A Group of Deterioration Curves of Bridges Estimated by Extended EM Algorithm

Abstract

This study proposes a method to separate data into several groups and perform regression analysis for each group at the same time statistically by using EM (Expectation and Maximization) algorithm or its extended formulation to nonlinear regression problem. With using the proposed method, a group of deterioration curves is estimated based on database of health index of 341 bridges by visual inspection in Gifu prefecture, Japan. Ordinary EM algorithm can treat linear regression only, so that the algorithm is extended to nonlinear regression problems. In our proposed method, the hidden parameter and standard deviation of error are estimated following ordinary EM algorithm, whereas nonlinear coefficients are estimated by using Particle Swarm Optimization, which is one of global optimization methods. Deterioration curves of the four bridge groups are estimated with the inspection database by the proposed algorithm.