- Email: [email protected]
Continuous Control Issues Concerning Operation Improvement of Small-Scale Biomass Boilers
Proceedings of the 18th World Congress The International Federation of Automatic Control Milano (Italy) August 28 - September 2, 2011
Continuous Control Issues Concerning Operation Improvement of Small-Scale Biomass Boilers Jonne Haapa-aho*. Timo Korpela*1. Tomas Björkqvist*. Jan Hrdlička** Viktor Plaček***. Stanislav Vrána***. Bohumil Šulc*** * Tampere University of Technology, Dept. Automation Science and Engineering, P.O. BOX 692, 33101 Tampere, Finland (email: [email protected], [email protected], [email protected]. Corresponding author1, Tel: +358 40 849 0063) **Czech Technical University in Prague, Dept. Fluid Dynamics and Power Engineering, Technická 4, 16607 Prague, Czech Republic (e-mail: [email protected]) ***Czech Technical University in Prague, Dept. Instrumentation & Control Engineering, Technická 4, 16607 Prague, Czech Republic (e-mail: [email protected], [email protected], [email protected]) Abstract: To meet the increasing environmental and efficiency requirements, the possibilities to improve the performance of a 25 kWth wood pellet boiler by utilizing PLC and feedback control are investigated. The process is first stabilized by improving the grate sweeping sequence, which originally disturbs the process. Prior to continuous combustion control development, the process is analyzed and identified. After sequential control improvements the combustion behaves well but tends to drift. A PID controller was designed to enable drifting compensation. It is shown that improved grate sweeping sequence and continuous feedback control provide a major improvement for system performance cost effectively. Keywords: Renewable energy systems, process automation, PID control, sensors, biomass, wood pellets.
1. INTRODUCTION During the last years, biomass combustion in small-scale (< 300 kW, according to EN 303-5) heating boilers in houses and public buildings has become more and more widespread. The main fuel in interest is pellets made of wood. Small-scale boilers for combustion of the pellets have undergone certain development in construction and nowadays their officially presented efficiencies reaches up to 95 % (Lackner et al., 2010). However, the reality always depends on the person operating the boiler. The devices are normally installed by persons who have modest knowledge in combustion and control systems. Additionally, the end users are usually persons having no idea about boiler operation. Therefore, it often happens that wrong manual operations of a boiler cause significant decrease in efficiency and increase of emitted gaseous pollutants. For this reason, there are two strong motivations for development and application of advanced control systems. Due to economical reasons, the boiler should reach as high efficiency as possible. Even if the increase should make few percent points, the economy is always crucial for the household operators of the boilers. Then, there is strong tendency to establish tightened emissions limits even for the small-scale boilers that now are not valid for this range of boilers. On the other hand, when developing advanced control the costs of necessary instrumentation must be taken into account because the increased boiler price must be comparable with the prize of similar products. Despite the strong interest in biomass, the biomass boilers have some handicaps compared to also very common natural gas fired boilers for households. Currently, mostly used discrete thermostatic “on-off” regulation does not allow 978-3-902661-93-7/11/$20.00 © 2011 IFAC
power output modulation, which is possible for the gas boilers. For the pellets boiler, there are several important consequences. When the heating water reaches the desired set-point, the boiler is shut down to a so-called “afterburning” mode. This rapidly decreases efficiency and increases emissions while the heating water temperature undesirably still increases. From a comfort point of view, also another situation happens. When a higher power load is required, e.g. in a case of need of a rapid warm up, the boiler cannot adapt the power output to the required rate. As a result, there is a need to change the on-off control systems to continuously operating systems that would be able to achieve simultaneously controlled values of the required heat output and optimal conditions for the combustion process from both the emissions and efficiency point of view. However, there is a large variation in types and designs of the biomass boilers which complicates the control system design. Still, the extra costs caused by automation should be as low as possible. Due to versatility in combustion solutions, it is quite common that once the control system is designed, it must be tuned case by case. Therefore, simple tuning methods would lower the threshold of increasing the automation level of the systems. This paper focuses on combustion control to achieve low emissions and high efficiency. Papers covering combustion control of wood pellets are few but are discussed e.g. by Padinger (2001), Ruusunen et al. (2009) and Šulc et al. (2009). However, combustion control of biomass fired systems is discussed e.g. by Good & Nussbaumer (1998), Bauer et al. (2010), Liao & Dexter (2004) and Morán et al. (2006). An overview of the small scale combustion control is presented in Korpela et al. (2009b).
7035
10.3182/20110828-6-IT-1002.01968
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
2. BOILER INSTRUMENTATION AND CONTROL PLATFORM DEVELOPMENT The small-scale boiler used for the experiments was of type “Verner A25”, which is commercially available. The boiler is designed for combustion of wooden and alternative (e.g. grain, hay) pellets of diameter 6 – 8 mm. Nominal power output of the boiler is 25 kW for wooden pellets. For all experiments 6 mm softwood pellets were used. The boiler consists of a lined combustion chamber located in the bottom section and a system of heat exchangers in the upper section. A steel grate with primary air inlet from beneath is placed in the combustion chamber and secondary air inlet holes are located in the side walls of the chamber. The primary/ secondary air ratio is manually adjustable by a flap. The pellets are fed from a storage container by a screw feeder to the grate. By using the original control unit of the boiler, feeding of the pellets takes place periodically with preset periods of screw movement and idle state. During the operation, also the grate is periodically moved (swept) in the way that the pellets are moved towards end side of the combustion chamber. In optimal case, the pellets are completely burned out when they reach the end of the chamber. The ash is afterwards collected in the ash container. The combustion air is fed into the boiler by an air fan, which is originally impulse-controlled and allows manual adjustment to four different air flow rates. The boiler has been afterwards equipped with several measurement and indication points, as shown on the Fig. 1. In the original setting, none of these additional sensors is available, as there are originally only two sensors detecting temperature of outlet water and flue gas. The experimental setup has involved measurement of temperatures at the front of the combustion chamber (T1), temperature after first convective section (T2), flue gas temperature (T3), temperature of inlet (T4) and outlet water (T5), oxygen and CO concentrations in the flue gas and flow rate of the combustion air. The temperature T2 is the one used later in this paper for analysis and control. The boiler was originally equipped with manufacturer’s control unit that only allows very limited intervention in the process. A newly designed switchboard was therefore connected parallel to the original control unit. The switchboard contains circuits assuring protection against overload, short-circuit and forbidden combination of inputs, power sourcing, noise shielding, central earth and emergency stop function. The switchboard is also equipped with RexWinLab-8000 control and data acquisition unit, which is developed along with the switchboard at Department of Instrumentation and Control Engineering at the Czech Technical University in Prague. The RexWinLab-8000 is based on PAC (Programmable Automation Controller) Wincon 8000 series, but the control firmware of PAC is replaced by software named REX Control developed at Institute of Cybernetics at University of West Bohemia in Pilsen. The REX Control allows using of advanced control algorithms commonly unavailable in standard controllers. It has also its own graphical user interface based on similar principles as Simulink produced by MathWorks. Also realtime communication with Matlab/Simulink is possible.
Fig. 1. Scheme of the boiler with the instrumentation. The new configuration allows preparing experiments in advance on a standard PC. Any control algorithm synthesis can be easily realized in the graphical development environment, which is very similar to the well known Simulink. Data from all sensors placed in the boiler sensors are time-synchronized and centralized to PAC. During the experiment PAC sends measured data to Simulink running on a remote computer. This allows the operator to monitor and potentially interfere the combustion process during the experiment. Data is periodically saved in Simulink for further off-line analysis. The described development of the switchbox with RexWinLab-8000 changed the standard factory boiler to an experimental base that allows preparation, monitoring, interfering, data acquisition and analysis during the experiments in real-time. 3. COMBUSTION PROCESS EVALUATION 3.1 Optimization of grate sweeping sequence Besides other influences, operation of the boiler depends on grate sweeping. If the fixed pre-programmed control units are used, disturbing periodical strong peaks in the combustion process may occur after the grate sweeping, as shown on Fig. 2 in the left. This phenomenon was not only the source of higher concentration of CO but it also made complications in the identification and control. The idea of solving this issue was to shorten the grate sweeping period making the grate to travel shorter distance, thus less disturbing the process. The sweeping period was shortened from original 10 minutes to 2 minutes. However, this change caused insufficient removal of ash from the grate. Therefore, other experiments have to be done to determine the right distance of grate sweeper travel. The peaks in all measured quantities of combustion process with this new configuration practically disappeared; see Fig. 2 in the right.
7036
O2 (%)
Temperature (C)
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
540 520 500 480 460 440
5
10
15 20 Time (min)
25
30
5
10
15 20 Time (min)
25
30
5
10
15 20 Time (min)
25
30
12 10
3.2 Process analysis Even if the improved grate sweeping was stopped on purpose, variables kept on fluctuating on a low level. This can be seen in Fig. 3, which presents a sample in time domain with open loop control showing temperature, O2 and CO. Therefore, the reason for fluctuations was studied by process analysis. The T2 (Fig. 1) measurement was the fastest and most sensitive measurement available in the process experiments, so its behaviour was investigated by auto- and crosscorrelation analysis with optimized grate sweeping applied. In the auto-correlation curve as a function of time delay in Fig. 4 (upper) it can be seen that there are side peaks on both sides of the main peak. These side peaks occur at some 120 second intervals, which is the time difference between two grate sweeping movements. However, the side peaks have a low correlation level, which indicates that the fluctuation of temperature is not dominated by the periodic grate sweeping movement and therefore the main reason for fluctuation comes elsewhere. The same minor influence can also be seen in the time domain signal (Fig. 3), where the grate sweeping procedure is not always followed by a rise in temperature but occasionally followed even by a drop.
Fig. 3. Temperature after the first heat exchanger T2 (upper), O2 (mid) and CO (low) as function of time. Instants of grate sweeping movement are presented in every figure. It can be deducted from Ruusunen et al. (2009) that changes in draft can be seen in air flow measurement. As there were no significant correlation between the air flow measurement and the temperature signals, it can be assumed that the temperature fluctuation is not explained by draft or air flow changes. The fuel feed fluctuation cannot be measured directly. However, the periodicity of 20 seconds, which was the fuel feed period, could been seen in cross-correlation curve of the temperature and the CO signals. This effect was minor in size, but this indicates that the ON/OFF fuel feed has an existing but minor effect on combustion in this process. As the major fluctuation of combustion temperature cannot be directly connected to draft or any of the control variables, i.e. pellet feed, combustion air feed, and grate sweeping, the conclusion is that the fluctuation is mainly due to internal non measured interactions. As the local combustion intensity on the grate is not caused by manipulating variables and cannot be compensated with the actuators available, the feasible solution is not to excite the feedback control with the disturbance but to filter it away. 1 Auto-correlation (-)
The grate sweeping causes peaks in emissions since there is usually not enough secondary air to burn all the gases from the momentarily accelerated pyrolysis. Thus, this effect could further be compensated by changing the air staging around the sweeping as presented by Korpela et al. (2008), but due to constant air distribution this was not possible in this boiler.
50 40 30 20 10
0.5
0
-0.5 -500
-400
-300
-200
-100
0 100 Delay (s)
200
300
400
500
-400
-300
-200
-100
0 100 Delay (s)
200
300
400
500
0.5 Cross-correlation (-)
Fig. 2. CO and temperature responses and grate sweeping time instants with original (left) and improved (right) grate sweeping algorithm.
CO (ppm, 10 % O2)
8
0
-0.5
-1 -500
Fig 4. Auto-correlation of temperature T2 (Up). Crosscorrelation of O2 in relation to temperature measurement (Down).
7037
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
4. CONTROL DESIGN 4.1 Filtering of disturbances As it can be seen in Fig. 3, the same disturbances occur in oxygen concentration and temperature in the upper end of the combustion chamber. Hence, it was reasonable to use the same filtering with both variables. Since it was quite clear that grate sweeping causes minor disturbances enduring tens of seconds, it was concluded that the filter should have the capability of removing such low frequency disturbances. A few different filters were tested which were: a running mean filter with time window of 150 s carried out by a discrete transfer function (1), a running median filter with time window of 120 s (2), a genuine running mean filter with time windows of 120 s (3) and 150 s (4). Such filters have dynamics that decrease the performance of the control system. 4.2 Step responses and process identification Step response tests were conducted for both process inputs, the air feed and the fuel feed in order to enable process identification. First order transfer function with a time delay (FOTD) was fitted to both upwards and downwards step data of the experiments. Since it was concluded that such kind of filtering was necessary due to the disturbances, the step response data was filtered with the same filter that was used in the control experiment before fitting the transfer function to it. When the measurement filter was changed, the process model was recalculated. Process models were obtained by conventional means and by exploiting Matlab Identification Toolbox. When Identification Toolbox was used, the input and output data were pre-processed by removing means and clearly disturbed data. Process models that were obtained with each filter are mentioned in Tables 2-4 in Appendix 1. Step response test for the air feed was conducted by running the boiler on the nominal heat output level (25 kW). The frequency of the fan motor was changed from 35 to 43 Hz and then back to 35 Hz. In the air feed step response experiments it was noticed that the dynamics, especially in O2, seemed to differ depending on the direction of the step change. When fitting the FOTD model to the data, the time constant of the process model was longer in the case of upward than downward step. Time delay of the process was somewhat equal regardless of the direction of the step change. The response of temperature to an air feed increase was almost immediate, although the temperature gradually came down after the initial rise. The fast response is due to the fact that the residence time of the hot gases in the combustion chamber was shortened. The gradual fall in the temperature is due to the cooling effect of the increased air feed into the combustion chamber. Step response experiments were also conducted with the fuel feed. In the first step the fuel feed rate was decreased from 100% to 90% of the nominal heat output. The second step raised the fuel feed rate back to nominal power level. O2 level’s response to an increase in the fuel feed was in the
opposite direction. There was a significant time delay from the fuel feed change to the reaction in the oxygen level and the response was clearly slower than in the case of air feed change. Also, the time constant was significant. Yet, when comparing to the temperature during the same step, the time delay was shorter. As a result, there were substantial dynamics from the fuel feed to the temperature as time delay was c. 5 minutes and the time constant was c. 15 minutes. 4.3 Determination of input-output pairs for control The chosen variables to be controlled were oxygen level in the flue gas and temperature after first convection section (T2 in Fig. 1). The oxygen level provides information of the state of combustion. The gas analyzer measuring the oxygen concentration had a time delay of c. 23 s which causes difficulties from the control point of view. Temperature provides information about the intensity of the combustion and gives an indication of the amount of fuel that is fed. It would be beneficial to use a temperature sensor as close to the combustion chamber as possible, but to avoid disturbing effect of heat radiation from direct flame to the thermocouple, the temperature was chosen to be measured from the upper end of the combustion chamber. The oxygen level can be controlled by the air feed. This offers the most efficient and the fastest way available to affect the O2 level. Despite that, there were still substantial dynamics from the air feed to oxygen. In addition, time delay of the gas analyzer decreased the performance that could be gained by this control strategy. Also a drawback is that the air feed has a significant effect on combustion on the grate and thus disturbing the combustion process, which in turn causes fluctuations into the oxygen level. Due to this strong interaction to other variables, it is necessary that the air feed changes are to be made calmly. Hence, this excludes the control of fast disturbances, but slow drifting can still be avoided by exploiting this strategy. Thus, a feedback control based on O2 measurement and air feed was utilized. The O2 level can also be controlled by the fuel feed. Since the response of O2 to a change in fuel feed was in the opposite direction (negative process gain), the controller’s gain got also a negative sign as can be seen in Table 3. In the step response experiments it proved to be a significantly slower way to affect oxygen level than with air feed. It is thus impossible to control fast changes in the oxygen level by the fuel feed but again the drifting of the process is avoidable. The fuel feed affects also the temperatures in the boiler but comparing to the air feed, the effect is much slower and smoother, thus disturbing the process less than the changes in the air feed. Therefore, controlling fast changes in the temperature by fuel feed is not possible but the elimination of slow drifting is. Though, changing the air had almost an instant effect on the chosen temperature, controlling this temperature by air feed would, in the end, drive the process either to a point where the oxygen would run out or a point where the fuel bed would shrink until a point where the temperature would not be high
7038
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
enough to maintain the combustion. In either of these cases fire would extinguish. Due to these facts it was not reasonable to utilize feedback control of air feed to control temperature. 4.4 Tuning methods Standard PID controllers were used during the experimental runs. Tuning of the controllers was done with AMIGO method (Åström & Hägglund, 2006) and the method proposed by Åström and Murray (2009). Also pole placement method was used (e.g. Åström & Hägglund, 2006). Conventional Ziegler-Nichols method provided significantly too high controller gains. Also the controller gains given by the used tuning rules were too high for the purpose, due to internal disturbances of the process and due to the fact that also these methods are developed mainly for the servo control case. There, the role of the disturbances is usually minor. Instead, combustion control is to a large extent a regulation control case, where control is used for disturbance compensation. The process models from the identification were used with the tuning formulas. The PID parameters that were used are presented in Tables 2-4 in Appendix 1. 5. CONTROL ANALYSIS 5.1 Open loop control The open loop control statistics are illustrated in Table 1 in Appendix 1. The same process inputs can result in quite different operating areas in terms of O2 and temperature which verifies that there is a tendency of drifting in the process. This kind of behaviour is undesirable and calls for feedback control. In the oxygen a difference of one percentage unit could be seen, which will cause significantly differing combustion conditions. Also difference of c. 10 °C in the temperature could be noticed. This change itself is insignificant but it indicates the change in process behaviour. Yet, the open loop behaviour of the combustion process was rather stable after the optimization of the grate sweeping sequence which can be seen in the deviations in Table 1. In the open loop experiments, the boiler was run on the nominal power level (25 kW). On this power level, the combustion got more restless in the terms of CO production when the O 2 concentration in the flue gas reached an average concentration of 10.1%. The first indication of this can be seen in the standard deviation of CO rather than in the mean value of CO. The same restlessness cannot yet be seen in the deviations of the O2 or the temperature. 5.2 Control of oxygen by air feed The fastest way to affect oxygen concentration was by air feed but as was discussed earlier, changes in air feed disturbed the combustion process quite remarkably indicating strong coupling between the variables. Another problem with this control strategy was the long delay of the gas analyzer. As it can be seen in Table 2, the desired setpoints were always achieved and maintained. Yet, when comparing the
standard deviations of the oxygen control experiments with 100% power to the ones from the open loop control experiments, such low deviations could not be obtained with the feedback control. Thus the oxygen behaviour with this feedback control strategy was more restless which in turn causes larger deviation into other variables as well. This is to some extent due to the long time delay of the oxygen measurement and the substantial filtering of the measurement signal. Another reason for the larger deviation was that the primary and secondary air was supplied by the same fan. Thus, the undesired effects from adding more primary air when only increased amount of secondary air was wanted worsens the results severely. Compared to the case with 100% power level experiments, the feedback control made the process more restless on a lower power level (c. 76% of the nominal) as well. The PI parameter values used during this experiment were obtained by using the AMIGO PI tuning method. It provides the smallest controller gain values from the formulas but in spite of this, the gain was still slightly too high. As the controller gain was manually reduced the control got calmer and due to that the deviations started to approach the ones of the open loop system. Though the controller’s acting on the process was reduced and the lower gains were enough to prevent the process from drifting. A reason for higher emissions during the lower power level run was the fact that the 10.5% setpoint was too low for this power region. Due to limitations set by the measurement equipment (long gas analyzer delay) and the equipment of the boiler (one fan), the purpose of the feedback control was only to prevent the process from drifting rather than active disturbance control. 5.3 Control of oxygen by fuel feed The response of the oxygen measurement to changes in fuel feed was quite slow and calm. Thus, feedback control with this input leads to slow and calm behaviour of the controlled system which suits well to drifting prevention. During the experiment, the setpoint was maintained well and the deviation of oxygen level was even smaller than in the case of the open loop system as can be seen when comparing the results in Tables 1 and 3. The calm control can also be seen in the small deviation in the controller’s output. Since the air feed during this strategy is kept constant (if constant power is desired), the behaviour of the combustion is very stable and it approaches the behaviour of open loop system, which can be seen in the standard deviations of CO and temperature. When comparing this control strategy to the oxygen control by air feed on the same power level, the standard deviation of the oxygen level was smaller and the standard deviation of the temperature was significantly smaller with this control strategy. Also the CO deviation was slightly smaller. Due to the slow effect of this input to the oxygen level, the time delay of the gas analyzer was insignificant. Therefore this control strategy worked better than the control with air feed.
7039
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
5.4 Control of temperature by fuel feed Control of the temperature by the fuel feed resulted in quite stable combustion behaviour given by slow dynamics from the fuel feed and constant air feed. During the experiments, the temperature setpoint was always maintained, as can be seen in Table 4. The deviations of temperature, oxygen and CO were also about the same size as in the open loop system. The CO emissions were quite low during the experiments due to optimal oxygen level. The variation in the controller’s output in the last two experiments in Table 4 was quite small indicating slow overall system behaviour. Since the fan’s input was kept constant during the experiments, the oxygen level changes as a function of the added fuel. In order to maintain a reasonable oxygen level, a corresponding setpoint for the temperature must be found (Korpela et. al., 2008). By increasing the temperature setpoint, the oxygen level goes down and vice versa. This control strategy only calls for very simple instrumentation since there is only one thermocouple needed. Since the oxygen level correlates with the temperature (Fig. 4), it is therefore possible to control the oxygen level indirectly by setting the setpoint properly. The problem of finding correct temperature setpoints for every power level can be quite tricky since the oxygen level is a function of air feed and fuel feed and the dependences in between are often nonlinear. The selection of a suitable temperature setpoint is discussed in Korpela et al. (2009b). When comparing the last two experiments in Table 4, there is a power raise conducted by raising the temperature setpoint and the input for the air fan. The oxygen level in both of the experiments remained the same. 6. DISCUSSION As seen in the process analysis, the autocorrelation of the combustion temperature showed a minor periodicity at an interval of 120 seconds (Fig. 4). The obvious reason is the grate sweeping at the same interval which indicates a remaining effect on combustion despite the implementation of the optimized grate sweeping sequence. The low relative values of correlation at the interval peaks tell that the influence is minor compared to other fluctuations of the combustion temperature. The same minor influence could also be seen in the time domain signal (Fig. 3). As the major fluctuation of combustion temperature could not be connected to draft or any of the control variables, i.e. pellet feed, combustion air feed and grate sweeping, the conclusion is that the fluctuation is mainly due to internal non modelled interactions. It is not excluded that these actions do have an excitation in one or some of the control variables but in this case the influence has a stochastic nature which cannot be seen in correlation analysis. Observing the fluctuations, an idea has appeared that there is an influence of a relatively large combustion chamber and a relatively low primary air supply to the grate area. This might imply that the combustion intensity on the grate is quite low and therefore the variation in pellet and primary air flow distributions over the grate become significant. Thus, the
amount of stored chemical energy on the grate can vary. At this situation it is likely that the local combustion intensity becomes self amplifying and the cumulative result is that the combustion temperature fluctuates in periods of tens of seconds which from a control point of view can be seen as an internal disturbance. Even though the disturbance has a stochastic nature it is most probable that the seeds for the disturbance come from differences in rest positions of the grate after sweeping and uneven pellet feed. However, there is no supporting evidence of the size influence of the combustion chamber. As the local combustion intensity on the grate cannot be compensated with the actuators available, the feasible solution is not to excite the feedback control with the disturbance but to filter it away as was done in the empirical part of the work. The result is a feedback control, which stabilizes the combustion on long term which compared to common environmental requirements gave good results. A natural way to improve the situation would be to change the burner design to decrease the amount of pellets under combustion and to enable possibility of controlling the air distribution towards increase of local primary air, which requires re-designing of the combustion air system. As a result, there would be less room for self amplifying combustion. Additionally to a more even combustion at open loop control, the solution could give more possibilities for feedback control on short term as air feed is one of the actuators, but this was not tested in the experimental work. On the other hand, fewer pellets under combustion could make the process more vulnerable to unequal pellet feed, which would be seen as fluctuation in all main combustion values. A central challenge would then be to even the pellet feed, which is a common issue in most of the biomass combustion processes. The fundamental task of the control system presented in this paper is the drifting compensation. However, further improvement of the system performance could be gained if there was a fast sensor for monitoring quality of combustion (e.g. lambda probe or new type of CO measurement) and secondary air feed was conducted by a separate fan. Then the improvement potential of the short term performance of the process by feedback control would be increased. In the case of sophisticated instrumentation also the benefit for advanced control, e.g. feed forward compensation of pellet feed and cascade, multiple input – multiple output (MIMO), neural networks and fuzzy control, is brought up. These kinds of control methods are utilized in some high-end products on the market but the performance and detailed operation of these systems are not published. Furthermore, application of such kinds of advanced approaches is more specific for large-scale boilers. It is still questionable, are the benefits that could be gained by adding such instrumentation enough in small-scale combustion where the behaviour is already quite steady or should the efforts be put into stabilizing process design, as was done with the optimization of the grate sweeping sequence. Generally, the benefit from the advanced control can also be gained in extended lifetime of the boiler and its more efficient operation, e.g. fuel saving.
7040
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
7. CONCLUSIONS
REFERENCES
This research shows that due to an already stable process the improvement potential of the short term behaviour of combustion with feedback control is limited. However, the utilization of PID control assures that a boiler is always run in a suitable operating area despite that combustion processes have a tendency to drift in the long term due to different variations in the process. With feedback control, the drifting can be prevented and thus efficiency can be maintained high and emissions low.
Bauer, R. Gölles, M., Brunner, T., Dourdoumas, N., Obernberger, I. (2010) Modelling of grate combustion in a medium scale biomass furnace for control purposes. Biomass and Bioenergy, 34, 2010. pp. 417 – 427. Good, J., Nussbaumer, Th. 1998. Efficiency improvement and emission reduction by advanced combustion control technique (ACCT) with CO/lambda control and setpoint optimization, Biomass for energy and industry, 10th European conference and technology exhibition, June 8– 11, 1998, Würzburg, Germany, pp. 1362 – 1365. Korpela, T., Björkqvist, T. & Lautala, P. (2008). Durable feedback control system for small scale wood chip combustion. Proceedings of World Bioenergy 2008, Conference & Exhibition on Biomass for Energy, 27–29 May 2008, Jönköping, Sweden. pp. 224–230. Korpela, T., Björkqvist, T., Lautala, P. (2009a). Control strategy for small-scale wood chip combustion. IFAC Symposium on Power Plants and Power Systems Control, 5–8 July, 2009, Tampere, Finland. Korpela, T., Ruusunen, M., Björkqvist, T. & Lautala, P. (2009b). Control structures of a multivariable process applied to small scale biomass combustion. Automaatio XVIII Seminaari 17–18.3.2009, Helsinki. Finland. SAS julkaisusarja 36. Suomen Automaatioseura. 8 s. Lackner, M., Winter, F., Agarwal A. K. editors 2010. Handbook of Combustion, Wiley-VCH Verlag. Vol. 4., pp. 141 – 169. Liao, Z., Dexter, A. (2004) The potential for energy saving in heating systems through improving boiler controls. Energy and Buildings. 36, 2004. pp. 261–271 Morán, J., Granada, E., Míguez, J., Porteiro, J. (2006) Use of grey relational analysis to assess and optimize small biomass boilers. Fuel Processing Technology, 87, 2006, pp. 123– 127. Padinger, R. (2001). Regelungstechnik für die Hausheizung der Zukunft, Project report, Joanneum Research, Austria, IEF-b-12/01 Ruusunen, M., Korpela T., Björkqvist, T. 2009. The effect of control parameters to the quality of small-scale wood pellet combustion. In: Savolainen, M. (ed.). Bioenergy 2009, Sustainable Bioenergy Business, 4th International Bioenergy Conference, Book of Proceedings, 31.8.– 4.9.2009, Jyväskylä, Finland. FINBIO publication 45. pp. 507–515. Šulc, B., Vrána, S., Hrdlička, J., Lepold, M. (2009). Control for ecological improvement of small biomass boilers, IFAC Symposium Power Plants & Power Systems, 5–8 July, 2009, Tampere, Finland. Åström, K., Hägglund, T. (2006). Advanced PID Control. ISA - Instrumentation, Systems, and Automation Society, Research Triangle Park, NC 27709. 461 p. Åström, K., Murray, R. (2008). Feedback Systems – An Introduction for Scientists and Engineers. Princeton University Press. Electronic edition, Version v2.10b. 396 p.
The utilization of continuous control by PID controllers and/or more sophisticated control methods requires utilization of PLC (programmable logic controller). This will increase the total costs of the boiler slightly but the obtainable benefits are significant. In addition to drifting prevention, the possibility of continuous power modulation, search for optimal controller parameters from the efficiency and emission point of view, adaptability for fuel quality changes and suppression of short term fluctuations are benefits that continuous control can provide. In general, short term performance of the process defines the safety margins needed around the setpoint. The benefit that is gained from a good short term performance of the process is that it enables adjustment of the setpoint of the control system closer to a limit value but still avoid crossing it. In this case, better short term performance, i.e. smaller fluctuation of oxygen level, allows adjustment of the oxygen setpoint closer to the limit value, where CO starts to grow due to lack of air, without crossing it. Running the process with lower oxygen would improve the efficiency. Thus, the short term performance can be said to be directly linked to efficiency of the boiler. However, the stochastic and nonlinear nature of the combustion process makes the process identification quite hard since the dynamics of the process can vary significantly depending on the operating area. This complicates utilization of classical tuning methods since they are developed for deterministic processes. Yet, the tuning methods provide reasonable rough estimates of the controller parameters. Hence, the PID controller gains given by standard tuning rules were somewhat too high for the purpose due to internal disturbances of the process. When the controller gains were manually brought down, the short term performance of the controlled process started to approach the short term performance of the open loop system. Yet, the main purpose of the applied control was to prevent the combustion from drifting and this goal was achieved. ACKNOWLEDGEMENTS This cooperative work was supported by Fortum Foundation and the Ministry of Education of the Czech Republic within the project No. MSM6840770035 “Development of environmentally friendly decentralized power systems”, which are gratefully acknowledged.
7041
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
APPENDIX 1
Table 1. Open loop system statistics. Test 1 2 3 4
O2 mean±std 10.6±0.6 10.9±0.6 10.1±0.6 11.1±0.6
CO mean± std 28.3±5.7 33.5±5.2 30.8±9.5 27.0±6.8
Tex mean± std 518.6±8.3 519.6±8.1 533.4±7.4 528.0±6.5
Power (% of nom.) 100 100 100 100
Fan (Hz) 35 35 35 35
Table 2: O2 control by air feed. In the filter column: 1 = running mean filter with 150s time window (discrete transfer function), 2 = running median filter with 120s time window, 3 = running mean filter with 120s time window, 4 = running mean filter with 150s time window. O2 setpoint = 10.5% in every experiment. u stands for fan input in Hz. PID parameters: K/Ti/Td
contr. mean±std
CO mean± std
u mean± std
Tex mean± std
Power (% of nom.)
2.4 / 125 / 0
10.5±0.7
31.4±11.7
34.5±0.7
535.2±9.4
100
5.5 / 100 / 18
10.5±0.6
32.7±12.6
33.4±1.1
516.5±12.5
100
Å&M PI
3.1 / 174 / 0
10.5±0.6
31.5±6.8
35.3±1.4
532.8±15.3
100
AMIGO PI
5.6 / 327 / 0
10.4±0.6
32.5±9.3
34.9±1.9
500.2±21.2
91
AMIGO PI
2.4 / 201 / 0
10.5±0.7
27.8±8.7
35.2±0.8
508.7±10.1
91
0.388/484/80
AMIGO PI
4.0 / 356 /0
10.4±0.7
49.9±30.1
31.4±1.7
441.2±14.4
76.3
-
-
0.75
3.0 / 356 / 0
10.5±0.7
56.3±16.3
32.1±1.1
455.5±9.8
76.3
-
-
2.0 / 356 / 0
10.5±0.6
50.0±12.8
32.3±0.7
457.8±7.4
76.3
-
-
2.7 / 484 / 0
10.4±0.6
47.8±13.5
32.2±0.8
456.8±7.5
76.3
4
0.388/484/80
0.5 Pole placement 50% AMIGO PI
2.0 / 356 / 0
10.6±0.7
38.3±7.8
34.3±0.5
485.3±6.7
87
Filter
FOTD model: K/Tp/Td
Tuning method
1
0.338/150/41
1
0.338/150/41
AMIGO PI AMIGO PID
2
0.325/216/87
3
0.325/466/67
3
0.325/240/67
4
Table 3: Oxygen control by fuel feed. In the filter column: 4 = running mean filter with 150s time window. Setpoint for the O2 was 10.5 %. u stands for the total fuel feed period (s) which is the sum of constant T on = 3s and controlled Toff. Filter
FOTD model: K/Tp/Td
Tuning method
PID parameters: K/Ti/Td
contr. Mean± std
CO mean± std
u mean± std
Tex mean±std
Fan (Hz)
4
-89.5/923/187
AMIGO PI
-0.00133 /719/0
10.5±0.5
32.5±6.8
21.7±0.7
513.5±6.0
36
Table 4: Temperature control by fuel feed. In the filter column: 2 = running median filter with 120s time window, 3 = running mean filter with 120s time window, 4 = running mean filter with 150s time window. u stands for the total fuel feed period (s) which is the sum of constant Ton = 3s and controlled Toff.
Filter 2 3 4
FOTD model: K/Tp/Td 1489/94 9/303 1489/91 6/305 1489/91 6/313
Tuning metho d Å&M AMIG O PI AMIG O PI
PID parameters K/Ti/Td 8.37*10-4/ 722.5/0 4.29*10-4 /703/0 4.16*10-4 /797/0
Set point
contr. Mean± std
CO mean± std
u mean± std
O2 mean± std
Fan (Hz)
Power (% of nom.)
480
482.2±9.7
28.2±9.5
25.8±1.4
12.5±0.7
35
77.4
500
497.8±6.7
27.2±6.8
24.0±0.5
11.2±0.6
35
83.3
510
511.5±7.9
33.8±5.2
22.5±0.4
11.3±0.5
38
88.9
7042
Continuous Control Issues Concerning Operation Improvement of Small-Scale Biomass Boilers Jonne Haapa-aho*. Timo Korpela*1. Tomas Björkqvist*. Jan Hrdlička** Viktor Plaček***. Stanislav Vrána***. Bohumil Šulc*** * Tampere University of Technology, Dept. Automation Science and Engineering, P.O. BOX 692, 33101 Tampere, Finland (email: [email protected], [email protected], [email protected]. Corresponding author1, Tel: +358 40 849 0063) **Czech Technical University in Prague, Dept. Fluid Dynamics and Power Engineering, Technická 4, 16607 Prague, Czech Republic (e-mail: [email protected]) ***Czech Technical University in Prague, Dept. Instrumentation & Control Engineering, Technická 4, 16607 Prague, Czech Republic (e-mail: [email protected], [email protected], [email protected]) Abstract: To meet the increasing environmental and efficiency requirements, the possibilities to improve the performance of a 25 kWth wood pellet boiler by utilizing PLC and feedback control are investigated. The process is first stabilized by improving the grate sweeping sequence, which originally disturbs the process. Prior to continuous combustion control development, the process is analyzed and identified. After sequential control improvements the combustion behaves well but tends to drift. A PID controller was designed to enable drifting compensation. It is shown that improved grate sweeping sequence and continuous feedback control provide a major improvement for system performance cost effectively. Keywords: Renewable energy systems, process automation, PID control, sensors, biomass, wood pellets.
1. INTRODUCTION During the last years, biomass combustion in small-scale (< 300 kW, according to EN 303-5) heating boilers in houses and public buildings has become more and more widespread. The main fuel in interest is pellets made of wood. Small-scale boilers for combustion of the pellets have undergone certain development in construction and nowadays their officially presented efficiencies reaches up to 95 % (Lackner et al., 2010). However, the reality always depends on the person operating the boiler. The devices are normally installed by persons who have modest knowledge in combustion and control systems. Additionally, the end users are usually persons having no idea about boiler operation. Therefore, it often happens that wrong manual operations of a boiler cause significant decrease in efficiency and increase of emitted gaseous pollutants. For this reason, there are two strong motivations for development and application of advanced control systems. Due to economical reasons, the boiler should reach as high efficiency as possible. Even if the increase should make few percent points, the economy is always crucial for the household operators of the boilers. Then, there is strong tendency to establish tightened emissions limits even for the small-scale boilers that now are not valid for this range of boilers. On the other hand, when developing advanced control the costs of necessary instrumentation must be taken into account because the increased boiler price must be comparable with the prize of similar products. Despite the strong interest in biomass, the biomass boilers have some handicaps compared to also very common natural gas fired boilers for households. Currently, mostly used discrete thermostatic “on-off” regulation does not allow 978-3-902661-93-7/11/$20.00 © 2011 IFAC
power output modulation, which is possible for the gas boilers. For the pellets boiler, there are several important consequences. When the heating water reaches the desired set-point, the boiler is shut down to a so-called “afterburning” mode. This rapidly decreases efficiency and increases emissions while the heating water temperature undesirably still increases. From a comfort point of view, also another situation happens. When a higher power load is required, e.g. in a case of need of a rapid warm up, the boiler cannot adapt the power output to the required rate. As a result, there is a need to change the on-off control systems to continuously operating systems that would be able to achieve simultaneously controlled values of the required heat output and optimal conditions for the combustion process from both the emissions and efficiency point of view. However, there is a large variation in types and designs of the biomass boilers which complicates the control system design. Still, the extra costs caused by automation should be as low as possible. Due to versatility in combustion solutions, it is quite common that once the control system is designed, it must be tuned case by case. Therefore, simple tuning methods would lower the threshold of increasing the automation level of the systems. This paper focuses on combustion control to achieve low emissions and high efficiency. Papers covering combustion control of wood pellets are few but are discussed e.g. by Padinger (2001), Ruusunen et al. (2009) and Šulc et al. (2009). However, combustion control of biomass fired systems is discussed e.g. by Good & Nussbaumer (1998), Bauer et al. (2010), Liao & Dexter (2004) and Morán et al. (2006). An overview of the small scale combustion control is presented in Korpela et al. (2009b).
7035
10.3182/20110828-6-IT-1002.01968
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
2. BOILER INSTRUMENTATION AND CONTROL PLATFORM DEVELOPMENT The small-scale boiler used for the experiments was of type “Verner A25”, which is commercially available. The boiler is designed for combustion of wooden and alternative (e.g. grain, hay) pellets of diameter 6 – 8 mm. Nominal power output of the boiler is 25 kW for wooden pellets. For all experiments 6 mm softwood pellets were used. The boiler consists of a lined combustion chamber located in the bottom section and a system of heat exchangers in the upper section. A steel grate with primary air inlet from beneath is placed in the combustion chamber and secondary air inlet holes are located in the side walls of the chamber. The primary/ secondary air ratio is manually adjustable by a flap. The pellets are fed from a storage container by a screw feeder to the grate. By using the original control unit of the boiler, feeding of the pellets takes place periodically with preset periods of screw movement and idle state. During the operation, also the grate is periodically moved (swept) in the way that the pellets are moved towards end side of the combustion chamber. In optimal case, the pellets are completely burned out when they reach the end of the chamber. The ash is afterwards collected in the ash container. The combustion air is fed into the boiler by an air fan, which is originally impulse-controlled and allows manual adjustment to four different air flow rates. The boiler has been afterwards equipped with several measurement and indication points, as shown on the Fig. 1. In the original setting, none of these additional sensors is available, as there are originally only two sensors detecting temperature of outlet water and flue gas. The experimental setup has involved measurement of temperatures at the front of the combustion chamber (T1), temperature after first convective section (T2), flue gas temperature (T3), temperature of inlet (T4) and outlet water (T5), oxygen and CO concentrations in the flue gas and flow rate of the combustion air. The temperature T2 is the one used later in this paper for analysis and control. The boiler was originally equipped with manufacturer’s control unit that only allows very limited intervention in the process. A newly designed switchboard was therefore connected parallel to the original control unit. The switchboard contains circuits assuring protection against overload, short-circuit and forbidden combination of inputs, power sourcing, noise shielding, central earth and emergency stop function. The switchboard is also equipped with RexWinLab-8000 control and data acquisition unit, which is developed along with the switchboard at Department of Instrumentation and Control Engineering at the Czech Technical University in Prague. The RexWinLab-8000 is based on PAC (Programmable Automation Controller) Wincon 8000 series, but the control firmware of PAC is replaced by software named REX Control developed at Institute of Cybernetics at University of West Bohemia in Pilsen. The REX Control allows using of advanced control algorithms commonly unavailable in standard controllers. It has also its own graphical user interface based on similar principles as Simulink produced by MathWorks. Also realtime communication with Matlab/Simulink is possible.
Fig. 1. Scheme of the boiler with the instrumentation. The new configuration allows preparing experiments in advance on a standard PC. Any control algorithm synthesis can be easily realized in the graphical development environment, which is very similar to the well known Simulink. Data from all sensors placed in the boiler sensors are time-synchronized and centralized to PAC. During the experiment PAC sends measured data to Simulink running on a remote computer. This allows the operator to monitor and potentially interfere the combustion process during the experiment. Data is periodically saved in Simulink for further off-line analysis. The described development of the switchbox with RexWinLab-8000 changed the standard factory boiler to an experimental base that allows preparation, monitoring, interfering, data acquisition and analysis during the experiments in real-time. 3. COMBUSTION PROCESS EVALUATION 3.1 Optimization of grate sweeping sequence Besides other influences, operation of the boiler depends on grate sweeping. If the fixed pre-programmed control units are used, disturbing periodical strong peaks in the combustion process may occur after the grate sweeping, as shown on Fig. 2 in the left. This phenomenon was not only the source of higher concentration of CO but it also made complications in the identification and control. The idea of solving this issue was to shorten the grate sweeping period making the grate to travel shorter distance, thus less disturbing the process. The sweeping period was shortened from original 10 minutes to 2 minutes. However, this change caused insufficient removal of ash from the grate. Therefore, other experiments have to be done to determine the right distance of grate sweeper travel. The peaks in all measured quantities of combustion process with this new configuration practically disappeared; see Fig. 2 in the right.
7036
O2 (%)
Temperature (C)
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
540 520 500 480 460 440
5
10
15 20 Time (min)
25
30
5
10
15 20 Time (min)
25
30
5
10
15 20 Time (min)
25
30
12 10
3.2 Process analysis Even if the improved grate sweeping was stopped on purpose, variables kept on fluctuating on a low level. This can be seen in Fig. 3, which presents a sample in time domain with open loop control showing temperature, O2 and CO. Therefore, the reason for fluctuations was studied by process analysis. The T2 (Fig. 1) measurement was the fastest and most sensitive measurement available in the process experiments, so its behaviour was investigated by auto- and crosscorrelation analysis with optimized grate sweeping applied. In the auto-correlation curve as a function of time delay in Fig. 4 (upper) it can be seen that there are side peaks on both sides of the main peak. These side peaks occur at some 120 second intervals, which is the time difference between two grate sweeping movements. However, the side peaks have a low correlation level, which indicates that the fluctuation of temperature is not dominated by the periodic grate sweeping movement and therefore the main reason for fluctuation comes elsewhere. The same minor influence can also be seen in the time domain signal (Fig. 3), where the grate sweeping procedure is not always followed by a rise in temperature but occasionally followed even by a drop.
Fig. 3. Temperature after the first heat exchanger T2 (upper), O2 (mid) and CO (low) as function of time. Instants of grate sweeping movement are presented in every figure. It can be deducted from Ruusunen et al. (2009) that changes in draft can be seen in air flow measurement. As there were no significant correlation between the air flow measurement and the temperature signals, it can be assumed that the temperature fluctuation is not explained by draft or air flow changes. The fuel feed fluctuation cannot be measured directly. However, the periodicity of 20 seconds, which was the fuel feed period, could been seen in cross-correlation curve of the temperature and the CO signals. This effect was minor in size, but this indicates that the ON/OFF fuel feed has an existing but minor effect on combustion in this process. As the major fluctuation of combustion temperature cannot be directly connected to draft or any of the control variables, i.e. pellet feed, combustion air feed, and grate sweeping, the conclusion is that the fluctuation is mainly due to internal non measured interactions. As the local combustion intensity on the grate is not caused by manipulating variables and cannot be compensated with the actuators available, the feasible solution is not to excite the feedback control with the disturbance but to filter it away. 1 Auto-correlation (-)
The grate sweeping causes peaks in emissions since there is usually not enough secondary air to burn all the gases from the momentarily accelerated pyrolysis. Thus, this effect could further be compensated by changing the air staging around the sweeping as presented by Korpela et al. (2008), but due to constant air distribution this was not possible in this boiler.
50 40 30 20 10
0.5
0
-0.5 -500
-400
-300
-200
-100
0 100 Delay (s)
200
300
400
500
-400
-300
-200
-100
0 100 Delay (s)
200
300
400
500
0.5 Cross-correlation (-)
Fig. 2. CO and temperature responses and grate sweeping time instants with original (left) and improved (right) grate sweeping algorithm.
CO (ppm, 10 % O2)
8
0
-0.5
-1 -500
Fig 4. Auto-correlation of temperature T2 (Up). Crosscorrelation of O2 in relation to temperature measurement (Down).
7037
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
4. CONTROL DESIGN 4.1 Filtering of disturbances As it can be seen in Fig. 3, the same disturbances occur in oxygen concentration and temperature in the upper end of the combustion chamber. Hence, it was reasonable to use the same filtering with both variables. Since it was quite clear that grate sweeping causes minor disturbances enduring tens of seconds, it was concluded that the filter should have the capability of removing such low frequency disturbances. A few different filters were tested which were: a running mean filter with time window of 150 s carried out by a discrete transfer function (1), a running median filter with time window of 120 s (2), a genuine running mean filter with time windows of 120 s (3) and 150 s (4). Such filters have dynamics that decrease the performance of the control system. 4.2 Step responses and process identification Step response tests were conducted for both process inputs, the air feed and the fuel feed in order to enable process identification. First order transfer function with a time delay (FOTD) was fitted to both upwards and downwards step data of the experiments. Since it was concluded that such kind of filtering was necessary due to the disturbances, the step response data was filtered with the same filter that was used in the control experiment before fitting the transfer function to it. When the measurement filter was changed, the process model was recalculated. Process models were obtained by conventional means and by exploiting Matlab Identification Toolbox. When Identification Toolbox was used, the input and output data were pre-processed by removing means and clearly disturbed data. Process models that were obtained with each filter are mentioned in Tables 2-4 in Appendix 1. Step response test for the air feed was conducted by running the boiler on the nominal heat output level (25 kW). The frequency of the fan motor was changed from 35 to 43 Hz and then back to 35 Hz. In the air feed step response experiments it was noticed that the dynamics, especially in O2, seemed to differ depending on the direction of the step change. When fitting the FOTD model to the data, the time constant of the process model was longer in the case of upward than downward step. Time delay of the process was somewhat equal regardless of the direction of the step change. The response of temperature to an air feed increase was almost immediate, although the temperature gradually came down after the initial rise. The fast response is due to the fact that the residence time of the hot gases in the combustion chamber was shortened. The gradual fall in the temperature is due to the cooling effect of the increased air feed into the combustion chamber. Step response experiments were also conducted with the fuel feed. In the first step the fuel feed rate was decreased from 100% to 90% of the nominal heat output. The second step raised the fuel feed rate back to nominal power level. O2 level’s response to an increase in the fuel feed was in the
opposite direction. There was a significant time delay from the fuel feed change to the reaction in the oxygen level and the response was clearly slower than in the case of air feed change. Also, the time constant was significant. Yet, when comparing to the temperature during the same step, the time delay was shorter. As a result, there were substantial dynamics from the fuel feed to the temperature as time delay was c. 5 minutes and the time constant was c. 15 minutes. 4.3 Determination of input-output pairs for control The chosen variables to be controlled were oxygen level in the flue gas and temperature after first convection section (T2 in Fig. 1). The oxygen level provides information of the state of combustion. The gas analyzer measuring the oxygen concentration had a time delay of c. 23 s which causes difficulties from the control point of view. Temperature provides information about the intensity of the combustion and gives an indication of the amount of fuel that is fed. It would be beneficial to use a temperature sensor as close to the combustion chamber as possible, but to avoid disturbing effect of heat radiation from direct flame to the thermocouple, the temperature was chosen to be measured from the upper end of the combustion chamber. The oxygen level can be controlled by the air feed. This offers the most efficient and the fastest way available to affect the O2 level. Despite that, there were still substantial dynamics from the air feed to oxygen. In addition, time delay of the gas analyzer decreased the performance that could be gained by this control strategy. Also a drawback is that the air feed has a significant effect on combustion on the grate and thus disturbing the combustion process, which in turn causes fluctuations into the oxygen level. Due to this strong interaction to other variables, it is necessary that the air feed changes are to be made calmly. Hence, this excludes the control of fast disturbances, but slow drifting can still be avoided by exploiting this strategy. Thus, a feedback control based on O2 measurement and air feed was utilized. The O2 level can also be controlled by the fuel feed. Since the response of O2 to a change in fuel feed was in the opposite direction (negative process gain), the controller’s gain got also a negative sign as can be seen in Table 3. In the step response experiments it proved to be a significantly slower way to affect oxygen level than with air feed. It is thus impossible to control fast changes in the oxygen level by the fuel feed but again the drifting of the process is avoidable. The fuel feed affects also the temperatures in the boiler but comparing to the air feed, the effect is much slower and smoother, thus disturbing the process less than the changes in the air feed. Therefore, controlling fast changes in the temperature by fuel feed is not possible but the elimination of slow drifting is. Though, changing the air had almost an instant effect on the chosen temperature, controlling this temperature by air feed would, in the end, drive the process either to a point where the oxygen would run out or a point where the fuel bed would shrink until a point where the temperature would not be high
7038
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
enough to maintain the combustion. In either of these cases fire would extinguish. Due to these facts it was not reasonable to utilize feedback control of air feed to control temperature. 4.4 Tuning methods Standard PID controllers were used during the experimental runs. Tuning of the controllers was done with AMIGO method (Åström & Hägglund, 2006) and the method proposed by Åström and Murray (2009). Also pole placement method was used (e.g. Åström & Hägglund, 2006). Conventional Ziegler-Nichols method provided significantly too high controller gains. Also the controller gains given by the used tuning rules were too high for the purpose, due to internal disturbances of the process and due to the fact that also these methods are developed mainly for the servo control case. There, the role of the disturbances is usually minor. Instead, combustion control is to a large extent a regulation control case, where control is used for disturbance compensation. The process models from the identification were used with the tuning formulas. The PID parameters that were used are presented in Tables 2-4 in Appendix 1. 5. CONTROL ANALYSIS 5.1 Open loop control The open loop control statistics are illustrated in Table 1 in Appendix 1. The same process inputs can result in quite different operating areas in terms of O2 and temperature which verifies that there is a tendency of drifting in the process. This kind of behaviour is undesirable and calls for feedback control. In the oxygen a difference of one percentage unit could be seen, which will cause significantly differing combustion conditions. Also difference of c. 10 °C in the temperature could be noticed. This change itself is insignificant but it indicates the change in process behaviour. Yet, the open loop behaviour of the combustion process was rather stable after the optimization of the grate sweeping sequence which can be seen in the deviations in Table 1. In the open loop experiments, the boiler was run on the nominal power level (25 kW). On this power level, the combustion got more restless in the terms of CO production when the O 2 concentration in the flue gas reached an average concentration of 10.1%. The first indication of this can be seen in the standard deviation of CO rather than in the mean value of CO. The same restlessness cannot yet be seen in the deviations of the O2 or the temperature. 5.2 Control of oxygen by air feed The fastest way to affect oxygen concentration was by air feed but as was discussed earlier, changes in air feed disturbed the combustion process quite remarkably indicating strong coupling between the variables. Another problem with this control strategy was the long delay of the gas analyzer. As it can be seen in Table 2, the desired setpoints were always achieved and maintained. Yet, when comparing the
standard deviations of the oxygen control experiments with 100% power to the ones from the open loop control experiments, such low deviations could not be obtained with the feedback control. Thus the oxygen behaviour with this feedback control strategy was more restless which in turn causes larger deviation into other variables as well. This is to some extent due to the long time delay of the oxygen measurement and the substantial filtering of the measurement signal. Another reason for the larger deviation was that the primary and secondary air was supplied by the same fan. Thus, the undesired effects from adding more primary air when only increased amount of secondary air was wanted worsens the results severely. Compared to the case with 100% power level experiments, the feedback control made the process more restless on a lower power level (c. 76% of the nominal) as well. The PI parameter values used during this experiment were obtained by using the AMIGO PI tuning method. It provides the smallest controller gain values from the formulas but in spite of this, the gain was still slightly too high. As the controller gain was manually reduced the control got calmer and due to that the deviations started to approach the ones of the open loop system. Though the controller’s acting on the process was reduced and the lower gains were enough to prevent the process from drifting. A reason for higher emissions during the lower power level run was the fact that the 10.5% setpoint was too low for this power region. Due to limitations set by the measurement equipment (long gas analyzer delay) and the equipment of the boiler (one fan), the purpose of the feedback control was only to prevent the process from drifting rather than active disturbance control. 5.3 Control of oxygen by fuel feed The response of the oxygen measurement to changes in fuel feed was quite slow and calm. Thus, feedback control with this input leads to slow and calm behaviour of the controlled system which suits well to drifting prevention. During the experiment, the setpoint was maintained well and the deviation of oxygen level was even smaller than in the case of the open loop system as can be seen when comparing the results in Tables 1 and 3. The calm control can also be seen in the small deviation in the controller’s output. Since the air feed during this strategy is kept constant (if constant power is desired), the behaviour of the combustion is very stable and it approaches the behaviour of open loop system, which can be seen in the standard deviations of CO and temperature. When comparing this control strategy to the oxygen control by air feed on the same power level, the standard deviation of the oxygen level was smaller and the standard deviation of the temperature was significantly smaller with this control strategy. Also the CO deviation was slightly smaller. Due to the slow effect of this input to the oxygen level, the time delay of the gas analyzer was insignificant. Therefore this control strategy worked better than the control with air feed.
7039
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
5.4 Control of temperature by fuel feed Control of the temperature by the fuel feed resulted in quite stable combustion behaviour given by slow dynamics from the fuel feed and constant air feed. During the experiments, the temperature setpoint was always maintained, as can be seen in Table 4. The deviations of temperature, oxygen and CO were also about the same size as in the open loop system. The CO emissions were quite low during the experiments due to optimal oxygen level. The variation in the controller’s output in the last two experiments in Table 4 was quite small indicating slow overall system behaviour. Since the fan’s input was kept constant during the experiments, the oxygen level changes as a function of the added fuel. In order to maintain a reasonable oxygen level, a corresponding setpoint for the temperature must be found (Korpela et. al., 2008). By increasing the temperature setpoint, the oxygen level goes down and vice versa. This control strategy only calls for very simple instrumentation since there is only one thermocouple needed. Since the oxygen level correlates with the temperature (Fig. 4), it is therefore possible to control the oxygen level indirectly by setting the setpoint properly. The problem of finding correct temperature setpoints for every power level can be quite tricky since the oxygen level is a function of air feed and fuel feed and the dependences in between are often nonlinear. The selection of a suitable temperature setpoint is discussed in Korpela et al. (2009b). When comparing the last two experiments in Table 4, there is a power raise conducted by raising the temperature setpoint and the input for the air fan. The oxygen level in both of the experiments remained the same. 6. DISCUSSION As seen in the process analysis, the autocorrelation of the combustion temperature showed a minor periodicity at an interval of 120 seconds (Fig. 4). The obvious reason is the grate sweeping at the same interval which indicates a remaining effect on combustion despite the implementation of the optimized grate sweeping sequence. The low relative values of correlation at the interval peaks tell that the influence is minor compared to other fluctuations of the combustion temperature. The same minor influence could also be seen in the time domain signal (Fig. 3). As the major fluctuation of combustion temperature could not be connected to draft or any of the control variables, i.e. pellet feed, combustion air feed and grate sweeping, the conclusion is that the fluctuation is mainly due to internal non modelled interactions. It is not excluded that these actions do have an excitation in one or some of the control variables but in this case the influence has a stochastic nature which cannot be seen in correlation analysis. Observing the fluctuations, an idea has appeared that there is an influence of a relatively large combustion chamber and a relatively low primary air supply to the grate area. This might imply that the combustion intensity on the grate is quite low and therefore the variation in pellet and primary air flow distributions over the grate become significant. Thus, the
amount of stored chemical energy on the grate can vary. At this situation it is likely that the local combustion intensity becomes self amplifying and the cumulative result is that the combustion temperature fluctuates in periods of tens of seconds which from a control point of view can be seen as an internal disturbance. Even though the disturbance has a stochastic nature it is most probable that the seeds for the disturbance come from differences in rest positions of the grate after sweeping and uneven pellet feed. However, there is no supporting evidence of the size influence of the combustion chamber. As the local combustion intensity on the grate cannot be compensated with the actuators available, the feasible solution is not to excite the feedback control with the disturbance but to filter it away as was done in the empirical part of the work. The result is a feedback control, which stabilizes the combustion on long term which compared to common environmental requirements gave good results. A natural way to improve the situation would be to change the burner design to decrease the amount of pellets under combustion and to enable possibility of controlling the air distribution towards increase of local primary air, which requires re-designing of the combustion air system. As a result, there would be less room for self amplifying combustion. Additionally to a more even combustion at open loop control, the solution could give more possibilities for feedback control on short term as air feed is one of the actuators, but this was not tested in the experimental work. On the other hand, fewer pellets under combustion could make the process more vulnerable to unequal pellet feed, which would be seen as fluctuation in all main combustion values. A central challenge would then be to even the pellet feed, which is a common issue in most of the biomass combustion processes. The fundamental task of the control system presented in this paper is the drifting compensation. However, further improvement of the system performance could be gained if there was a fast sensor for monitoring quality of combustion (e.g. lambda probe or new type of CO measurement) and secondary air feed was conducted by a separate fan. Then the improvement potential of the short term performance of the process by feedback control would be increased. In the case of sophisticated instrumentation also the benefit for advanced control, e.g. feed forward compensation of pellet feed and cascade, multiple input – multiple output (MIMO), neural networks and fuzzy control, is brought up. These kinds of control methods are utilized in some high-end products on the market but the performance and detailed operation of these systems are not published. Furthermore, application of such kinds of advanced approaches is more specific for large-scale boilers. It is still questionable, are the benefits that could be gained by adding such instrumentation enough in small-scale combustion where the behaviour is already quite steady or should the efforts be put into stabilizing process design, as was done with the optimization of the grate sweeping sequence. Generally, the benefit from the advanced control can also be gained in extended lifetime of the boiler and its more efficient operation, e.g. fuel saving.
7040
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
7. CONCLUSIONS
REFERENCES
This research shows that due to an already stable process the improvement potential of the short term behaviour of combustion with feedback control is limited. However, the utilization of PID control assures that a boiler is always run in a suitable operating area despite that combustion processes have a tendency to drift in the long term due to different variations in the process. With feedback control, the drifting can be prevented and thus efficiency can be maintained high and emissions low.
Bauer, R. Gölles, M., Brunner, T., Dourdoumas, N., Obernberger, I. (2010) Modelling of grate combustion in a medium scale biomass furnace for control purposes. Biomass and Bioenergy, 34, 2010. pp. 417 – 427. Good, J., Nussbaumer, Th. 1998. Efficiency improvement and emission reduction by advanced combustion control technique (ACCT) with CO/lambda control and setpoint optimization, Biomass for energy and industry, 10th European conference and technology exhibition, June 8– 11, 1998, Würzburg, Germany, pp. 1362 – 1365. Korpela, T., Björkqvist, T. & Lautala, P. (2008). Durable feedback control system for small scale wood chip combustion. Proceedings of World Bioenergy 2008, Conference & Exhibition on Biomass for Energy, 27–29 May 2008, Jönköping, Sweden. pp. 224–230. Korpela, T., Björkqvist, T., Lautala, P. (2009a). Control strategy for small-scale wood chip combustion. IFAC Symposium on Power Plants and Power Systems Control, 5–8 July, 2009, Tampere, Finland. Korpela, T., Ruusunen, M., Björkqvist, T. & Lautala, P. (2009b). Control structures of a multivariable process applied to small scale biomass combustion. Automaatio XVIII Seminaari 17–18.3.2009, Helsinki. Finland. SAS julkaisusarja 36. Suomen Automaatioseura. 8 s. Lackner, M., Winter, F., Agarwal A. K. editors 2010. Handbook of Combustion, Wiley-VCH Verlag. Vol. 4., pp. 141 – 169. Liao, Z., Dexter, A. (2004) The potential for energy saving in heating systems through improving boiler controls. Energy and Buildings. 36, 2004. pp. 261–271 Morán, J., Granada, E., Míguez, J., Porteiro, J. (2006) Use of grey relational analysis to assess and optimize small biomass boilers. Fuel Processing Technology, 87, 2006, pp. 123– 127. Padinger, R. (2001). Regelungstechnik für die Hausheizung der Zukunft, Project report, Joanneum Research, Austria, IEF-b-12/01 Ruusunen, M., Korpela T., Björkqvist, T. 2009. The effect of control parameters to the quality of small-scale wood pellet combustion. In: Savolainen, M. (ed.). Bioenergy 2009, Sustainable Bioenergy Business, 4th International Bioenergy Conference, Book of Proceedings, 31.8.– 4.9.2009, Jyväskylä, Finland. FINBIO publication 45. pp. 507–515. Šulc, B., Vrána, S., Hrdlička, J., Lepold, M. (2009). Control for ecological improvement of small biomass boilers, IFAC Symposium Power Plants & Power Systems, 5–8 July, 2009, Tampere, Finland. Åström, K., Hägglund, T. (2006). Advanced PID Control. ISA - Instrumentation, Systems, and Automation Society, Research Triangle Park, NC 27709. 461 p. Åström, K., Murray, R. (2008). Feedback Systems – An Introduction for Scientists and Engineers. Princeton University Press. Electronic edition, Version v2.10b. 396 p.
The utilization of continuous control by PID controllers and/or more sophisticated control methods requires utilization of PLC (programmable logic controller). This will increase the total costs of the boiler slightly but the obtainable benefits are significant. In addition to drifting prevention, the possibility of continuous power modulation, search for optimal controller parameters from the efficiency and emission point of view, adaptability for fuel quality changes and suppression of short term fluctuations are benefits that continuous control can provide. In general, short term performance of the process defines the safety margins needed around the setpoint. The benefit that is gained from a good short term performance of the process is that it enables adjustment of the setpoint of the control system closer to a limit value but still avoid crossing it. In this case, better short term performance, i.e. smaller fluctuation of oxygen level, allows adjustment of the oxygen setpoint closer to the limit value, where CO starts to grow due to lack of air, without crossing it. Running the process with lower oxygen would improve the efficiency. Thus, the short term performance can be said to be directly linked to efficiency of the boiler. However, the stochastic and nonlinear nature of the combustion process makes the process identification quite hard since the dynamics of the process can vary significantly depending on the operating area. This complicates utilization of classical tuning methods since they are developed for deterministic processes. Yet, the tuning methods provide reasonable rough estimates of the controller parameters. Hence, the PID controller gains given by standard tuning rules were somewhat too high for the purpose due to internal disturbances of the process. When the controller gains were manually brought down, the short term performance of the controlled process started to approach the short term performance of the open loop system. Yet, the main purpose of the applied control was to prevent the combustion from drifting and this goal was achieved. ACKNOWLEDGEMENTS This cooperative work was supported by Fortum Foundation and the Ministry of Education of the Czech Republic within the project No. MSM6840770035 “Development of environmentally friendly decentralized power systems”, which are gratefully acknowledged.
7041
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
APPENDIX 1
Table 1. Open loop system statistics. Test 1 2 3 4
O2 mean±std 10.6±0.6 10.9±0.6 10.1±0.6 11.1±0.6
CO mean± std 28.3±5.7 33.5±5.2 30.8±9.5 27.0±6.8
Tex mean± std 518.6±8.3 519.6±8.1 533.4±7.4 528.0±6.5
Power (% of nom.) 100 100 100 100
Fan (Hz) 35 35 35 35
Table 2: O2 control by air feed. In the filter column: 1 = running mean filter with 150s time window (discrete transfer function), 2 = running median filter with 120s time window, 3 = running mean filter with 120s time window, 4 = running mean filter with 150s time window. O2 setpoint = 10.5% in every experiment. u stands for fan input in Hz. PID parameters: K/Ti/Td
contr. mean±std
CO mean± std
u mean± std
Tex mean± std
Power (% of nom.)
2.4 / 125 / 0
10.5±0.7
31.4±11.7
34.5±0.7
535.2±9.4
100
5.5 / 100 / 18
10.5±0.6
32.7±12.6
33.4±1.1
516.5±12.5
100
Å&M PI
3.1 / 174 / 0
10.5±0.6
31.5±6.8
35.3±1.4
532.8±15.3
100
AMIGO PI
5.6 / 327 / 0
10.4±0.6
32.5±9.3
34.9±1.9
500.2±21.2
91
AMIGO PI
2.4 / 201 / 0
10.5±0.7
27.8±8.7
35.2±0.8
508.7±10.1
91
0.388/484/80
AMIGO PI
4.0 / 356 /0
10.4±0.7
49.9±30.1
31.4±1.7
441.2±14.4
76.3
-
-
0.75
3.0 / 356 / 0
10.5±0.7
56.3±16.3
32.1±1.1
455.5±9.8
76.3
-
-
2.0 / 356 / 0
10.5±0.6
50.0±12.8
32.3±0.7
457.8±7.4
76.3
-
-
2.7 / 484 / 0
10.4±0.6
47.8±13.5
32.2±0.8
456.8±7.5
76.3
4
0.388/484/80
0.5 Pole placement 50% AMIGO PI
2.0 / 356 / 0
10.6±0.7
38.3±7.8
34.3±0.5
485.3±6.7
87
Filter
FOTD model: K/Tp/Td
Tuning method
1
0.338/150/41
1
0.338/150/41
AMIGO PI AMIGO PID
2
0.325/216/87
3
0.325/466/67
3
0.325/240/67
4
Table 3: Oxygen control by fuel feed. In the filter column: 4 = running mean filter with 150s time window. Setpoint for the O2 was 10.5 %. u stands for the total fuel feed period (s) which is the sum of constant T on = 3s and controlled Toff. Filter
FOTD model: K/Tp/Td
Tuning method
PID parameters: K/Ti/Td
contr. Mean± std
CO mean± std
u mean± std
Tex mean±std
Fan (Hz)
4
-89.5/923/187
AMIGO PI
-0.00133 /719/0
10.5±0.5
32.5±6.8
21.7±0.7
513.5±6.0
36
Table 4: Temperature control by fuel feed. In the filter column: 2 = running median filter with 120s time window, 3 = running mean filter with 120s time window, 4 = running mean filter with 150s time window. u stands for the total fuel feed period (s) which is the sum of constant Ton = 3s and controlled Toff.
Filter 2 3 4
FOTD model: K/Tp/Td 1489/94 9/303 1489/91 6/305 1489/91 6/313
Tuning metho d Å&M AMIG O PI AMIG O PI
PID parameters K/Ti/Td 8.37*10-4/ 722.5/0 4.29*10-4 /703/0 4.16*10-4 /797/0
Set point
contr. Mean± std
CO mean± std
u mean± std
O2 mean± std
Fan (Hz)
Power (% of nom.)
480
482.2±9.7
28.2±9.5
25.8±1.4
12.5±0.7
35
77.4
500
497.8±6.7
27.2±6.8
24.0±0.5
11.2±0.6
35
83.3
510
511.5±7.9
33.8±5.2
22.5±0.4
11.3±0.5
38
88.9
7042