Greater efficiency by on line coal measurement20 October 2000
Using fuel feed pipes as a waveguide allows coal flow to be measured by microwaves. This permits the targeting of specific problem areas. From test data generated in power stations a strategy can be developed to optimise the combustion process in iterative steps at individual control points. Hans Georg Conrads, Promecon GmbH, Barleben, Germany
The idea of optimising coal fired boilers by on line measurement and control of the coal flow has been around for many years. The option of controlling a boiler on an individual burner basis promises better boiler efficiency along with better emission control. The main targets for this sort of optimisation are the reduction of NOx, CO, O2, particulate emissions and flue gas losses.
Compensating for pulverised fuel (PF) flow imbalance is often a difficult task. But carrying out on line measurement of PF mass flow shows up opportunities in several major areas among the adjustable parameters of a boiler that can be used to optimise the combustion process.
A technical platform that can be used to generate the on line data is Promecon's PfFLO system. This is an application that allows the measurement of coal concentration, coal and air velocities, and coal mass flow, in a coal conduit. The hardware can also be used to measure the secondary air as well as the carbon in fly ash content. These additional packages are called AirFLo and AshFLO. The PfFLO platform comprises a full SCADA package that allows the recording and management of further fundamental boiler data such as NOx, CO and O2 content. The concentration of the pulverised fuel is measured by using the burner pipe as a microwave guide. The wave guide properties in this case are dependent only on the dielectric load, i.e. the density of the PF in the measured section.
The velocity of the pulverised coal is measured by a cross correlation method. Two pairs of sensors (Figure 1) are mounted at right angles (to give readings across the pipe in horizontal and vertical directions) at a known distance along a pipe. From each of the sensors a series of stochastic signals (so called 'signatures') is acquired where the signal irregularities relate to stochastic processes such as small density variations within the pipe. The signals from the sensors show similarities; ideally they are identical and shifted by the time the PF needs to get from one sensor to the next. One can calculate this time shift with cross correlation. As the distance between the sensors is fixed the PF velocity in the pipe section can be accurately calculated.
PfFLO provides an absolute measurement of the coal flow through the conduits of a burner. The absolute measurement is used as the basic concept because it has several advantages for the user that are only available through this approach. Each pipe is measured individually. The measured values in one pipe will not be influenced by the rest of the pipes, so a faulty measurement in one pipe does not affect other pipes and the obtained values are still valid.
The AirFLO application package uses the cross correlation technique of the PfFLO to determine the secondary air velocity in air ducts. By using the cross sectional area of the duct as well as the temperature and pressure information the air mass flow can be determined on line. AirFLO and PfFLO can be combined to provide full fuel and secondary air measurement information on-line.
AshFLO is an application package that runs on the same hardware platform as PfFLO. By using a microwave sensor the dielectric properties of the fly ash are determined and out of this the UBC content is derived.
Sensors can be located at various points in the process. One location chosen for current applications is the intermediate storage after the precipitator. The multiplexing hardware of the system can handle the output of many sensors. AshFLO has been applied to various boiler applications in Europe and proved against standard laboratory methods.
The application allows the measurement of feed forward values of the combustion process (coal flow, air flow) as well as the measurement and recording of feedback values such as the unburnt carbon (UBC) or values such as CO, NOx and O2 content.
The general targets for optimisation using the PfFLO platform are the operation of the primary air control (PA control), the pressure drop over the coal burner pipes and the secondary air distribution settings, the main parameters that can be adjusted in a modern boiler. There are several factors that limit the range of transport velocity of coal. So the general problem of coal transport velocity is that often it cannot be varied as much as desired, because the above constraints must be met.
Figure 2 shows a typical problem in a vertical burner pipe with a very low coal transport velocity. Below the critical velocity of 16 m/sec the coal flow starts to pulsate, which results in variations of the coal velocity. As can be seen there are lower value velocity spikes from time to time. These spikes disappear at higher loads owing to the higher PA setting and hence the higher velocity.
A similar phenomenon occurs at lower velocity in horizontal pipes (Figure 3). The pipes with the lowest transport velocity also show spiking values. However these spikes are going to higher velocities. The reason for this behaviour is coal deposits (coal layout) on the bottom of the pipe. These deposits slowly reduce the cross sectional area, sending velocities up. In horizontal pipes there are two factors that must be noted:
• In general the minimum velocity of coal in a horizontal duct cannot be controlled by the PA fan settings because at low settings the cross sectional area of the horizontal pipe section will be reduced by coal deposits until the critical velocity is reached. However in an actual application this is happening in a mode of strong velocity pulsation and is extremely unstable. As can be seen the dunes cause velocities of up to 40 m/sec. The results are pulsations in the burner that can lead to CO problems.
• The formation as well as the dissolution of coal deposits is a rather slow process. Once a large dune has build up, it cannot be blown out by a short purge. In some long horizontal pipes it takes a long time to purge the dunes out. This is especially the case when the mill is fan limited as the air at full cross sectional area of the pipe cannot reach the critical velocity that would carry the PF all the way through the horizontal. But the formation takes place over a long time. PfFLO can detect the early stage of rope formation by using microwaves of different polarisation angles to "scan" the bottom of the pipe run and detect early stages of coal lay out long before this layout causes larger velocity changes.
The primary objective should be to have a stable coal transport in the pipes. This can only be guaranteed with a minimum transport velocity. But low velocities as well as high ones can indicate problems with the combustion. For example, a boiler might be producing too much UBC with new low NOx burners. The reason can be high coal velocities causing the classification to be poor and hence the particle size to be too large. On the other hand velocities that are too low may cause coal deposits and hence an unstable stoicheometry in the flame. So both extremes in coal transport velocity can indicate the same problem in the boiler. In this example for horizontal pipes the problem was cured by throttling the high velocity pipes, thereby increasing the pressure drop over the low velocity pipes. The tendency to coal layout was significantly reduced.
A very similar problem is the one of pressure drop over the individual burner pipes. Generally there should be equal velocities, which means equal pressure drops over all burner pipes. This goal is often hard to achieve because the pipes from one mill may have different lengths and depending where their destination burner is they have a different number of bends, resulting in deviations in pressure drop. By orificing, these differences can be eliminated. A typical example is shown in Figure 4. The three pipes off one exhauster show velocities ranging from 25m/sec to 32m/sec. Pipe 3 in particular shows a very low velocity. The reason for this is that during clean air testing while setting up the mill an orifice was chosen that throttled the pipe too much. A wider orifice in the slow pipe reduced the pressure drop and allowed a more even velocity distribution and reduced CO levels from 1000 to 24 ppm
PfFLO can be used to reduce the imbalance of coal velocity in the pipe and hence find an optimum setting for the operation of the boiler. In practice the coal velocity deviates significantly from the velocity of the carrier gas (primary air). Most of the primary air settings are done with 'clean air tests', which show the velocity distribution of air over the pipes. Therefore the picture that a clean air test provides does not always reflect what is happening when coal is actually in the pipes. As can be seen in Figure 5 the particle velocities of individual coal pipes can change dramatically and deviate from the PA velocity owing to a varying relationship of particle velocity to PA velocity. The diagram shows a decrease in load (that is, in PA velocity, marked by the broken line) that results in an increase of the coal velocity in two pipes of a burner group even though the velocities in the remaining pipes follow the load change.
Controlling the pressure drop pipe to pipe on line using a variable orifice is a promising solution because in many cases overall adjustment of the PA fan is not successful – some pipes are already overshooting on velocity while others suffer from coal layout caused by low velocities.
Secondary air settings
The on line control of the secondary air is the primary target of many power stations as the secondary air settings directly influence the stoicheometric ratio at the burner. Results with this method have been very encouraging, particularly as regards NOx, CO and O2 levels. Figures 6 and 7 show the results achieved in a 10 burner boiler using PfFLO a well as on line secondary air flow measurement (IBAMS by Air Monitor Corporation) to balance the stochiometric ratio in the boiler and level out the combustion. As can be seen the NOx as well as O2 but mostly CO even out making it possible to decrease the overall level of O2 in the boiler and increase efficiency. The test was verified using a grid of 8 measurement points for O2, NOx and CO across the boiler in a total of 3 locations, giving a total of 24 measurement points throughout the boiler.
Also shown are the calculated burner stoichs as measured with PfFLO and the air flow measurement (IBAM) which reveal a similar profile to the CO readings in uncontrolled mode. Figure 6 shows the stoichiometry of the burners in uncontrolled mode, Figure 7 when the secondary air is controlled using the coal mass flow signals of PfFLO. In controlled mode the burner stoich levels out. So does the NOx, the O2 and the CO. Figure 8 shows the O2, NOx and CO levels in uncontrolled mode, Figure 9 in controlled mode. The same test showed that when controlling the boiler in the absolute mode (not using the feeder signal but just the coal flow signal) the best optimisation was achieved. This clearly indicates that (with no gravimetric feeders) the slope of the feeder table as well as the density of the coal has an impact on proper boiler control.
Generally the improvement in combustion can be verified using the CO as well as NOx and O2 measurements. On line measurement of the UBC is possible using the AshFLO application package. All parameters can be recorded and processed for boiler optimisation in the SCADA package that communicates with the PfFLO platform via a serial driver protocol. Beside manual optimisation or the direct control of secondary air in a simple control loop, knowing how to find the optimum settings of a boiler is of value as every little improvement has a great effect on savings. Hence an optimiser can be considered.
The results achieved so far with the technology of coal mass flow measurement clearly show the potential of optimisation. This evolves most clearly in cases where there is an obvious problem with a mill or a burner such as coal layout. In particular problems with burners are not easy to detect from the back end of the boiler by looking at factors such as NOx and CO content. Sometimes it is possible to track down an individual mill although very often just one burner of that mill is the problem, not the mill itself. But an on line measurement enables the user to get closer to the causes of a specific problem.
Once it has been established that on line measurements can improve combustion by controlling the main target parameters it is interesting to know what the optimum settings are. In recent years there has been success using software programs to find the best settings for a boiler. These optimisers use a set of controllable inputs and non controllable outputs to model the boiler heuristically. This is done either by neural net techniques or by linear/non linear equation solving. These models attempt to find an optimum setting for the boiler and after a certain learning time can advise on what the best settings are. Some of these programs are available as closed loop applications which allow the user to control the boiler directly.
The problem of complexity
The problem with optimisers is the complexity of information. Generally, optimisers use iterative techniques to refine their model on an ongoing basis, which means that they are always lagging behind the current state of the boiler, (for example the everyday wear and tear on the boiler) and they also have a certain noise level in regard to the accuracy of their predictions. As long as the learning phase or the speed of adaptation is short compared to the changes in the boiler the inaccuracies in prediction are small.
With an increasing number of inputs the complexity and also the reaction time of the optimisers to deviations between model and reality grow larger and accuracy levels deteriorate, mainly because the number of inputs causes the optimisation cycle time to grow in a non linear way. This also means that for example adding 40 coal flow and 40 air flow signals, plus on line UBC, as feedback values into an optimiser will not necessarily enable the system to optimise the boiler using the new information.
Deriving key parameters
In many cases it is not the measurement value itself that needs to be fed to an optimiser because the impact of it can be controlled by conventional algorithms. But the question to be solved by the optimiser is a derived value of the measurement that allows the optimisation of the control of that value.
For example if the coal and air flow levels of all burners can be measured then the question is – how can an optimum stochiometric ratio (SR) be achieved in the boiler? A typical problem that can occur here is that controlling the SR on a burner by burner basis is not that easy, because each adjustment of a burner influences another burner. Take the common observation that in a wall fired arrangement the secondary air of a middle burner can be completely switched off and still the burner stays ignited. This is caused by the air leakage of the adjacent burners in the row. So the middle burner gets its oxygen from the adjacent burners and in turn their SR is also affected by turning off the secondary air of the middle burner. This effect has a direct impact on secondary air control strategy.
The important variable to discover is the air leakage factor. Assuming that it is a variable then the effect on the correct air balance can be demonstrated by the block diagrams shown in Figure 10. The coal distribution is the same in all examples. The air distribution is modelled to that it allows an air leakage of 15 per cent to each adjacent burner. The SRs are shown for an air leakage factor of 0, 15 and 20 per cent. As can be seen there is an optimum for the SR depending on how this factor is chosen. The difficulty is that usually this factor is unknown and must be established first. This can be done using an optimiser.
Instead of all 40 fuel and air flows an optimiser should have only this parameter as a controllable input, thereby reducing the number of inputs from 40 to 1. The actual control of the individual air settings is performed by PfFLO and the air leakage model on the SCADA and can run over the plant DCS. The optimiser will find out at which air leakage factor the on line air to fuel control has its optimum. Furthermore the leakage factor could be a function with key parameters such as the vertical and horizontal leakage as well as the relationship between leakage and secondary/ primary air settings. In this case there might be up to 4 or 5 controllable input parameters to the optimiser and it will be able to make much faster predictions using derived parameters.
For T fired boilers particularly this model will certainly be interesting as the path of air and fuel is quite different from that in wall fired units. The best strategy for controlling various fuel to air distribution patterns may be by using parameters derived from the fuel and air flow information.
There are other parameters that can be derived such as the optimum velocity ratio between the coal particles and the secondary air. This optimum ratio may not be a constant but may vary with different coal types and loads etc. Also in this case there is a complexity of parameters to be monitored such as the minimum critical velocities of the coal. This can be performed by the PfFLO, and the derived parameter sent to the optimiser reduced to a key value that can be put into the model.