A novel bridge structure damage diagnosis algorithm based on post-nonlinear ICA and statistical pattern recognition

Monitoring the health of bridges and diagnosing the damage is vital for government and related institutions in Japan because of frequent earthquakes and the oceanic climate. This paper develops a bridge structure health monitoring system (BSHM), which includes data acquisition and analysis. A two-stage structure damage detection algorithm based on post-nonlinear independent component analysis (ICA) and statistical pattern recognition is proposed to analyze the acquired data and evaluate the health of bridges. First, an improved post-nonlinear ICA algorithm is proposed for denoising, and a data-sample matching based data normalization scheme to reduce the effect of varying environmental and operational condition. Thereafter, fast Fourier transform (FFT) is used to detect the damage. Based on the first stage, a statistical pattern recognition damage detection algorithm, including a new damage sensitive index D_SPR, is proposed to determine the severity and location(s) of damage. In addition to the algorithm, this paper presents several simulations and experiments, including a detection experiment that applies artificial damage to a real bridge to show that our design choices are indeed effective.