Research On The Computational Intelligence-based Damage Identification Strategy For Simply Supported Bridge

The health of bridge structure plays a crucial role of the national economy, people's lives and property since bridge structure is an important component of traffic facilities. A lot of bridge structures have been broken due to the rising traffic volume in recent years and the effect of environment factor and natural disaster. The carrying capacity of bridge is reduced and the safety, applicability and durability of bridge structure do not meet safety standards. As a result, it has great practical significance to monitor the health of bridge realtimely, find out the damage of structure, assess its safety and develop appropriate maintenance and repair strategies to improve the operational efficiency of the structure.Damage identification of simply supported girder bridge structure is carried out in the paper based on national high technology research and development plan (863 project) named'Research of large range road disaster parameter monitoring and identification and early warning system in seasonal frozen region. The damage identification theory is based on dynamic behavior of structure. Neural networks, support vector machines, fuzzy reasoning and other computational intelligence techniques are used for structural damage identification, which achieve good results. The successful solution of some difficult problems in the area provides some new ideas for the study of damage identification of simply supported girder bridge.Main research works in this paper are as follows:1. A strategy for damage detection of simply supported girder bridge is put forward through numerical simulation and analysis of structure. The strategy is based on step by step identification method, the curvature difference is used to determine the damaged area of the structure and to identify the the suspected damage elements. The training samples and the test samples of neural network are constructed according to those suspected damage elements, then the accurately damage locations are determined and the extent of the damage can also be obtained. The simulation results show that the fingerprint of modal curvature difference contains information of the damage location, it can be used for structural damage location. However, sometimes it is necessary to consider the first few orders of the modal curvature difference comprehensively so as not to misjudge. BP neural network and the RBF network are applied respectively for damage identification of the structure and comparative analysis is conducted. The results show that the learning and convergence speed of RBF neural network is superior to BP neural network. The RBF neural network's accuracy of identification for damage location is similar to the BP neural network, but its identification result of damage degree is slightly better than the BP network.To sum up, the step by step identification method is simple and practical and it can be used for the damage identification of reinforced concrete girder bridges. Neural network can identify the location and extent of damage accurately.The modal curvature difference method can be used to pre-judg the the general damage area considering the computational complexity of neural network, which can greatly reduce the number of required samples for neural networks and improve the work efficiency effectively.Finally, gaussian white noise is added in the test sample to analyze its impact on the BP neural network. The results showed that BP neural network is able to identify the damage location when level of noise is within 20%. But in the process of damage degree identification, the results of BP neural network are generally satisfactory when the noise level is less than 3%, but it can not meet the engineering requirements when the noise level reaches 5%.2. Basic on the theory of SVM, steel-concrete composite beams and multiple simply supported girder bridge.are chosen as object to simulation. It is verified that it's practial to use SVM to do damage location identification and damage degree identification.c ? svc is used as the algorithm for damage location identification and ? ?svras the algorithm for damage degree identification through method of regression analysis. Radial basis function is chosen as kernel function of support vector machines.The numerical simulation results of composite beam bridge show that support vector machines have a good ability to identify the location and extent of damage, Subsequently, the model of multiple simply supported girder bridge is built, and the vulnerable element of stucture is chosen to study. The damage location and the extent of the damage are identified. Respectively using the traditional support vector machine method and support vector machine cross-validation method optimized by genetic algorithm and ant colony algorithm .The simulation results show that these methods have good accuracy and can achieve damage identification for single and multiple suspected damage elements.3. Based on fuzzy reasoning theory, a strategy have a good robustness against model error and measurement noise is put forward taking into account of the defective of anti-noise ability of neural networks, and the damage identification system for. simply supported girder bridge is established.A fuzzy strategy of separated section is proposed, which can reducethe duplication of rules. Different damage state is corresponded to a set of unique input variables which means each damage state is determined by an unique damage rule. This shows that the fuzzy system is a good system for model classification and it is suitable for structural damage identification.The rate of change of natural frequency is select as the input parameters of fuzzy systems and the degree of structure damage is selected as the output parameters. Based on the change rate of modal frequency, the fuzzy reasoning damage identification system is established. The simulation results show that the memory of the systerm is good and have 100% accuracy, all the training samples'damage can be identified effectivly; for single location damage, the accuracy of the fuzzy reasoning system can reach more than 85% and for two- location damage, the accuracy can reach more 80 %; anti-noise analysis shows that the anti-noise level the of the fuzzy reasoning system based on the frequency can reach 15% or more, which means it has strong anti-noise ability.Considering the limitation of natural frequency of in dealing with symmetric problem, the vibration mode ratio is chosen as input parameters of fuzzy systems. Through choosing appropriate input parameters, the fuzzy reasoning damage detection system is establisedt based on model vibration mode The memory of the systerm is good and have 100% accuracy, all the training samples'damage can be identified effectivly; the numerical simulation results of the characteristical condition show that the fuzzy reasoning system based on vibration mode have good reasoning ability of damage identification; anti-noise analysis shows that the its anti-noise level can reach 15% or more, which means it has strong anti-noise ability.