Interested in linking to "Going, Going, Gone–Part 1"?
You may use the Headline, Deck, Byline and URL of this article on your Web site. To link to this article, select and copy the HTML code below and paste it on your own Web site.
By Greg McMillan and Stan Weiner
Greg McMillan and Stan Weiner bring their wits and more than 66 years of process control experience to bear on your questions, comments, and problems. Write to them at email@example.com.
Stan: I expect the process industry in the U.S. has lost at least 25% of its expertise. For mature products whose manufacturing has gone overseas, 75% or more of the specialists have left, leaving only a shell of technical capability.
Greg: A number of major companies estimate that almost 50% of their engineers will be eligible to retire in the next five years, but the stock market bottom may delay retirement, so there is a window of opportunity to capture this expertise before it's gone forever.
Stan: When we retired, there was no attempt to save our expertise. Over 100 of the best minds in process modeling and control retired, and the company made no attempt to retain even a snippet of their knowledge. The lesson here is: if the people who make the decisions on staffing don't understand the value of your expertise, you will be given a package to leave. Hmm, maybe that is what you want.
Greg: There was a misguided attempt to put technical expertise online by the use of expert systems. About a dozen people concentrated on this for about five years. Seemed like a great idea, until you realize that the flexibility of just adding ad hoc and heuristic rules leads to a disorganized and fuzzy knowledge base. So we added fuzzy logic. Need I say more? The expert systems didn't transfer knowledge. There was one incident where an expert system identified an equipment problem, but the operators said they could have figured it out anyway.
Stan: Then we went with the idea of data mining with neural networks. Just think of all the info there must be in the gobs of historian data. Just dump it into software, and out will come wonderful and insightful results. Believe it or not, some people bought this line from the software supplier, and even thought neural networks could predict a composition or quality that was not measured in the field or the lab. About a half dozen specialists worked for five years on putting the most powerful neural network software online. Just as with expert systems, only the architect seemed to be gaining process knowledge. As soon as the specialist left the unit, the system was turned off. The half million dollars spent on expert and neural network software was eclipsed by the $10 million spent on engineering.
Greg: On the other hand, the few ventures into model predictive control (MPC) were successful with some big results. On the downside, an inordinate amount of time was spent learning, setting up special computers and interfacing the MPC software.
Stan: Nearly all of the 5% cost of goods (COG) that was saved per year by process control improvement came from tuning controllers, shifting set points, improving measurements and valves, implementing feed-forward and adding feedback control to batch columns and batch reactors.
Greg: The most successful engineers making process control improvements had a combination of process, modeling, control, configuration and instrumentation skills. They could conceive an idea, estimate its value, detail implementation, and take the improvement all the way from concept through startup. The best had the deepest process understanding. Not only is this a model for the most effective process control engineer, it may be the only way forward for chemical and petroleum companies that cut back on specialists and rely more on outside contractors and vendors. Except for small projects, the process control engineer becomes a coordinator of outside services.
Stan: The optimum engineer has to understand which measurements and valves are winners and which are losers, so the design isn't based solely on lowest price and the project doesn't end up with a “performance challenged" automation system (à la package equipment).
Greg: He must also understand process relationships and process metrics for batch and continuous processes to focus on what's important and the value of an improvement. This necessitates a basic understanding of material and energy balances, mass transfer, process gains and process costs. Process simulations can provide this knowledge if they're easy to use and the groundwork has been done in terms of physical properties, components and parameters. To estimate the value of reducing process variability, the engineer must know how to transfer variability from key process outputs to the appropriate process inputs, be able to estimate the attenuation of oscillations by volumes, and move the set point of key operating conditions to capture the value (see the "Cost and Source of Oscillations, Parts 1-4," April 20–May 11, 2009, entries on www.modelingandcontrol.com). Dynamic simulations could help develop the concept considerably and prototype the implementation if the actual DCS configuration is used. The idea of duplicating all the many capabilities and nuances of a DCS (even just the PID) in a dynamic simulation is ridiculous from a standpoint of effort and fidelity. The engineer should know how to configure to take advantage of the power of the DCS.
Stan: The optimum engineer must understand measurement and valve dynamics; otherwise, thermocouples for exothermic reactor control will be installed in glass-lined baffles and on-off piping valves posing as control valves will be bought. Most dynamic simulations don't get into these details that will make or break or blow up an installation.
Greg: My pet peeve is that control valves in simulations don't have the installed flow characteristics, stick-slip and backlash modeled. Also, controller tuning is thought to be just a side issue. Many process consultants and professors apparently don't realize that the integrated error for load upsets is proportional to the integral time and inversely proportional to the controller gain. The detuning of the controller is equivalent to introducing dead time into a loop. A seriously detuned controller becomes so sluggish it approaches manual control. If the disturbances are really small or extremely slow, which is generally the case for bioreactors, the effect of tuning is not seen, which is why the biopharmaceutical industry doesn't understand all the fuss about tuning.
Stan: Universities could help engineers get started with these skills by teaching batch and continuous process first principles, online dynamic process yields and COGs, the dynamics of process and automation systems, process variability, and the best technology in measurement and valves. Unit operation and computer labs should use industrial software for modeling and control of batch and continuous processes, most notably the DCS, so the students are introduced to the terminology and capability of the today's industrial control systems.
Greg: Next month we will discuss with Nick Sands what can be done to save and promote process control expertise.