How to write good expression to check if any of few tags have Bad or N/A value??

What problem are you trying to solve or what end goal are you trying to accomplish? 

What would be the output of that analysis if it were to work? 

Generally, the end result will be a control lamp on the Pi Vision screen, which signals that among the tags included in the analysis, there are those with the value "Bad" or N/A.

I don't think an analysis is necessary to do this. In PI Vision you can set some multi state symbols to give you this information.

If you want to pursue this in an analysis, can you share the configuration of the table lookup data reference (MP5_SoSe_Input_Tags)? 

You can get this to work (depending on the TLDR configuration) using a similar formula: 

MapData('Attribute1', If BadVal(TagVal(TagNum($val),'*')) then 1 else 0)

Thanks Sebastien for hint but for some reasons I can't use the table name:

Can you tell me what am I doing wrong??

You can't access a table directly from inside an analysis. You will have to use an attribute with a table lookup data reference and an array value type to store all the table values. As I mentioned previous, I don't believe this is necessary to achieve your end goal, and if the table is quite large you may run into some performance issues. 

Many thanks Sebastien! No I am much much more close to achieve my goal!

Right now I get the array with just 0 and 1 values.

Now I want to look into table and if I found just one "1" value I want to save biniary value into Noinput attribute.

You can also use a dedicated AF Element & Attributes (instead of a table) + Rollup Analysis to count the number of Good values (Assuming all tags contain Numeric values !). As Sebastian already wrote in his post, you can use a multi state symbols in PI VISION to easily reflect the "cumulative" quality status based on the analysis results .

For example:

I agree with the other commenters since hard coding tag names into analyses and large internal table queries are generally not a scalable ways to use AF, but if you are set on using the approach above, you can use the Total function on the array to do an event-weighted sum of the values or the Max function to find the maximum value in the array.  The total would tell you how many bad values you have, whereas the maximum would tell you whether or not you have any bad values.

This is a much better solution in my opinion. Much easier to manage, and less heavy on the PI Analysis Service which is good if the system needs to scale to a large volume of analyses.