An SHM approach unifying data normalization and damage detection is presented. The proposed methodology is based on an adaptive kernel spectral clustering model. An online calibration procedure is devised to adapt the initial hyper-parameters. Two different damage indicators are introduced.