I am designing process control strategy in a batch system for a cascade control temperature master feeding a jacket temperature control slave. It is for an existing process, and the jacket temperature is measured only on the jacket outlet. This adds approximately two minutes of dead time to the jacket temperature response. Nobody believes this dead time will slow my ability to tune the master control at the same levels that I could by using an inlet temperature. Do you have any information on how much slower my master would have to be tuned with this dead time than if it were eliminated by installing a temperature instrument on the inlet to the jacket? I am guessing that it would at least double my response time, but suspect it would be much more dramatic.
Locating the Reactor Temperature Sensor
Dead time of the loop is cut by reducing the distance between the source of an upset and the point where the controller measures the variable. In the case of a chemical reactor, the upset can originate either from the process, due to changes in heat release, or can be caused by a change in the pressure or temperature conditions of the utilities. If the cascade master detects the reactor temperature, and its cascade slave measures the jacket coolant outlet temperature, both potential sources of upset are detected (See Figure 1 below). Naturally, to minimize the dead time in the coolant loop, the water must be recirculated.
FIGURE 1: REACTOR TEMPERATURE CONTROL LOOPS
Here's how to do a temperature control loop for nearly all exothermic chemical reactors.
If you expect stable utility conditions, the source of the upset will probably be the heat release of the chemical reaction. If your control system is configured as shown in Figure 1, the dead time will be minimized by detecting the jacket outlet temperature. This is true of exothermic chemical reactors for all phases of both batch and continuous operation, including heat-up, reaction and stripping.
If you expect dominant upsets to be caused by changes in the properties of the utilities, such as steam pressure or cooling/chilled water temperatures, move the slave sensor to the jacket inlet or add a third-level cascade for best performance. In that case, you have to make the innermost slave loop the fastest of the three control loops.
The cascade master should always be a PID controller provided with external reset from the slave measurement to guarantee bumpless transfer. If the reactor dead time exceeds half its time constant, you should replace the regular PID algorithm with a PIDTd algorithm.
The slave controller must be at least four times faster than the master, and the inner slave, four times faster than the outer slave. Usually the slave can be proportional only with a proportional band of about 15 percent. One way to improve the speed of the slave loop is to use a PD algorithm by adding a little derivative on error (not measurement) to the proportional action.
The valve positioner is a position controller, so it should also be four times faster than the slave which manipulates it. The characteristics of the valve(s) should be equal percentage.
“. . . but suspect it would be much more dramatic.”
You can bet on that. The problem is small only if the heating/cooling times are much longer than a control response time several times the jacket flow dead time. In this case the loop will most likely be run in manual, and the control system not used.
With appreciable dead time, the secondary loop gain must be set low enough to avoid cycling, and the reset time must be several times slower than the dead time. Then it is necessary to de-tune the primary loop to be slower and with less gain than the jacket loop to have any chance of stable operation. Note also that any less-than-perfect control valve response will add control problems. It all multiplies, as you suggested. Any dead time—any inaccuracy anywhere in the loop—will reduce controllability.
A detail not given in the problem is whether the flow in the jacket is pumped at a constant rate and with a more-or-less predictable dead time. The flow varying with the heating/cooling situation will create a near-impossible situation. If the product has any value at all, or if quality is sensitive to the temperature-cycle control or time, then there should be no question but that the jacket inlet temperature must be measured. The only problems I have ever seen in using jacket inlet line temperature occurred with a poorly mixed heating/cooling fluid. In that case the indicated temperature was extremely noisy. Adding an elbow and just a little pipe solved that problem.
With jacketed vessels, installing the “Pfaudler nozzle” to swirl the incoming fluids around the jacket will greatly improve heat transfer, reduce heating time constants and generally make control easier. I have seen older vessels where these nozzles somehow vanished over time or new installations where they were omitted. Instrument people need to look at the whole system.
Cullen Langford, P.E.
I would like to know what type of inlet temperature control loop you would have had, if you could have had an inlet temperature controller (TC). I presume you would either be modulating an inlet heat exchanger or modulating a mixing valve. Since these two types of inlet temperature control would have different response times, it is difficult to estimate how much difference the response of a jacket outlet TC would be. But in either case, with two-minutes’ dead time, the secondary will have a significantly slower response than an inlet TC. This might affect the tuning of the primary TC, since it must be tuned to give a slower response than its secondary.
I would even consider a three-level cascade, with the master cascaded to the jacket outlet TC, which is in turn cascaded to an inlet TC. Then any disturbances to the inlet temperature (sticking valve, variations in supply temperature, etc.) would be handled in the lowest level loop, rather than propagating through to the jacket temperature.
One other point: Consider using a derivative on error—not just on measurement—for the jacket temperature controller. Then when the master requires an adjustment to the jacket temperature controller, that TC will produce a faster response than with no derivative, or derivative on measurement.
Harold Wade, Wade Associates, Inc.