On the use of kernel PCA for compensation of environmental effects on natural frequency estimates

Modal based Structural Health Monitoring (SHM) systems detect damage or degradation phenomena from stiffness changes in the monitored structure. However, experimentally estimated dynamic properties are also influenced by environmental and operational variables (EOVs) in different, often non-linear ways. The use of kernel PCA is here proposed to compensate the environmental influence on dynamic parameters. In particular, a procedure for compensation of the EOV influence on natural frequencies based on kernel PCA and providing residues in the input space is discussed. Simplicity and computational efficiency are the main strengths of the method. As an additional advantage, it also makes possible the automatic selection of the user-defined parameters in kernel PCA. The basics of the method and results from a sample application are reported, pointing out its promising applicative perspectives.