I have data (series data) for two types of faults

I wanted to use the neural network to practice it

Then I test it using new data and the network categorizes the data and gives us whether it is for the first or second damage.

But the problem is that when I test it with false data or not very similar, it classifies it .

How do I make the network not categorize abnormal or unknown data?

Is there a better way to classify this type of data ?

The proposed network below.