Spatial Steel Structural Damage Alarming Based On Statistical Pattern Recognition

Large spatial steel structures, usually constructed for the landmark stadiums and exhibition centers in the cities, have large sizes, unique shapes and novel designs. To ensure safety, comprehensive health monitoring and damage alarming are needed for the large spatial steel structures. This paper studies extracting structural damage-sensitive index from the acceleration time-histories, separating the influence of environmental changes and monitoring signal noise, and finally identifying the structural work status statistically. The detailed research plan is as follows:First, the study looks into the establishment of ARMA models by the acceleration time history. By theoretical analysis for the fit order of ARMA models, the order is decided by the number of vibration modes. A method based on genetic algorithm is used for ARMA model order determination. First, determine the scopes of ARMA model order according to the order of AR model, then use the genetic algorithm to search the fit order of ARMA model. The numerical example of a K8-type gridshell shows that the proposed method reduces the computation and then fastens the determination of the ARMA model order.To feature the structural work status, the structural damage-sensitive indexes such as the AR coefficients and structural impulse responses are investigated respectively. The corresponding methods of statistical discriminate technique are proposed based on principal component analysis and hypothesis test. The numerical examples of the simply supported beam and the K8-type gridshell show that, the AR coefficients used as the structural damage-sensitive indexes can alarm slight damage of the beam and relatively serious damage of the gridshell structure. The impulse responses used as the structural damage-sensitive indexes can effectively alarm slight damages of the gridshell structure.Based on the numerical model of Water Cube structure, the study investigates the robustness of the impulse responses to monitored signal noise. The numerical example shows that the proposed index can alarm the damage of a certain degree with good robustness. To separate the environmental influence, the method based on support vector machine regression is used to function the impulse response of temperature and wind speed. Finally, support vector machines are trained and training residual errors are used to establish statistical process control charts.