We can rule out any mechanical mode around 17.7 kHz causing instability. Any mode around 17.7 kHz does not have good overlap with 3rd order mechanical mode. It most probably aliased mode from ~ 47.7 kHz. Interestingly, according to my FEA model of ETM, there are three candidates: 47.767 kHz, 47.811 kHz, and 47.827 kHz mode. These modes corresponds to the mode number #542, #544, and #545, respectively. Mode shapes are attached. All of them have reasonably high overlap factor:  
#     freq, kHz       TEM    Overlap ITM    Overlap ETM  
542    47.767          02     0.028            0.026
544    47.811          02     0.057            0.050
545    47.827          02'    0.026            0.024

'-see figure
The overlap with 2nd order confirms that one of these modes must ring up. The parametric gain for tuned ETM is shown in attached figure. It is not possible to pinpoint  which  of these modes is observed. According to the overlap value there is slightly better chance that this mode belongs to ITM and it's mode #544. However, mode #542 is the closest to the observed frequency (I do not expect to see error of computed frequency larger than 0.1% at 50 kHz range).  Using ring heater we should be able to rule out at leas one of the three modes.