Is Automated PID Tuning Dependable?
It is surprising that more than 30% of all factory PID control loops are operating in manual rather than automatic mode. To make matters worse, of those loops operating in automatic mode, 65% produce less variance when in manual. This reality is possibly due to a dearth of personnel having sufficient know-how to tune and optimize a PID loop, which is one reason so many vendors provide automated...
It is surprising that more than 30% of all factory PID control loops are operating in manual rather than automatic mode. To make matters worse, of those loops operating in automatic mode, 65% produce less variance when in manual. This reality is possibly due to a dearth of personnel having sufficient know-how to tune and optimize a PID loop, which is one reason so many vendors provide automated configuration features. Unfortunately, these are all not created equal, and utilities from various vendors provide much different results. We will describe these features in the context of temperature control units, with the results of a performance comparison between several anonymous vendors.
Some controllers offer different degrees of aggressiveness in executing a set point change. Make your choice based on how fast the process needs to change against the product's ability to tolerate being overheated.
There are two levels of sophistication in this technology, commonly referred to as auto-tune and adaptive-tune (or continuous tune). Auto-tune is generally offered as a standard feature and relies on characterizing the system through artificial manipulation of control output power levels. however, this is generally impractical to do during production operation. The software watches the process and analyzes its response while applying either a step increase in fixed power level, or cycle-tune using oscillating power levels. The step-tune approach performs best when started from a stable, constant process. The cycletune is not as dependent on initial conditions, but the process must be able to handle the amplitude of the necessary oscillations.
Adaptive-tune generally implies an increase in cost and requires more intelligence built into the controller. It must be able to monitor the process during production operation and judge when and how to modify PID parameters to maintain stability as conditions require. It characterizes the process without having to manipulate power levels artificially, as is done in both step and cycle style auto tune.
These capabilities reduce puzzling over PID parameters. A user can just let the controller tune itself. However, the ability of major vendors to deliver satisfactory control after tuning, however, was the topic of our recent investigation. The findings may shed light on reports of unacceptably poor performance found industry-wide.
Identifying differences in control performance requires a platform capable of simulation and accurate measurement. A PC with high resolution I/O was developed to interface with the controller, simulate a wide variety of industrial plant conditions, and provide easily repeatable test results. It reads the controller’s power output levels while running a plant simulation that calculates the proper signal to feed back to the controller’s input sensor. As the PC continuously graphs process value and power level data, differences between products become obvious looking at the performance curves.
The dynamic range in test loads is designed to be wide enough to bring out differences in products’ ability to control fast and slow loads, but remain compatible with their sampling and update rates. The plant model is purely first order in nature. A primary characteristic of a first order load is its time constant. In these tests separate heat plant and cool plants are used, and each can have a different time constant. Heat plant time constants are 10, 60, and 300 seconds, referred to as the fast, medium, and slow loads. Cool plant time constants are 15, 60, and 90 seconds.
Another characteristic inherent to a typical industrial plant is dead time, the duration between applying power and the point at which a response is detected at the sensor. The ratio of dead time to time constant is the controllability ratio. This ratio is relevant because as it increases the PID algorithm is inherently less effective. For design of loads used in testing, values of 10 and 20 are selected to represent typical industrial processes.
Settling time is a measure of the controller’s ability to stabilize the process. Technically, it’s the time required for the process to approach and remain within 1% of set point after an event such as a change in set point or disturbance. For slow processes its magnitude is especially meaningful, since it says how long personnel must wait before commencing production. Consequently, for this article settling time is used as a primary basis for comparison.
To isolate auto-tune and adaptive-tune capabilities, all controllers were tuned as advised by product documentation and literature. Identical start conditions were applied, with a steady ambient temperature.
Comparing these performance curves against those on the preceeding page show that some auto-tuning routines are better than others.
Running the auto-tune tests
After auto-tuning and characterizing the load, a controller should be able to stabilize the process quickly in the face of a disturbance or set point change. Some products offer more than one auto-tune setting, presumably to allow users to customize control response. One offers standard and robust settings, another offers settings for under-, critical-, and over-damping. This latter product happened to be the only one that can stabilize all loads; and did it for all damping settings.
There were some notable departures from the anticipated responses (see graphic):
Vendor 1 suffers from a large offset from the set point on the fastest load. This offset should go to zero as the process approaches the set point, but it never does.
Vendor 2’s two auto tune settings (shown as 2A and 2B) give wildly different outcomes: one stabilizes the process after a very short settling time, but the other results in unacceptable oscillations.
Vendor 3 eventually stabilizes all three loads after a set point change, but oscillates. This ringing becomes even more apparent during recovery after a process disturbance. A propensity to ring potentially means increased sensitivity to noise and disturbances, i.e., the process moves in the direction of instability rather than stability.
Vendor 4’s performance after auto tune is even more oscillatory in nature than Vendor 3.
The response of the unit (first graphic) with three damping settings is acceptable since it tunes itself to deliver smooth, stable control for all load conditions. The three tune settings, under, over, and critical damping, can be adjusted using a knob to provide an intuitive response. Under damping is associated with fast rise times and some overshoot. Critical damping is slower to rise but exhibits little or no overshoot. Over damping gives slowest response but guarantees no overshoot which can be desirable in a critical process.
The other automatic configuration feature is adaptive tuning. It differs from auto-tune in that the controller must characterize and tune the process invisibly while keeping it under control. This capability requires advanced algorithms that are generally proprietary, and it is not widely available. Of the vendors evaluated, only two offer
Medium load set point changes are not as difficult for the auto tuning functions and generally settle out after a long enough period of time, although there are exceptions.
this feature. Evaluating its effectiveness requires observing a controller’s response to a sudden shift in load dynamics when the process is under control at steady state. For this test, the dynamic change applied to the load was an abrupt reduction in load mass by 65%. No controller without adaptive-tune was able to control the load after a change of this magnitude. However, both controllers with adaptive-tune capability were able to calm and stabilize the process without user intervention.
While a direct ranking of controllers by name is beyond the scope of this article, it is clear that some are better than others with respect to their ability to provide a novice user with good control right out of the box. There were clear differences between the performance of self-configuration features from vendor to vendor.
In a plant environment, if auto-tuning the process does not provide stable control, an expert can usually tinker and tweak until stability is achieved. However, users do not always have such a luxury. Auto-tune functions must be effective since PID experts are scarce and product quality is normally directly related to process variation.
The findings of this investigation are consistent with and may even explain the reports of large numbers of industrial loops performing below expectation. A multitude of elements constitute a modern process controller; communications, user interface menus, advanced electronics, formal ratings, and certifications. The most basic functions of controllability should not be neglected since stabilizing the processes is a controller’s primary role.
Greg Baker is pursuing an MSE at SJSU in control and involved in algorithm development at Watlow. Reach him at email@example.com.
Case Study Database
Get more exposure for your case study by uploading it to the Control Engineering case study database, where end-users can identify relevant solutions and explore what the experts are doing to effectively implement a variety of technology and productivity related projects.
These case studies provide examples of how knowledgeable solution providers have used technology, processes and people to create effective and successful implementations in real-world situations. Case studies can be completed by filling out a simple online form where you can outline the project title, abstract, and full story in 1500 words or less; upload photos, videos and a logo.
Click here to visit the Case Study Database and upload your case study.