Large optical telescopes such as the revolutionary Gemini South are subject to numerous loads, but the most critical are those due to wind effects. These cause structural vibrations that degrade the overall performance. In order to dynamically characterize the system and optimize the controller design, Dr. Peter Avitable of the University of Massachusetts traveled to Chile to perform both operational and modal tests.Test Setup
The telescope was equipped with 75 accelerometers arranged on the primary and secondary portions of the structure along with 24 pressure transducers on the face of a dummy mirror. An 88 channel LMS Roadrunner was used for data acquisition and LMS CADA-X TMON for data reduction. Since almost 1km of cable was used to reach all of the sensor locations on the telescope, the first step was to perform a diagnostic test to assure that all channels were measuring data correctly.
Data collection
Test data was collected in individual tests for more than 50 configurations of the telescope system. The majority of the data collected was operational data due to various wind velocities and geometry configurations of the telescope. Several tests were also conducted using an impact excitation for the development of FRFs for modal characterization. Time datawas acquired at 200Hz (10 times higher than most of the expected frequencies to be studied) and written directly to the Roadrunner hard drive.
Impact testing
Since excitation at the base of the structure was not possible, an impact excitation at the extremity of the telescope (where the structure is most flexible) was considered the logical alternative. This approach assumed that the excitation at the flexible extremity of the system would naturally tend to excite the modes of the system. Impact locations were selected at two points in the two horizontal directions at the top of the secondary portion of the telescope. A series of consistently placed impacts spaced close to 20 seconds apart were measured for 25 repetitions, resulting in 500s of data on each of the 75 channels.
The impact tests described above were performed during moments when the wind had subsided sufficiently that collection of operating data was not expected to produce any meaningful results. However, when it was windy, operating time data was acquired for about 300s, with the telescope in a variety of configurations.
On-site data reduction
In order to make certain that reasonable measurements were acquired, one impact data set was reduced so that the extracted spectra and resulting FRS were of sufficient accuracy to continue with the collection of additional data. In general, the measurements were reasonable but not of the quality needed for extraction of modal parameters. Upon review, very obvious "background" effects could be seen within the measured data, caused by the normal plant activities and people walking on the structure during the measurements. Another very important task that needed to be performed using the preliminary processed data was to quickly reduce the measured data using simplistic peak picking techniques to ensure that all the channels were configured properly prior to collecting additional data. This quick data reduction and assessment is an extremely important step of the acquisition process. If there are any problems, they can be quickly resolved before collecting additional data. Having guarded for the problems detected by the preliminary investigations, the data collection proper could start.
Operational and modal analysis
The time data that was acquired and saved on disk was first inspected for any inconsistencies; data channels were either adjusted through filtering techniques or removed from the data set. Operating data reduction could then proceed using cross-spectral measurements relative to a number1of references - selected where significant response was noted for the majority of the shapes to be extracted. Only the tests with the most significant response were used for the initial investigation of the operating modes.
Operating modal deflections were computed for a number of frequencies. The majority of these frequencies coincided with the frequencies observed from the modal tests, as expected. Using a standard peak picking approach, operating deformation patterns were obtained for a number of frequencies and for a number of different references. A significant amount of additional processing was performed using singular value decomposition techniques to assist in the identification of reference locations and the extraction of operating modes. As expected, the telescope’s operating deflections consisted of the typical bending, nodding and torsion modes.
The modal parameter estimation process was performed using FRFs computed from the impact time data. While many references were available, not all were used for the extraction of each of the poles of the system. The major modes of interest in the telescope were extracted from the measured frequency response data. Again, the modes consisted of the typical bending, nodding and torsion modes expected in this type of structure.
Correlation of modal and operating data
The experimental mode shapes and the operating modes were correlated using the MAC function. Again, many different tests and configurations were evaluated and the results of one typical case are shown. There was good agreement between the modal data and the estimated operating modes of the telescope.
Observations
A good deal of effort was expended in the measurement and reduction of data for the Gemini Optical Telescope. While some data reduction was performed on-site to assure that adequate measurements were being made, the majority was reduced long afterwards. The use of captured time data permitted the more in-depth evaluation1of the data collected. If all the data collected was processed so as to capture frequency averaged data at the site (without having the time data available), then there would be no possibility of further processing the data in a multitude of different scenarios. Having the possibility of processing the time data in a variety of ways allows for tremendous flexibility, which would not be possible if only averaged frequency data was obtained. LMS thanks Dr. Avitable for his help in reducing this article from a much more detailed paper.
