6 A. delli Carri et al. (A) chan#1 (L) chan#30 10-10 10-5 10-10 1 500 1000 1500 2000 frequency [Hz] 3000 4000 5000 3500 0 γ2 xy γ 2 xy 1 0.5 0018mV 0180mV 1800mV 4500 2500 0 500 1000 1500 2000 frequency [Hz] 3000 4000 3500 0 1 0.5 4500 5000 2500 |H| |H| Fig. 1.7 FRFs and Coherences of LDV point 1 and LDV point 30 -4 -2 0 0 10 20 30 25 15 realisation# 5 0 10 20 30 25 15 realisation# 5 m s 2 (A) stationarity plot ×10-3 4 10 8 6 4 2 Fig. 1.8 Accelerometers stationarity plots If gaussian input voltage is provided, a gaussian output force is expected from the shaker. This is not generally the case due to shaker-structure interactions. A gaussian time history passing through a non-linear system always generates a nongaussian output [1]. Stationarity checks are performed to assess the quality of the data and determine if the system is changing over time (the structure might be settling on supports or getting to the operating temperature). The stationarity plots for both accelerometers and lasers are found in Figs. 1.8 and 1.9. By inspecting the mean and standard deviation of the channels, it can be observed that the accelerometric channels are quite stationary in mean, but the structure had to settle at around the 15th realisation for a stable standard deviation. The laser means and standard deviations are more scattered around an average. Since the laser channels were not captured at the same time it is complicated to discern if the ensemble data is ergodic or even stationary, as some channels might have been captured before the structure had any time to settle. Since the quality metric chosen is the multiple coherence, the data was sanitised by detrending. This operation has no impact on coherence, that only measures the input-output causality, but will hinder any calculation and retrieval of frequency response functions.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==