Neural Networks for Pattern Recognition
Neural Networks for Pattern Recognition
| Hi-index | 0.00 |
Integrating structural health monitoring (SHM) data with reliability analysis procedures provides a novel approach for bridge condition assessment since reliability is an important performance measure of structural condition and reliability-based procedures have the capability of accommodating uncertainties in measurement data. Because the strain response acquired from a bridge under in-service environment is usually a result of multi-load effect such as traffic (highway, railway, or both of them) and wind (monsoon or typhoon), it cannot be characterized by a standard probability distribution model adequately. In the present study, a reliability-based approach to structural condition assessment using mixture distribution models is proposed. With the Weibull distributions being the component density functions, the expectation maximization (EM) algorithm in conjunction with the Akaike information criterion (AIC) is implemented for iterative solution of the optimal number of components and the parameters in finite mixture modeling of peak stresses which are derived from long-term strain monitoring data. Because of using the mixture distribution models, the proposed method is capable of handling monitoring data of any structure and accurately evaluating the reliability indices of monitored structural components. In the case study, the proposed method is applied to assess the in-service structural condition of the deck trusses of the instrumented Tsing Ma Bridge (TMB) under various load combinations such as monsoon, typhoon, with and without railway traffic; and the efficiency of the finite Weibull mixture model for characterizing the statistical properties of peak-stress data with multiple engendering effects and the convergence of the EM-based iteration algorithm for model estimation are validated.