Introduction

There is no argument that dissolved gas analysis (DGA) and partial discharge (PD) detection are two of the most useful and key methods to detect insulation damage in, and impending failure of, a transformer. However, real-time monitoring implementations are on the extreme end of cost-per-unit monitored. For a 500 MVA GSU or grid-level transformer, the cost can be justified. But for the majority of transformers in our systems, there simply is no justification for the expense.

Both DGA and PD detection can also be performed on an as-needed basis. This is the most common form of DGA, which is typically performed monthly, bi-yearly or yearly. As-needed PD detection, however, entails outages to install and remove equipment, and thus is far less common on a shorter term. For both, granularity is high and far from real-time monitoring.

Given the widespread and increasing application of Power Quality (PQ) monitors, it would seem we may have a lower-cost path to provide online transformer condition assessment.

Some Critical Facts

Before we discuss what and how to detect, let us first note a major caveat that encompasses all PQ detection methods. That is, PQ monitoring rarely has the direct information needed to indicate an impending transformer failure, even with an Artificial Intelligence (AI) implementation. Why? The insulation condition metrics are simply not there. DGA senses heightened levels of dissolved gases that reveal thermal or arcing damage to insulation. BAM! PD has heightened levels of discharge when a dielectric is compromised. BAM! Again. But captured voltage and current signals in the range of typical PQ monitors do not have either “Pfffft . . . Wow, that was deflating.” So, things appear to be stacked against us. To overcome this, we need another approach. We need to formulate methods that supply an accurate indication of damage and loss of life from what we can detect. Unfortunately, there is limited direct industry knowledge for us to leverage on this matter, and so we must pull information from many sources and lean on relevant experience and knowledge of the various PQ conditions that present themselves. And in full disclosure, we must note that these methods are secondary evaluations at best.

Now, is a secondary indicator of any value? I would argue, absolutely, when used correctly. My Oura ring senses I may have had a “disruptive air flow event” by looking at blood oxygen levels, breathing regularity, heart rate, and movement. Is it a sleep apnea diagnosis? No. But is it helpful? Absolutely. I have guidance without an invasive sleep trial. Also, I can wear my Oura ring night after night for months. I can gain a much better long-term picture of my sleep health instead of a conclusion based on one nervous night in a sleep trial. In the same way, PQ events and conditions can be monitored over weeks, months, and years to gain an overall picture of external conditions that compromise our transformer’s health and life. And, who knows, we may even capture the smoking gun. DGA and PD detection have little to no ability to do this.

To wrap up this section, we should ask: Can a PQ monitor catch an internal condition leading to failure? Most likely no. By the time internal conditions manifest externally, we are probably incurring an internal fault. We can certainly capture this, but it is not predictive. This is the blunt reality of external PQ monitoring. We can neither detect gases in the oil nor high frequency PD on a medium or high voltage winding. But let’s continue on. There are external events and conditions that reveal much of what we need to know.

Transformer PQ Detection

The PQ conditions discussed next are not in any particular order, nor have they been supplied with a method to accumulate suspect conditions into an overall loss-of-life metric. It would certainly be great to have that here. However, those details are for another article. It will take the remainder of my space here to just enumerate what a PQ monitor can detect and form into a loss-oflife metric. So, as the Genie says, please “ruminate while I illuminate the possibilities”.

Long-Term Over-Voltage: Long-term over-voltages applied on transformer windings degrade the insulation. Plain and simple. But there is a voltage level where damage begins to shorten life, and increasing voltage levels where damage increases. There is little information available, and so much work is needed to develop consistent curves for a loss-of-life metric.
Typical causes are:

• Ferranti rise along a long-unloaded line
• Sustained harmonic resonance
• Loss of load condition with bus capacitors left online
• Incorrect voltage regulator control action
• Incorrecct capacitor bank control action

Short-Term Over-Voltage: This is similar to long-term over-voltages, but our metrics will consider higher voltages over a shorter period.
Typical causes are:

• Dynamic harmonic resonance (Figure 1 and Figure 2)
• AVR overshoot after monitor starting or fault conditions
• Ungrounded system voltage escalation
• Ground fault neutral offset
• Ferro-resonance
• Inverter back-feed into an isolated system

Figure 1. 226% peak over-voltage affecting secondary 690V winding.

Figure 2. 145% over-voltage on main 34.5 kV collection system bus, affecting all 34.5 kV primary windings, amplified 12th harmonic currents affecting main 50 MVA transformer.

Transient Peak Over-Voltage: This mirrors the previous two conditions, but our metrics consider voltage peaks. Since dielectrics fail via peak voltage stress, we may find more help specifying damage limits and methods to update a loss-of-life metric.
Typical causes are:

• Capacitor Switching
• Transient recovery voltage
• Switching transient
• Lightening induced voltage

High Frequency Noise – dV/dt: This condition can exist on systems where fast switching converter modules are present (Figure 3). When insulation is exposed to higherfrequency and fast-changing voltages (high dV/dt), dielectric impurities and voids see an increase in electric field intensity and manifest internal breakdowns. When these accumulate and form internal tracking paths, failures can occur. There is even less information available on this relating it to accumulated loss of life, but the condition is very real and impacts many systems.
Typical causes are:

• Unfiltered converter switching noise
• Converter noise coupling into AC systems from improperly installed shielding

Figure 3. Damaging high frequency noise on secondary winding at VFD startup

Through Fault Current: Secondary faults produce through-fault currents that impact transformer windings thermally and mechanically (via compression, insulation wear, and displacement). These can be compared against frequent-fault damage curves specified in IEEE C57.109 (Figure 4). When a damage curve is exceeded, a loss-of-life condition can be reported and a loss-of-life metric updated. Typical causes that exacerbate this condition are:

• Primary protection not protecting the transformer damage curve
• Protection settings with minimal “protection” favoring continuation of process
• Improperly oversized fuses
• Lack of maintenance and care of the electrical distribution system

Figure 4. Transformer damage curve (TX-1) shown with upstream fuse protection

Excessive Transformer Energization: Transformer energization creates high currents which impart both mechanical and heating stress to the windings. If these become excessive, damage and loss-of-life will occur. Some manufacturers specify time between energizations, or a limit on energizations, but data is sparse. There are no direct methods relating excessive energizations to a loss-of-life index.
Typical causes are:

• Re-energization immidiately after shutdown leading to higher surges due to core residual conditions

• Re-energization due to upstream protective device nuisance tripping during energization
• Lack of knowledge on the danger of reperat energization attempts durging unplanned outages to get equipment back online

K-Factor: K-Duty can be calculated using UL 1651 with measured harmonic currents. If the K-Duty exceeds the transformer’s K-Rating, this would present an overload condition that leads to loss-of-life. Alternatively, harmonic derating of a transformer can be calculated using IEEE C57.110, and if loading exceeds the derating, this will also present an overload and loss-of-life condition.
Typical conditions that cause such overloads are:

• Non-K-Rated transformers applied to load centers with significant VFD load
• Resonant conditions from harmonic sources and capacitor banks
• High harmonic load conditions
• High fundemental base loading with additional harmonic load

Fundamental Overload: In addition to K-Factor, fundamental frequency overloading will lead to overheating and loss-of-life.
Typical causes are:

• Load transfers to already heavily loaded transformers
• Miscalculation of expected loading
• Ignoring low power factor loads that have higher current draw

Temperature via Thermal Modeling: Heating from both fundamental and harmonic currents can be combined with measured ambient temperature and a transformer thermal model to estimate the transformer’s hotspot temperature (Figure 5). If the temperature exceeds ratings, this would present a loss-of-life condition.

Figure 5. Transformer hot-spot temperature simulation during load cycling.

Zero-Sequence Current: Zero sequence current incurs zero sequence flux that couples with the steel of the tank and other internal components (Figure 6). This causes additional heating and possible damage to the transformer. If zero sequence currents are above manufacturer’s recommended limits, overheating will occur with a loss-oflife.
Typical causes are:

• Extended faults with high zero-sequence content
• Severe and continous load imbalance
• Loads using power supplies with 3^rd harmonic content

Figure 6. Zero-sequence flux coupling with tank and other internal components.

Vibration: Most PQ meters do not accommodate accelerometer inputs. However, if they do, we can monitor the vibration of a transformer. There are expected normal levels of vibration, and if these are exceeded, a loss-oflife metric can be updated accordingly.
Typical conditions leading to excessive vibration are:

• Load impacting with unexpected effects on transformer mechanics
• Harmonic resonant conditions and repeating harmonic resonant conditions
• Heavy unbalanced loading conditions
• High inrush conditions from core residual conditions

Sound Level: Most PQ monitors do not accommodate microphone inputs. However, if they do, we can install microphones to monitor the overall sound levels of the transformer. There are expected levels of noise from no-load up to full load, and if these are exceeded, a loss-of-life metric can be updated accordingly.
Typical conditions leading to excessive noise are:

• Faliure of components within the transformer
• Winding damange having loosened the windings
• Screws having come loose withing or on the exterior
• Core damage allowing more core movement

Transformer Saturation: Transformer saturation creates a core overheating condition. If the core temperature is not directly monitored, then an estimate of core heating can be obtained from currents flowing in the windings.
Typical causes of saturation:

• Excessive transformer energizations (see discussion above)
• Rectifer bridge failures drawing current with unequal half cycles (Figure 7)
• Resonances involving even harmonics

Figure 7. Rectifer current waveforms causing severe transfomer saturation and core overheating.

Conclusions

From the list of PQ conditions just noted, it is clear that PQ meters can have the measured values and ability to see almost all external electrical conditions and some internal conditions affecting a transformer. In many ways, we can see more than DGA and PD detection regarding external causes. Thus, PQ monitors have the ability to fill a major need in detecting potential transformer issues. Hopefully, it is clear now that if developed, specified, and implemented correctly, PQ monitors can effectively provide a secondary loss-of-life metric and transformer condition assessment.