A latent variable approach for mitigation of environmental and operational variability in vibrationbased SHM – A linear approach

One of the major challenges in the development of vibration-based structural health monitoring is the influence of environmental and operational parameters (EOPs) on damage-sensitive features (DSFs). With multiple methods already available in the literature, a challenge remains on the development of robust compensation methods when EOPs are partially measurable and noisy. Considering this limitation, this study proposes a novel environmental and operational variability (EOV) mitigation procedure based on the estimation of latent factors from available EOPs and DSFs to extract a comprehensive baseline model of DSFs. The obtained latent factors contain information on measured and un-measured sources of variation that can be used to infer noise-free estimates of the observed DSFs. In turn, the difference between measured and predicted DSFs is used as a robust damage indicator. In this work, the proposed methodology is applied in the form of an encoder-decoder architecture implemented via Principal Component Analysis (PCA) on Power Spectral Density (PSD) spectra extracted from vibration responses of a small-scale wind turbine blade. Two variations of the proposed method are considered, the first one considering only DSFs, and the second adding the ambient temperature as the single EOP. Our analysis demonstrates that the extracted latent factors are in both cases correlated with the temperature, and that the perfect damage detection performance can be achieved upon careful selection of the number of latent factors.