Closing a loop across a heat exchanger


Thread Starter

Monte Sheppard

Typical of three situations:

A valve has been placed in parallel with one side of a heat exchanger to divert flow around the heat exchanger, thereby reducing the amount of refrigerant entering the heat exchanger. On the other side, the temperature of a process flow is measured downstream of the heat exchanger to send a signal to the diverter valve on the opposite side.

It occurs to me that there could be very long delays between the change in valve position and a resultant change in temperature in the process gas/liquid. Stability of a closed loop in this situation could be difficult to obtain.

Any suggestions?

david mertens

I think you are right, my suggestion would be to add two temperature transmitters that measure the in and out temperatures of the coolant. This delta T could be used to control the valve. A second PID controller uses the product temperature and the output of this controller is used to control the delta T of the coolant (cascaded control). I think this way succesfull control can be obtained. If this should not be sufficient, try model predictive control, the results will be astonishing (and the costs to set it up also, but the gains may be enormous).
Not sure that I follow your suggestion for cascade control exactly. This is a PLC implementation, so am working with the limits of the PID instruction command. Haven't tried cascade control before. Can you expand? Also, do you have any links to predictive control solutions?

Koen Verschueren

I don't know if you can have a continuous flow of refrigerant trough the heatexchanger? If that's no problem you could place the valve in the other side of the heatexchanger. With the valve you can bypass a part of the heated process flow. You can place the temperature sensor where the cooled and
uncooled flows are joined. Maybe you have to use a static mixer and measure the temperature after the mixer.
We use diverting valves (Fisher YD) for the glycol loops on some of our chillers to maintain a constant flow rate in the loop, but the refrigerant for the chillers is still maintained with a standard thermostatic expansion valve.

The watch out is that you can freeze the chiller barrel if you divert to much and also there is the possibility of slugging the compressor with liquid if there isn't enough hot gas bypass. The minimun loads need to be determined so that the loop doesn't get to cold and that the compressors can unload enough.

Good luck
The problem with realizing this idea is that we cannot calculate a set point for the loop that controls delta T of the coolant. Actually, we can measure it after we bring the system into a steady state but those measurements will be valid only for this particular mode. If the product flow or even the ambient temperature changes those measurements are not valid any more. The principle is that: once the error signal of the main control loop reaches zero level we have brought the process variable to the desired value and the controller should produce a certain set point signal for the other loop (in a steady
state), which is a desired delta T value of the coolant. But we don't know this desired value. This case is more complicated than, for example, a level control loop with a flow control loop cascaded with it, where the relationships are obvious. I believe that there are the following 3 options (maybe more):

1. Simple. You build a conventional PID control loop. Time delays are significant and the resultant system will be a system with a large dead time.
As a result, the gains of your PI-controller that meet the stability conditions will be low and the control will be fairly slow but this may be satisfactory, particularly if the modes of operation do not change often or significantly.

2. Expensive. If you use MPC that will give you a better control quality but will it be a reasonable approach to a comparatively simple piece of equipment ?
You need to weigh the cost increase against the quality gain.

3. Simple with some elements of intelligence. You can try to improve the first option by adding some intelligence to it, for example, a Smith predictor or an auto-tuner. In that case, if you have a variable product flow you can add a flow transmitter (you probably already have it and need just a signal) to estimate the dead time, which is a function of the product flow. Other process variables can also be helpful. The more information you use the more precise but more expensive control you obtain. There are lots of various opportunities here that should be weighed against the cost.

Dr. Igor Boiko
Consulting in Control is available
(including modeling, simulations and control design)
[email protected]
Just a small addition/clarification to my previous comment regarding the cascaded connection.
An integral action within the main loop can eventually bring the controller output to the desired set point but, I am afraid, a small integral action will not do this job well while a large integral action will strongly affect the
Igor Boiko

david mertens

I don't know the PID controller your PLC provides, but basically what you need is an input for the SP (external or automatic setpoint), an input to force the controller in SP tracking and MV tracking and an output that says if the
controller is in automatic/external or manual/internal. The controller controlling the valve (called slave controller) gets its automatic (or external) setpoint from the output of the master controller (controlling the
output temperature of the product). The slave controller has the difference in coolant temperature as PV (e.g. -100 to +100) and the valve position as output (0 - 100%). The master controller has the actual product temperature as PV input (e.g. -10 to 150 degr) and the expected correction as MV output (-100 to +100 degrees). The MV of the master is connected to the SP of the slave. If the slave is in manual/internal, the master and slave controller must be forced in
setpoint tracking and the master also in output tracking to keep the MV equal to the setpoint of the slave controller. (This is to avoid bumps when switching the operating modes of the controllers). To tune this loop, first tune the slave controller, then the master controller
with the entire loop in cascade. Some information on MPC can be found at "": and in some of the ongoing discussions in this forum. It requires very specialised people to set up, you would normally have a specialised firm do the project on a no gain no pay basis.
Thanks to all who replied. I am better armed to pursue design changes with our process engineers. Will keep a record of the replies for future reference or possible contact, as applicable.

David W Spitzer


Which control strategies might be appropriate is dependent upon the fluid state(s), fluid temperatures/pressures, fluid properties, equipment size, control objectives, operation, economic benefits/penalties associated with
stability... for your heat exchanger. There are many ways to do things, but usually only a small number of good ways. I suggest a detailed review of any potential control designs. You might consider hiring someone to help with your particular issues.


David W Spitzer, PE
8 Perth Avenue
Chestnut Ridge, NY 10977
845.623.1830 (phone/fax)

---------SPITZER AND BOYES, LLC-------------
"Consulting from the engineer
to the distribution channel"
This appears to be a problem with the response time due to the heat transfer between the coolant flow and the process fluid and the
intervening heat exchanger material.
I suggest the approach is right but on the wrong fluid! By-passing the coolant doesn't overcome the heat transfer delay between the coolant
and the process fluid and especially, the heat exchanger material.
We faced this problem in a viscosity analyzer and the solution, based on an idea by Mark shelley consists of diverting part of the process
flow around the heat exchanger, not the heat exchange fluid. The recombined process fluid flow enters a static mixer and is homogenised
virtually instantamneously. A temperature measurement at the exit of the static mixer is used to modulate the by-pass flow. Thus the
temperature of the process fluid is capable of being accurately controlled because the control response is very fast.
Of course, as the flow rate through the heat exchanger changes its temperature will tend to change due to the coolant flow so a second
temperature sensor in the process fluid in outlet from the heat exchanger is used to modulate the coolant flow. This is a relatively slow response. However the combinatination of fast response controlling the mixed stream and slow response in the coolant flow control results in very accurate temperature control in the process fluid even when the sample flow temperature at the inlet of the system can change
significantly and rapidly.
To see a schematic visit "": select the UK site, select viscosity and then click download. The download file you need is
"Density and Viscosity systems" I think the example is in scheme 2.