Separating recoverable from non-recoverable gas turbine degradation

A gas turbine loses output and gains heat rate over time through two broadly different routes. Compressor fouling, where deposits accumulate on blade surfaces and change the aerodynamics, is largely reversible by washing. Deterioration of the hot section and of blade and seal geometry, through erosion, oxidation, creep and increased clearances, is largely not.

The distinction decides the next action and the next spend. Recoverable loss argues for a wash, which costs a short outage and some water. Non-recoverable loss argues for an inspection or for accepting a permanently lower baseline until the next major overhaul. Confusing the two produces either an unnecessary outage or a slow loss nobody investigates.

It is the same attribution problem as separating stack degradation from balance-of-plant losses on an electrolyser, and it is solved the same way: correct the measurements, look at the pattern rather than a single parameter, and check the instruments before believing any of it.

A gas turbine's output and heat rate depend strongly on ambient temperature, pressure and humidity, on inlet and exhaust pressure losses, and on the operating mode. Raw performance data across a year is dominated by weather, and a trend fitted to uncorrected data is largely a trend in the seasons.

Correction to reference conditions is therefore the first step and the one most often done badly. The correction curves come from the OEM, and the parameters they cover, the range over which they are valid and the treatment of anything outside that range are all worth establishing before a conclusion is drawn.

A practical check: plot corrected output against ambient temperature. If a slope remains, the correction is incomplete and any degradation number derived from it carries that error. This is cheap to do and settles a surprising number of arguments before they start.

Compressor fouling reduces air mass flow and compressor efficiency together. The characteristic pattern is falling corrected mass flow with a falling pressure ratio, output down, heat rate up, and compressor discharge temperature rising relative to what the pressure ratio would imply.

Hot section deterioration and increased clearances show differently. Where turbine efficiency has fallen, exhaust temperature rises for a given firing condition, and the relationship between fuel flow, output and exhaust temperature moves in a way that fouling does not reproduce.

Neither is diagnosable from a single parameter. Output alone is ambiguous, heat rate alone is ambiguous, and it is the combination across mass flow, pressure ratio, discharge temperature and exhaust temperature that carries the answer. This is precisely the case where a physical model earns its place, because the constraint that ties those quantities together is what makes an inconsistent set of readings visible.

Filtration condition belongs in the same analysis. A differential pressure rising across the inlet filters produces a loss that looks like fouling in the output data and is fixed by a different action entirely.

A drifting thermocouple, a fouled pressure tapping or a fuel flow meter reading progressively high all produce smooth, plausible degradation trends. They are among the most common causes of a degradation investigation that ends with nothing found.

The defence is mutual consistency rather than individual calibration. Mass and energy have to balance: a measured fuel flow at a stated heating value implies a heat input, which implies a relationship between output and exhaust energy. When the set cannot be reconciled, something in the set is wrong, and that is a different conclusion from the machine deteriorating.

Calibration history for the instruments used in the analysis should be part of the analysis, not an afterthought. A performance conclusion drawn from an instrument last calibrated four years ago is a conclusion about the instrument.

Every wash the plant has ever done is a controlled test of how much of the accumulated loss was recoverable. The recovery achieved each time, measured on corrected data before and after, bounds the recoverable fraction directly.

Plotted over several years the pattern is informative in its own right. If recovery per wash is shrinking while the pre-wash loss is similar, non-recoverable deterioration is accumulating underneath the fouling. If recovery is holding steady, the baseline is intact and the loss is being managed.

It also produces the number needed for the next decision: the rate at which fouling reaccumulates under current site conditions, which is what sets the economically correct wash interval rather than the interval in the manual.

This only works if the before-and-after measurements were taken under comparable conditions and retained. Most plants have the wash dates and not the performance data around them, which is a retention decision rather than a measurement one.

For an online wash, the decision is whether the recoverable loss has accumulated enough that the fuel saving over the next interval exceeds the small output penalty during the wash. That calculation needs the reaccumulation rate and the current marginal fuel cost, both of which the plant has.

For an offline wash the outage cost enters, which makes the timing a scheduling question as much as a performance one, and usually means aligning it with something else already planned.

For non-recoverable loss the question is different: whether the deterioration rate justifies bringing forward an inspection, and what the data suggests about which section is losing. That is an evidence question a plant is much better placed to answer with a corrected history than with a snapshot.

And where the machine sits under an availability or output guarantee, the split is also the attribution question in that argument, which is evidenced the same way as any other: continuous, corrected, timestamped and retained. The general shape of that is set out in evidencing availability guarantee compliance.