CAPFTemp curve from operational data

asked 2022-04-22 16:37:44 -0500

updated 2022-04-25 14:03:40 -0500

Hi All!

Has anyone successfully generated the CAPFTemp performance curve using operational data? I'm looking at chillers in particular, but this could apply to any EnergyPlus object that uses a biquadratic curve to characterize capacity as a function of temperature.

The idea of variation in available capacity tends to be a little squirrely to nail down even when manufacturer data is available, which, in the best case scenario, tends to be efficiency as a function of PLR and ECWT (but PLR is always based on reference capacity). The concept seems to be even trickier to nail down with real-life operational data for several reasons:

  • BAS point availability and quality varies significantly from building to building.
  • Real operational bands are limited (most chillers aren't being operated at their max while other conditions move through a range of temperatures).

I'm interested to know what data points people have used to characterize available full-load capacity, and whether it was from BAS or some sort of research setup. Occasionally I might see a BAS point named something like "ChillerCurrentEnteringDrawRla (%)" that may indicate actual compressor speed, but it's often difficult to tell which points are actual measurements and which are calculated from other points.

So far all I've been able to come up with is using mCpdT through the evaporator (temperature data is typically easy to come by, but flow is often not available) for all instances where the compressor is at or near max. If compressor data is not available (likely), reduced capacity could be indicated by loads near rated capacity, but unit is not making setpoint. Increased capacity could just be any measured load that exceeds rated. Obviously time scale of the data throws a big wrench into all of this as well, but I'm ignoring that for now.

Anyways, just curious if anybody has a general strategy when it comes to characterizing CapFT performance in this way.

Thanks for your time!

edit retag flag offensive close merge delete


We derive curve sets for DX , VRF and heat pumps frequently - but haven't had the need yet for chillers. We use EnergyPlus, and I'm guessing you use something else. The principles should be very similar, however:

  1. Understand the inputs used by the curve (E+ uses condenser water inlet temp and evap water outlet temp)
  2. Obtain that data from the manufacturer over the range of expected operation
  3. Tabulate the data, then run an Excel LINEST function. Happy to share our (DX) tool if you think that would be helpful.
Jim Dirkes's avatar Jim Dirkes  ( 2022-04-23 08:43:19 -0500 )edit

Hi Jim! Thanks for the info!

Eventually I will use the EnergyPlus CurveFitTool for the actual biquad curve generation. My question is more about a scenario where catalog data is not attainable, but operational data exists. This could be for a chiller, DX coil, or any E+ object that uses the CAPFTemp curve to adjust available full load capacity.

So I'm wondering, in that scenario, what data points people may have used to ensure the unit is running at its max, but capacity is being altered based on LCHW/ECW conditions.

I'll update my question to try and be more clear. Thanks again!

codybond's avatar codybond  ( 2022-04-23 10:56:54 -0500 )edit

Good luck with that effort! Calibrated sensors for temp and flow? Not likely. Calibrated to each other within a couple tenths of a degree for entering & leaving temps. VERY unlikely! The "real world" doesn't normally care about anything that accurate. I went through the effort many years ago for a university's multi-building chilled water loop and learned some interesting things.... For example, two sensors in an ice bath cannot be expected to read within 0.1F unless they're exactly next to each other. Similarly, two sensors on a work desk (must be next to each other AND under a cloth)

Jim Dirkes's avatar Jim Dirkes  ( 2022-04-23 15:53:34 -0500 )edit

Oh, I've definitely seen this myself and hold similar skepticism. This is more as an exercise, and considering a scenario where no data exists other than current operation. Ultimately there would be model calibration tweaks based on measured power at the end, but I'm interested in seeing if the shape can be improved with some data synthesis, even if it is based on mediocre/poorly maintained sensors. Thanks again, Jim!

codybond's avatar codybond  ( 2022-04-24 10:10:23 -0500 )edit