Autotuning PID capabilities in temperature controllers solve most problems if used properly, but they do not always work as desired. Sometimes, a good old-fashioned manual approach is the solution.
Proportional-integral-derivative control methods — more commonly known as PID control — have been used in industrial applications for as long as anybody can remember. Over the decades, control engineers have developed tuning techniques to get loops to behave as desired. Dozens — if not hundreds — of ways exist to determine the “magic settings” to get a loop to remain stable while quickly responding to process upsets and setpoint changes.
Industrial process temperature control loops offer some unique challenges. When a control engineer wants to tune a flow or pressure loop, the control device can be adjusted with an infinite number of steps, for all practical purposes. A valve usually can be tweaked just the right amount to make the process variable hug the setpoint.
Temperature loops generally do not work that way. Because the final control element often is an electric heater, it offers two settings: 0 and 100 percent. Maintaining a process at some intermediate point requires turning the element on and off, so getting a 50 percent output means the element is alternately turned on and off for the same period of time. In theory, the thermal inertia of the process makes the cycling invisible, but we’ll discuss the reality in greater depth later.
The control element also only works in one direction. While a valve controlling a flow loop can be opened or closed to increase or decrease the flow, most temperature applications have no counter-control capability. It is possible to heat the process, but there may be no provision for cooling it beyond ambient heat dissipation or introducing lower temperature feedstock coming into the process. Of course, some applications may have both cooling and heating elements, but these are less common. In most situations, if the controller makes the process material too hot, the only remedy is waiting, which makes overshooting an issue.
Understanding PID Control Concepts
Anyone working in manufacturing for 10 years or more has seen major changes as more companies have reduced their engineering and technical staff. Gone are the days when plants had staffs of engineers available to take on technical issues to ensure instrumentation was operating at peak efficiency. Nowadays, you may be one of the few process automation professionals remaining at your company. Likely, you are responsible for keeping the process running as well as taking care of field devices.
The ability to tune a PID loop manually is an art that is quickly becoming scarce. But, it is one of those skills — like driving a car with a stick shift — that can be unquestionably helpful in the right circumstances. The ability to tune a thermal PID or other loop is even rarer.
Many of the newest controllers are vastly more advanced than units from just a few years ago. Autotuning functions are available on bare-bones, 1/32 DIN controllers, and there are even several tuning strategy options to keep unruly loops under control. In spite of all this sophistication, however, these automated methods simply cannot tame some loops.
All autotuners are not created equal. Sometimes, the problem is the nature of the loop itself, but other times, the autotuning method available for a given controller is not suited to the specific application. This is when the skills to tune a loop manually are necessary.
As mentioned, most universal temperature and profile controllers provide an autotuning function as a standard feature. The controller automatically measures the process characteristics and tries to calculate the best PID parameters. Autotuning generally will succeed in setting optimum or at least workable PID parameters in most applications, so using it should be your first approach. If it works, you are done. But, if an autotuning error occurs, or if no improvement in control is observed, you must tune it manually.
Thermal-Loop Cycle Time
There are a few unique aspects of thermal loops you should keep in mind.
Cycle time is the total length of time for the controller output to complete one on/off cycle. With a temperature loop, in most situations, the heating element is either running at maximum heat output or off (figure 1).
Cycle time is used when controller output is configured for time-proportional PID. It is typically implemented by a relay or voltage-pulse output. The percentage of the cycle time during which the output is on is the same as the output calculated by the PID. If output display value is 50 percent, and cycle time is set at 20 seconds, the heating element will be on for 10 seconds and off for 10 seconds.
Reducing the cycle time results in faster cycling and finer control; however, it also reduces the life of the final control element and often the switching device. So, the cycle time should be as long as possible without creating oscillation. The output of the heating element relative to the amount of material will help make this determination. This parameter typically is set after autotuning or during the manual tuning process — in both cases, after the dead time has been calculated. Many loops, tuned with autotune, may not perform properly or appear to oscillate when they are actually tuned properly but have the wrong cycle time. Unless the cycle time is set properly, the loop will never exhibit the desired stability.
PID Parameters for Industrial Process Controls
The proportional band (PB) is the range over which the output is adjusted from 0 to 100 percent. PB is expressed as a percentage of the full operating span of the controller. If the manual reset is at 50 percent, the proportional band centers on the setpoint. For example, an operating range of 0 to 1000°F with a proportional band of 5 percent would equal a proportional band of 50°F, and so a setpoint of ±25°F.
Integral action (I), also known as reset, is added to proportional action to overcome steady-state offset or error from the setpoint. It responds to the error signal of the feedback system just as proportional action does, but it responds to the magnitude of the error by summing the error over time and adjusting its effect on the proportional band over time.
A large change in the load on a system will cause the process variable (PV) to experience a large deviation from the setpoint. For example, if a heavy load is placed into a hot preheating furnace, the temperature of the furnace will drop before the control system can increase the output to add more energy to the furnace and the load. The proportional action alone may not be enough to recover from the disturbance in a timely manner, especially if the temperature is within the proportional band. As the error between the setpoint and process variable becomes smaller, the position of the final control element gets closer to the point required to maintain a constant value.
Derivative action (D), also known as rate action, responds to rapid changes in the error signal. It will anticipate the rise or fall of the process variable and automatically adjust the proportional band to minimize overshoot or undershoot. Derivative action makes additional adjustments to the proportional band relative to the rate of change of the error signal.
When the process variable is steady, the derivative action is zero. When the value of the process variable is changing rapidly, the derivative signal is large. The derivative signal has a major effect on the output of the controller (figure 2).
In this way, a larger control signal is produced when there is a rapid change in the process variable because the final control element receives a larger input signal. The net result is faster responses to load changes, with overshoot/undershoot limited or prevented.
Given the one-way nature of most temperature loops, avoiding overshoot can be critical. However, as a practical matter, temperature loops do not generally change very quickly, so derivative action may not be necessary, and many temperature loops will operate well using only PI factors.
Dead time is defined as a measurable time delay, in minutes or hours, before a response in the process variable is observed due to an output change (figure 3). Most ovens, furnaces and tanks of liquid have relatively long dead times compared to other types of processes. When the heating element goes on, there can be some time — perhaps a long time — before it is possible to observe its action, furthering complicating control.
When Autotuning Works Well
Effective autotuning should allow you to choose how the process responds by using damping (retarding action) on process change. When a controller responds to a setpoint change, the degree of damping will determine how fast it tries to reach the new setpoint (figure 4). An over-damped process will change slowly, gradually moving the process variable to the new setpoint. If the process is sensitive to overshoot, this might be a desirable approach.
An under-damped approach will begin moving to the new setpoint much more quickly, and it will probably overshoot, only to have to come back. Because there is probably no counter effort, returning to the setpoint might take some time. If the product is sensitive to high temperatures, it could be ruined in the process.
The ideal situation in most applications is a fast change without excessive overshoot. The process variable finds the new setpoint quickly and settles without oscillations. When this happens, the autotuning system is doing its job. Sometimes, autotuning will not work, so a shift to manual is required.
Manual Tuning with a Long Dead Time
This kind of situation is common with temperature loops because dead time is usually a significant factor. This approach is best used where a recorder is connected to the controller. This way, the influence of proportional-setting changes can be seen over time, and the dead time can be measured.
Begin by setting the PID parameters to the following:
- P (PB): 5 percent.
- I: 0 percent.
- D: 0 percent.
Start the furnace or process with a setpoint that will allow the process variable to stabilize with an output between 25 and 75 percent. With the I and D values at 0 percent, the process variable will probably stabilize with a steady-state deviation or offset from setpoint. In other words, the process variable will settle near the setpoint but not quite there.
If there is a disturbance or setpoint change, a properly set P value will allow the process variable to oscillate and settle out to steady-state condition in what is called one-quarter wave decay. If the process variable oscillates continually, increase the value of P from 5 percent to some larger number until the process variable settles to a steady state. Generally, adjustments to P need to be significant to see a change — on the order of halving or doubling.
Note whether there are regular cycles at this temperature by observing the graph over time on the recorder. A cycle, or oscillation, may be as long as an hour, so patience is necessary.
If there are no regular cycles in the process or an upset, cut the P value in half (figure 5). Narrowing P by halving it helps maintain tighter control. Allow the process to stabilize and check for oscillations. If oscillations are not detected, continue to divide the P value by two until oscillations are obtained. Fine tune by increasing or decreasing the P setting by smaller increments until the process oscillates and then reaches steady state, just as the Figure 5 graph shows.
Once the dead time is measured, set the I value at five times the dead time. Dead time is defined as the period between an output change and a noticeable change in the process variable. For example, assume a process variable of 200°F, the process in manual at 25 percent and the control stable. Increase the output by 50 percent to drive the process variable to a new setpoint of 250°F. If after two minutes, the process variable begins to climb, we have determined the dead time is 120 seconds. The I value then is set as seconds per repeat; therefore, it should be set at 600 seconds.
At this point, the cycle time value can be determined. If your process is using time-proportional PID with either a relay or voltage-pulse output, a good rule of thumb says cycle time should be set for 25 percent of the dead time. So, using the above example of 120 seconds for dead time, the cycle time should be set around 30 seconds.
This is not a hard-and-fast rule. Lower values will have no effect on the process variable, but they can cause unnecessary wear on the final control element. But, be aware that values much larger than a properly calculated cycle time (in this case, 30 seconds) can introduce cycle-time oscillation to the process at steady state.
D, if used at all for a slow-moving loop, is usually set at 25 percent of I; therefore, if I is set at 600 seconds, the D value would be 150 seconds.
In conclusion, end users are demanding control instrumentation be easier to use and, at the same time, more capable. Multiple autotuning algorithms and features as well as overshoot and hunting suppression allow these new-generation controllers to handle the toughest processes and satisfy end user requirements. With difficult temperature loops, consider the fine art of manual tuning and achieve tight, steady-state heating in your process.