It is definitely the exact opposite of this. Even though I understand why you would think this.

The thing with systems like these is they are mission critical, which is usually defined as failure = loss of life or significant monetary loss (like, tens of millions of dollars).

Mission critical software is not unit tested at all. It is proven. What you do is you take the code line by line, and you prove what each line does, how it does it, and you document each possible outcome.

Mission critical software is ridiculously expensive to develop for this exact reason. And upgrading to deploy on different systems means you'll be running things in a new environment, which introduces a ton of unknown factors. What happens, on a line by line basis, when you run this code on a faster processor? Does this chip process the commands in a slightly different order because they use a slightly different algorithm? You don't know until you take the new hardware, the new software, and the code, then go through the lengthy process of proving it again, until you can document that you've proven that this will not result in any unusual train behavior.