neural network's input parameter the structural single and multiple damages locations and degree can be accurately identified, and only the first mode shape is needed, so the actual measuring workload is greatly decreased, it shows this method is simple and feasible in project. According to the above analysis, the development prospect and problems of artificial neural network's application to structural damage detection are presented.