do you use puls-echo or transmission mode? Share wave or long. wave? Is there an acoustic standoff/delay?
Assuming that you know the velocity of sound in your medium (e.g. long wave 3700m/s) and thickness of your probe you can calculate the time for wall echoes.
If you use shear wave (appx. 2250m/s) then angle of diffraction has to be considered.
If the signal in your attached file doesn`t contain the wall echo, then the first signal at appx. 500µs appears to be the entry signal into your medium.
The repetitive signals (path: transducer-crack-transducer-crack...) have a distance of 1000µs towards each other. And we see double packets of signals with 180° phase shifting. Phase shifting probably caused by reflecting at the crack.
You should try cross correlation with a reference signal to achieve better results.
The attached signal sims to be a CODA wave wich is a resultant after multiples scattering at the surface of the transucer. It contains shear waves and preassure waves. You should read about "Coda wave interferometry" technique to detect damage.
The short answer for your question is: you should use the whole signal.
Considering a coda wave, the signal contains randomly localiced information about the structure. Normaly to detect damage the people use 2 signals one after and one before of some small strultural change. An then they compare them using the coda wave interferometry to localize the damaga or the structural change.