Effects of measurement noise on modal parameter identification

In the past decade, much research has been conducted on data-driven structural health monitoring (SHM) algorithms with use of sensor measurements. A fundamental step in this SHM application is to identify the dynamic characteristics of structures. Despite the significant efforts devoted to development and enhancement of the modal parameter identification algorithms, there are still substantial uncertainties in the results obtained in real-life deployments. One of the sources of uncertainties in the results is the existence of noise in the measurement data. Depending on the subsequent application of the system identification, the level of uncertainty in the results and, consequently, the level of noise contamination can be very important. As an effort towards understanding the effect of measurement noise on the modal identification, this paper presents parameters that quantify the effects of measurement noise on the modal identification process and determine their influence on the accuracy of results. The performance of these parameters is validated by a numerically simulated example. They are then used to investigate the accuracy of identified modal properties of the Golden Gate Bridge using ambient data collected by wireless sensors. The vibration monitoring tests of the Golden Gate Bridge provided two synchronized data sets collected by two different sensor types. The influence of the sensor noise level on the accuracy of results is investigated throughout this work and it is shown that high quality sensors provide more accurate results as the physical contribution of response in their measured data is significantly higher. Additionally, higher purity and consistency of modal parameters, identified by higher quality sensors, is observed in the results.