In this work, we propose a data-driven method to perform detection and quantification of damage in guided wave data from Structural Health Monitoring. The data is from a full-scale panel mimicking a critical region of a Floating Production Storage and Offloading storage tank. For this, signals were initially recorded from the structure in the pristine condition to create a baseline database. Subsequently, different damage levels were introduced in the panel; we took new measurements for each damage level. Our data-driven method consists in performing a dimension reduction in these datasets using independent component analysis. The weights vectors of the independent component analysis were used to identify if a given signal came from a pristine or a damaged condition through a procedure involving statistical correlation. We obtained average accuracy rates above 90% for damage detection, with average Type I-error rates below 10% and average Type II-error rates below 35%. In addition, it was possible to quantify the severity of the damage at all depth stages.