I'm trying to laminate a carbon part that's essentially a Y-shaped tube where the outside surface needs to be a dimensionally defined mold surface - inside dimensions don't matter much. Ideally it would be made with prepregs but that's not an option for me at this time so I'm trying to come up with an alternative method. Initially I was planning to make it by wrapping and wet laminating a printed water soluble core and then clamping that between two female molds (also 3d-printed). I have made one project with this method before and it worked surprisingly well but it's a bit tricky to pull off - the main challenge is getting an even laminate thickness so there is compression everywhere while still being able to close the female molds without any carbon being pinched in between them.

So I came up with this idea that's illustrated in the attached image: the soluble core is wrapped in dry fibers and closed inside two female molds. Everything is then placed into a vat of epoxy and vacuum is pulled at the top of the mold hopefully infusing the dry fibers with resin. Would make layup a lot easier  and less messy if it works. Molds breaking during demolding isn't an issue, this would be a one-off part. Has anyone here tried something like this before, is it feasible? I haven't done any infusion work, just wet layup stuff with vacuum bagging, so there might be some problem I'm not seeing. Alternative manufacturing method ideas are welcome too!

There's potential, but I can see so many issues already. The biggest one is that infusion needs to remain under vacuum until full cure. If you only have the resin bath to create the vacuum chamber, then the resin is either just going to be drawn through the system until it is empty and you lose all vacuum, or it is going to gel around the mould and leave you with a nice diorama of the 3D model you've drawn... 

It is also a massive waste of epoxy. You would only need a fraction of what was in the resin bath to wet out the carbon, the rest goes in the bin.

If your internal dimensions aren't critical, why not just vac bag the pipe? Pre-pregs are nice and would make your life easier, but they aren't necessary, you could easily achieve the same thing with wet-lay vac bagging, you just likely won't get the perfect surface finish straight from the mould. You could also infuse it, but I would recommend against this - removing the consumables from the inside would be a hair short of impossible.

There is a inherent problem with vacuum infusion between rigid walls and my best explanation is a bit hand wavy, but i never managed to get a good void free part, without pressuring the resin as well. So this would then be light vrtm but printed moulds will not handle the pressure. 

I disagree with the last post.
Printed mould will easily handle the pressure, as long as the filler pattern is thight enough and the wall thickness is sufficient. Remember that the vacuum does not press things like a mechanical press would do. It just holds things together by removing the medium between them. The pressure on the part is the same as without the vacuum. This does not count for the inside of the mould, where a vacuum cavity has to be sustained. Here the usual 10t/m^2 apply from the inside of the mould's filler cavitys. So keep the relatively small and it should be fine. One thing to look into might be variable infill rate.

However, your resin bath idea won't work, because you can't build up vacuum before the resin starts to flow. So the will be resin flowing through partially evacuated reinforcement, resulting in dry spots and pinholes.
Just bag the mould and place spirals on each exit. keep the flow rate very low and you should get a perfect result. I made several different y tubes exactly that way.

I was referring to the printed moulds not withstanding typical pressures of 2...3 bar of pressurized resin when working with vrtm, which is not the same as the hydrostatic pressure of a vacuum bag.

Sorry, i misread your post. Light RTM is processed without pressure on the resin, wich is the difference to regular RTM.
Yes, overpressure will be difficult for printed moulds. You could do it, but those moulds would not fit well with the lost plug idea. But you can get good results without pressuring the resin.

Thanks for the help, makes sense that you have to first suck the air out and then let in the resin. I think I'll try oekmont's suggestion and bag the molds and infuse it from the bottom. I'm going to try with a small test section first. I have some SP115 epoxy (non infusion resin) already past it's expiry date - I wonder if that would work or is the low viscous infusion resin a must for any kind of infusion application?

Yes, low viscosity is crucial for good fibre penetration. But most regular epoxys are not that much more viscous than infusion resin and work ok. Middle viscous resin won't work.

If the mixed viscosity on the data sheet is higher than about 550cps, it wont work well. Infusion resins are around 150-300cps, low viscosity hand laminating resins are about 400-600cps, and regular laminating resins are usually above that and won't infuse well.

I made a test piece during the weekend. I printed the female molds from PLA and the core from PVA and instead of spirals I printed a part witch resin channels guiding the flow from the resin hose to the two bottom sections of the part. My Sp115 resin seemed to work fine - to make it more viscous I heated the mixed epoxy in a warm water bath. The trial was partly a success: the inside surface turned out perfect but the outside didn't. The surface areas where the print layer imprints are visible worked as intended but the majority of the surface didn't. I'm not completely sure why - one reason was probably because I didn't use enough fibers so the core was really loose inside the molds. Also I forgot to wax the molds so I think the surface epoxy sticked to the molds instead of to the fibers everywhere. But there doesn't seem to be any dry spots which is good I guess. I didn't degass the resin before infusion so there are some air bubbles in the good spots. Any thoughts?