Abstract
We examine three research topics to infer reliable Green's functions (Gfs) of vibration of torsionally coupled buildings. The first is waveform deformation of Gfs associated with rotation due to the lack of low-frequency content in incident rotational motions. To correct it, we propose two methods: an addition of band-cut-filtered uncorrelated noises to the incident rotational motions and weighting zero-lag autocorrelation of the incident rotational motions. A comparison of Gfs corrected by the two methods revels that the addition of the noises is better than the weighting. The second is the effect of wind on building vibration. A temporal fluctuation of building vibration is much more similar to that of incident microtremors than a temporal fluctuation of wind. Furthermore, Gfs inferred from microtremor vibration acquired during a time span of low-level wind blow and high-level incident microtremors are almost same as those inferred from microtremor vibration acquired during a time span of the reverse. These results mean that wind does not affect the Gfs under condition that a mean wind velocity is less than 2m/s, and incident microtremors are more than 0.002 m/s2. The third is data-length to infer reliable Gfs. Non-dimensional averages of root mean square error of Gfs inferred from 3- hour-length data reduce to one third of that of Gfs inferred from half-hour-length data, which suggest that we use data-length above 3 hours to infer reliable Gfs. Practically, it is recommended to use microtremor recordings acquired in the nighttime from 23:00 to 5:00.
