Push the Limits
A recent poll of advanced process control (APC) experts revealed unanimous agreement, users can expect APC applications to deliver sustainable, measurable benefits, but only when APC is built on a solid basic process control system (BPCS) foundation that:Wait a minute... if those three conditions are met, then the process would be as optimized as is possible, and APC wouldn't be needed! W...
A recent poll of advanced process control (APC) experts revealed unanimous agreement, users can expect APC applications to deliver sustainable, measurable benefits, but only when APC is built on a solid basic process control system (BPCS) foundation that:
Develops control strategies from an understanding of the process and its properties;
Ensures sensors, transmitters, and final control devices are working properly; and
Ensures control loops are tuned and operating in automatic mode.
Wait a minute... if those three conditions are met, then the process would be as optimized as is possible, and APC wouldn't be needed! Well, yes and no.
Yes , the process would likely be operating with a minimum amount of variability, but no it's unlikely it would be optimized to the point where economic, productivity, and quality measurements were being pushed to their upper limits.
Operating with only the BPCS requires operators to continuously analyze the process and tweak loop setpoints to reach economic, productivity, and quality constraints. Perhaps if there were an operator for each processing unit only responsible for pushing the control system against constraints it would be possible to achieve an optimized process without applying APC; but how likely is that? Most control room operators have a plethora of duties and prefer to operate the process with a comfortable cushion.
The 'cushion' mentality raises two interesting questions.
What's a two or three percent operational cushion cost?
How much would it cost to add 2% capacity to your plant?
Those questions have been asked and answered for olefin processes. According to data collected by Solomon Associates, olefin plants using APC tend to have fewer unplanned slowdowns and shutdowns and generally operate about 2% better than olefin plants without APC. Two percent may not sound like much, but adding that much capacity to the average olefin plant costs about $10 million.
Only when APC is built on a solid control
system foundation can it push the limits of
quality (6s), productivity (#'s), and
economic ($) constraints.
Depending on the marketing literature you read, APC appears in a variety of forms including multivariable control, dynamic matrix control, and model-, general-, or horizon-predictive control. Adding to the confusion, APC can be deployed as centralized or decentralized applications.
Regardless of name or deployment, APC techniques establish a controller that cancels out or compensates for the natural dynamics of the process, using a desired speed of response. Notice APC does not eliminate variability-the best APC can do is reduce variation amplitude.
Whether APC should be centralized or decentralized requires considering two things:
How fast can the process change?
How many variables need to be controlled?
Centralized APC is generally deployed on its own application server over the top of the BPCS, manages 20 to several hundred controlled variables, and tweaks control-loop setpoints using the control network to interface with controllers in much the same way as operators tweak setpoints from an operator interface terminal.
Decentralized APC follows architectures of distributed control systems, programmable logic controllers, or hybrid control systems, and places APC application algorithms in the same controller as regulatory control, sequences, and interlocks. Processes-such as compressor surge control, boiler combustion and steam control, and some units of refining and pulp and paper processes-can change very quickly and thus require the APC be deployed as a decentralized solution.
Regardless of APC deployment, it's important to keep in mind APC adds another level of cascade control over the BPCS and brings with it the same control design considerations as any cascade application.
Where does control belong?
Everyone agrees, you can't build anything lasting on a poor foundation, and since the BPCS forms the foundation for building APC applications, it's important to ensure that:
Control strategies are developed from an understanding of how the process reacts to disturbances;
Sensors, transmitters, and final control devices are working properly; and
Control loops are tuned and operating in automatic mode.
Control strategies can be developed to react to a process disturbance (feedback control), or to anticipate a process disturbance (feedforward-control).
All feedback control implementations have one thing in common; they react to a controlled variable disturbance after the disturbance is detected. Feedback control is simple to understand and most control systems provide a library of feedback control algorithms to choose from, complete with pre-built operator faceplate displays.
Feedforward is also simple to understand, but less commonly applied. Feedforward control measures a disturbance and introduces a dynamically compensated corrective action to the control algorithm before the disturbance affects the controlled variable. (See Feedforward is like insider trading sidebar.)
On occasion an APC application is discovered to be directly manipulating a final controlling device or valve. Experts caution this philosophy should be used sparingly, but nothing is a hard and fast rule. For example, operators sometimes use a manual loader to send a fixed output signal to a final control device and have learned exactly what to expect for each 5% change between 20% and 50%. It's also common to learn that under normal operating conditions, operators change the output signal a few percent once or twice a shift.
When these, or similar, conditions exist, let the APC mimic what the operator does using direct digital control. Learn what operators watch and how they decide when to change the output signal, how long to wait for a response, and then develop a corresponding algorithm in the APC application.
However, don't forget to carefully scrutinize the timing issues of applying direct digital control as part of a centralized APC application. Remember, BPCS-based control loops can read sensor inputs, execute the control algorithm, and change the output signal in milliseconds, and still not always cancel process disturbances. That makes it easier to understand how introducing a centralized APC's additional time lags in reading inputs, performing calculations, and getting the new output value to the final control device can amplify, rather than cancel variability.
Linearization is another area where experts agree it's far more important to ensure the loop-sensor, transmitter, BPCS and APC controllers, and final control device-are linear in the range of normal operation than to jump through hoops to achieve linearization across the entire measurement range.
The message is clear, BPCS and APC must be considered as a tightly integrated solution, and placement of control is not either/or. Understanding the process and its dynamics will guide controller algorithm placement. That doesn't mean BPCS loops are relegated to simple PID; quite the contrary. All available control tools, such as lead, lag, ratio, feedforward, cascade, adaptive/notch gain, limiters, high/low selectors, multipliers, adders, subtractors, etc., should be appropriately applied to create and deploy the most robust control solution possible based on knowing how the process performs.
Optimization is not a myth
To many practicing process control engineers and operators, APC's promise of optimization is a myth that works only in theory and can't be successful in a world where sensors plug, pumps and valves cavitate, and pipes leak. Maybe that was true in the past, but the analysis and modeling tools used to develop, deploy, operate, and maintain APC systems have significantly improved.
Today installing and expecting an APC system to actually maximize industrial productivity is no longer a myth, it's very doable.
In his book 'Optimization of Unit Operations' (Chilton, 1987), Bela Lipták defines optimization as, 'The integration of process control know-how to maximize industrial productivity.'
Mr. Lipták's definition of optimization could be met if process operations could:
Avoid unplanned shutdowns caused by fouling, coking, plugging, and scaling;
Avoid equipment constraint violations such as overspeed, cavitation, vibration, and surge;
Implement optimum setpoint sequencing and ramping during feedstock or equipment switchovers;
Continually monitor process stability;
Push a process to economic, productivity, and quality constraints and then maintain that level of performance for days, weeks, or months:
Ensure 'real' alarms are identified and managed by reducing or even eliminating nuisance alarms; and
Provide operators with accurate and timely information about equipment performance so operators can become skilled caretakers of equipment.
An integrated BPCS/APC solution is capable of achieving all these things, but it requires approaching the solution with management's commitment, knowledgeable and committed people, healthy field instrumentation and control devices, regulatory control built on process knowledge, realistic and measurable improvement goals, and an appropriate APC application.
So, what would it cost to add 2% capacity to your plant? I'd wager it's more than installing an integrated BPCS/APC solution.
Avoid creating a MONSTER
Basic process control and advanced process control (BPCS/APC) systems tend to grow and evolve over time. Follow these checkpoints for successful control system evolution through the life cycle. Ignore them, and the system may mutate into a beast to be served.
Be sure claimed benefits can be measured and achieved;
Ensure sensors, transmitters, controllers, and final control devices are in good working condition;
Where possible, implement APC to manipulate the highest-level control loops;
Keep the operator in mind for those times when the APC is off-line;
Integrate the BPCS and APC to achieve the best operator on their best day, every day;
Spend time conducting plant testing and analysis; this will pay off in good controller design;
Use sound judgement in developing and deploying inferential measurements;
Understand controller tuning and the parameters used;
Keep it simple; and
Remember, there is no substitute for understanding the process.
Feedforward is like insider trading
There is no substitute for knowing exactly what's going to happen before it happens. On Wall Street that's called insider trading; in process control it's called feedforward.
For example, a steam boiler contains a steam drum that is partially filled with water. The drum is heated by the boiler combustion system and the water is turned to steam, thus drum water level is a critically controlled variable.
Typically, drum level control is implemented as a cascade control with the feedwater controller. Both are implemented as PID feedback controllers, known as two-element feedwater control. (See Three-element Feedwater Control diagram.)
As long as steam leaving the boiler matches the flow of water entering the drum everything is fine. But when the steam flow suddenly increases or decreases a phenomenon known as shrink/swell occurs to the water in the steam drum.
When steam flow suddenly increases, the pressure inside the steam drum reduces and the level of the water rises (swells) slightly. The drum level transmitter interprets this as an increase in level and sends a signal to the feedwater controller to reduce flow-exactly the opposite of what should happen.
Very quickly steam pressure recovers and the 'swell' returns to its original condition. However the amount of water in the steam drum has reduced to satisfy the increased steam demand. That means the drum level controller must undo the change made based on swell and start adding water to meet the new steam flow demand.
Had steam flow decreased, the pressure on the water would have temporarily increased causing a 'shrink' in drum level and the opposite set of actions and reactions would have occurred. Adding steam flow as a feedforward value to the output of the drum level controller compensates for the shrink/swell phenomena, stabilizing the critical variable drum level. This is called three-element feedwater control and it's working quite nicely in thousands of boiler applications around the world.
Experience indicates numerous processes could benefit by adding feedforward to an existing control strategy, and unlike insider trading, you won't go to jail for using it. (See CE , Oct. '00, p. 104)
APC expert sources
Responses from experts on this topic concurred in many areas. The following individuals provided insights for successful BPCS/APC implementations.
Aspen Technology's (Houston, Tex.) Mark Darby;
Control Consulting's (Houston, Tex.) Mike McCarty;
EnTech Control's (Toronto, Canada) Bill Bialkowski;
ExperTune's (Hubertus, Wis.) John Gerry;
Fisher-Rosemount's (Austin, Tex.) Lou Heavner;
Foxboro's (Foxboro, Mass.) Harry Forbes;
Honeywell's (Phoenix, Ariz.) Clive Duebel, John Roffel, and John Escarcega;
James-Mangan Automation's (Carson, Calif.) Dan Roessler;
Solomon Associate's (Dallas, Tex.) George Birchfield; and
Techmation's (Scottsdale, Ariz.) David Ender.
Select the right tuning or things get worse
There are three basic controller algorithms, ideal, parallel, and series, but digital implementations of these three types has produced hundreds of variations with a variety of tunable parameters using different units and time bases.
Historically, articles describing how to calculate controller-tuning parameters assumed:
A certain type controller algorithm was being used; and
The controller needed to cancel the effects of disturbances as quickly and efficiently as possible.
The first explains why some calculation methods work better for some people than others-the calculation method needs to match the algorithm, units, and time base. The second requires more thought and attention in an integrated BPCS/APC solution. Keep in mind the APC is going to 'tweak' setpoints to bring the process ever closer to constraint limits. That means controller tuning must cancel disturbances and respond to setpoint tweaks in a smooth, non-oscillatory manner.
Frequently, tuning parameters applied to control algorithms for cancelling disturbances are too aggressive for setpoint changes and may cause the controlled variable to overshoot the setpoint and violate constraint limits. The common solution is to reduce the controller gain, but that can sacrifice disturbance rejection.
There are several ways to achieve good setpoint and disturbance response, including:
Configuring proportional action to work on the measurement value, not on the calculated error;
Passing the setpoint value through a lead/lag filter before applying it to the control algorithm;
Applying setpoint weighting; or
Simply selecting the appropriate tuning technique.
Choosing the 'appropriate' tuning technique is where obtaining expertise in process analysis and tuning methods can pay big dividends.
For example, lambda (pole canceling) tuning has gained significant attention in the past few years, but may not be appropriate for all applications.
Consider an example, provided by ExperTune, where the process has a gain of 1, dead time of 0.2 minutes, a time constant of 10 minutes, with an ideal type PID (proportional, integral, derivative) controller.
Disturbance tuning calculations will yield a proportional band of 8.2 and integral of 3 minutes per repeat. Fast lambda (lambda = 10 minutes) tuning calculations yield a proportional band of 100 and integral of 10 minutes per repeat. For this example, disturbance tuning will be 4,000% faster than lambda tuning. However, if the process dead time were 1, the difference is much less, and if the process dead time were 10, the two methods produce almost identical results.
Bottom line, there is no substitute for:
Knowing the process;
Understanding the pros and cons of various tuning methods; and
How to best apply one to the other.