Classifying Linear and Nonlinear Structural Damage Using Frequency Domain ARX Models

Abstract

Structural health monitoring can be viewed as a problem in statistical pattern recognition involving operational evaluation, data cleansing, damage identification, and life prediction. In damage identification, damage features derived from available input-output and output-only time and frequency data are used to detect, locate, and quantify damage in structural dynamic systems. A new set of damage features and their implementation for damage detection and quantification are discussed in this article. These features are the autoregressive and exogenous coefficients in a frequency domain data model and can be used to distinguish between linear and nonlinear types of damage. In this work, autoregressive coefficients are used to characterize nonlinear damage and exogenous coefficients are used to characterize linear damage states. The ability to distinguish between linear and nonlinear types of damage and healthy system nonlinearities is critical when diagnosing structural health because damage initiation and growth are fundamentally nonlinear processes. It is shown that absolute damage severity can sometimes be determined solely from the degree of linearity-nonlinearity in the system. Experimental data from a three-story building model is analyzed using these features and some important application issues are discussed.