Machine-learning-driven automatic application of the stochastic subspace identification method

Vibration-based operational modal analysis (OMA) methods have been proven effective in identifying dynamic properties of existing structures and infrastructures under operational conditions. Nevertheless, the provision and installation of continuous monitoring systems for long-term structural health monitoring (SHM) purposes potentially applicable to the entire infrastructure networks or to the regional scale of existing vulnerable building heritage require significant economic planning efforts. Nowadays research trends are oriented toward developing effective automatic OMA (AOMA) methods for setting up novel and efficient long-term SHM solutions. The current study illustrates a new recent paradigm for the automatic output-only modal identification of linear structures under ambient vibrations called intelligent automatic operational modal analysis (i-AOMA). The proposed approach relies on the covariance-based stochastic subspace identification (SSI-cov) algorithm and effectively integrates a machine learning intelligent core, i.e. a random forest (RF) classifier, in a conceptually two steps procedure, i.e. an explorative phase and an intelligently-driven phase. The i-AOMA procedure provided a new framework that requires a minimum intervention to the user and is potentially able to deliver uncertainty measures of the modal parameters' estimates based on the explored SSI-cov control parameters. An application on a shear-type RC frame building typical of existing heritage in Italy is herein discussed and reported.