Signal pattern-recognition for damage diagnosis in structures

Abstract

A signal-based pattern-recognition approach is used for structural damage diagnosis with a single or limited number of input/output signals. The approach is based on extraction of the features of the structural response that present a unique pattern for each specific damage case. In this study, frequency-based features and time-frequency-based features were extracted from measured vibration signals by Fast Fourier Transform (FFT) and Continuous Wavelet Transform (CWT) to form one-dimensional or two-dimensional patterns, respectively. Three pattern-matching algorithms including correlation, least square distance, and Cosh spectral distance were investigated for pattern-matching. To demonstrate the validity of the approach, numerical and experimental studies were conducted on a simple three-story steel building. Results showed that features of the signal for different damage scenarios could be uniquely identified by these transformations, and suitable correlation algorithms could perform pattern matching that identified both damage location and damage severity. Meanwhile, statistical issues for more complex structures as well as the choice of wavelet functions are discussed.