Joint Damage Diagnosis Of Framed Structures Using Neural Network Technique

Since more and more existing structures are to be surveyed and strengthened, structural damage diagnosis has become one of the advancing fronts of civil engineering researches. The dynamic characteristics of structures are in close relationship with structural parameters, and structural damage will cause dynamic characteristics shifts correspondently. Therefore, if the mapping relationship between structural damage and dynamic characteristic shifts can be established, the damage can be diagnosed using dynamic measurements of the structures. The neural network technique has shown great superiority in structural damage diagnosis for its strong non-linear mapping ability, rapid computation and anti-interference capability. But there are still some problems to be solved such as selection of neural networks, determination of structural damage indicator and incompletion of measurement.In this paper, some joint damage diagnosing approaches of framed structures are discussed, and some efficient neural networks other than BP networks are explored. The strain mode shift is chosen as structural damage indicator, and a new approach for joints damage diagnosis of framed structures using artificial neural networks is presented. The noise injection training technique is introduced to enhance the anti-interference capability of the neural network.The computation of dynamic characteristics of framed structures with joint damages is proposed in this paper, and the joint fixity factor is defined to describe the severity of joint damage, then the finite element modal of the frame can be established. The relationship between joint damage and dynamic characteristic shift is acquired by analyzing the established modal. The strain mode theory is introduced and the experimental measurement method of strain mode is also discussed.A two-step approach for joints damage diagnosis of framed structures using artificial neural networks is presented in this paper. The fist step is to judge the damaged sub-area of the structure, which is divided into several sub-areas, using probability neural networks with neural frequencies shift ratio input, and the next step is to diagnose the exact damage location and extent using RBF neural network with the second element end strain mode of the damaged sub-area input.Conclusion can be drawn from the researches in this paper that probability neural network is efficient to judge damage location approximately, and that radial basis functions network is an excellent tool for structural damage diagnosis. The strain mode is sensitive to structural damage, and joint damages of framed structures can be diagnosed successfully using Element End Strain Mode shift. The proposed two-step approach can reduce the experimental measurement amount greatly and increase reliability of diagnosis results, which make it possible for damage diagnosis of complicated and large-scaled structures.