HYDROGEN-H2 PHASE A (PPM): 1300.0 — Partial discharge in oil
05-02 05:00
rule
CRITICAL
METHANE-CH4 PHASE A (PPM): 650.0 — Thermal fault above ~300 C
05-02 05:00
dga
MEDIUM
DGA fault PD: Partial discharge (corona)
IEC Rules Catalog (reference)
Parameter
Unit
Warn
Critical
Standard
HYDROGEN-H2
ppm
100
700
IEC 60599
METHANE-CH4
ppm
120
400
IEC 60599
ETHYLENE-C2H4
ppm
50
200
IEC 60599
ACETYLENE-C2H2
ppm
3
35
IEC 60599
CARBON MONOXIDE-CO
ppm
350
900
IEEE C57.104
MOISTURE-H2O
ppm
10
30
IEC 60422
TOP OIL TEMP
°C
85
105
IEC 60076-7
WINDING TEMP
°C
98
120
IEC 60076-2
AMBIENT TEMP
°C
40
50
IEC 60076-2
OLTC OIL TEMP
°C
80
100
IEC 60214-1
BOTTOM OIL TEMP
°C
75
95
IEC 60076-7
① Rule-Based · IEC 60599 / 60076
Compares each sensor's latest reading against fixed IEC limits — no
learning, instantly explainable, highest trust. At/above the warning limit is flagged; at/above
critical it's urgent.
Parameter
Latest
Warn
Critical
Status
HYDROGEN-H2 PHASE A (PPM)
1300.0 ppm
100.0
700.0
CRITICAL
METHANE-CH4 PHASE A (PPM)
650.0 ppm
120.0
400.0
CRITICAL
② Statistical · Z-Score Consensus
Learns each sensor's own normal range, then flags points by a
2-of-3 consensus of global Z-score, rolling Z-score, and IQR — catching drift and spikes even while
still under the IEC limit.
Sensor
Anomalous pts
of total
Rate
HYDROGEN-H2 PHASE A (PPM)
4
30
13.3%
METHANE-CH4 PHASE A (PPM)
4
30
13.3%
③ Isolation Forest · Multivariate ML
An unsupervised model that learns the joint normal pattern
across all sensors at once — so it can flag a combination that's abnormal even when every sensor
alone looks fine (e.g. winding hot while load is light).
MULTIVARIATE CONTEXT · 33/100
Singled out the 2 most-unusual of
30 multivariate snapshots — it always isolates the oddest sensor
combinations by design, so this is a context signal, not a standalone verdict
(the overall priority above fuses all four methods). The flagged snapshots deviate most on:
(multivariate = all sensors viewed together, not one at a time.)
HYDROGEN-H2 PHASE A (PPM)+2.3σ
METHANE-CH4 PHASE A (PPM)+2.3σ
Likely drivers (approximate) — Isolation Forest gives no exact per-sensor
attribution; ranked by mean deviation of the flagged snapshots vs the window normal.
④ DGA · Duval Triangle + IEC Ratios
Identifies the type of fault from dissolved gases — corona,
sparking, arcing, or overheating — using the Duval Triangle and IEC 60599 gas ratios.
CO
0
ppm
H2
1300
ppm
CH4
650
ppm
CO2
0
ppm
C2H2
0
ppm
C2H4
0
ppm
C2H6
0
ppm
The three combustible gases are reduced to their
relative % of CH₄ / C₂H₄ / C₂H₂ (they sum to 100). That single point is plotted on the
Duval Triangle — the zone it lands in names the fault:
PDPartial discharge (corona)
D1Discharge of low energy (sparking)
D2Discharge of high energy (arcing)
T1Thermal fault below 300 C
T2Thermal fault 300-700 C
T3Thermal fault above 700 C
DTMixed thermal and electrical fault
IEC 60599 ratio cross-check
R1 = C₂H₂ / C₂H₄0.0
R2 = CH₄ / H₂0.5
R5 = C₂H₄ / C₂H₆0.0
Duval Triangle
PD
Partial discharge (corona)
IEC 60599 ratios
ND
No fault detected / not determinable
Two independent methods, shown side by side; the DGA score uses the
more severe of the two. ND = no fault determinable.