methodology to use to do impact analysis in a distributed system?
When in doubt, try it out.
Stop developing in production. Make your changes first on a development build. Watch what actually happens when you update in a safe place that isn't going to bring down the system.
This will be a fair bit of work but you'll find it paying off again and again as every change becomes less and less of a nail biter. You'll need to learn how to deploy to production in a reliable way but once you master that this wild beast will seem more tame.
As for your other points:
1 first this documentation will not be synchronized with the code.
Never assume any documentation is synchronized with the code. Documentation is not a picture of what is happening. It's war stories told of days gone by. Use it to get into the head of the designer not to check on what is or will happen. For that look at the code.
2 some data is mutated during its life in the system (ie. concatenation of various data to make a unique id etc...)
You mean like a composite key? That's not that weird and I don't see how it's relevant. It doesn't surprise me that data changes. Just means your system isn't immutable. Not sure what your concern here is.
3 I cannot automaticaly filter my excel file and found all the application using a certain datatype and how...
This bit flat out scares me. I've seen people capture their ignorance by thrusting tons of effort into making massive impressive spreadsheets that clearly show the amount of work that's been done yet doesn't do a single useful thing.
It's not that spread sheets are useless. But they can be a deep rabbit hole to get lost in. Any spreadsheet thats usefulness takes more than 5 minutes to explain to others is not a tool. It's a time consuming hobby.
I started to do some work in excel but I am feeling that this is a no go
Since you haven't been able to clearly explain what this work is I'm forced to agree with you.