Nope, I noticed another difference. The downwind evaporator doesn't draw from txv1, it draws from the flash tank. Why not have it draw from Txv1? then you have 2 flow paths: 1 thru downwind evap to intermediate pressure at the ejector mixing port and 1 thru the ejector mixing nozzle. Slightly superheated gas mixes with saturated liquid at ejector and is pointed at upwind evaporator.

It looks to me as if you added txv3 to aid the operation of the ejector. without it, there would be very little pressure on txv2 and the downwind evaporator. Txv3 basically forces the ejector to work by applying suction. HMMMMM, is your ejector not sucking enough?

I'm not trying to be a richard here, I just do this sort of thing for a living. I take blueprints, sketches, and drawings and the like, and try to turn them into something that is real and useful. Engineers have a way of taking something simple and twisting it into something useless or impossible to build or operate. As you probably know, what they start out with (on paper or computer) is usually revised many times before the project is finished. From this experience, I have developed high skill in finding and removing rats, wrenches, and traps. Please don't take it personal.

If the second TXV sources from the first TXV, there would be less flow to operate the ejector. And thus a lower efficiency. I can assure you the ejector works since the second evaporator gets colder than the first.

The "aspect ratio" of the phase separator is much taller than what the drawing indicates. The bulb for the third TXV is located pretty low so there's plenty of room for the liquid and vapor to separate. Also, the flow rate through that TXV is relatively low due to the low pressure drop across it. The first TXV is 1/2 ton as the design calls for, but the second and third TXVs are 5 ton as the larger bore is needed to get enough flow at the low pressure drop.

It does in fact draw from the cold and push into the hot. The dsPIC controls the water pump speed based on hot water temperature. It uses the local reading to run the PID loop, then uses the remote reading to make small corrections to the local setpoint. If the remote reading is unavailable or out of range, it simply reverts to only using the local reading.

Hey NiHaoMike! I haven't been on this forum in some time, but since you linked to it from my Youtube video on ejectors, I became quite interested in your work. Where does the system stand at this date? Have you continued working with the ejector? Do you have any data to suggest it's reliability / effectiveness?

Some more pics would be grand!

Best wishes

Here's a picture of the control electronics:
http://i58.tinypic.com/2j162ox.jpg
There's a 250MHz Jessica Simpson network processor running OpenWRT and an 80MHz dsPIC running FreeRTOS. I initially had plans to implement a web UI, but I ended up just SSHing into it and running some Python scripts. The PID loops have been tuned and it works really well, with the power use going down to about 270W.

I didn't bother to build a remote sensor as the local sensors worked well enough. The sensors I used are thermal "diodes" (really diode connected transistors) read by TMP513s, taped to the copper pipes with insulation over them. After making a PT table for R433b (had to interpolate from available data), the readings were spot on to what my gauges measured. The ejector gets several C of delta T when the system is in dehumidification mode.

I have since moved to a place where I don't get to use it much.

Very cool. Where we live definitely dictates what is suitable. You seem to be very well adept with the control and data collection side of things. That is where my abilities fall short. I'm going to be looking into the Phidget system as a means to do just that. I understand it to be fairly simple for a mechanically minded guy like me.

I just use simple thermocouples right now with a four channel readout. I wasn't happy with the response time or the readings I get at large delta Ts with ambient, so I pursued a method of installing the thermocouples directly into the lines, with great success, I believe. I describe my method in this video, if it is of any use to you:

https://www.youtube.com/watch?v=PaL7r5paiZ4