Does omitting leading principal components fully remove EOV influence?

Determine whether the implicit unsupervised principal component analysis strategy that removes environmental and operational variable effects by omitting the leading principal components can completely mitigate the influence of environmental and operational variables (e.g., temperature variations) on system outputs in structural health monitoring.

Background

In structural health monitoring, principal component analysis is widely used for data normalization, with a common practice of excluding the first few principal components under the assumption that they capture environmental and operational variability such as temperature. However, selecting how many components to omit is nontrivial, and it is unclear whether this implicit strategy truly eliminates environmental effects.

The paper advocates a supervised, conditional PCA approach that explicitly models conditional covariances with respect to measured confounders, arguing that this yields better removal of environmental influences. Nevertheless, the authors explicitly note that it remains uncertain whether the conventional unsupervised approach of omitting leading principal components can fully mitigate the effects of environmental and operational variables.