Chapter 14 - Signal Processing and Pattern Recognition for Structural Health Monitoring with PWAS Transducers

Abstract

This chapter examines two aspects of considerable importance for the structural health monitoring (SHM) process: (i) signal processing and (ii) damage identification/pattern recognition algorithms. The SHM community is constantly exposed to various aspects of signal processing, spectra collection, data processing and analysis, pattern recognition, and decision-making. This chapter reviews some of these aspects. Both the state-of-the-art and general principles, on one hand, and specific examples, on the other hand, are presented. However, an exhaustive presentation of these vast topics is not attempted, since it would be beyond the scope and possibilities of a single chapter. The signal-processing subject is described in terms of two major algorithmic paths: the short-term Fourier transform (STFT) and the wavelet transform (WT). Both these approaches fall into the larger class of time-frequency analysis. The damage identification/pattern recognition process is discussed in the context of the analysis of spectrum features (e.g., resonance frequencies, resonance peaks). The existence of adequate algorithms would allow one to classify the spectral data into classes according to the damage state of the structure. An elementary classification problem would consist of distinguishing between “damaged” and “pristine” structures. More advanced techniques would allow classifying the structure’s health by assessing the progress of damage based on its severity and identify the damage location. Neural nets (NN) are presented as efficient classification algorithms. In particular, probabilistic neural nets (PNN) are discussed in some detail for SHM applications. Specific examples are presented and discussed in the context of the electromechanical (E/M) impedance method, which permits the direct determination of the structural modal spectrum at high frequencies.