Spotting system data shifts before they trigger outages
Unplanned downtime in industrial settings can cost up to £194,000 an hour, and that’s before considering the ripple effects through the grid. In excitation systems, the first signs of trouble often show up as subtle changes. Here, Ryan Kavanagh, director at Excitation & Engineering Services, explains how to read system trend data more effectively and spot early‑warning signs before they escalate.
Excitation systems rarely fail without giving clues.
The first signs of trouble often appear quietly in the data:
- A rotor winding temperature creeps up by a degree or two, barely noticeable at first
- The field current begins to rise as the system works harder to keep output steady
- What looks like a minor fluctuation is in fact a cycle building towards thermal runaway, where more heat demands more current, and more current generates more heat
Not every symptom shows in the same way. Sometimes the voltage line on the graph holds steady, yet the field current beneath keeps on climbing: a balance that points to stress hidden deeper in the system.
Other times, the power output tells the story. A machine that once reached its rated capacity with excitation comfortably below the nominal limit now struggles to get there, requiring excitation levels close to—or even exceeding—its nominal levels.
Even the shape of analogue measurements can reveal problems. Instead of a clean, flat trace, small oscillations begin to appear. In static excitation systems, the signs can be physical before they are digital. Excessive brush gear wear produces excessive carbon dust, showing that uneven current distribution is already taking its toll.
None of these changes are random glitches. Each one is the system’s way of signalling that something inside is under strain and catching them early can make the difference between a simple fix and a costly outage.
Why are these signs missed?
With so many responsibilities, operators are already working against a tide of information.
Modern SCADA systems generate thousands of data points, far more than the simple dials and gauges of older equipment, and it isn’t always clear which ones matter. Subtle changes often sit buried in the background, or worse, they don’t appear on screen at all unless specific trends are being tracked.
Even when the data is visible, it can be easy to dismiss. A small temperature swing looks harmless, a slight rise in current feels within tolerance. Without advanced monitoring that cross-checks variables, such as comparing current against temperature shifts, those early clues slip quietly past.
Trend data, however, gives operators a point of comparison.
By looking at today alongside yesterday, or this year against the last, we can build a clear picture of what “normal” really looks like. Even this analysis is challenged by numerous variables, though:
- Ambient temperature
- Humidity and weather
- Duty cycles
- Load
Artificial intelligence is being deployed for analysis to provide intelligent condition monitoring.
“Machine learning algorithms have successfully been used to extract complex nonlinear relationships from multi-dimensional sensor data [in rotating electrical machines],” said Zachariades and Xavier, Sensors (Basel), 2025.
“AI techniques such as expert systems, fuzzy logic, and neural networks have demonstrated superior capabilities in identifying subtle fault patterns and adapting to dynamic operating conditions.”
Continuous trending offers the richest insight, but regular checks can still reveal the story. Asking a simple question, “Is the field current the same this week as last?”, can highlight the first signs in stress.
In one recent case, a hydro power station failed to make nominal voltage on startup. With support, the pattern was traced back to an inter-turn fault in the rotor winding, an issue that may have otherwise gone unseen.
Trend data turns vague worries into measurable signals. Instead of something feeling ’off’, operators gain evidence they can act on.
Where possible, continuous monitoring adds another layer of protection. Faults don’t always unfold slowly, and real-time trending acts as a safety net when changes happen fast.
When a reading looks unusual, the key is not to panic. Watch it closely, build up the evidence and be ready to escalate. Shifts in current, temperature or voltage patterns are all reasons to bring in a specialist. Plant operators don’t need to become excitation experts, nor should they be expected to.
What should operators do with their trend data?
Even when technicians can spot unusual trends early, interpreting what they mean isn’t always straightforward. Service providers like EES can step in at this stage, using higher-resolution monitoring to reveal behaviour that standard plant instrumentation may miss.
For example, one power station noticed a subtle imbalance in excitation transformer currents. EES engineers employed power system modelling and simulation techniques, as well as gathering higher resolution data using a chart recorder, to compare simulations with measured data and to analyse behaviour under various conditions.
Using tools like MATLAB Simulink, they identified a fault in the thyristor gate pulse transformer, something that routine checks hadn’t detected. This approach allowed for a detailed analysis without interrupting operations, leading to a precise diagnosis and timely intervention.
In another case, at a CHP plant experiencing oscillatory voltage behaviour, EES service engineers deployed a high-speed recording oscilloscope. This allowed them to capture voltage waveforms at kHz sampling intervals, significantly higher than standard control systems could record.
By combining them with their own instrumentation, they were able to monitor voltage fluctuations in unprecedented detail and trace the root cause. Working alongside the generator manufacturer, they confirmed the fault pattern and planned a controlled shutdown, averting a potential failure and preventing unplanned downtime.
Support can take many forms, from targeted troubleshooting to periodic tests over time. By turning early-warning data into clear, actionable insight, operators gain the confidence to act, bridging the gap between “something looks off” and knowing exactly what to do next.
By making trend data part of routine checks, operators can stop small anomalies from snowballing into outages. Service providers can help interpret these signs, enabling action before issues escalate.
Learn more about our operational support for synchronous machine owners here.
References: Zachariades, C. and Xavier, V. (2025) ‘A review of artificial intelligence techniques in fault diagnosis of electric machines’, Sensors, 25(16), p. 5128. doi:10.3390/s25165128.
Recent Comments