In order to eliminate cycling, make sure that each inner loop is acting faster than its outer loop. The positioner should be about 10 times faster than the slave temperature controller, and the slave should be 10 times faster than the cascade master. To increase the speed of response, in critical cases, I also use derivative in both the master and the slave controller.
You should also consider protection against reset wind-up that can occur if the cascade master is switched to manual. We protect against this by using external reset (Figure 1). Lastly, carefully follow the configuration shown in Figure 1, except the 50%-50% range split for the valves. If your heating is more effective than your cooling (because of higher ΔT), increase the cooling effectiveness by shifting the split point with a slight overlap towards cooling.
A: This is not a cascade system. In a cascade system, the master temperature controller would set the setpoint of a coolant exit temperature controller that manipulates the valves, configured as recommended in Figure 1. The master controller must have the integral mode if offset is to be eliminated, as the heat load changes with temperature and with time in a batch reactor.
If your controller gain setting is, in fact, 70, this is incredibly high, making the calculation of a bias implausible unless the controller temperature measurement span is incredibly large (e.g. 0 ºF to 1,000 ºF). If the measurement or controller scale range is not that large, the controller output will be bouncing between its output limits. Remember, the controller works with signals in percent.
If the reaction is exothermic, the process could even be unstable in its steady state because it is not self-regulating.
I do not recommend lambda tuning for any process. Both primary and secondary controllers should be tuned as tightly as possible in response to load changes. Derivative is required in both controllers and on the controlled variable only. See my paper, "Exothermic Reactors: the Stable, the Unstable, the Uncontrollable," Chemical Engineering, March '04, pp. 54-59, and in Process Control and Optimization, Ch. 8.10.
A: This application should have a cascade loop, so the secondary controller adjusts the split-range valves. I think you have an exothermic reactor. Exothermic reactants can be unstable at the normal operating point (Figure 2.)
I don't believe lambda tuning is the way to go, because (1) it presumes that the process has self-regulating characteristics, which an exothermic reactor does not, and (2) it only applies the proportional and integral modes. To overcome the open-loop instability of the process, the derivative in the master controller should be on the measurement, but in the secondary controller, it should be on the error. Both should have integral that would eliminate disturbances in the supply temperatures.
Most people split the valve ranges at 50% of the controller output, but your heating valve opens faster than the cooling valve because the strength of the heating effect is greater than the cooling effect. A better configuration can be to shift the split point in the direction of the weaker effect—in this case, cooling—assigning say, 65% to heating and 35% to cooling.
A: If your controller gain setting is in fact 70, this gain is incredibly high, making the calculation of a bias implausible, unless the measurement span is 1,000º or more. If that is not the case, with this gain setting, the controller output will be bouncing between its limits.
For PD control, the bias parameter of the PID block should be set to eliminate the offset, so if the controller output is 70% when the temperature has settled, and the error is 3º above set point on a 50º to 100º scale, with a controller gain setting of 2 of a reverse acting controller, then the estimated bias is:
Bias = settled output + offset percent error (process variable % – setpoint %) x controller gain x controller action sign (-1 for reverse) = 70% + (6%) x (2) x (-1) = 58%.
The estimated bias adjustment must be tested with close observation. Whenever the PID is switched from manual to automatic, the bias is automatically corrected by an invisible amount by DeltaV to provide a bumpless transfer.
It is unlikely that you have a linear constant process gain. Valves generally have a nonlinear installed characteristic; the process is nonlinear; and heating and cooling have different process dynamics (e.g., speeds of effect). For these and other reasons, PID rather than PD control is preferred unless the reactor is exothermic and has a runaway response (accelerating response) when the controller is in manual.
If PID is used, the integral time should be large for slow integrating processes, which yours probably is. You could start an integral setting of 10 times your derivative time if you are not presently experiencing oscillations. If the present response falters or oscillates, your rate setting is high.
If the approach to setpoint is smooth, but oscillates after you reach it, your gain setting is high. If, with these settings, the overshoot is not excessive, you can decrease the integral time. Low gain settings on integrating processes cause slow rolling oscillations and overshoot, while too high a gain causes instability. Therefore (P)(I) should exceed 4/(process gain).
Since the valves are nonlinear and the process dynamics for heating and cooling are different, you should use different tuning settings based on which control valve is being throttled. The adaptive control feature in DeltaV Insight can be used to identify (Version 9.x) and schedule tuning settings (Version 10.x) as a function of controller output.
If the heating effect of your installation is twice as large as the cooling effect, set the split range point so that the portion of the controller output for throttling the heating valve is twice as large as the portion for throttling the cooling valve.
You should contact your DeltaV representative to help you use the DeltaV PID block options and advanced capability such as DeltaV Insight.