6 Gregory K. Mcmillan: N A D K
6 Gregory K. Mcmillan: N A D K
McMillan
[1,2,4,5,10]. A user-selected step change is injected into the PID output initially
and any time the process variable reverses direction and crosses the setpoint and
the corresponding noise band. The controller action is used to determine if the re-
versal in the process variable is in the correct direction to drive the process vari-
able back to setpoint. The ultimate period (Tu) is the oscillation period. Equation
14-8 is used to compute the ultimate gain (Ku) from the PID output step size (d)
and the process variable amplitude (a) corrected for the noise band (n). Figure 14-
1 shows the relay oscillation method with a large change in the process variable
(PV) for illustrative purposes. For processes with large time constants, the PV am-
plitude (a) is so small, the oscillation is barely perceptible and the oscillation pe-
riod is about 4 deadtimes. Since the more important PID loops, such as tempera-
ture have a large process time constant, the auto tuner provides a test that is less
disruptive and faster than an open loop test that is waiting to reach a new steady
state to identify the process time constant. The time constant identified in relay os-
cillation method is not very accurate. Thus, when the relay oscillation method,
tuning settings based on the ultimate period and ultimate gain are more accurate
than those that require knowledge of the process time constant.
Fig. 14-1 Relay Oscillation Method Offers Fast Tuning Test [24]
4d
Ku = (14-8)
π + a2 − n2
The PID gain is the ultimate gain multiplied by a 0.25 factor [23,25]. The PID
integral time is the ultimate period multiplied by 1.0 factor for self-regulating and
10.0 for non-self-regulating processes [10,23,25]. The PID rate time is the ulti-
mate period multiplied by 0.1 when derivative action is beneficial [10,23,25]. If
14 Industrial Applications of PID Control 7
the ultimate period is less than 3 times the dead time, the rate time should be 0
since the loop is deadtime dominant (deadtime is significantly greater than the
largest time constant in the loop) [10]. If the ultimate period is greater than 4 times
the deadtime, rate time should be used to prevent a runaway since the process may
have positive feedback and an unstable open loop response. These factors are gen-
erally in the direction to provide a non oscillatory PID response that is more robust
(more resistance to excessive oscillations from changes in process dynamics). The
Ziegler-Nichols factors were designed to provide a quarter amplitude response
(amplitude of each succeeding oscillation is ¼ the amplitude of last oscillation).
Most publications on tuning based on the ultimate period and ultimate gain use the
Ziegler-Nichols factors leading to improper conclusions on smoothness and ro-
bustness of the tuning method [10].
Since an adaptive tuner uses current tuning settings to compute process dynam-
ics as the starting point for its search, the number of tests required to get an adap-
tive model with a high fidelity rating can be minimized by first running the On-
Demand tuner with a requirement of just 2 or 3 cycles. Since the cycle period is on
the average the ultimate period, the test is usually faster than an Adaptive Tuning
test, especially for the overly conservative (sluggish) tuning commonly found in
industrial PID controllers that have not been tuned by an automated method.
The step size in the output for On-Demand and Adaptive Tuning should be at
least: 5 times the noise band, the trigger level of a wireless device, and the dead
band and resolution-sensitivity of the control valve [2]. Note that these step
changes will not show the deadtime from wireless update times and valve back-
lash and stick-slip. For wireless devices, about half of the default update rate
should be added to the identified deadtime [14,23,25,29,30,31,35].
Fig. 14-2 Models Enable Adaptive Level Control of Conical Tank [18]
Instead of integrating the error, the feeding back of the controller output or ex-
ternal reset signal through a filter block and adding it to the contribution of the
proportional and derivative modes creates an integral mode action where the filter
time constant is the integral time setting [4,25]. When the error is zero, the output