In the prototype stage, usually we only have a few transducers, let's say, 3 transducers. What is the best way to perform statistical analysis in this case? Just don't do it.

So, what can we learn from just a few samples? You can create specifications on critical performance parameters. That is to say, from a small sample size you can understand what you want (or don't want), but you can't determine what you will get.