Lay a sheet of paper flat, spanning an open space between two stacks of books. It will sag in the middle.
Now fold the paper longways twice, making three flat planes, dividing the sheet into three sections, two vertical, the center one flat, U shaped in cross section.
Now it will span the gap with much less sag. The side verticals stiffen the deck, much like beams stiffen the deck of any bridge. When you fully understand why this works, you understand why a load carrying beam cannot be curved when spanning an open space between two supporting piers or columns. The beams have to be straight.
From there, all else follows. You can have straight beams and a curved deck, or straight beams and straight sections of deck, each successive section between piers taking an angle with respect to the previous section, but either way, the beams have to be straight.
That said, you can heat up a styrene Warren truss model in hot water or with a blow dryer and flex it around a curve (with the deck detached), and it may carry the weight of model trains, but it is not prototype, and I don’t recommend doing it to support models you care about.
If there’s an exception to every rule, then two trusses or plate girders curved between piers, could concievably be stabilized by using frequent X bracing in the cross section, to hold the beams immovable with respect to each other, and by anchoring all beam/pier attachment points against both downward and upward thrusts and loads, but again, I do not recommend the practice. Keep the beams straight, and you and your models will be a lot happier.