Repeated differential trips on a 40 MVA, 132/33 kV grid transformer
A 40 MVA, 132/33 kV, YNd11 transformer was retrofitted with a numerical differential relay rated 1 A, with pickup 0.20 pu, bias slope-1 of 25% and second-harmonic restraint of 15%. The CTs fitted are 250/1 A on the 132 kV side and 600/1 A on the 33 kV side. During commissioning the relay's CT-primary settings were left at the factory default of 1 A on both sides, and the load test was done at only 8 MW, where nothing operated. Three weeks later the relay tripped at 33 MW, 0.92 pf lagging, recording Idiff = 0.42 A and Ibias = 0.84 A with only 3% second harmonic. A second trip occurred when a 33 kV feeder fault of 8 kA was cleared by the feeder relay, this time with Idiff = 1.9 A and high bias. Insulation resistance, turns-ratio measurements and dissolved gas analysis are all normal, and the transformer is presently carrying 20 MW without incident.
1. Using the recorded currents, decide whether the transformer has an internal fault. Quantify the CT mismatch and locate the relay's operating point relative to its characteristic.
Read the worked answer
First establish the base currents. HV full-load current = 40e6/(1.732 x 132e3) = 175 A, so the HV CT secondary at rating is 175/250 = 0.70 A. LV full-load current = 40e6/(1.732 x 33e3) = 700 A, so the LV CT secondary is 700/600 = 1.167 A. With the relay left at 1 A on both sides it applies no normalization, so a 40% mismatch (1.167 vs 0.70) exists between the two ends.
Check the trip point: 33 MW at 0.92 pf = 35.9 MVA = 0.897 pu. HV secondary = 0.70 x 0.897 = 0.628 A, LV secondary = 1.167 x 0.897 = 1.047 A. Differential = 1.047 - 0.628 = 0.419 A, bias = (0.628 + 1.047)/2 = 0.838 A. Threshold = 0.20 + 0.25 x 0.838 = 0.409 A. Since 0.419 > 0.409, the relay operates exactly as recorded.
This is a setting error, not a fault: the differential current tracks load (at 20 MW it is about 0.25 A against a threshold of 0.33 A, which is why it now holds), the second harmonic is only 3%, far below inrush, and insulation, ratio and DGA are clean. A genuine internal fault would show a large Idiff essentially independent of load and would not self-clear. Note also that the mismatch (40%) exceeds the bias slope (25%), so the scheme can never be stable for through currents.
2. Explain why the relay also operated for the external 33 kV feeder fault, and why a higher bias slope would not be the correct cure.
Read the worked answer
For the external fault the LV CT sees 8000/600 = 13.3 A secondary, i.e. about 11.4 times its rated secondary current, and the HV CT sees the corresponding 8 kA at 132 kV reflected as roughly 8 kA x (33/132)/250 = 8.0 A on a 0.70 A base. Both ends are driven well into the linear region where the 40% ratio mismatch appears directly as spill current.
The bias characteristic only restrains spill that is smaller than the slope percentage of the bias. Here the inherent spill is 40% of the through current, so any slope below 40% will eventually be exceeded as the through current rises; the observed Idiff of 1.9 A at high bias is exactly that. Raising slope-1 to, say, 50% would make the relay stable for external faults but would desensitize it for genuine internal faults with low infeed, which is unacceptable for a transformer of this size.
Secondary issues compound the problem during the external fault: the 8 kA primary with DC offset and possible remanence can saturate the LV CT if its class and connected burden are marginal, producing waveform distortion that the relay may see as differential current. The correct diagnosis is that CT ratio normalization is wrong, and the CT class/burden and the vector-group compensation must be verified rather than masked by raising the slope.
3. What would you change, and what tests would you perform before re-energising the transformer?
Read the worked answer
Settings: enter the true CT primary ratings (250 A HV, 600 A LV) so the relay normalizes both sides to the transformer base, or set explicit matching factors of 1.0 and 0.60. Verify the vector-group compensation - for YNd11 the LV currents must be rotated by 30 degrees and the zero-sequence component removed, normally done numerically with both CT sets star-connected. Then set pickup 0.30 pu, slope-1 30%, slope-2 80% and second-harmonic restraint 15-20%.
Tests before energising:
- Secondary injection: inject the recorded 33 MW load pattern (0.628 A HV, 1.047 A LV) and confirm the relay is stable with Idiff near zero; then simulate an external 8 kA through fault and confirm stability across the full bias range.
- Primary injection or a load test at 50-100% rated current with a clamp-on ammeter on both CT secondaries, confirming the measured spill is under 5% after correction.
- Inrush test: energise with the LV open and confirm the second-harmonic restraint blocks; check the second-harmonic content is above 15% at 0.1 s.
- Verify CT class and connected burden (including lead resistance) for the 8 kA through-fault level, and confirm the CTs are not shared with any other burden in the circuit.
- Record a 24 h load test at 80% load and review the oscillography before returning the relay to service.