Practical Clustering Approaches for SHM

The evolution of the dynamic properties of a structure over time can reveal significant insights into its health state. This is a main assumption in modal based Structural Health Monitoring (SHM), and it can be achieved by applying automated Operational Modal Analysis procedures. However, these come with substantial challenges, including handling noise and inherent measurement errors in the data, managing non-stationary dynamics due to varying operational conditions, addressing mode mixing when modal frequencies are closely spaced, working with limited data, and dealing with the complexities resulting from structural nonlinearities. As a result, the success rate in modal parameter identification is inevitably affected, and spurious or missing estimates can come with the right ones. Clustering techniques offer a practical solution to discriminate between actual modal parameter estimates and spurious estimates in the context of automated OMA and modal parameter tracking, and to automatically clean the resulting time series for SHM purposes. An approach based on DBSCAN is here proposed and applied to a real dataset for validation purposes.