Nonlinearity detection using new signal analysis methods for global health monitoring

Statistical pattern recognition has recently emerged as promising and effective set of complementary methods in structural health monitoring to assess the global state of structures. The aim of this paper is to detect nonlinearity changes resulting from damage by some efficient signal analysis methods. The primary idea behind these methods is to use raw measured vibration time-domain data without applying any feature extraction technique associated with the statistical pattern recognition paradigm. Firstly, statistical moments and central tendency measurements are applied as damage indicators to determine their changes due to damage occurrence. Subsequently, cross correlation and convolution methods are used to measure the similarity between the vibration time-domain signals in the undamaged and damaged conditions. The main innovation of this study is the capability of proposed signal analysis methods for implementing the nonlinear damage identification without extracting damage-sensitive features. In the following, numerical and experimental benchmark models are employed to demonstrate the performance of proposed methods. Results show that nonlinearity changes lead to a reduction in the values of cross correlation and convolution methods caused by damage. Moreover, some of the statistical criteria on the basis of the exploratory data analysis are applicable tools for the global structural health monitoring.