---

The presence of environmental and measurement noises and ignoring the input effects are the main sources of error in system identification using ambient vibration test results. Therefore, reducing uncertainty or noise levels from the records has always been one of the main goals of the new techniques in the field of ambient vibration. Among the modal analysis techniques, stochastic subspace identification is considered as a powerful technique. In this study, the modal analysis method based on canonical correlation analysis in stochastic subspace is presented that identifies dynamic properties in optimized space instead of data space by extracting ortho-normal vector of data space. The advantage of this method, due to the nature of canonical correlation analysis, is lower noise which results in greater accuracy in estimating modal properties. Moreover, the presented process is faster due to the smaller space of identification compared to the previous methods. To validate the proposed method, an analytical model of two-dimensional frame excited under Elcentro earthquake acceleration and also the results of ambient vibration tests carried out on the Alamosa Canyon Bridge are used. The results indicate that this method eliminates more noise than other subspace methods and moreover it is faster in solving practical problems. The computation of dynamic properties, natural frequencies and mode shapes, of Alamosa Canyon Bridge with 30 sampling sensors, space matrix size of 750 and 50 excited modes are carried out in less than 150 seconds with a quad-core 2.30 GHz processor.

---

Finite element model is the conventional method used for static and dynamic analysis of widely used structures such as dams and bridges, since it is cheap and requires no special tools. Nevertheless, these models are not able to describe the accurate behavior of structures against dynamic loads because of simplifying assumptions used in numerical modeling process, including loading, boundary conditions and flexibility. Nowadays, modal testing is used to solve these problems. The dynamic tests used to identify civil structures’ system usually include forced, free and environment vibration tests. Considering either unknown nature of inputs or failure to measure them, some methods have been developed to analyze the results of dynamic tests which are based on measuring only output data and are known as operational modal analysis. Some of such methods are Peak Picking (PP), Frequency Domain Decomposition (FDD) and stochastic subspace methods. However, unknown nature of applied forces, the presence of environmental noise and measurement errors contribute to some uncertainties within the results of these tests. In this article, a modal analysis is presented within a stochastic subspace which is among the most robust and accurate system identification techniques. In contrast to the previous methodologies, this analysis identifies dynamic properties in optimized space instead of data space by extracting ortho-normal vector of data space. Given the optimum nature of the proposed method, more accuracy in detection and removal of unstable poles as well as high-speed analysis can be served as its advantages. In order to evaluate the proposed method in terms of civil systems detection, seismic data (being among the most real and strong environmental vibrations) and steady-state sinusoidal excitation (which is among the most precise forced vibration tests) were used. In the first step, 2001 San Fernando earthquake data were analyzed using SSI-CCA and SSI-data methods, the results of which are presented in the following. Data processing rate in the SSI-CCA method is almost twice that in SSI-data method which is because of processing in an optimum space while lowering the use of least squares method to compute system vector. Furthermore, there is one unstable pole in the results of the proposed method while 4 noisy characteristics were recognized in the results of SSI-Data method. Estimated damping ratios comprised the major difference observed in this analysis using above-mentioned two methods. Modal damping ratios estimated by the proposed method were 60% closer to the previous results when compared to those of the previous subspace method. Mode shapes of both subspace methods with MAC value of 92% and 75% for the first and the second modes, respectively, are well correlated with each other. Due to lack of access to the mode shape vectors of Alves’s method, it was not feasible to calculate the corresponding MAC value. In the following, forced vibration test results of Rajai Dam conducted by steady sine excitation in 2000 and analyzed by a method known as four spectral, are re-processed Using the SSI-CCA method. As results indicate, using the proposed method the first three modes are obtained that were not on the preliminary results. In addition, other modes are of great fit with the values of the finite element.