This action might not be possible to undo. Are you sure you want to continue?

# FAULT DETECTION IN COAL MILLS USED IN POWER PLANTS Peter Fogh Odgaard ∗ Babak Mataji ∗∗

∗

Department of Control Engineering, Aalborg University, Aalborg, Denmark, odgaard@control.aau.dk ∗∗ Elsam Engineering A/S, Kraftværksvej 53, DK-7000 Fredericia, Denmark, bama@elsam-eng.com

Abstract In order to achieve high performance and eﬃciency of coal-ﬁred power plants, it is highly important to control the coal ﬂow into the boiler in the power plant. This means suppression of disturbances and force the coal mill to deliver the required coal ﬂow, as well as monitor the coal mill in order to detect faults in the coal mill when they emerge. This paper deals with the second objective. Based on a simple dynamic model of the energy balance a residual is formed for the coal mill. An optimal unknown input observer is designed to estimate this residual. The estimated residual is following tested by measured data of a fault in a coal mill, it can hereby be concluded that this residual is very useful for detecting faults in the coal mill. Keywords: Fault Detection, Coal-ﬁred Power Plants, Optimal Unknown Input Observer

1. INTRODUCTION The performance of a coal-ﬁred power plant is highly dependent on the supplied coal ﬂow into the boiler. In order to achieve good performance of the power plant, this coal ﬂow should be known and controllable. In other words the performance requirements of the power plant can be transfered to the coal mill which delivers the pulverized coal to the burners inside the power plant boiler. One thing is to guarantee the required coal ﬂow, in terms of suppressing plant disturbances. It is, however, another thing to monitor the coal mill for detecting faults. In order to avoid additional failures and decreased performance in the power plant unit, it is important to detect any fault as soon as possible, and hopefully as the fault emerges. In most coal mills it is not possible to measure the pulverized coal ﬂow out of the mill. The coal ﬂow into the mill is only indirectly mea-

surable. Implying that the early fault detection based on detection of drops in the coal ﬂow of the coal mill, is not so simple as it seems. A critical example on a fault in the coal mill is caused by a blocking in the raw coal inlet pipe, a coal mill is illustrated in Fig. 1. This blocking of the raw coal inlet will with time result in a stopped coal ﬂow from the coal mill to the boiler, and since the actual coal ﬂow into the boiler cannot be measured, the fault in the coal mill cannot be directly detected. Dynamic modeling and nominal control of these coal mills have been the topic of numerous of publications. Some examples dealing with modeling of coal mills are (Rees and Fan 2003), (Zhang et al. 2002) and (Tigges et al. 1998). Controllers for the coal mill are designed in (Rees and Fan 2003) and (Hasselbacher et al. 1992). High order dynamic models and observer design for coal mills

Such an optimal unknown input observer is as consequence designed for estimating the residual/fault signal and the temperature. E. A coal mill is a harsh environment in which it is diﬃcult to perform measurements. The primary air is mixed by cold outside air and heated outside air. as well as the temperature at the classiﬁer. If this extra energy input could be estimated. This state is driven by an unknown input which is ﬁltered through the fault model. The temperature controller is often required to keep temperature constant at 100◦ C in order to evaporate the moisture content in the coal. it would be a useful residual for monitoring the coal mill. An illustration of the coal mill. which is heated . However. This fault detection method is subsequently applied to data sampled during a fault in a coal mill.g. The temperature of the primary air is used to control the temperature in the coal mill at the classiﬁer. 2004). The coal mill is illustrated in principles in Fig. THE COAL MILL The work presented in this paper. A threshold method is following applied to the residual with the purpose of detecting the fault in coal mill. with the intension of detecting eventual faults in the coal mill. since it can be used to separate the possible faults into two classes which help in the fault isolation. since a control loop is formed to keep the measured temperature at 100◦ C. A fault in the coal mill is in this setting an extra energy input to this model. However. The temperature of coal dust ﬂow after the classiﬁer is a good indicator of the “health” of the coal mill. and the temperature as output. the actual coal ﬂows in and out of the coal mill are not measurable. Coal particles which in the pulverizing have been small enough will pass through the classiﬁer and out through the outlet pipes into the boiler. The pulverized coal is then blown up and the moisture content is evaporated by the hot primary air. Meaning that the fault can be seen by combining the temperature and the energy in the primary air. the primary air ﬂow and temperature are. the interest has not previously been addressed on the monitoring of the coal mills with the purpose of detecting any emerging faults. as well as rotational speed of the classiﬁer.are the topics in (Fukayama et al. this means that all the variables are not measurable. is based on a Babcock MPS 212 coal mill used at Elsam’s Nordjyllandsværktet Unit 3.1 Control and measurements References to coal ﬂow and primary air ﬂow are given by the general power plant controller. This described estimation problem can be solved by the use of an optimal unknown input observer. it might be useful to monitor an energy balance state space model of the coal mill. The coal is fed to the coal mill through the central inlet pipe. see (Chen and Patton 1999). 2. the measured temperature cannot be used to detect faults directly. However. I. 2. This residual can be estimated by introducing a state representing the fault in the energy model. which has the energy ﬂows in and out of the coal mill as inputs. it is as well interesting to investigate the sign of this energy balance. However. The coal is pulverized on the rotating grinding table by the rollers.e. the method proposed in the paper is so generic that it can be applied to other types of coal mills. by the furnace. Pulverized coal Rotating Classifier 11 00 11 00 1 0 11 00 1 0 111 000 1 0 111 000 1 0 111 000 1 0 111 000 1 0 1 0 1 0 111 000 11 00 111 000 11 00 11 00 111 000 1 0 111 000 1 0 11 00 1 0 11 00 1 0 1 0 11 00 1 0 11 00 11 00 11 00 11 00 11 00 1 0 1 0 1 0 11 00 11 00 Roller 11 00 1 1 0 0 1 1 0 0 1 1 0 0 Grinding table Raw coal Inlet Roller Primary Air Figure 1. 1. since a non-intendant change in the in and out ﬂows of the coal mill will result in a variation in the temperature. The ratio of these air ﬂows are used to control the temperature of the primary air ﬂow. This experiment shows the potential of the proposed residual estimated based on the energy balance model of the coal mill. in order to evaporate the moisture in the coal.

faults in primary air supply both the fan and the temperature controller. This combined heat coeﬃcient does not deal with the fact that the speciﬁc heat of water and steam . it is assumed in this paper for simplifying the model. is following deﬁned as Hst = Cw + Lsteam /100.2 Faults A number of diﬀerent faults can occur in the coal mill. are diﬀerent. Qair is the energy in the primary air ﬂow . and ﬁnally all the sensors. Pmotor denotes the power delivered by the grinding table. Ts is the surrounding temperature. It is also assumed that the input coal ﬂow is equal the output coal ﬂow. where Cw is the speciﬁc heat of the water. ˙ mm · Cm · T (t) = Qair (t) − Qcoal (t) − Qmoisture (t) + Pmotor (t). TPA (t) is the temperature of the inlet primary air. T (t) is the mill temperature at the classiﬁer. it is diﬃcult to validate the details in the response due to the way the signals are sampled. The dynamic non-linear model is subsequently given by ˙ mm Cm T (t) =mpa (t)Cair (TPA (t) − T (t)) ˙ + mc (t)Cc · (Ts − T (t)) ˙ + γ(t)mc (t)Cw · Ts ˙ − γ(t)mc (t)Hst · T (t) ˙ + Pmotor (t). mc (t) is the ˙ coal mass ﬂow. The model response is compared with measurements as well as the response of a linearized model in Fig. Pmotor denotes the power delivered by the roller motors. The non-linear model (2) is subsequently linearized and transformed into a state space representation. mpa (t) ˙ is the primary air mass ﬂow in and out of the mill. and Qmoisture is the energy in the coal moisture.104 non−linear model measurements linear model 102 Qair Pmotor T mm Qcoal Qmoisture degrees C 100 98 Figure 2. and Lsteam is the latent heat. γ(t) is the ratio of moisture in the coal. A more detailed model which takes diﬀerent coal ﬂows into account might result in better results with respect to even earlier detection of the faults. Even though this assumption is only entirely true for steady state. An illustration of energy balance in the coal mill. A dead band on one per cent is applied to these measurements meaning that the signals shall have changes of a given size before this change is sampled. in which T is the temperature in the mill. ENERGY BALANCE MODEL OF THE COAL MILL A simple energy balance model of the coal mill is derived based on (Rees and Fan 2003). Cair is the speciﬁc heat of air. see (3). A plot of the non-linear and linear model response compared with measurements of a step response on the coal mill. 3.5 2 2. as illustrated in Fig. and if the fault leads to a decrease in the output coal ﬂow from the coal mill. 2. and Pmotor (t) is the power delivered by the mill motor. the motor power is also neglected from this state space model since it is The heating and evaporation of the moisture in the coal is modeled by a combined heating coeﬃcient. In this model the coal mill is seen as one body with the mass mm . From this ﬁgure it can be seen that the responses of both models are quit similar to the large dynamical changes as the measurements show. However.5 x 10 4 Figure 3. The temperature is due to the control loop is kept at 100◦ C. (1) where: Cm is the speciﬁc heat of the mill. Cc is the speciﬁc heat of the coal. where T is the temperature in the mill. Cw is the speciﬁc heat of the moisture. All these are faults which it are urgent to detect. etc. The energy balance is given by (1). Qcoal is the energy in the coal ﬂow. Some examples on critical faults are: choking of the raw coal inlet. Hst parameter combining the latent heat of the steam and speciﬁc heat of the water. All parameters in this model are found in data books except mm · Cm which is identiﬁed based on measurements of a step response on the coal mill. 96 94 92 90 0 0. The latent energy of the evaporation dominates the energy required for a few degrees heating of the moisture. (2) 3. The combined heat coeﬃcient. however the model error due to heat of steam to a couple of degrees above 100◦ C is neglectable in this context. Hst . this can lead to drop out of the entire power plant unit. 2. Qair is the energy in the primary air ﬂow.5 1 samples [n] 1. and Qmoisture is the energy in the coal moisture. Qcoal is the energy in the coal ﬂow.

Bq . Tm [n] = Cd ¯ Q[n] (15) where Ad . The model represented by (8-13) is discritizied before an observer is designed to estimate the states in the model. i. 0 −p Bq = B1×(1···3) . 1 . an optimal unknown input observer would be an obvious observer to use for estimating the residual. Ed are the discritizied representations of Aq . and hereby the residual. see (14-15). ¯ mPA [n] ˙ ¯ ¯ T [n] T [n ˙+ 1] ¯ + Bd TPA [n] m = Ad ¯ ¯ Q[n] Q[n + 1] (14) ¯ m [n] ˙ . ¯ mPA (t) ˙ ¯ TPA (t) ˙ ¯ ¯ (3) T (t) = AT (t) + B · ¯ mc (t) + q(t). Qn (t) is the generic unknown input which is low-pass ﬁltered in order to represent the residual. (4) ¯ where a given signal ◦ is linearized by ◦ = ◦ − ◦o .e. is that the estimator gain is recomputed at each sample meaning that the model can be changed such that the point of operations can be updated. Tm (t) = Cq ¯ Q(t) (9) (7) to balance the model). r(t) is the normal distributed measurement noises.o ˙ mm · Cm B = C · (T − T ) + γ · (C · T − H · T ) s o o w s st o c mm · Cm mc. By combining these two model inputs to one unknown input. y[n] = Cn x[n] + r[n]. For discrete time systems with unknown inputs and disturbances which can be represented by x[n + 1] = An x[n] + Bn u[n] + En d[n] + q[n]. ¯ T [n] ¯ + r[n]. The schemes for computing the matrices in the optimal unknown input observer can be seen in Appendix A.o · (Cw · Ts − Hst · To ) ˙ mm · Cm (6) C = I. ˙ γ (t) ¯ ¯ ¯ Tm (t) = CT (t) + r(t).o · (Cc + γo · Hst )) ˙ ˙ A= . an estimate of this unknown signal can be used as a residual for detection faults in the coal mill. q(t) is the normal distributed process disturbances. which represents both the fault energy and coal moisture. Cq .much smaller than the other powers in the equation. and subsequently design a Kalman estimator for the transformed system. The unbalance in the model can be represented by an additional fault input. 4.o − To ) mm · Cm Cair · mPA. ˆ x[n + 1] = z[n + 1] + Hn+1 y[n + 1]. This model is a system with an unknown input. Eq = 1 (12) (13) where B1×(1···3) denotes the three ﬁrst elements of B. . in order to estimate the need energy/fault signal it is more convenient to represent it by an internal state in the model. Cd . mm · Cm (5) T Cair · (TPA. ¯ T (t) ¯ + r(t).o · Cair − mc. (16) (17) an optimal unknown input observer of the following form can be derived z[n + 1] = Fn+1 z[n] + Tn+1 Bn u[n] + Kn+1 y[n]. Bd . c + Ed Qn [n] + q[n]. The simplest representation is a ﬁrst order lowpass ﬁlter which is included in the model. and 1 A (10) Aq = p .1 Optimal unknown input observer The optimal unknown input observer is described in (Chen and Patton 1999). 0 (11) Cq = C 0 . 4. However. It is in addition a possible measures of the moisture in the coal which is not valid during dynamic behavior of the coal mill. Eq . ¯ mPA (t) ˙ ˙ ¯(t) ¯ (t) T T ¯ ¯ ¯ m = Aq Q(t) + Bq TPA (t) Q(t) (8) ¯ mc (t) ˙ + Eq Qn (t) + q(t). ◦o is the operation point of ◦. A positive side eﬀect of this. It can be assumed that the moisture content is changing much slower than a fault can occur or emerge. Tm (t) is the measured temperature and (−mPA. RESIDUAL GENERATION The energy balance given by (2) and in a linear state space version in (3-7) would in case of a fault deviate from the coal mill which indicates an unbalance in the model. This results in the linear model given by (8-13). disturbances and measurement noises. Q(t) is the estimated residual (need energy ﬂow The basic idea in this observer is to eliminate the dependency of the unknown input from the estimation error by matrix transforms. (18) (19) where p is the pole of internal residual model.

in the way that the observer estimates the energy signal in the fault free situation well. for detecting faults in the mill. a detection based on method 2 is more certain to actually be a detection of a fault. i. in terms of variations in the coal ﬂow. 6. can for this given fault be seen in Fig. fd [n] = mc [n] · Cc · (Ts − T [n]) ˙ 0 else were. before the threshold is applied. A plot of the signal used in fault detecˆ tion method 1. and that method 2 did detect the fault 1 sample later than the visual inspection. 7. such that it gives a detection of the beginning of the fault as early as possible. 5. The ﬁrst method is to use a simple threshold β.5 energy 5000 1 0. From this ﬁgure it can be seen that the observer estimates the fault signal due to the coal moisture content well. A state is introduced representing the fault. In the following fd [n] denotes a signal which is equal 1 in case of a fault and 0 elsewhere. see (21).5 0 0 −5000 0 200 400 600 Samples [n] 800 1000 1200 −0. This means that a fault is deˆ tected based on this rule if Q[n] is larger than the threshold β. It is a fault where the coal inlet pipe has been stopped by the raw coal.e.5 2 10000 energy 1. as well as the internal fault model parameter p are all found by trial and error. EXPERIMENTS The introduced fault detection methods based on the estimated residual can detect a number of diﬀerent faults in the coal mill. CONCLUSION A simple energy balance model is formed of a coal mill with the purpose of monitoring the coal mill. see Fig. The sample frequency in this experiment is 60s. 1116. mc [n]·CQ[n]s −T [n]) . fd [n] = (20) 0 else were.2 Detection rules In this section two methods for detecting the faults based on the residual are described. and not due to a variation in the coal ﬂow or in the moisture content in the coal.3 Fault isolation Earlier in this paper it was mentioned that the sign of the energy balance can be used to separate the possible faults in two groups. and it is hereby much more certain that a given detection is due to a fault in the coal mill and not a disturbance e. Comparison of the estimated energy due to the coal moisture and the static calculated of this energy. The variance of the disturbance and measurement noises Q[n] and R[n]. This ˙ c ·(T normalization has the eﬀect that variations due to process variations is out compensated. In practice only data of one given fault type is available.20000 observed calculated 3 x 10 4 observed 15000 2. 4. and ˆ compare the absolute value of Q[n] with a threshold β as in (20). The detection signals of both detection methods are illustrated in Fig. The other suggested method is to normalize the residual by the energy ˆ ﬂow due to the coal ﬂow. 4. however. The monitoring problem can be viewed as a system with an The diﬀerences between these two methods are that the ﬁrst method reacts faster on a variation ˆ in Q[n]. Another method partly removes the inﬂuence by ˆ coal moisture from Q[n] by dividing it with the energy ﬂow contribution from the coal. For both methods the threshold β is found. . 5. (21) Figure 5. and it is hereby concluded that the observer and model is well tuned. ˆ Q[n] 1 if > β.g. ˆ The estimated residual Q[n]. A stopped coal ﬂow will result in a positive energy balance. from where it can be seen that method 1 detects the fault earlier than the visual inspection did. By visual inspection of the data from the coal mill the beginning of this fault can be detected at sample 1115. at sample 1110. ˆ 1 if Q[n] > β. with the constraint that false detection of faults are avoided. This means that the sign of the energy unbalance can be used for fault isolation if it is combined with additional information. Q[n]. 4.5 1100 1150 1200 1250 1300 Samples [n] 1350 1400 1450 Figure 4. based on experimental data.

G. A plot of the signal in fault detection ˆ method 2. 0 −1 −2 energy −3 −4 −5 −6 1100 1150 1200 1250 1300 Samples [n] 1350 1400 1450 Figure 6.6 fd[n] 0. grant no2002-603/4001-93. Y. An adaptive state esitmator for pulverizer control using moments of particle size distribution. Walzensch¨sselm¨hlen als kompou u nenten moderner feuerungstechnik. This method. Q.. Thresholds are chosen such that they do not result in false detection and at the same time detect the fault as early as possible. Zhang. 148–157. K. + (2) Compute Hn+1 = En (Cn+1 En ) ... Oluwande. K. Ed. (7) . (3) Compute −1 T T 1 . VGB KraftwerksTechnik 78. Zhou (2002). Using the estimated residual results in a detection of fault 5 samples (5 minutes) earlier than a detection based on visual inspection. ﬁrst ed. Lang and G.D. Lausterer (1992). unknown input and disturbances and measurement noises. The method is following applied to a fault example where the coal inlet pipe is stopped by the coal. Fn+1 = An − Hn+1 Cn+1 An − K1 Cn . and Kn+1 = Kn+1 + 2 Kn+1 .8 0. Now compute z[n + 1] = Fn+1 z[n] + Tn+1 Bn u[n] + Kn+1 y[n] ˆ and x[n + 1] = z[n + 1] + Hn+1 y[n + 1].4 0.. 797–811. Matts and X. J. ′ T . K. OPTIMAL UNKNOWN INPUT OBSERVER A necessary and suﬃcient condition for the existence of a solution to the given observer problem is in (Chen and Patton 1999) given as: An optimal unknown input observer solution exists if and only if: rank (Cn+1 En ) = rank (En ). ˙ c ·(T Method 1 Method 2 1 0. Shimohira and H. T (5) (6) 7.1 observed REFERENCES Chen.). Y.. H0 = 0.2 1100 1105 1110 1115 Samples [n] 1120 1125 1130 Appendix A. however. Tigges. 77–88. Fan (2003).Q. An optimal unknown input observer is subsequently designed. Kluwer academic publishers. Rees. ACKNOWLEDGMENT The authors acknowledge the Danish Ministry of Science Technology and Innovation. R.2 0 −0. Bischoﬀ and T. IEEE Transaction on Control System Technology 12.H. results in detection of the beginning of the fault 1 samples (1 minute) later than the visual inspection did. Compute Pn+1 = Pn −K1 Cn Pn A1 n+1 n+1 and following ′ T +Tn+1 Qn TT + Pn+1 = A1 Pn+1 A1 n+1 n+1 n+1 T Hn+1 Rn+1 Hn+1 . Wang. Modelling and control of pulverised fuel coal mills. X. n+1 1 K2 n+1 = Fn+1 Hn . M. In addition a normalized residual is used for the detection as well. IEEE Transactions on Energy Conversion 17. mc [n]·CQ[n]s −T [n]) . Flynn. Patton (1999). Institution of Electrical Engineers. A plot of the fault detection based on the two proposed methods. Robust modelbased fault diagnosis for dynamic systems. The proposed observer based method has showed to be a promising method for detecting faults in coal mills. Jie and R. Hirasawa. Automatisierungstechnik 40. z0 = x0 − + C0 E0 (C0 E0 ) y0 . Kanemoto (2004). ﬁrst ed. Hasselbacher. ′ Figure 7. for support to the research program CMBC (Center for Model Based Control). and F.W. In: Thermal power plant simulation and control (D. Set n = n + 1 and jump to step 2. Steinhage (1998). K1 n+1 = An+1 Pn Cn Cn Pn Cn + Rn (4) and Pn+1 = Pn − K1 Cn Pn A1 n+1 n+1 Compute Tn+1 = I − Hn+1 Cn+1 . Wu. 549–555. G. W. N. D. J. The computation of the matrices in the observer is also given in (Chen and Patton 1999) as: (1) Set initial values: P0 = P (0). Fukayama. Reglerentwurf f¨r ein kohlenm¨hlenu u modell unter ber¨cksichtingung der veru faharenstechnischen randbedingungen. Coal mill modeling by machine learning based on onsite measurements.