@Peter Nachtwey: if you are mocking anyone about flailing, then why are you bringing steam into this?
When 20Torr was first mentioned, it was mentioned as a vacuum, i.e. relative to atmospheric, i.e. the absolute pressure inside the vessel, above the liquid, is (~760 - 20) = 740Torr. That lowers the boiling point by less than 1°C (cf.
here). So unless the incoming water is near atmospheric boiling point to begin with, it is not boiling and there is no "steam," in the common sense of the word, in the vessel; there is certainly water vapor, but it should not be called steam.
If the only things that can penetrate the pressure vessel, other than perhaps the existing vacuum measurement (draw?), are electrical signal and power wiring, then a system of floats may be the best option.
If it is acceptable to penetrate the drain piping, outside the vessel, near the pump suction for a pressure tap (atmospheric - suction differential), we can derive and estimate the liquid level from that plus the vacuum measurement; whether that estimate will be accurate enough is TBD, but I suspect it will be okay:
- We only need the high limit when the pump is not running, when the frictional and dynamic head losses in the drain pipe are zero, so it is a direct static head calculation.
- We only need the low limit when draining and the pump is running, and the actual issue is NPSH. So a pressure measurement at the pump suction is actually the direct, most relevant measurement; it would be pointless to convert, with non-zero frictional and dynamic losses, that direct pressure measurement to level and comparing the level result to some minimum level setpoint, because using a low level to stop a drain pump only works because level is a cheap proxy for NPSH.
Sidebar: I am looking at my dad's 1967 ASME Steam Tables on my bookshelf, complete with Mollier diagram inside the back cover, the formulation from which my dad converted from Fortran to Basic on both IBM PC and Timex 2068. He had me fix one of the successive substitution convergence routines, which failed when using single-precision floating-point (4 bytes on the PC; 5 bytes on the Timex) in one regions of the H-S diagram. When Dad, with an automotive tech associate's degree from LIATI Farmingdale, interviewed for the GE Test Program for technicians in the 1950s, he was asked what he knew about the Mollier diagram. He said, "I can spell it." The interviewer replied, "That's better than some of my engineers."