Modeling and Control

My core dump of myths went on longer than intended because there were so many stuck in my brain, it was so freeing, and it was so easy to unload them when pressed for time.

My recent spike in work load has put me in danger of being kicked out of various retirement associations including the Retired Automation Professionals (RAP). Apparently, my excursion into 40+ hour weeks means I not retired. Personally I think this is a bad rap. I still take 12 weeks off a year to visit friends, relatives, national parks, and beaches. I am having too much fun being part of the future to give it all up. Plus there are the benefits of active membership in the Adorable Automators Association as noted in my April 2008 Control Talk. http://www.controlglobal.com/articles/2008/118.html

The myths provided a reality check and a basis for looking forward to new tools that address many of the issues raised. Friendlier and more proficient versions of all of the process control technologies are being embedded as standard tools in a DCS. Now you can explore, discover, prototype, justify, and deploy process control improvements in the same configuration environment sued for the basic control system. Adaptive tuning, a rich spectrum of PID enhancements, fuzzy logic control, loop performance monitoring, model predictive controllers, neural networks, on-demand tuning (auto tuners), and process dynamics identification are presently embedded. Soon online data analytics (multivariate statistical process control) and process modeling capability will be added. Terry Blevins (principal technologist) and Mark Nixon (chief architect) had this vision at Emerson Process Management and made it happen in DeltaV.

This comes at a turning point in industry where most of expertise in the application and understanding of the value of control opportunities are becoming full time members of RAP. Also, most of the opportunities are now overseas. The change in demographics is obvious when you look at the weekly questions on process automation submitted to Liptak where 95% of the questions are from overseas and cover a wide range of practical and essential application issues.

When I was helping the Instrumentation Systems and Automation (ISA) society in the early stages of the development of the Certification of Automation Professionals (CAP) program, I realized that it was difficult to find a book that addressed the day to day needs. I realized that publications in our field including my own were at too high a level and assumed too much for the new workplace where mentors and company training programs are scarce. Also books on process control were too mathematical and theoretical and the books on instrumentation were often a rote description of the principles of operation. Not much was offered on selection, application, installation, performance, and maintenance for the extensive range of process types and conditions needed by relatively inexperienced professionals to do their job. The Automation Book of Knowledge (ABoK) developed as part of CAP and various handbooks by Liptak, Boyes, and myself help but much remains to be done. Toward this goal, I am looking forward to working with Terry Blevins and Mark Nixon to provide a hands-on learning source/guide employing a full suite of embedded technologies in a virtual plant. This book focuses on the opportunity of doing a better job of process control for both batch and continuous processes.

Technorati Tags: adaptive control | controller tuning | data analytics | fuzzy logic | loop performance | process control | process dynamics | process models | model predictive control | neural networks |

I dug up the following myths from my April 2006 Control Talk column in Control Magazine. I am into recycling and going green. In fact these myths may be a bit moldy.

(26) Auto tuners can compute controller tuning settings with an accuracy of more than one significant figure. Act surprised when unmeasured disturbances, load changes, valve stick-slip, and noise cause each result to be different. Look forward to the opportunity to play bingo with the second digit.

(27) You can just dump all your historical data into a neural network and get wonderful results. Forget about the same stuff that cause auto tuners to have problems and use variables drawing straight lines because anything that smooth or well controlled must be important. Use the controlled variables (process variables) instead of the manipulated variables (controller outputs). Don’t try to avoid extraneous inputs or identification of the control algorithm instead of the process. If you want to purse a career in data processing, use every variable.

(28) Models can predict a process variable that is not measured in the field or lab. Great way to spur creativity in training a neural network and validating a first principal model plus it has the added bonus of the model never being wrong. Wait till your customers figure out something is wrong with the composition of your product. Discount as hearsay any suggestions that even the best models need periodic correction.

(29) To reduce variability in process outputs (temperatures and compositions), keep all the process inputs (flows) constant. Keep believing that you can fix both the process inputs and outputs and don’t accept the notion that process control must transfer variability from process outputs to process inputs to compensate for disturbances.

(30) Positioners should not be used on fast loops. This was true for the good old days of pneumatic positioners and analog controllers. Surely, digital positioners with tuning settings and digital control system scan times can’t make the original theoretical concerns less important than the practical issues of real valves. If you would rather believe the controller outputs are the actual valve positions, and just want valve problems to slip by, save some bucks on your project and only put positioners on slow loops. Just don’t stick around for start up.

Technorati Tags: | loop performance | process control | process dynamics | process models | neural | networks | soft sensors | valve dynamics |

We conclude with the following myths from Appendix D in my guide Models Unleashed - Virtual Plant and Model Predictive Control Applications published by ISA, 2004.

(21) You need an advanced degree to do advanced control. Not so anymore. New software packages used to form a virtual plant automate much of the expertise needed and eliminate the need for special interfaces. The user can now focus mostly on the application and the goal.

(22) Dynamic simulations and model based control are only applicable to continuous processes. Since most of the applications are in the continuous industry, this is a common misconception. While it is true that a steady state simulation is not valid since there is by definition no steady state in batch, dynamic simulation can follow a batch as long as the software can handle zero flows and empty vessels. Model predictive control (MPC), which looks at trajectories, is suitable for the optimization of fed batch processes during particularly important points in the batch cycle. The opportunities to improve a process’s efficiency by MPC are about 25% for batch compared to 5% for continuous operations.

(23) You need consultants to maintain models and advanced control systems. No longer necessarily true. The ease of use of new software allows the user to get much more involved, which is critical to make sure the plant gets the most value out of the models. Previously, the benefits started to drop as soon as the consultant left the job site. Now the user should be able to tune, troubleshoot, and update the models.

(24) You don’t need good operator displays and training for well designed advanced control systems. The operators are the biggest constraint in most plants. Even if the models used for real time optimization and model based control are perfect, operators will take these systems offline if they don’t understand them. The new guy in town is always suspect, so the first time there is an operational problem and there is no one around to answer questions, MPC systems are turned off even if they are doing the right thing. Training sessions and displays should show the individual contribution of the trajectories of each controlled, disturbance, and constraint variable to the observed changes in the manipulated and optimization variables.

(25) You need to know your process before you start a model based control system application. This would be nice, but often the benefits from a model stems from the knowledge discovery during the systematic building and identification procedures. Frequently, the understanding gained from developing models leads to immediate benefits in terms of better set points and instruments. The commissioning of the MPC is the icing on the cake and locks in benefits for varying plant conditions.

Technorati Tags: dynamic simulation | model predictive control | loop performance | process control | process dynamics | virtual plant |

To blog or not to blog, that is the question. When on vacation, I choose not to. Now I am back and ready to blog on. Here are some myths from my days (and nights) at Monsanto improving pH loops.

(16) More frequent buffer calibrations improve pH measurement accuracy – unless an electrode needs to be removed to be cleaned, it is better to leave it in the process and do an adjustment based on a statistical average of process samples. The buffer is not representative of the process in terms of liquid junction potentials and the glass electrode response especially when the process has salts or strong acids and bases. The reference electrode can take up to a day to reach equilibrium with the process when it is reinserted. Also, wiping the glass electrode or soaking it in cleaning solutions reduces its life expectancy. Finally, the effect of temperature on pH is not found via a buffer which leads to the next myth.

(17) The pH electrode temperature compensator corrects for changes in pH with temperature – the standard temperature compensation is the change in millivolts per pH unit per the Nernst equation. The change in the actual solution pH due to the change in dissociation constants with temperature requires another correction. Many smart pH transmitters now have this solution pH temperature compensation but it is up to the user to find the relationship. Information on how the dissociation constants change with temperature and how this would affect a complex solution is scarce to nonexistent so it requires a pH sample’s temperature to be varied in the lab. For the simple case of water and a strong base (e.g. caustic) where the effect is dominated by the change in the water dissociation constant with temperature, the error is about -0.03 pH per degree C.

(18) A second pH electrode always improves measurement accuracy and reliability – a second electrode offers more questions than answers since they never agree unless you are lucky enough to have a problem that a smart transmitter can detect. Every operator has a favorite electrode based on some interesting war stories but its anybodies guess as to which is right. Also, since electrodes can fail or develop errors in almost any direction, a decision to select the best electrode requires some extra intelligence such as changes in electrode resistance or a comparison of response times. At Monsanto we used middle signal selection of three electrodes to automatically and inherently ride out a failure of any type and reduce the noise and short term errors for concentration gradients.

(19) The rangeability of a control valve is per catalog specifications – the rangeability statements by valve suppliers usually does not take into account the stick-slip near the closed position. Many rotary valves carry impressive rangeability statements but don’t deliver the goods because the control at low flows is a big saw tooth from high seal friction and breakaway torque.

(20) A valve with a positioner is a good throttling valve – the response of a control valve depends upon the entire package (valve, seat, seal, actuator, packing, positioner, and feedback mechanism). Putting a positioner on an on-off valve (e.g. block or isolation valve) does not make it a throttling control valve. In fact the feedback mechanism may be lying to the positioner in which case the smart diagnostics mean nothing. For more info on this check out the article “Improve Control Loop Performance”

I have “Final Four” on my mind and I am off to see if my alma Mata Kansas can upset UNC but before I go, if you ever come to Austin check out the Oasis for sunsets and its new Starlight bar for star gazing over Lake Travis. I spent last night at the Oasis Starlight listening to the Eggmen (a great Beatles cover band) and having flashbacks to the 1960s and 1970s. Too bad I didn't wear my plaid bellbottoms.

Technorati Tags: loop performance | pH | process control | process dynamics valve dynamics |

My head is spinning with myths. The examples seem endless.

(11) The process dynamics can be identified with the controller in auto without any perturbation – a loop startup or shutdown, change in set point, or an injection of a pulse or step in the controller output is needed to sort out the process dynamics from the controller dynamics. Bob Otto (Monsanto Fellow) alerted me to this dilemma 20 years ago. Cecil Smith published an article about 5 years ago making the same point. Now if you have inside knowledge of the process gain, it may be possible to find the process time constant from estimates of the dead time. However, even sophisticated process simulations have difficulty in providing an accurate process gain. So why not face up to the situation and benefit from perturbations? Batch processes often have plenty of perturbations because there is a loop startup for each batch and changes in the loop set point and output by the batch sequence. New closed loop process identification tools such as DeltaV Insight can do a good job of taking advantage of these batch opportunities. Continuous processes time often run at the same set point long periods of time and require a periodic injection of a pulse into the output. Fortunately, the size and duration of the pulse can be rather innocuous in most cases.

(12) Model predictive control is not suitable for batch operations – the reason frequently given is that the process is too nonlinear but that’s old news. The world of industrial process control is nonlinear. The real issue is more the one direction integrating response for bioreactors and some chemical reactors. If you make a translation of the controlled variable from composition to slope of the composition profile, an MPC can control a trajectory of a key component over a key portion of the batch. While the MPC applications in continuous processes dominate the literature, there have been successes in batch temperature and composition control with far greater benefits (e.g. 25% increase in capacity by 25% reduction in cycle time).

(13) Rate is more trouble than it is worth – don’t try controlling a severely exothermic reactor without rate unless you like exercising the plant relief system and alarm system. In general, for temperature control, rate should be set equal to 1/5 the process time constant or thermowell lag, whichever is largest.

(14) You need to start and end at a steady state to tune a loop in manual – there is no steady state for integrating and runaway processes. Even self-regulating processes may be moving with the controller in manual from frequent upsets. The short cut tuning method in my Good Tuning - Pocket Guide works well for these less than ideal conditions.

(15) Process dynamics are in the process – the most common loops are flow and pressure and for these loops most of the dynamics are in the automation system (valve, transmitter/sensor, and the PID execution and signal filtering). The process delay and nonlinearity is negligible compared to that of a control valve and the process lag is negligible compared to measurement filters. The term “process dynamics” is a misnomer. A better term would be “open loop dynamics” so people would better realize the dynamics are often in the automation system. The reality is I can’t get people despite 20 years of publications to think “open loop gain” instead of “process gain” and “open loop time constant” instead of “process time constant”, and “total loop dead time” instead of “process dead time.” Maybe I need some catchy phrases like “don’t blame it all on the process” or “the only loop with only process dynamics is in your college textbook.”

Technorati Tags: controller tuning | loop performance | process control | process dynamics valve dynamics |

Before I leave to enjoy a crawfish boil and Cajun music, here are some more myths.

(6) Loop oscillations can always be decreased by reducing the controller gain – too low of a controller gain can cause nearly sustained slow rolling oscillations in an integrating process (e.g. level or gas pressure) and instability in a runaway process (e.g. exothermic reactor temperature). Decreasing the gain makes the problem worse for these cases. There are also limit cycles that persist regardless of controller gain.

(7) Limit cycles can always be stopped by eliminating valve stick-slip – limit cycles can also be caused by output limits, IO card resolution limits, deadband in integrating processes and cascade loops, and extreme nonlinearities, such as the pH titration curve.

(8) The installed valve resolution and deadband meet catalog specifications – often tests by manufacturers are for hand tight packing at positions remote from the seating friction. Also, temperature and fouling can make the installed performance worse.

(9) The most accurate type of pH sensors are used most often – the most popular sensors are the ones that require the least amount of maintenance, such as references with solid electrolytes, even though these may require more time to equilibrate and have a more variable junction potential. The flowing liquid junction reference for the right materials of construction and electrolyte is generally the most accurate but the least used type of pH electrode in industry because of the need to pressurize and refill the reservoir.

(10) Thermocouples are faster than RTDs – while this would be true for a bare element, nearly all the installations in the chemical industry I have seen use thermowells. The fit, design, and materials of construction of the thermowell have a far greater effect than the sensor on the speed of response of the temperature measurement.

Technorati Tags: controller tuning | loop performance | process control | process dynamics valve dynamics |

Myths are a fertile topic maybe because of all the fertilizer in process control. You can make almost any point you want, by changing what are often obscure details on process and automation system dynamics. For example, you can show a variable speed drive can do better or worse than a control valve. The results can easily be swayed by VSD settings (e.g. deadband and rate limiting), VSD options (e.g tachometer feedback and vector control), and valve type\accessories (e.g. throttling sliding stem or rotary isolation valve and digital dual relay positioner versus pneumatic spool positioner). For insights into the relative merits of the VSD versus control valve in terms of control loop performance, check out the February Control Talk column in Control magazine titled "Deal or No Deal.”

I promised to post this week the development of equations that are a myth buster. The equations show there is an implied dead time greater than the actual dead time in most loops because the controller is tuned to be slower than what is shown in academic articles and papers. Control loop performance does not appreciably deteriorate until the actual dead time exceeds the implied dead time. The equations go on to provide an estimation of the peak and integrated absolute errors for the implied and actual dead times for step disturbances. The effect of the slowness of real life load disturbances can be roughly included by adding the load disturbance time constant to the process time constant in the equations for the peak and integrated errors. The first page appeared in a blog and Control Talk column in 2006. This updated document has better explanations/nomenclature and adds a second page for the estimation of the peak and integrated errors. When I bounce out of negative free time, I will do an application note to study the accuracy and implications of the equations. Next week we will continue on with mythology.

ScanTimeEffectonPeakandIntegratedErrors

Technorati Tags: controller tuning | loop performance | process control | process dynamics |