A Unified PID Control Methodology

PID control is the workhorse of process automation. At this week’s AIChE 2013 Spring Meeting and 9th Global Congress on Process Safety, Control Talk blog‘s Greg McMillan and Eastman Chemical’s Héctor Torres presented their paper, A Unified PID Control Methodology to Meet Plant Objectives.

Here is the presentation:


Greg and Héctor describe this unified approach:

A unified approach to tuning has been found that enables a common and simplified method for setting PID tuning parameters. Key features can be used to eliminate the need for retuning to deal with different dynamics and objectives. This paper shows a methodology that integrates a unified tuning approach and key features that minimizes implementation and maintenance efforts. A step by step method will be used that will address the myriad of dynamics and objectives in PID applications.

They explain the characteristics of the four types of process responses: dead time dominant self-regulating response, lag dominant self-regulating response, integrating response, and runaway response. The tuning requirements are so different for each type of response that process automation professionals can get confused.

The authors note that automation systems including sensors, final control elements and controllers can introduce excessive discontinuities, lag, noise, and dead time. For example, a control valve may have excessive friction causing a time delay in opening or closing.

They highlight the unified tuning approach:

The key breakthrough in thinking is to use the lambda tuning rules for integrating processes for lag dominant self-regulating and runaway processes. The lambda tuning rules automatically prevent the product of the PID gain and reset time from being less than the twice the inverse of the integrating process gain that would result in slow rolling oscillations. The main decision is whether to maximize the transfer of variability or maximize the absorption of the variability. Key features are used rather than tuning to address the automation system difficulties and to meet different process objectives.

You’ll want to read the paper to understand the considerations in the decision to transfer or absorb variability, understand the key features in PID to address these decisions, and the step-by-step procedure through the unified methodology.

Greg and Héctor conclude:

The use of key PID features, a unified PID tuning method, and an adaptive tuner enables an effective application of PID control for wide spectrum of process responses. Drastically different sources of variability, automation system difficulties and changing process objectives do not require PID retuning. PID features can be used to tailor the PID response to different application requirements on a much more understandable basis by the user. The role of the PID can be expanded to include process optimization besides regulation.

MP3 | iTunes

Audio clip: Adobe Flash Player (version 9 or above) is required to play this audio clip. Download the latest version here. You also need to have JavaScript enabled in your browser.

One comment so far

  1. I saw a note from Greg to his mentees (http://automation.isa.org/2012/07/isa-mentor-program-a-guide-to-success-2/) and asked if I could append it as a comment:

    “I am proud of the presentation and paper because it addresses nearly all of the
    issues with PID and eliminates the contention that the reset time is too slow for self-regulating process with a large time constant (lag) to dead time ratio. By denoting these lag dominant processes as near-integrating, the reset time becomes close to what was recommended by Greg Shinskey and most specialists from 1960s to the 1990s.

    The lambda tuning rules for dead time dominant self-regulating processes on the other hand work better than the traditional rules. The traditional rules stick with setting the reset time as a factor of the dead time for all processes. Shinskey and I modified the factor as a function of the lag to dead time ratio to make the reset time an order of magnitude less for dead time dominant processes. The real solution is to use the reset time equal to the lag approaching integral only control.

    To summarize:
    1. For lag dominant processes self-regulating processes (lag > 4 dead times), the reset time is proportional to the dead time (Lambda integrating process tuning rules).
    2. For all other self-regulating processes, the reset time is equal to the time
    constant (lag) (Lambda self-regulating process tuning rules).

    Note that some tuning rules (e.g. IMC and Fertik) have the reset time set equal to a factor of the time constant plus the dead time. This works OK for moderate ratios of lag to dead
    time but does not work as well as the above (1) and (2) split of lambda tuning rules for the lag dominant and dead time dominant processes.

    I updated the AIChE presentation to show the ultimate period and actual period for the slow rolling oscillations causes by too low of a PID gain.

    I also added notes in the presentation and paper that while a lambda equal to the max dead time provides the best disturbance rejection, a more practical lambda is twice the max dead time due to unknowns. Even if you have an adaptive tuner, there can be missing dead time and process changes. The most notable source of missing dead time is from valve or VSD dead band and resolution limits because the step changes in PID output used to identify dynamics zip right through the dead band or the resolution limits whereas the PID has to work through them.”

Leave a Reply