Regarding the natural frequencies are depends on the analytical study or you can identify them by a software of simulation of these frequencies, also you can confirm these frequencies drawn by the acquisition of the sensor by looking directly on the resonance peaks characterized by high amplitudes.
Just to add a soft distinguishing, I would like to remind you that peaking frequencies are slightly different from the true resonant frequencies associated to an eigenmode. This because in the real world you might have very tight modeshapes that participate to a common peak; only modal analysis can identify the underneath modeshapes. Please have a look at my works about optical measurements, in particular the latest Article Full field optical measurements in experimental modal analys...
Mode shapes and natural frequency can be derived not only with hammer excitation but also using also shaker excitation, plucking (a string with damage is pulled until it snaps) or Operating Modal Analysis.
You find examples of most of these techniques (not plucking) here
It depends on what sort of structure you are working on. If the structure is a simple one, impact hammer testing is a suitable choice. But if you work on a large complex structure, using shaker excitation may be the best.
Anyway, the way of setting up the structure in experimental modal analysis is a crucial part, especially, when you are working on a structure made of thin metal sheets.
You could try and find materials about 'Modal Analysis'. It's has been studied for decades and there are a few great textbooks focuing on that. Those materials should answer your question in terms of the theory and the experimental procedure.
Measuring the natural frequencies of components/sub-structures/structures is relatively easy but measuring accurate mode shapes of those things is very challenging and requires certain particular expertise. Usually, FE models are required for guidance on the measurement.
Beware of the complexity of your part and its modal content. When the overlapping of natural modes and its similarity is high, better use experimental modal analysis with global approaches. This because a single peak might be participated by different eigenmodes. You might find tough examples in the output papers of my TEFFMA projects or related researches: https://www.researchgate.net/project/project-FP7-PEOPLE-2011-IEF-298543-TEFFMA-Towards-Experimental-Full-Field-Modal-Analysis
Therefore, test your part, because any FE model might be affected by errors, to be tuned on your specific part.