Research Of Fault Diagnosis For The Roof Latticed Truss Structure Base On Data Fusion Techniques

The system for roof space latticed truss structure (SLTS) in Shenzhen citizen center is selected as a practice engineering background in this paper, the structure is a new political center and symbolic architecture in Shenzhen which overall structural area is 2.1 million square meter, the roof is one of the largest space lattice in the world, which is 486 meter long and 154 meter wide. In the one hand, there are many advantages in SLTS such as high space stiffness, moderate stability, good earthquake resistance capacity, sound bearing capacity, In the other hand, due to magnitude dimension, subjected to random load, exposing in bad environment, SLTS are affected easily by the natural disaster. Structure health monitoring is a method using spot and non-damage equipment, getting structure internal information and analyzing all kinds characters of structure response, then acquiring the structures change due to damage and degeneration. How to monitor and diagnose the capability of structure and predict the damage before the disaster is become a research issue now. To make it be more safety, the health monitoring and structural damage detection system of SLTS is designed based on multi-sensor data fusion.This thesis project introduces the basis theory and concerning information of data fusion. Such as adaptive weighted data fusion techniques, comparability coefficient techniques, Back-propagation neural networks theory, Dempster-Shafer theory and so on. In addition, a new combination rules based on the D-S framework is put up when it is invalidation after highly evidence conflict.Firstly, to make more accurate estimation of the SLTS lies on the accurate signal coming from the sensors of the system. So, how to distinguish and deal with the fault of sensors become important. The result shows that adaptive weighted method of data level fusion is well to compute the strain of components belong to SLTS, and realize the management of faulted sensors. As is in order to improve the anti-jamming ability and assure veracity of collecting data. Then comparability coefficient techniques be used to identify the mode of strain field, and finished the health monitoring function by calculate the corresponding strain field according to the mode at the moment. Secondly, the omen of components will be diagnosed to discover what damage has occurred by using collateral back-propagation neural network in miniature. Thirdly, the new combination rules and Dempster-Shafer theory be used to deduce the last conclusion at the decision-making level. And the fusion's result will be more accurate and less uncertain because of the correlation data coming from most of sensors generally, which provide valuable redundance to make diagnosis of the system. The data fusion techniques as above improve the reliability of the SLTS greatly, and conform to secure demands of the system.This thesis has finished the debugging of the health monitoring and structural damage detection system by java language. The test result shows that the designed system is in reason and up to par.