Greg: The key to how well a steam system can do its job depends heavily upon boiler response. Steam header pressure controllers can be properly tuned for fast response, and use feedforward signals and half decouplers to minimize disruptions in a header and between headers from large changes in steam usage and generation by production units. The deadtime in these header loops is often a matter of a few seconds, largely coming from the control valve response time. Ultimately, how much you can reduce the cost of steam (how well you can maximize use of waste fuels, minimize use of purchased fuels, maximize use of steam from cogeneration units, and prevent venting steam) depends on how fast the boilers can respond.
Stan: Scott Pettigrew, senior energy consultant at Emerson, uses his 29-plus years of experience to give us the inside scoop at how boilers can do much more than is commonly expected. Scott, how do you minimize the use of purchased fuels?
Scott: We can use boilers running on waste fuel to take all the swings in the plant steam demand within minutes. The starting point is good flow measurements and computations on a mass flow basis. It turns out that the variation in BTU value for a given mass flow value often is not more than a few percent. One big exception is hydrogen in waste gas fuels. A Coriolis flowmeter together with a specific gravity meter on such waste fuels can provide an inferential measurement of the hydrogen content. In general, Coriolis flowmeters are great in terms of providing the most accurate mass flow measurement with the greatest rangeability, as well as density measurement with incredible precision. However, for solid fuels, very large lines or other applications where Coriolis flowmeters are not practical, strategies can provide the missing information as long as the flow measurements are relatively repeatable. In general, 95% of the time the purchased fuel (e.g., fuel oil and natural gas) can be at its minimum.
Greg: The air flow measurement needs to be repeatable over the total range of air flow demand. While venturi meters or insertion flow tubes are desirable for air lines in smaller boilers, larger boilers typically have ducts with annubars or averaging pitot tubes. The performance of these sensors can be increased by duct modifications (e.g., bell mouth duct fitting) to increase the pressure drop, and adding upstream straightening vanes to make the velocity profile more uniform and consistent. Both of these improvements improve the signal-to-noise ratio. High- and low-range differential pressure transmitters can improve the rangeability of the flow measurement, given a good signal to noise ratio.
Stan: How do you deal with wide spectrum of waste fuels?
Scott: We can swap out one fuel for another by building into the control strategy the deadtime and time constant associated with each fuel, so the changes in fuels are coordinated and almost seamless in terms of steam generation. We can meet changes in steam demand much faster than expected. For example, we can make steam flow changes of 20% to 25% per minute, outrunning the drum level controller unless special efforts are made there.
We integrate the air and fuel together in the control strategy. We don’t need or want to use empirical curves on air flow versus fuel flow. Such curves are prone to test and inherent errors due to a great dependence upon operating conditions, most of which were not sufficiently specified or even known for the curves generated. At any rate, better performance is attained and wasted time avoided by not using these curves.
Greg: How do you get the right air flow?
Scott: We ask for a detailed composition of each waste fuel, realizing the compositions can change. Based on stoichiometry and first principles, we develop an initial combustion heat value (e.g., BTUs per lb), air requirement and excess air requirement.
Depending on the waste fuel phase, we have a starting point for the excess air required in the combustion zone to provide the desired oxygen concentration in the stack. For example, solid, liquid and gaseous waste fuels may require 28%, 25% and 20% excess air, respectively, to achieve the oxygen concentration setpoint. Early on in the project, measurements of carbon monoxide in the stack (often by handheld meters) are used to determine the best oxygen setpoint.
The amount of required air flow per pound of waste fuel to achieve an oxygen setpoint acts as a target for excess air. The oxygen controller output can be used to continually correct the total air demand dealing with actual heating value of the fuel. The amount of combustion air consumed at a given load can be used to correct for changes in heating values. The excess air is optimized as a function of load to achieve the best possible oxygen setpoint to improve efficiency as much as possible. Additional strategies can also be used to further minimize the oxygen trim.
Greg: In general, the difference between a first principle model and actual plant operation is seen in differences in the manipulated flow and specifically, the correction of a feedforward signal by a feedback controller. A simple integral-only controller has been used to gradually make the feedforward correction nearly zero. For a feedforward summer, when the feedback controller provides a plus or minus 50% correction, a zero correction corresponds to a feedback controller output of 50% (as noted in the 5/22/2016 Control Talk Blog, “Control Strategies to Improve Process Capacity and Efficiency -Part 1."
Virtual plants can be adapted by using Model Predictive Control (MPC), where model parameters are the manipulated variables, to minimize the difference between flows in the actual and virtual plant (as noted in the 7/26/2016 Control Talk Blog “Control Strategies to Improve Process Capacity and Efficiency - Part 3" and the Control November 2007 article Virtual Control of Real pH.Given that the actual and virtual plants have the same setpoints and tight control, the fidelity of the model is seen in how well the model’s PID manipulated flows match the corresponding flows in the plant.
Stan: Where do you see the biggest improvements?
Scott: The improvements are particularly impressive for boilers using bark, a common waste fuel from pulp and paper plants. Bark is not measured on a mass flow basis. We use feeder speed and correct the BTU per percent speed online. We have done lots of boilers that use only bark. We can minimize the undesired swings from the type of bark and loading on the feeder, and maximize the desired swings to meet changes in steam users and generators on the headers.
If 50% or more of the solids are ¾-in. or larger, we can do a lot to swing loads by just changing air flow. By simply changing undergrate air flow, we can greatly change heat release. It is kind of like a bellows on a camp fire. Using undergrate air on the bed leads immediately to a burst of flame. Reducing undergrate air flow will quickly dampen the fire resulting in a very fast load reduction. The limitation may well be the ability of the induced draft (ID) fan to keep up with load changes. If there is an excessive amount of fines (e.g., sawdust), you don’t maintain the desired amount of solids on the grate and the swing capability is much more dependent on feeder speed. A bigger bed helps.
Even after 29 years, I am amazed at how fast increasing and decreasing air flow can change steam generation. This is important because paper machines can stop and restart due to sheet breaks, drastically and suddenly changing the steam load.
Greg: How do you tune boiler loops?
Scott: We take a holistic approach to the steam system. All features in the control strategies must be effectively employed, factors adjusted and loops tuned with the right objectives in the right order. Checkout, commissioning and startup is where is all comes together.
The furnace pressure and drum level loops must be tuned first for fast response. In general, we try to use more gain and less integral action to provide a more immediate response and better sense of direction afforded by the proportional mode. We are willing to have some overshoot and minor oscillation to gain tighter control. Tighter control of these loops minimizes the upsets to the combustion control and steam header control loops from swings in waste fuel and load.
We don’t want the air and fuel flow controllers to outrun the furnace pressure controller. We see a typical problem of excessive rate limiting in the variable-frequency drive (VFD) making the response of the ID fan too slow, the result of an overriding concern of fast speed changes causing motor overload. For properly designed motors and newer VFDs, rate limiting is too conservatively set. With the older motors often seen in existing boilers, users are hesitant to do anything that might cause a motor overload trip.
Rate limiting should not be done to slow down loops. These rate limits can cause reset windup and continual cycling. Loop action can be coordinated by tuning.
Greg: If rate limiting is used, external reset feedback (dynamic reset limit) must be turned on and fast readback of the manipulated speed or position used as the external reset feedback signal to prevent integral action changing the PID output faster than the manipulated variable can respond. Many controllers do not have this feature, and fast readback of speed or valve position is often not available.
The tuning settings that minimize the peak and integrated errors for load upsets generally have a slight bit of oscillation, as long as the oscillations never get greater than quarter amplitude decay for the worst combination of final control element dynamics.
For control valves, we want a linear installed flow characteristic, an 86% response less than 4 seconds, deadband less than 0.3%, and resolution better than 0.2% for small steps (e.g., 0.5%) and medium size steps (e.g., 5%). This performance is best achieved by using diaphragm actuators, sliding-stem globe valves and low-torque butterfly valves. For large steps (e.g., 20%), the stroking time becomes a consideration and a volume booster on the positioner output may be required. For more on these requirements see the March 2016 article “How to specify valves and positioners that do not compromise control” and the associated white paper “Valve Response – Truth or Consequences.”
For final control elements that are motors, we want the same or better dynamic response by manipulation of motor speed. This performance is best achieved by using VFDs with pulse width-modulated (PWM) inverters, high-resolution input signal cards, proper drive setup, and motors designed to minimize overheating and overload. Proper drive setup means minimizing any deadband settings and rate limiting. A good rangeability, low-noise measurement with optimal filtering can enable deadband settings less than 0.2%. Motors should have at least a 1.15 service factor and Class F insulation. Overheating must be prevented by sufficient ventilation. For the totally enclosed fan-cooled (TEFC) motors normally used in process plants, the fan speed will decrease with motor speed if the fan is run off the same power as the motor. Overheating can occur at low speeds. For ID fans, the stack gas temperature is high and the problem is severe, limiting rangeability. The solution is to use a separate source of power for the fan to give a constant speed, and a large frame size to give better ventilation. For more on the VFD requirements for best rangeability and response, see the 5/10/2012 Control Talk Blog, “Checklist for Best Variable Speed Pump Performance."
Stan: Does drum shrink and swell cause level tuning problems?
Scott: If the temperature difference between the feedwater and boiler drum is less than 100 oF, shrink and swell does not prevent us from using the traditional three-element level control system and aggressive tuning. Here you can see the value of a feedwater heater or economizer. For firing rate increase, the swell from the increase in firing rate can be somewhat reduced by the increase in feedwater. If necessary, we can simply delay the feedforward to optimize its timing by simple insertion of a deadtime block. For example, the delay can help offset the swell from an increase in firing rate by the shrink from an increase in feedwater for an increase in steam flow demand. About 80% of drums can use three-element control, although we have seen recently an increase in drums where shrink and swell is causing drum level control to be a limiting factor in the boiler response. Sometimes the drum size is reduced to make the boiler price more attractive, making the shrink and swell as a percent of level worse.
Greg: What kind of problems do you see onsite?
Scott: I’d say 50 to 60% of fixes are just configuration issues such as scaling and connection of back calculate (BKCAL) signals. Implementation of feedforward external to the PID often leads to integral windup and a lack of bumpless transfer for mode changes. The fixes are simple, but require an experienced person. Most of the issues originate from lack of training and mentoring.
The control strategies are powerful and flexible, but parameters and connections must be right, and a holistic approach taken to the design and the tuning of controllers.
Top 10 reasons to love dead time
- (10) It makes you appreciate “alive time”
- (9) Offers incentive for universities to invent another deadtime compensator
- (8) Creates scenarios for 100 different tuning methods
- (7) Provides continual surprise to operators about manual actions taken
- (6) Promotes creative explanations of what will be happening in the process
- (5) Nurtures mystery due to unsuspected sources, such as valve deadband and resolution
- (4) Enables you to stir things up by seeking valve specs that require a valve actually respond
- (3) Makes you a hero when you reduce deadtime by clean electrodes and Coriolis meters
- (2) Makes you a hero when you reduce deadtime by tight fitting, tapered thermowells
- (1) Increases job opportunities for process control engineers