Beyond automation: Humans as process controllers
While automated control systems can keep our manufacturing processes running efficiently, new variables are entering into the production equation that are beyond what we can expect from PID. In an interview, Peter Martin explains forces pushing a return to humans as controllers to accommodate these changes, at least for a while.
Peter Martin, PhD, is vice president, business value solutions for Invensys Operations Management. He has emerged as something of a control strategy futurist, looking for how our concepts of process control need to evolve as business-related demands on manufacturing change. Control Engineering contributors Vance VanDoren and Peter Welander asked the questions.
CE: Over the last year or two, you have made comments about how process industries are changing, and suggested that you expect a larger role for humans in control functions. With the growing importance of automation, this seems counter-intuitive. Where do you see this headed?
What we see going on is that automatic control (and manual control) have been applied to controlling the efficiency of plants for many, many years. It’s been going on for a long time, and we’re pretty good at it. Over the years, we’ve been able to replace human decisions with automatic decisions, especially in a more real-time world where automatic controls can make decisions much better, more effectively, and more quickly. I don’t see a reversal of that. What I do see going on is that the critical business variables in these plants are starting to change in real time. So, for example, 15 years ago, companies used to be able to develop contracts with their electricity suppliers for a year at a time, essentially relegating the price of electricity to a constant for the contract period. With the price as a constant, all you really needed to control was consumption, and by reducing consumption, it directly translated into profitability. With the opening of the power grid and deregulation, all of a sudden the price, not only the usage of electricity, but the price is changing more frequently than it ever has. In fact in the U.S., on the open market, the price changes about every 15 minutes.
Historically, we’ve applied control theory to the plant floor, and we’ve applied management theory to the business. That made some sense because all of the critical business variables didn’t change within a given month. You could use monthly information from SAP or Oracle, and you’d be getting measures of energy costs, material costs, and product value, all of those things that were fairly stable and could be managed with monthly data. Today with the opening of the power grid and the domino effect that it’s caused, all of a sudden we’re seeing not just electricity costs, but the price for natural gas changes every 15 minutes. Similar things are happening to some of the materials used in production process, especially heavy process industries. If you watch the price of critical metals like copper, they might change multiple times per minute. If one of those is a raw material in your manufacturing process, that’s a new dynamic introduced to your manufacturing process—a high-speed change that’s never been there before. And if you’re going to play that correctly, perhaps the products that are being produced should be adjusted to reflect that and reflect market demand. The three critical business variables—production value, energy costs, and material costs—are changing multiple times daily, and companies are still trying to manage them. It’s not that I see a lessening of applying control theory on the plant floor, but rather I see controlling the business side of industry as more of a real time control problem than it’s ever been, and we believe that we have to apply real time control theory to those critical business variables.
That said, it’s not easy to apply something like PID control because you don’t get the natural periods of the loops in the business side that you get in the process side. Therefore, what we see going on is that humans have to jump in and be controllers of the critical profitability variable almost in the same way they were 100 years ago for the process variables. Back then we set operators on a hand valve looking at a gage, and we said to the operator, “Look, when the needle goes this way, turn the valve this way, and when the needle goes the other way, turn the valve the other way.” We used humans as controllers. The interesting thing about it is that humans did a pretty good job. What I see going on, in terms of control theory, is that humans are getting involved more in controlling profit. A lot of people think, “You’re talking about business managers.” But no, I’m talking about operators and maintenance people. When an operator changes the set point of a loop, let’s say a temperature loop from 400 to 410 degrees, from a business point of view, that either added value or destroyed value. There’s no other alternative. That type of change is either creating or destroying economic value. Just like in the old days of manual process control, if we can stick operators in front of a gage that will show them what the impact is, in terms of the business, of every activity and action they take, then over time the operators can learn how to take actions and how to perform activities that will drive the most value. That’s where I see much more manual control than we would have seen 10 years ago, but not at the process level, it’s at the business level. But it’s the same people—it’s the operators or the maintenance people learning how their actions and activities impact the profitability of the plant. In reality, that is feedback control.
It’s the difference between control and management. Management is when you can’t control something. If you can control it, do. If you can’t control it, manage it. We’re getting to the point in business where the traditional management constructs, like using monthly reports to manage your business, are truly becoming obsolete. It’s not that we don’t need the monthly reports, but you can’t use that same monthly data to manage the performance of your operations, because the operations are moving so fast that the speed of the business precludes running it monthly. So if you can’t run it monthly, you have to run it in the time frame in which it changes, which is essentially becoming real time. Then the people that become the business managers, who are the manual controllers of profitability, are the operators, maintenance people, supervisors, and engineers.
CE: But won’t all these parameters taken together threaten to exceed the ability of human operators to keep up? Won’t we have to develop some kind of automatic control at the management level just the way we did in the 1930s at the process level?
I absolutely think that is going to have to happen. That’s the bad news. We all need to be looking for how to do automatic control of all these variables. Unfortunately, it’s not as simple as applying PID. We have to be looking for that automatic control algorithm. The good news is that in most parts of the world, governments have jumped in and regulated the time in which these variables can change. For example, if you look at the price of electricity on the open grid, in the U.S., it changes every 15 minutes. That interval is not because of any business or physical reality but because the government says, “You can’t change any faster because we can’t keep up with it.”
When we first started looking at this, everybody said, “Yes, this can happen for energy, but it’s never going to happen for raw materials because people have too much inventory, and the inventory itself will slow things down.” That tends to be true. The inventory does add a capacity buffer effect, but what I’m seeing going on right now is business managers who understand what’s going on have two dynamic problems: One is controlling the business, the second is controlling the physical process. We see a lot of people rethinking the physical processes themselves.
North West Redwater Partnership is building a new refinery near Edmonton. We haven’t built a new refinery in North America in decades, so why are they doing that? Today, when you produce crude from the oil sands in the Athabasca range, you put it in a pipeline and pump it down to the U.S. or wherever you’re shipping it to be refined. During that trip it ends up being in the pipeline for three days, and during that period the price of that crude may have changed 120 times, and there’s not a thing a business manager can do about it. What we see going on is that the whole concept around storage and inventory has to be altered along with the whole trend. I was at a Momentive Chemical plant in Deer Park, Texas, recently, and it’s fascinating: they have no on-site storage. They buy their raw material off a pipeline from the refinery next door. All of a sudden we’re finding more and more processes that are changing because the real-timeness of these variables. Everything is going to become faster and faster.
I think we’re starting a new era for control engineers where they’re going to have to look at this problem and figure out how we can do predictive control or model-based control of business variables. You’ll have profit control cascading to efficiency control, and we’ll have a new type of closed-loop controller. There are some really fun challenges.
CE: The tricky part will be finding the right control algorithm.
You can’t use PID because you don’t have a natural period in the business variable loop. We’ve come up with model-based control and other things that are really pretty sophisticated. Maybe some of these other algorithms, expert algorithms, or neural-net models may make some sense going forward. Down on the plant floor, you can always default to PID, and we do, because it’s relatively easy and effective. Maybe when we get up to business control, because you can’t default to PID, we might see some of the great research that’s been done over the last couple of decades show its applicability with business variables rather than process variables.
One way might be to look at it as a real-time optimization problem of sorts, where you’re trying to balance production value, energy cost, and material costs that are constrained by safety and environmental considerations. The problem with linear or non-linear programming today is that you typically have to pick an objective function and relegate all your other objectives to constraint functions. I’m not sure that will be dynamic enough. There is some new work being done in multi-objective linear programming and optimization that I think holds a lot of promise. You’re trying to balance three objectives, production value with energy cost and material cost, so that may be the direction for closed-loop business control.