Combustion Dynamics Monitoring for Gas Turbines Using AI

What Combustion Dynamics Are

Heat release and pressure waves feeding each other inside the can.

Combustion instability happens when changes in the flame's heat release line up with pressure waves in the combustor. A 1995 NASA paper on premixed combustors describes this coupling, and notes that lean premixed designs, used for low NOx, are prone to it. When the system has too little damping, the flame adds energy to the pressure wave and the pulsation grows. The paper lists the results: damaging vibration of combustor hardware, worse emissions, lower efficiency and, in a lean flame, blowout. A GE patent on dynamics monitoring adds that repeated pulsation can fatigue the combustor and has been linked to failures of fuel nozzles, cans and transition pieces. Our combustion specialists can review what your units show today.

Frequency Bands and What They Point To

Different tones tell you different things about the flame.

Dynamics are measured by a pressure sensor on each can and split into frequency bands. Each band tends to point at a different cause. The ranges below come from two GE patents, each describing a specific combustor. They are examples, not limits. One patent describes a lean blowout band of about 10 to 25 Hz, a cold tone of about 70 to 120 Hz and a hot tone of about 120 to 160 Hz, and notes that as the fuel-air mix gets leaner, cold tone rises and hot tone falls. The other describes low frequency rumble up to 50 Hz, mid frequency at 100 to 250 Hz and high frequency screech above 250 Hz, which it says can, at high amplitude, cause component failure in as little as a few minutes. Your OEM sets the bands for your combustor. Map yours in a working session.

Ranges are from GE patents US7210297 and US8125646 and are combustor specific. Use your OEM's bands and limits.

Can-to-Can Variation

Cans on the same unit do not burn exactly alike.

In a multi-can combustor, each can gets its own share of fuel and air. A GE patent on finding hot and cold chambers notes that the fuel-air ratio can differ slightly from can to can, and that small changes in flame temperature can cause large changes in emissions. These differences do not average out. One lean can may be near blowout while another runs hot, and the unit average still looks fine. This is why each can needs its own trend, not just a unit total. Exhaust thermocouples give a second view. Because the gas swirls on its way through the turbine, a cold spot in the exhaust does not sit directly behind the can that made it, and the patent notes that mapping one to the other needs care. Compare your cans with our engineers.

Tuning Drift or Hardware Damage?

The same alarm can have two very different causes.

DLN combustors are usually tuned at commissioning and after major maintenance, and many units are retuned as the seasons change. GE Vernova, for example, describes its F-class AutoTune as a way to end seasonal retunes. Between tunes, ambient conditions, fuel and wear move the combustor away from its tuned point. That is tuning drift. Hardware damage is different: a cracked liner, a worn seal or a damaged nozzle changes one can. The patterns below are typical rather than certain, and an engineer confirms the cause. Get help reading your trends in a short session.

NOx Drift and Flashback Risk

Emissions and dynamics are two views of the same flame.

NOx depends strongly on flame temperature, so a slow NOx rise at the stack often means part of the combustor is running hotter than when it was tuned. Reading NOx and CO next to the per-can dynamics helps show which cans are moving. Flashback is a different risk. It happens when the local flame speed is faster than the speed of the fuel-air mix, so the flame moves back into the premixer, where it can damage hardware. The OEM protection system guards against it. The AI's role is narrower: looking for changes in dynamics, emissions and temperatures around events, and showing them to engineers. Ask our team how this fits your protection scheme.

How AI Models Dynamics per Band and per Can

The baseline moves with load and weather. The model moves with it.

A fixed alarm level cannot tell a normal summer value from an early warning. The model learns what each band in each can looks like at each load, ambient temperature and fuel condition, from your own history. It then flags a can when it moves away from that expected value, or away from its neighbours, before any protection limit is near. It suggests a likely cause and a response. It does not change tuning or settings. Engineers decide. See an example in a live demo.

Voice Alerts and Work Orders

The right person hears about it, and the follow-up is recorded.

A dynamics alert is only useful if someone acts on it. Control room operators hear a short voice alert naming the unit, the can, the band and what the AI suggests. The combustion engineer gets the detail. A tuning or inspection work order opens in your CMMS for review, so the alert does not vanish at shift change. When the work is done, the finding is recorded against the alert. Over time this builds a history of which patterns were tuning, which were hardware and which were a faulty sensor. Ask our team how alerts are routed.

Combustion Dynamics Monitoring Checklist

Eight checks before dynamics monitoring goes live.

Use this checklist for one unit first. Tick items already in place. Get help with any gaps from our setup team.

Combustion Dynamics Mistakes to Avoid

A quiet average can hide one loud can.

These are the patterns that most often weaken combustion dynamics programs. A short dynamics review can find them.

How iFactory Monitors Combustion Dynamics

Predictive AI and Voice Alerts, on your own network.

iFactory's Predictive AI models dynamics per frequency band and can, and flags tuning drift. Voice Alerts tell the control room, and a tuning or inspection work order opens. It runs on an on-prem AI server inside your plant.

Tuning, protection settings and inspection decisions remain with your engineers and OEM. Figures on this page are worked examples.

Turnkey AI: Delivered, Connected and Live in 6–12 Weeks

You do not build this. It arrives ready.

iFactory ships as a pre-configured NVIDIA AI server with the software pre-loaded, racked and ready. Rack it, plug in power and Ethernet, and the AI is live on your network. Our team handles cabling, network setup, PLC, SCADA, DCS and CMMS integration, dynamics and emissions signal mapping, operator training and 24×7 remote monitoring. Data stays on your network. For a scope matched to your plant, request a turnkey quote.

Frequently Asked Questions