Structural Abnormality Detection Using Statistical Pattern Recognition Technique

A new method to detect structural abnormality is proposed in the thesis by using statistical pattern recognition technique. Two statistical patterns are firstly established based on the measured data come from normal structure and abnormal structure respectively. Then two statistical patterns are distinguished from the most effective differential standard indices to achieve the goal of effectively detecting structural abnormality as well as deleting the influence of the environment. The most important issues, such as system order determination of ARMA model, accuracy of parameter estimation, deduction of system parameters and selection of the most effective differential standard indices, have been extensively studied. The main work and conclusions of this thesis includes: 1. After reviewing of the theoretical background and physical meanings of time series models and considering of the feasibility of statistical pattern recognition, ARMA model is chosen to use as the statistical pattern. 2. A computer program is developed to determine the system order of ARMA model based the minimum eigenvalue method. A two-stage based method is presented to estimate the accuracy of model parameters. The program and proposed method have been validated by both numerical example and tested beam. 3. Principal Component Analysis is proposed to carry out the effective curtailment of the multivariables where a new variable maintaining the most important information of the primary multivariables is selected, so that therefore a judgment on the status of the structure can be made solely and swiftly. 4. Three indices are studied to represent the difference between statistic patterns. Comparison has been performed among these indices through the results of both numerical simulation and experiments on simple beams. It is demonstrated that the mean value control chart is a good and sensitive index to detect the discrepancy between statistic patterns.