You are on page 1of 6

Paper accepted for presentation at 2003 IEEE Bologna Power Tech Conference, June 23th-26th, Bologna, Italy

An algorithm for detecting CT saturation using the secondary current third-difference function
Yong-Cheol Kang, Sang-Hee Kang and Peter Crossley, Members, IEEE
evaluates the mean of the error and the mean and variance of the current amplitude [3]. The error is derived on the assumption that if the current is a perfect sinusoid, the summation of the current and its second-order derivative should be zero. When a CT is saturated, the secondary current distorts and contains points of inflection at the start and end of each saturation period. The third-difference function converts these points into pulses that can be used to detect saturation. This paper describes an algorithm for detecting CT saturation based on third-difference function of the secondary current. It also describes the effect of a low-pass filter and remanent flux in the core on the algorithm. The algorithm was implemented on a prototype system based on a digital signal processor. The system was experimentally tested in the laboratory using signals from a small CT with a turns ratio of 100:5. II. A CT SATURATION DETECTION ALGORITHM Fig 1 shows a simplified equivalent circuit of a CT. The primary current i1(t) is:-

Abstract When a current transformer (CT) saturates, the wave-shape of the secondary current distorts and contains points of inflection at the start and end of each saturation period. These can be observed as discontinuities in the first-difference function. The second- and third-difference functions convert the points of inflection into pulses that can be easily used to detect saturation. An algorithm for detecting CT saturation based on the thirddifference function is described in this paper. An anti-aliasing low-pass filter is required to prevent aliasing and reduce the effect of noise. Inadvertently, the filter softens or smooths the current signal and reduces the peaks in the third-difference function. The performance of the algorithm was validated by changing the remanent flux in the core, considering the effect of a low-pass filter and also experimentally using a test system with a real CT. Test results indicate that the detector can correctly detect each saturation period when the secondary current is severely distorted. Index Terms-- CT Saturation, Points of Inflection, Difference Functions, Remanent Flux

I. INTRODUCTION During a fault the secondary current signals may be distorted by current transformer (CT) saturation, which could result in the mal-operation of a protection relay. Thus, protection relays are designed to cope with the effect of partial saturation, but the operating performance is a compromise between dependability and security, and an appropriately designed CT must be used. A method for detecting the onset of CT saturation was suggested in [1]. It relies on the collapse in the current when the CT enters saturation. However, difficulties may arise if the current does not collapse to zero during saturation. An algorithm for calculating the core flux using the secondary current and then compensating the distortion in the secondary current was suggested in [2]. This method is based on the assumption that the remanent flux at the start of the calculation period is zero. A CT saturation detection method was proposed that
This work was sponsored by Korean Ministry of Science and Technology and the Korean Science and Engineering Foundation through the ERC program (Next-Generation Power Technology Center, NPTC). Yong-Cheol Kang is with NPTC and the Division of Electronics and Information Engineering, Chonbuk National Univ., Chonju, 561-756, Korea (e-mail: yckang@moak.chonbuk.ac.kr). Sang-Hee Kang is with NPTC and Department of Electrical Engineering, Myongji Univ., Yongin, 449-728, Korea (e-mail: shkang@mju.ac.kr). Peter Crossley is with School of Electrical and Electronic Engineering, the Queens University of Belfast, N. Ireland (email: p.crossley@ee.qub.ac.uk).

I [cos(t ) e t TP cos ], i1 (t ) = max 0, for t < 0

for t 0

(1)

where, Imax, Tp, and are the maximum fault current, the primary time constant and a fault inception angle. The secondary current i2(t) is:i2 (t ) = Ae t Ts + Be
t T p

C sin( t )

(2)

where Ts is the secondary time constant and tan = Ts [4]. i2(t) consists of two exponentially decaying terms and one sinusoidal term. The first and second terms exponentially decay with the time constant of Ts and Tp and the magnitude of sinusoid is: C = 2 I maxTs cos = 2 I max

Ts
1 + (Ts ) 2

(3)

i1
im

i2

Lm
Fig. 1 A simplified equivalent circuit of a CT

0-7803-7967-5/03/$17.00 2003 IEEE

The discrete-time version of i2(t) is derived using t = nT: i2 [n] = Ae nT Ts + Be


nT T p

C sin(

2 n ) N

(4)

where, T is the sampling interval and N is the number of samples per cycle. The first difference of i2[n] is: del1[n] = i2 [n] i2 [n 1]
p = A(1 eT Ts )e nT Ts + B(1 e p )e 2 n + ) C (2 sin ) sin( N N N 2

T T

nT T

(5)

(a) Undistorted and distorted secondary current due to saturation

When the power frequency is 60 Hz and N = 64, T = 0.26 ms. If Ts = 1 s and Tp = 0.02 s, the reduction rates of (1 eT/Ts) and (1 eT/Tp) for exponential terms will be 0.00026 and 0.0131, respectively. Since the time constants are large, the exponential terms in del1[n] are reduced to a negligible value. If i2[n] is in the form of (4) and the secondary time constant is large, the exponential term in del1[n] is insignificant. On the other hand, the magnitude of sinusoid in del1[n] is reduced to 2sin(/N)C = 0.098C, i.e. a decrease of about 90%. Consequently, if i2[n] is described by (4) and the time constants are sufficiently large, then del1[n] contains only a sinusoid and its magnitude is 0.098C. The second and third-differences of i2[n] are defined as:

(b) del1[n]

(c) del2[n]

del 2[n] = del1[n] del1[n 1]


del 3[n] = del 2[n] del 2[n 1]

(6) (7)

where, del2[n] and del3[n] are sinusoids with magnitudes of [2sin(/N)]2C = 0.009604C and [2sin(/N)]3C = 0.000941C, respectively. Hence, the magnitude of sinusoid in del3[n] is reduced to 0.1% of its value in i2[n]. If saturation starts at n = m+1 and ends at n = k+1 (where k > m), then k and m are the points of inflection. Let i21[n], i22[n] and i23[n] be the secondary currents before, during and after saturation. i21[n] and i23[n] have a waveform similar to (4). i22[n] is either similar to (4) or an exponential which collapses to zero during saturation. In either case, each of the three signals is continuous and does not change its form during each interval. Hence, i21[m] = i22[m] and i22[k] = i23[k] while i21[m+1] i22[m+1] and i22[k+1] i23[k+1], where i21[m+1] and i22[k+1] are not actual but virtual points on the extended line. Fig. 2 shows the secondary currents and the difference functions when a CT saturates. Fig. 2a shows typical undistorted (solid) and distorted (dotted) secondary currents. A. The feature of del1[n] Since the magnetizing inductance during saturation is much smaller than its value before saturation, i2[n] is distorted and contains points of inflection when a CT saturates. Since i2[n] has the same form as (4), the exponential terms in del1[n] can be ignored. Thus, del1[n] consists of sinusoids with different magnitudes in the saturated and unsaturated states. The discontinuities in del1[n] arise at m+1 and k+1, which corresponds to the start and end of saturation. As shown in Fig. 2b, del1[n] with no saturation is continuous whilst del1[n] with saturation is discontinuous at the saturation start and end.

Fig. 2

(d) del3[n] Secondary current and its difference functions

B. The feature of del2[n] If [2sin(/N)]2C is very small, del2[n] can be used as a detector. For example, if C = 100 A and N = 64, then [2sin(/N)]2C = 0.96 A. In the interval of n m (before saturation), del2[n] is a sinusoid and the values do not exceed 0.96 A. At n = m+1, i.e. the start of saturation: del 2[m + 1] = i22 [ m + 1] 2i21[m] + i21[m 1] (8)

The approximation in (9) is valid with a maximum error of 0.96 A. i21[m + 1] 2i21[ m] + i21[m 1] 0 Thus, the value of del2[m+1] can be approximated by: del2[m+1] i22[m+1] i21[m+1] (10) (9)

As mentioned earlier, i21[m+1] is a virtual point on the extended line and can be considered as the value at n = m+1 if the CT had not saturated. Therefore, del2[m+1], which is bigger than the current value of the saturation point. During the period m+2 n k (saturated state), the magnitude of

del2[n], which is also a sinusoid, is smaller than before saturation. At n = k+1 (the saturation end), del 2[k + 1] = i23 [k + 1] 2i22 [k ] + i22 [k 1] (11)

magnitudes of sinusoid in del3[n] are reduced more than that in del2[n]. Therefore, del3[n] can be used as a saturation detector since del3[m+1] and del3[k+1] are much larger than [2sin(/N)]3C. D. The criteria for CT saturation detection Saturation detection can be based on either del2[n] or del3[n]. The choice depends on the values of C and N. Generally, del3[n] when saturation begins or ends is much larger than [2sin(/N)]3C. Consequently, del3[n] is more effective than del2[n] and is used as the saturation detector in this paper. The detection criteria is based on (17). The samples that satisfy the inequality in (17) determine the start or end of the saturation period. The saturated or non-saturated status of the immediately preceding period is used to determine if the sample correspond to the start or end of saturation. del 3[n] > Th (17)

Similarly, del2[k+1] i23[k+1] i22[k+1], where i22[k+1] is a virtual point on the extended line. The error is less than 0.96 A and del2[k+1] has a large value. In the interval k+2 n, del2[n] is also a sinusoid. Consequently, if i2[n] can be expressed in the form of (4), the magnitudes of sinusoidal in del2[n], are smaller than [2sin(/N)]2C except where saturation starts and ends. The values of del2[n] at the start and end of each saturation period have spikes as shown in Fig. 2c. Thus, del2[n] can be used as a detector if del2[m+1] and del2[k+1] are much larger than [2sin(/N)]2C. C. The feature of del3[n] If [2sin(/N)]2C is only slightly smaller than del2[m+1] and del2[k+1], del3[n] can be used as a saturation detector. For example, if C = 200A and N = 64, then [2sin(/N)]3C = 0.188 A. In the interval of n m (before saturation), the values of del3[n], which is a sinusoid, are equal to or less than 0.188 A. At n = m+1 (the start of saturation),

where, Th = k 2 I f max [2 sin( N )]3 and Ifmax is the expected maximum fault current and k is a margin factor that depends on the low-pass filter and the required sensitivity. III. CASE STUDIES Fig. 3 shows a typical Korean 345 kV transmission system. The simulated fault is an A phase-to-ground fault at a distance of 2 km from the P bus. The fault inception angle is zero degrees and Ifmax is 75 A. A C400 CT (2000:5, R2 = 0.61 ) is used for the case studies and the burden is a resistance of 3.42 . The saturation point of (2.047 A, 1.512 Vs) is selected to generate hysteresis data using HYSDAT, an auxiliary program in EMTP. The secondary current contains a point of inflection at the inception of the fault. This results in a peak in del3[n], that may be detected as the start of saturation. To prevent maloperation, the algorithm is only activated if the current exceeds three times the rated secondary current for three successive samples. The performance of the algorithm was tested by varying the remanent flux from 80% to +80% of the flux of the saturation point. A. 0% remanent flux Fig. 4 shows the results for a 0% remanent flux without a filter. The dotted and solid lines are the ideal (unsaturated) and distorted (saturated) secondary currents, respectively. The secondary current is continuous and contains points of inflection at the start and end of each saturation period.
30 [GVA] 7 [GVA] 2 [km] Zs CT 98 [km]

del 3[m + 1] = i22 [m + 1] 3i21[m] + 3i21[m 1] i21[m 2] (12)


The following approximation is also valid with a max error of 0.188 A. i21[ m + 1] 3i21[m] + 3i21[m 1] i21[m 2] 0 (13)

Thus del3[m+1] i22[m+1] i21[m+1], which is nearly the same value as del2[m+1]. At n = m+2, del 3[m + 2] = i22 [m + 2] 3i22 [m + 1] + 3i22 [m] i21[m 1] (14) As i22[m+2] 3i22[m+1] +3i22[m] i22[m+1] 0 with a max error of 0.188 A, del3[m+2] i22[m1] i21[m1] 0. In the saturated period, m+3 n k, the magnitude of del3[n], which is also a sinusoid, is much smaller than before saturation. At n = k+1 (the saturation end), del 3[k + 1] = i23 [k + 1] 3i22 [k ] + 3i22 [k 1] i22 [k 2] (15)

Since i22[k+1] 3i22[k] +3i22[k1] i22[k2] 0 with a maximum error of 0.188 A, del3[k+1] i23[k+1] i22[k+1], which is nearly the same as del2[k+1]. Similarly, at n = k+2,
del 3[k + 2] = i23[k + 2] 3i23 [k + 1] + 3i23 [k ] i22 [k 1] i23 [k 1] i22 [k 1] 0

(16)

In the period k+3 n, del3[n] is also a sinusoid. As a result, the magnitudes of del3[n], which are sinusoids, are smaller than [2sin(/N)]3C except for the instants where the saturation begins and ends. The values of del3[n] at the start and end of each saturation period have spikes as shown in Fig. 2d. Consequently, del3[m+1] and del3[k+1] are nearly the same as the corresponding values of del2[n] whereas the

Fig. 3

The model system

The magnitude of the sinusoid in del2[n] is reduced to about 1.02 A (Fig. 4c); this is only slightly smaller than del2[n] at the start and end of each saturation period. Thus, del2[n] cannot be used as the saturation detector. The magnitude of the sinusoid in del3[n] is reduced to about 0.10 A (Fig. 4d); this is significantly smaller than del3[n] at the start and end of each saturation period. Consequently, del3[n] can be used as the saturation detector. The detector maintains the former state for 3 samples once (17) is satisfied. This prevents mal-operation due to asynchronism. In Fig. 4e, level 1 and 0 corresponds to the saturated and unsaturated states. The results clearly show that the algorithm successfully detects the periods of saturation. Fig. 5 shows the results, where the current is passed through a 1st order RC filter with a cut-off frequency of 1920 Hz (half the sampling frequency). They are very similar to Fig. 4 although the values of del3[n] at the start and end of saturation are slightly decreased. The above filter does not significantly affect pulses in del3[n] and the algorithm successfully detects saturation.

(b) del1[n]

(c) del2[n]

(d) del3[n]

Fig. 5 (a) Secondary currents

(e) Saturation detector output In the case of a 0% remanent flux (filtered, fc = 1920 Hz)

(b) del1[n]

(c) del2[n]

(d) del3[n]

Fig. 4

(e) Saturation detector output In the case of a 0% remanent flux (unfiltered)

B. +80% remanent flux Fig. 6 shows the results in the case of a +80% remanent without a filter. As expected, the remanent flux forces the first saturation earlier, although the location of subsequent saturation periods is unaffected, as compared with Fig. 4a. Thus, i2[n] in Fig. 6a is more severely distorted than Fig. 4a and also contains points of inflection at the start and end of each saturation period. In this example, i2[n] increases in value for several samples after the start of first saturation; this can be detected by observing the positive del1[n] values that occur immediately after the start of the first saturation period in Fig. 6b. Fig. 6 also illustrates that discontinuities arise at each point of inflection and are detected by comparing the pulses in del2[n] and del3[n] with the low amplitude sinusoidal background (see Fig. 6c and Fig. 6d). The values of del3[n] at the start and end of saturation are much smaller than those observed in Fig. 4, but are significantly larger than the background sinusoid. Thus, each saturation period is successfully detected as shown in Fig. 6e. The results indicate that the algorithm can detect the start and end of each saturation period even when i2[n] is severely distorted due to a high remanent flux. The results, where the currents are passed through the same filter as before are shown in Fig. 7. As in the previous case of Fig. 5, the filter does not give significant influence on the results and the algorithm successfully detects the start and end of saturation.

(a) Secondary currents

(a) Secondary currents Fig. 7

(e) Saturation detector output In the case of an 80% remanent flux (filtered, fc = 1920 Hz)

IV.

HARWARE IMPLEMENTATION

(b) del1[n]

(c) del2[n]

(d) del3[n]

Fig. 6

(e) Saturation detector output In the case of an 80% remanent flux (unfiltered)

When the saturation detection algorithm is applied to a real current signal, noise may distort the signal and affect the performance of the algorithm. The proposed algorithm is based on a third-difference function del3[n] that will amplify the noise signal and might affect the operating performance of the algorithm. This section describes a series of experimental tests on a small CT in the laboratory. Fig. 8 shows the configuration of the test system. The turns ratio of the CT is 100:5 and a large burden is connected to the secondary to ensure deep saturation. The algorithm was implemented on a TMS320C6701 digital signal processor. The sampling rate is 64 samples/cycle and the currents are passed through the 1st order RC filter (fc = 1920 Hz) to the 16-bit A/D converter. In these cases, Ifmax is 90 A and the threshold value is set at 1.8 A, which corresponds to k = 15. Fig. 9 and Fig 10 show two sets of experimental results. As the CT has a very small magnetization inductance, the current collapses to zero when the CT saturates. Thus, del1[n] during saturation is represented as an exponent. Thus, the number of samples, where the detector should keep the former state, is increased. However, since the current contains points of inflection, del3[n] is large at the start and end of each saturation period. Consequently, all the experimental tests, including the two quoted examples, support the conclusion that a prototype monitoring system, based on the algorithm described in this paper, can successfully detect the start and end of each saturation period.
R

(a) Secondary currents

R=0.05

LPF

S/H

A/D

TMS320 C6701

Display

(b) del1[n]
source 220V slidac Transformer 220:1

CT 100:5

magnetic switch

load

Fig. 8 (c) del2[n]

Configuration of the test system

(d) del3[n]

(a) Secondary current Fig. 10

(e) Saturation detector output


Operating signals obtained using prototype detector on experimental test system

V. CONCLUSION
(b) del1[n]

(c) del2[n]

(d) del3[n]

This paper describes a CT saturation detection algorithm using the third-difference function of the secondary current. The magnitude of the third-difference function at a point of inflection in the secondary current that corresponds to the start or end of a saturation period is significantly larger than the magnitude of the sinusoid seen before, during and after saturation. This sudden increase in magnitude is used to determine when the CT is saturated. The proposed algorithm was tested under various operating conditions including the remanent flux in the CT and the effect of an anti-aliasing low-pass filter. Results indicate that the algorithm successfully detects the start and end of each saturation period when a current is distorted due to a large remanent flux. A prototype monitoring system, based on the algorithm, successfully detects saturation in a real CT. VI. REFERENCES
[1] [2] A. G. Phadke, J. S. Thorp, Computer Relaying for Power Systems, Research Studies Press LTD., 1988, pp.185~186 Y. C. Kang, J. K. Park, S. H. Kang, A. T. Johns, R. K. Aggarwal, "An Algorithm for Compensating the Secondary Current of Current Transformers," IEEE Trans. on PWRD, Vol. 12, No. 1, January 1997, pp.116~124 T. Bunyagul, P. Crossley, P. Gale, "Overcurrent Protection Using Signals Derived from Saturated Measurement CTs," IEEE PES 2001 SM, July 15 ~ 19, 2001, Vancouver, BC, Canada S. H. Horowitz, A. G. Phadke, Power System Relaying, Research Studies Press LTD. 1992, pp.56-57 M. Kezunovic, L. Kojovic, A. Abur, C. W. Fromen, and F. Phillips, "Experimental Evaluation of EMTP-Based Current Transformer Models for Protective Relay Transient Study," IEEE Trans. PWRD, vol. 9, no. 1, January 1994, pp. 405-413

Fig. 9

(e) Saturation detector output Operating signals obtained using prototype detector on experimental test system

[3] [4] [5] (a) Secondary current

VII. BIOGRAPHIES
Yong-Cheol Kang (S' 1993, M' 1998) received the B.S., M.S. and Ph.D degrees from Seoul National University, Korea in 1991, 1993, and 1997, respectively. Now he is an assistant professor at Chonbuk National University, Korea. His research interest is development of new protection systems for power systems using digital signal processing techniques. Sang-Hee Kang (S' 1990, M' 1993) received the B.S., M.S. and Ph. D. degrees from Seoul National University, Korea in 1985, 1987 and 1993, respectively. Now, he is an associate professor at Myongji University, Korea. His research interest is to develop protection relays and protection systems for power systems using digital signal processing techniques and artificial intelligence. Peter A. Crossley (M 1996) is a Professor and Head of Power & Energy Group, the Queens University of Belfast, UK. He graduated with a B.Sc. degree from UMIST in 1977 and a PH.D. degree from the University of Cambridge, in 1983. During the period, 1977-1990, he worked at ALSTOM on the design and application of digital protection relays. He is a Chartered Engineer and a Member of the IEE and IEEE.

(b) del1[n]

(c) del2[n]

(d) del3[n]