@Brotfahrer from my perspective you would first have to formulate some sort of plan and maybe put your challenge into a sample workflow that would cover your whole set of problems as to ‘isolate’ them and help other people to understand what you want to do.
First not sure if it has been mentioned but there is a:
If you must use a function like LIKE you could try to restort to a small database and see if that does help (still the planning would be needed to extract the column heads and something like a positive reference of accepted/expected? column names).
This example uses H2 and BETWEEN but LIKE might also be an option.
Then of course there is a wide range of text analytics and similarity search functions and workflows in KNIME - but you might have to tread carfully how to employ such a thing.
And you will have to make certain decisions
my_column_1
my column_1
my column 1
MY_colmnn_a
Are they all the same, would you accept them oder would you just accept ‘extensions’ like _1 _a ? This very much depends on your data and your business case and again that would involve some planning ahead and formulating rules before resorting to some available technical solutions.