The following discussion is part of an occasional series, "Ask the Automation Pros," authored by Greg McMillan, industry consultant, author of numerous process control books, and 2010 ISA Life Achievement Award recipient. Program administrators will collect submitted questions and solicits responses from automation professionals. Past Q&A videos are available on the ISA YouTube channel. View the playlist here. You can read all posts from this series here.

Looking for additional career guidance, or to offer support to those new to automation? Sign up for the ISA Mentor Program.

See Part 1 here. See Part 2 here. See Part 3 here. See Part 4 here.

Here we continue our series with some modeling and control mistakes seen by Greg and Dr. Russ Rhinehart, professor emeritus at Oklahoma State University, that are widespread. For more knowledge, see the ISA-5.9 Technical Report on PID Algorithms and Performance. 

Greg McMillan’s Question 

What are the biggest mistakes you have seen in automation system design, configuration, calibration, installation, checkout, commissioning, and maintenance? What were the consequences and the fixes and what can be done to prevent future occurrences?

Greg McMillan’s Answer 

Four mistakes come to mind that are pervasive, dating back 40 years or more that don’t seem to be widely recognized. My hope is that we can move forward and realize these mistakes are the source of poor modeling and process control.

1. Lack of recognition that the ultimate enemy is deadtime, because it prevents the controller and operator from seeing and correcting for a disturbance until after it has adversely affected the process. The minimum possible peak and integrated error from a disturbance is proportional to the deadtime and the deadtime squared, respectively.

Key learning points: The biggest source of deadtime in flow, pressure, most level, and well mixed vessel temperature and pH control is in the automation system design and installation. This is good news because we can minimize the problem without making changes in mechanical or process design which is expensive and disruptive and not within our responsibility. We can make valves and variable frequency drives faster and more precise, reduce transportation delays by sensor and valve location, and make sensor lags, filter times, damping settings, update rates, cycle times, and execution rates faster.

2. Lack of recognition of the deadtime missing in dynamic simulations most notably associated with mixing, transportation delays, and automation system dynamics.

Key learning points: Mixing delays can be added via a turnover time based on agitation and recirculation flows, or by putting well-mixed volumes in series. Transportation delays may be added so changes in a process stream experience a delay that is the volume divided by total flow. Automation system dynamics can be added, realizing the impact of size and that the total deadtime is the summation of pure delays and the summation of effective deadtimes from lags. For gradual rather than step changes in signals, there is an additional deadtime which is the resolution or deadband divided by the rate of change in signal.

3. Lack of recognition that most controllers are tuned for a desired setpoint response whereas particularly for lag dominant self-regulating, integrating, and runaway processes they should first be tuned for a load disturbance for process performance.

Key learning points: If you first tune for a load disturbance by momentarily putting the controller in manual and making an output change, you can simulate the effect of a load disturbance and tune the controller to meet your objective, such as arrest time. You can then use setpoint weights or lead-lag settings to get the desired setpoint response without the need to go back and retune for the load disturbance.

4. Lack of recognition that deadtime compensation in a proportional–integral–derivative (PID) controller offers better performance for unmeasured disturbances than model predictive control (MPC) for deadtime dominant and all other loops. However, the deadtime setting must be accurate and the PID much more aggressively tuned to see the improvement. Also, the PID or MPC are both much more sensitive and adversely affected by a model deadtime larger than the actual deadtime.

Key learning points: A model deadtime 10% greater than the actual dead time can cause oscillations whereas a model deadtime less than the actual dead time just causes a sluggish response and inability to tune the controller aggressively enough to see the advantage of deadtime compensation. The improvement is much greater for lag dominant loops even though this is rarely cited since these loops can usually do quite well by aggressive tuning without deadtime compensation. A deadtime block inserted into the external-reset feedback of a PID rather than the use of a traditional Smith predictor provides a much simpler implementation (no need to set an open loop process gain and time constant) and does not disrupt the operator interface (original PV is retained as the controlled variable). The deadtime is updated with actual deadtime and a small filter is put on the PID output to smooth out small high frequency oscillations that may occur due to timing issues.

Russ Rhinehart’s Answer 

I started participating in process control as a new process engineer with zero control training, seeking to learn what I needed to know along the way. There were no control experts in my community, and as a novice, my activities included designing control systems on piping and instrumentation diagrams (P&IDs).

I don’t know what mistakes I made, because I did not know what I did not know. However, I did know that control was like magic; somehow it works. I thought I understood single loop control, but was unaware of interactions, override, ratio, and nearly everything commonplace. Textbooks on control just wanted me to learn differential equations, Laplace transforms, and frequency analysis, all of which was a misdirection of what I needed.

Much later as a professor with substantial experience teaching process control courses, I saw students making many errors on P&IDs. Even after teaching PID, ratio, and cascade with ISA Standard 5.1 examples, students’ signals went the wrong way, signal types were inconsistent, etc. I doubt that even the magical mystery behind control could have made their designs work! So, I suspect my designs long ago had the same kinds of errors.

The mistakes I’d say my employer made was not providing a control expert in the design stage, or not providing practical control training for us process engineers. The mistakes I made were seeking help from academic authors.