ON THE USE OF STATISTICAL PATTERN RECOGNITION METHODS FOR STRUCTURAL HEALTH MONITORING

摘要

In this paper, signal processing based damage detection algorithms proposed by the authors are compared. These algorithms involve local signal processing at the sensor level. The algorithms are based on the time series analysis of vibration data, and the features obtained are classified using pattern classification techniques. In two algorithms, the vibration signals are modeled as auto-regressive moving average (ARMA) processes whereas in the third, the continuous wavelet transform (CWT) is used to decompose the vibration signal. The feature vectors used in the algorithms are the first three auto-regressive (AR) coefficients and the energies of the transformed wavelet coefficients. In the first algorithm, a damage index is proposed based on first three AR coefficients. In the second and third algorithms, a Gaussian Mixture Model (GMM) is used to model the feature vector. A classification method for damage detection is obtained using the gap statistic, which determines the optimal number of mixtures in the GMM. For damage extent, the Mahalanobis metric between the baseline and damaged mixtures has been used. Also, it is seen that the angle between the subspaces formed by wavelet coefficients before and after damage, correlates well with damage extent. Application cases from the ASCE Benchmark Structure simulated data have been used to test the efficacy of the algorithms.