WhyInvestPROCESSCONTROL
- 1. ditorials, features, and letters to the editors of CONTROL magazine during the past several years have described and decried the state of commercial practice of process control engineering and automa- tion. Application and justification of analyzers, actuators, instrumentation, control systems, plant modeling, algo- rithms, control computers, information systems, project services, software, systems integration, and maintenance are fragmented, risky, and not sufficiently profitable. Many large technology suppliers are troubled by lack of profitable business growth, recognition, and appreciation of their value-added offerings. Academia has lost interest in adding to the mountains of mathematical publications gen- erated since 1960. Free literature and technical conferences have been available around the world for decades, yet, in plant man- agement circles, turmoil and confusion reign about the value (potential and realized) of deploying the enormous suite of technology available for control, optimization, scheduling, and IT. Management continues to say, “Show me where the money comes from, how much I get, with what risk.” Reports of potential benefits (1,3) have not led to realization. Investment in the field is shrinking; the field has become a business priority backwater. Many good peo- ple and companies have left. Something is amiss. Why? What’s wrong? Myths and Measures Why should we do control? What is the purpose of control- ling things? What is the unifying universal objective? How do we decide what things to control? How do we measure the performance of control; the value of controlling better? How do we know how successful we are at improved control and how delighted we are with our achievements? Too many say they do control because it’s good, necessary, cool, modern, everyone is doing it, fun, neat, important. These are simple myths. Too many say the purpose of control is to reduce fluctua- tions, minimize variations, minimize the integral of error squared, smooth things out, stabilize operations, reduce upsets, increase speed of response, move closer to limits, improve operations, improve quality, increase yield/capacity, save energy/utilities, help operators, increase reliability, improve safety, reduce emissions, cut maintenance costs, reduce manpower, etc., ad nauseam. More myths. These justifications are often repeated by people who believe the purpose of a process plant is to make product. But the purpose of all process plants, in all industries world- wide, is to add value to repay investors and governments (some do this by shutting down to stop hemorrhaging losses). The purpose of process control is to add value, create wealth, increase profits, and make more money. Always. Ethically and legally, of course. Humanity decided long ago to meas- ure its commercial values with money. Process control suffers from lack of an agreed-upon, meaningful measure of performance. An essential ingredient for baseball is universal agreement: if the ball goes left of the left field pole it’s a foul; to the right it’s a home run. We can- not start a Super Bowl in front of 100 million viewers with a debate on whether a touchdown is worth five or six points. The Olympic Games are truly great and interesting when the performance measures for broad jump, marathon, slalom, and high hurdles are thoroughly understood and agreed upon. Great sports are built on consensus on the per- formance measures for success. In the process industries, such a measure is expected net present value profit (ENPVP). Control engineers and busi- ness people should adopt as a primary purpose to identify, cap- ture, and sustain significant economic benefits from the process or system for their customers, investors, and them- selves by using the appropriate products, tools, techniques, R E P R I N T E D F R O M C O N T R O L , M AY / 2 0 0 2 TURBINE FLOWMETERS VS. PROCESS CONTROL? WHY INVEST IN PROCESS CONTROL?Understanding the Real Benefits in Quantitative Terms Is the First Step to Proving Payback. By Pierre R. Latour FIGURE 1. DETERMINE THE OPTIMUM SETPOINT QUANTIFYING THE RISK AND COSTS ASSOCIATED WITH OFF-SPEC FUEL OIL BY CALCULATING THE PROFIT PER BARREL (RED) LETS US SEE THAT, IN THIS EXAMPLE, SIMPLY MOVING THE MEAN OF THE PERCENT SULFUR DISTRIBUTION (BLUE) FROM 0.6% TO 0.758% CAN RAISE THE PROFIT RATE FOR THAT DISTRIBUTION (PINK) TO AN AVERAGE DAILY PROFIT OF $7,433 FROM $6,806. EE -0.20-0.20 0.200.20 0.600.60 1.001.00 1.401.40 1.801.80 % S 0205 F- Justify 6/10/02 9:32 AM Page 41
- 2. and services to maximize the ENPVP. The word “expected” has an important statistical definition; the words “net present value” have an important financial definition. The word “profit” has a significant modeling meaning. The Clifftent function (2) provides the rigorous means to measure financial value of dynamic performance. Process con- trol, maintenance, and IT now have their measure for winning. Control Risk to Make Money Control engineering and technology is a basic method for deploying knowledge to mitigate risk or uncertainty. It inte- grates knowledge of the process physical behavior, econom- ic impacts, disturbances, key measurable responses, inde- pendent adjustments, and optimal control theory for non- linear/multivariable dynamic systems to build computer- integrated systems of models, measurements, actuators, and control algorithms. Feedback control basically swaps variations in independ- ent variables we care about for variations in independent variables we care less about. That is how control mitigates and manages risk. That is the source of profit. The key phys- ical performance claim is the reduction in a properly speci- fied variance. All control system components should play a role in this reduction. How does this make money? The traditional answer for decades has been: Reduced variations in a key measured dependent response variable about its base case mean does not make money per se because the variations average out, but it is a necessary prerequisite to allowing us to move the mean somewhat in the profitable direction toward a limit or specification. This provides a steady-state average improvement like yield, capacity, or utilities. Multiply the physical gain by the right economic factor and benefit, $/day, is achieved. This classic universal approach is wrong. It is incomplete. It relies on at least four of the myths described above. A better calculation can be done using Clifftent. First, determine if the base case mean is near its desired target (setpoint) and, more importantly, whether the setpoint is optimally set to maximize ENPVP. Calculate what the optimum setpoint is, and the profit gain for moving the mean from its base case value to its base case optimum value. The low-sulfur fuel oil example below illustrates how this is done with Clifftent (2,4). People unfamiliar with Clifftent naturally skip this vital first step, and there may be as much easy profit in this step as in all subsequent control endeavors. Further, if one cannot determine the optimum setpoint of a proposed controlled variable, there is no (not little, but no) basis for controlling it. Clifftent proves this mathematically. Second, determine the ENPVP for reducing dynamic vari- ance by the proposed amount at the same setpoint. This always makes money, but people unfamiliar with Clifftent naturally skip this step and assume it has zero intrinsic value. Unfortunately, this myth misses typically half the provable tangible benefit. The literature must describe this as intangible—in other words, the benefit from good control is typically twice that claimed by the classical incomplete method. This is one of the great tragedies of process control. Clifftent is the rigorous method for quantifying the financial value of improved dynamic performance of any system. It completes the quali- ty work of Deming, Juran (5), and Crosby in the 1980s by quantifying the profit from quality control. Third, determine the maximum ENPVP and correspon- ding new optimum setpoint for the reduced-variance situa- tion. This gives the third profit gain component, which is close to that determined by the classical approach, with one major difference: It is now optimal. Fourth, the traditional method moves the mean an arbi- trary distance, ad hoc, non-rigorous, because no one bothers to model the financial consequences of violating the limit or spec. Most assume it’s forbidden, unknowable, infinite. Clifftent shows that this missing ingredient, the penalty of violating the limit or spec, is just as important as the eco- nomic credit factor for approaching the limit. That is the key. The example below illustrates how one can never opti- mize setpoints in the neighborhood of limits without knowl- edge of the penalty for exceeding the limit. WHY INVEST IN PROCESS CONTROL? FIGURE 2. EVALUATE PROCESS IMPROVEMENTS -0.20-0.20 0.200.20 0.600.60 1.001.00 1.401.40 1.801.80 % S REDUCING THE PERCENT SULFUR STANDARD DEVIATION FROM 0.23 TO 0.07 AND ADJUSTING THE SETPOINT TO 0.860% BRINGS THE PROFIT RATE FOR THE IMPROVED DISTRIBUTION (BROWN) TO AN AVERAGE DAILY PROFIT OF $9,333, CLOSE TO THE THE- ORETICAL MAXIMUM OF $10,000. 0205 F- Justify 6/10/02 9:32 AM Page 42
- 3. WHY INVEST IN PROCESS CONTROL? Clifftent Capabilities Clifftent provides a rational, rigorous method for setting lim- its, targets, and tolerances with uncertainty (2,4). It unifies statistics and process control. It does quantifiable risk man- agement. It properly sets dependent variable constraint val- ues for linear and quadratic programming (which often adds more value than the programming solution itself). It sets the operating limits for online process optimizers (which often adds more value than the optimizers alone). It sets limits for production and inventory schedulers (which often adds as much value as the scheduler itself). It incorporates the value proposition to key performance indicators used to justify large IT, MES, and value-chain management projects. (IT has suffered notoriously for decades from lack of a rigorous financial performance method.) Clifftent needs two input functions for each dependent controlled variable (CV): the statistical distribution and the steady-state profit rate as a function of the CV mean. The distribution forecast may be Normal Gaussian, Gamma, or arbitrary. The steady-state profit rate always increases (lin- early or not) from the left toward the limit or spec and decreases (linearly or not) toward the right away from the limit. It is shaped like a tent. Steady-state profit defines a tradeoff. It may have a dis- continuity at the limit, like a cliff. All value functions drop to negative values at left and right extremes. The method com- bines these two functions in a proprietary way to provide the actual profit as it varies with the mean of the distribution, ENPVP (2,4). The uncertainty distribution provides roundness to the real profit function; it is always a smooth hill. The hilltop locates the maximum ENPVP: that’s all there is. Slopes are critical. Cliffs are critical. Curvature is important. Distribution changes are significant. One side of the input profit function combines the physi- cal process model with its economic sensitivities. The other side connects the process to its surroundings: customers, suppliers, environment, and maintenance. So how do we select candidates to be control variables? They have a Clifftent function. Their value affects profits. The steeper the tent slopes, the more sensitive, critical, and inter- esting the variables are. The higher the cliffs, the more criti- cal and interesting they are. Forget about controlling variables with shallow tents and tiny cliffs. They are of no consequence. Low-Sulfur Fuel Oil Example Suppose we are in charge of making 10,000 bpd of low-sul- fur fuel oil (LSFO). Our product should contain no more than 1 wt% sulfur. Last year the average of our cargoes was 0.6% with standard deviation of 0.23%. A sulfur analyzer salesman claims we can reduce the deviation at least 30% with his $50,000 instrument. An advanced control solution provider claims he can reduce it 70% for $1 million. As we over-desulfurize below 1%, refinery models quanti- fy the increased costs (from lost yield, higher catalyst and hydrogen consumption, and sulfur plant H2S load) to be $5/bbl/%S plus some quadratic curvature. Our power cus- tomer contract for LSFO specifies it will accept off-spec car- goes with a penalty of $0.6/bbl plus $3/bbl/%S plus some quadratic curvature for spec exceedance. He must inject expensive 0.2% S diluent to his boilers to maintain SO2 emissions. If our LSFO were always at 1.0% sulfur our prof- it would be $1.0/bbl x 10kbpd = $10k/d. What to do? Figure 1 shows five curves. The horizontal axis is percent sulfur from 0-1.8%. The blue curve is the quality distribution, normal with mean = 0.6, standard deviation = 0.23. Red is the steady-state profit function: $1.0/bbl at 1.0%, a $0.6/bbl cliff to the right, further decline in product value with higher sulfur content, and profit decline to the left from higher desulfur- ization costs. The pink curve is ENPVP rate depending on quality mean, as the blue curve slides from left to right, incorporating the dis- tribution. We find our average profit is only $6,806 per day, far from the perfect $10,000. But if the mean were moved to 0.758 we would gain $627 per day to $7,433. While we incur more off-spec product, we gain more from operating costs. At this base optimum, 15.0% of product is off-spec and 0.43% is unprofitable. Next assume the advanced control supplier reduces the standard deviation to 0.07. The green curve is this narrower distribution forecast. The brown curve is the ENPVP rate depending on quality mean as green curve slides from left to right, incorporating the narrower distribution. Now profit jumps $1,446 to $8,880 per day at the new optimum 0.758 mean. This is the value from reduced variance only, improved smoothness, tighter control, better dynamic per- formance. The money comes from reduced operating costs on the left tail plus reduced spec violations on the right tail. This is real tangible money. It is always a combination of two or more tradeoff phenomena, never just one. If the cliff penalty were omitted, this benefit can never be quantified; it would remain intangible and hidden. Finally, we see the opportunity for more profit from our improved risk management capability. Move the mean far- ther right to the new hilltop at 0.860% sulfur to gain another $454 to $9,333 per day. This is 93.3% of $10,000 with perfect control. This money comes from reduction in operating costs larger than the additional penalty from more off-spec prod- uct. This tradeoff is optimized. Figure 2 shows the same five curves with the green distri- bution at its new optimum position, a mean of 0.860. Only 2.56% of product is off-spec and virtually none of it is 0205 F- Justify 6/10/02 9:32 AM Page 44
- 4. unprofitable at the new optimum. Note also the traditional steady-state benefit of $454 is only 454/(454+1446) = 24% of the total process control credit. Seems worthwhile to quantify the 76% pure dynamic control portion rather than leave it intangible. Clifftent proves it is usually better to play on the safe side (unless the cliff is small and the safe-side slope steeper than the unsafe-side slope, when it may be more profitable to play on the unsafe side). In 1980 the potential profit (benefit minus cost) for com- puter-integrated manufacturing for petroleum refining was $1/bbl crude x 65 million bbls/day, worldwide. In 2000 with cleaner gasoline and diesel it increased to $1.2/bbl x 75 mil- lion bbl/day (1). Adding petrochemicals, polymers, and natural gas doubles this potential for the hydrocarbon processing industry (HPI). That’s $65 billion per year profit from the HPI for somebody. The cost to capture this profit is less than half that (gross ben- efit = $1.5/bbl). Maybe some progress has been made, but surely we can do better. Pierre R. Latour, consulting chemical engineer, Clifftent Inc., Houston, may be reached at clifftent@hotmail.com. References 1. Latour, P.R., “Benefits of Modern Refinery Information Systems for Manufacturing Cleaner Fuels,” API Fuels Conference, 1995. 2. Latour, P.R., “Process Control: Clifftent Shows It’s More Profitable Than Expected,” Hydrocarbon Processing, Vol. 75, No. 12, pp 75-80. 3. Latour, P.R., “Does the HPI Do Its CIM Business Right?” Hydrocarbon Processing, Vol. 76, No. 7, July 1997, pp 15-16 and “Optimize the $19 Billion CIMFuels Profit Split,” Vol. 77, No. 6, June 1998, pp 17-18. 4. Latour, P.R., “Clifftent: Determining Full Financial Benefit From Improved Dynamic Performance,” Proceedings of the Third International Conference on Foundations of Computer- Aided Process Operations, AIChE Symposium Series No. 320, Vol. 94, 1998, pp 297-302. 6. Juran, J.M., “Juran on Leadership for Quality—An Executive Handbook,” Macmillan Free Press, New York, 1989. WHY INVEST IN PROCESS CONTROL? C 0205 F- Justify 6/10/02 9:32 AM Page 45