How to diagnose PID tuning problems

Greg: We’re fortunate to have an independent, advanced process control (APC) consultant, Vivek R. Dabholkar, share his insights on practical techniques to help us get smarter at proactively recognizing the appearance and causes of PID tuning problems. Vivek previously shared his expertise in March 2024’s Control Talk column, “Achieving plantwide, multivariable control.” 

Vivek began his career as a pyrolysis modeler/APC engineer at Stone & Webster Corp., before working for Setpoint, a consulting firm acquired by AspenTech. Later, he was hired as the first employee at Applied Manufacturing Technologies. In addition, he has 15 years of operating experience as a senior staff engineer at ExxonMobil’s olefins plant in Baytown, Texas, and at ExxonMobil and Saudi Basic Industries Corp.’s (SABIC) joint venture, Kemya, where he improved bottom-line profitability by improvising and implementing plantwide dynamic matrix control (DMC) by more than $35 million per year. Following a short stay at Shell Polymers’ new cracker in Monaca, Pa., he improved the performance of the composite at Dow Chemicals by retesting and reidentifying the cold side of an olefins plant in Canada, enabling it to push-feed against changing, backend plant constraints, making more than $10 million per year. He also improved Ineos’ olefins plant-2, saving more than $1 million per year by improving C2-splitter operations, substantially reducing bottoms ethylene loss, and stabilizing 900# steam header within three months. 

Vivek, what should we understand about PID tuning issues?

Vivek: The Internet is full of software, journal articles, magazine publications and books on PID control, yet the key factors in successful PID tuning are missed by many. Academia, and to a large extent the industry, is all about tuning rules. There’s even a handbook on PID tuning rules with more than 400 pages of rules. There’s a lot of disagreement and little guidance about the advantages and disadvantages of various rules.

Greg: I addressed these issues in my book, PID Tuning and Control Loop Performance, fourth edition, 2014 free to download at the https://www.prosera.com/store There are chapters on fundamentals, a united methodology and performance criteria, which I developed with Greg Shinskey. There are also chapters providing equations and test results for the effects of process dynamics, controller dynamics, measurement dynamics, valve and variable frequency drive dynamics, disturbances, nonlinearities and interactions.

What do you need from the start? 

Vivek: Strong commitment. Be ready to take your laptop to the control room, and engage with the operators on changes you’re planning to make. Nothing upsets them more than you making changes to PID tuning, setpoint (SP) or mode without telling them. They bear the consequences at odd hours when no one is around. Be sensitive to their concerns. 

Take ownership for the outcome. Make sure you’re trending SP/process variable (PV) /output (OP) and major disturbances affecting PID. Many young engineers are afraid to make sizeable changes to PID tuning, and it becomes a deterrent to progress. If you’re closely observing, you can always undo a change. So, if you’re planning to leave within couple of hours or before the weekend, don’t make changes to tuning (unless you’re authorized to observe, communicate and make necessary changes remotely over the weekend). Building trust with those who run the process often pays dividends in the long run. 

Most operating companies in our industry have poorly tuned PID controllers, so the first job for an experienced APC engineer is to fix the PID controllers. At one plant I worked at, steam header pressure (SHP) swung by 40 psig, either venting SHP steam to atmosphere or sending refrigerant flows to flare on high-suction pressures—and the plant-manager and CEO were aware of them. To top it off, a well-known consulting firm had implemented dynamic matrix control (DMC) on the olefins unit. After using P on PV algorithm, and retuning SHP control by using much larger controller gain, pressure variation was brought down to less than a psig. Operators were ecstatic after many years of neglect and abuse.

Greg: What about instrument scale range and signal filtering?

Vivek: If PV instrument scale range was increased or decreased recently, then the gain needs to be adjusted proportionately higher or lower, respectively. If not, the PID will be sluggish (in case of PV range increase) or unstable (in case of decrease of PV range) assuming the PID was properly tuned. If PV signal is noisy, then put a lightest filter on PV that addresses the issue, keeps the PID OP within the valve resolution limit, and is less than 10% of the PID integral time. 

Greg: How can we diagnose cause of continuous equal amplitude oscillations?

Vivek: Check how the PV responds to OP changes when PID loop is in AUTO. Plot them on the same plot. If you see, for a self-regulation process, a triangular pattern in OP and rectangular pattern in PV, then the valve is sticking (poor valve resolution). This pattern emerges due to fact that, even though PID OP is changing, the valve position in the field is not changing.

For a near-integrating and true-integrating loop, the PV has a triangular pattern, and the OP has a sinusoidal pattern. This can also happen after you tune the PID well due to excessive valve packing by the environmental department to reduce emissions. No amount of tuning will stop these limit cycles if there is integrating action in somewhere in the positioner, processes or controller. Increasing the integral time or decreasing the PID gain will increase the limit cycle period. The amplitude is set by the process gain.

Greg: How can you tell if the proportional, integral or derivative action is too aggressive?

Vivek: Observe how PV turns in relation to OP. If PV already has turned (local maximum/minimum) and OP is still going in the same direction, then the integral time is too short (excessive reset action), increasing integral time up by at least 50% (not by 2-5%). If OP turns ahead of PV, then one explanation is derivative action. Derivative can be used for temperature controllers after adequately filtering the signal, or else it’s counterproductive due to overreaction to noise. 

Another explanation is PD on PV algorithm with high controller gain, which is trying to bring PV to its SP as fast as possible without a turn in PV. The turn of PV in relation to OP observation comes from the late Prof. James Riggs of Texas Tech University. If the PV turns towards SP, then there’s no sense in continuing to increase/decrease OP through integral action. It’s like having a manager penalizing employees, even though their behavior has markedly changed. 

On the other side, if integral time is excessive, then PV would wander around on one side of SP. It’s like having an ineffective manager with zero control action. In case of evaluating responses to large or increasing disturbances, you may find OP not turning when PV turns. This is acceptable provided the delay in turning doesn’t cause an excessive increase in peak error or integrated error from unmeasured load disturbances. Also, wandering is fine if the separation between the PV and SP is negligible and within the PV’s resolution limits. The offset from proportional action is inversely proportional to PID gain, which means the offset is negligible for large PID gains needed in highly exothermic reaction temperature control. In case of evaluating responses to large disturbances, you may find OP turning much later when PV turns. This is acceptable if delay in turning isn’t excessive enough to cause excessive ringing.

The late Dr. Charlie Cutler verbally shared his wisdom with me on one occasion. PID Proportional action always opposes change in PV, regardless of direction towards or away from the SP. and integral action is trying to bring PV to SP but for SP changes or unmeasured disturbances proportional mode is taking proactive action to reach SP and integral action is ceaselessly trying to bring PV to SP even if it is going to cause overshoot. Integral action has no sense of direction and is never satisfied. Good PID tuning strikes balance between the two while handling load disturbances. In most industrial settings control issues are caused by too short an integral time. As noted in April 2025’s Control Talk column, “Examining PID tuning essentials,” in many composition, level, temperature and pressure control cases, the reset action was too high by a factor of 10-100. James Beall observed the same problem. This is an issue that needs to be addressed. As a result, first set integral time large to get it out of the way, and tune proportional gain. Once satisfied with disturbance rejection, reduce integral time to bring PV to SP, knowing there’s built-in control lag due to process lag time and deadtime. 

Greg: My article debunks several misleading concepts, such as control literature that incorrectly places disturbances on process outputs instead of process inputs, which creates misunderstandings about process time constants. The consequences of these misleading concepts and misunderstandings have been severe. They can be seen in tuning settings for SP changes that have a PID gain that’s an order of magnitude too low and have no rate action. I strongly advocate classifying self-regulating processes with a process time constant greater than four times the deadtime as near-integrating processes. They should also use integrating process tuning rules that deliver high gain and rate action designed for load disturbances. Tuning for load disturbances can be simply tested by momentarily putting the controller in manual and making an output change. 

How do you tune for setpoint changes? 

Vivek: Request SP changes on PID in the direction that’s acceptable to the operators. There’s usually always one direction available. While high gain is helpful in rejecting disturbance because it’s always opposing the change caused by disturbance, it will cause large overshoot/undershoot on SP change if the PID algorithm is in error. Instead, pressure and temperature PID controllers change PID algorithms to proportional (P) and derivative (D) on PV. Then you may increase controller gain by at least a factor of two or more for better disturbance rejection according to Brendan Minter, president of AMT. 

Get your subscription to Control's tri-weekly newsletter.

This has another advantage for the operator. The algorithm PD on PV is forgiving of fat-fingering during SP change, which could otherwise cause a large upset, such as on-suction pressure or tower-pressure SP changes, if operators correct their error within a reasonable time. Another advantage of using P on PV is when tower overhead flow feeds the reactor. It avoids proportional kick on setpoint change to tower overhead pressure. However, excessively cranking up controller gain will lead to non-monotonic changes in OP, such as undesirable feed flow to reactor. Usually, using P on PV is discouraged on secondary PID. However, in one case, a deethanizer’s reflux drum level was controlled by a temperature controller in the front-end feed (Shell and Linde ethane cracker). In this case, P on PV was configured on temperature control (TC) for better disturbance rejection, so we couldn’t use P on error. 

Greg: A good solution is to use a controller with 2 degrees of freedom (2DOF) structure, so you can adjust the SP weights for proportional and derivative action. The proportional and derivative action on SP change is their respective setpoint weights beta and gamma multiplied by the setpoint change. If you don’t have a 2DOF option, you can use a setpoint lead-lag advocated by Greg Shinskey. The controller is tuned first for good load disturbance rejection, and then 2DOF setpoint weights or setpoint lead-lag are adjusted to achieve a fast SP response with negligible overshoot.

What about cascade control? 

Vivek: Observe whether the secondary flow controller in TC-to-flow control (FC) or pressure control (PC)-to-FC cascade in different valve regimes. In most cases, you’ll need output characterization in the distributed control system (DCS) to account for large non-linearity in flow vs output. I prefer using variable gains with piece-wise linear (PWLN) function available in most DCSs. Use it with a logic block to check bad value (in case of updating PWLN gain characterization) before pushing it as a gain to PID controller. 

This approach is lot better than valve characterization. It offers many advantages, and users don’t have to worry about air-to-open vs close characterization. This can get tricky, for example, in case of an air-to-close furnace damper OP. In fact, I’ve seen a furnace shut down by DCS Engineer due to an incorrect configuration. Next, users don’t have to reconfigure existing loops with output characterization. You can change your mind on PWLN based on data available for valve OP range, without reconfiguring each time.

Greg: What are some advanced considerations?

Vivek: Hopefully, distillation column TC location is suitable for effective control of composition. Plot relevant composition with column temperatures available as a function of time. Choose temperature that distinctly captures composition movement.

If tuning distillation column TC, first assess whether TC is ramp or steady-state variable in open loop. For this, observe TC in closed-loop (auto) and make a SP change, and ask if the reboiler flow changes at nearly constant reflux and feed? If not (net gain zero), TC is a ramp, high PID gain Kc is needed, and integral time TI > ~ 10 mins. If you fail to make that “ramp” distinction and do an open loop test recommended by the textbooks, you’ll regret your decision. For TC as a ramp, treating it as steady-state variable will make process gain very large, and make Kc small and Ti very long. Good luck controlling TC with those settings.

In case of SHP control , where number of boilers cascaded to PC change, controller gain must be adjusted by a factor equal to the number of cascaded boilers when the PC was tuned divided by the number of boilers cascaded to the PC now. If not, you may either face sluggish control or instability in case of high gain with P on PV action.

Greg: Here are some diagnostics from my Process/industrial Instruments and Controls Handbook, sixth edition, McGraw-Hill, 2019. In items 1-3, the lambda in parentheses is the lambda tuning parameter that corresponds to the cited gain and phase margin.

1. Gain, phase margin = 3, 61o (λ = θ_o) for max load rejection for fixed, known dynamics
2. Gain, phase margin = 6, 76o (λ = 3θ_o) to deal with adverse changes in dynamics < 5
3. Gain, phase margin = 9, 81o (λ = 5θ_o) to reduce resonance and interaction
4. Ultimate period is 2x to 4x dead time for self-regulating processes, and is 4x dead time for other processes using first order plus dead time approximation 
5. Resonance can occur for disturbance oscillation periods 2x => 10x dead time
6. PID execution rate and filter time should be < 0.2x and 0.1x dead time, respectively
7. Oscillation periods < 4x dead time indicate PID gain or rate time that’s too large
8. Oscillation periods 5x => 10x dead time indicate a reset time that’s too fast
9. Oscillation periods > 10x dead time indicate control valve problems (backlash, stiction, slow stroking or rate limiting, and poor actuator and positioner sensitivity)
10. Oscillation periods approaching or exceeding 40x dead time indicate a PID gain and/or reset time that’s too low for near-integrating, integrating and runaway processes