The symptom walks into a meeting, not a boiler house: the fuel bill is creeping, the nameplate says 82%, the operator says "we run at about 75", and an audit report from three years ago says 78.6% to one decimal. Nobody can defend any of the numbers, because none of them came with a boundary, a window and a basis. This Clinic lesson is the cure: how to run an efficiency test on site, with plant instruments, that survives cross-examination — and how to catch the four ways a test day quietly lies. Lesson B4 owns what the number means and where the missing percent go; this lesson is about getting a number you can sign.
A test is a boundary and a window
Efficiency is not a property of the boiler; it is a property of a measured boundary over a stated window. Decide both before anyone touches an instrument. The boundary: fuel and air in, steam out, flue and blowdown and radiation out — and everything that crosses it during the window gets measured or deliberately held at zero. The window: long enough to smooth the plant's twitching (two steady hours is the honest minimum, after an hour of stabilisation you discard), at the load the boiler actually runs — not the flattering full-fire hour, and not the 30% crawl where radiation loss doubles its share (B4 explains why). Everything that follows is bookkeeping across that fence.
The direct method, run honestly
The arithmetic is one line — steam × heat gained, over fuel × heat content (B4). The test is everything around the line. Fuel goes over a platform scale into a marked bunker — never scooped loose off the pile mid-test. The steam integrator is read at start and end (no steam meter? run the same window on the feedwater meter with blowdown shut — D8 explains why that is often the more honest meter). Feed temperature is a thermometer reading, not a recollection. And every fifteen minutes someone writes down integrator, feed °C, stack °C, flue O₂, drum pressure and drum level — steadiness is data, not an impression.
Sixty-nine per cent, on a boiler the plant believes runs at 75. Before anyone orders a new boiler, cross-examine the test day itself.
The four ways a test day lies
Lie one — the fuel was not the fuel on the certificate. The 3,200 kcal/kg is the dry-season figure at ~10% moisture; the composite sample bagged during the test came back 15% wet. As-fired, that husk was nearer 2,950 kcal/kg — the boiler was being charged for heat the monsoon had already stolen. Rerun: 10,000 × 574 ÷ (2,600 × 2,950) ≈ 74.8%. Six points of "inefficiency" were water in the fuel. This is why the sealed half-hourly samples matter more than any instrument on the list.
Lie two — the drum lent you steam. The level bookmarks say the drum fell 25 mm across the window. On a shell with ~12 m² of water surface that is roughly 0.3 m³ — some 300 kg of steam that came out of storage, not out of the window's fuel. Honest output: 9,700 kg. Rerun: ≈ 72.6%. (A rising level cheats the other way, hiding steam the fuel really made. Either way the bookmark converts the drum from a liar into a line item.)
Lie three — the blowdown valve was passing. Found warm at the drain during the window. Heat left the boundary as hot water down a drain — real fuel, never steam — so the direct method silently charges it to the boiler. Shut it for the window or measure it; a passing valve is B6's problem wearing the test's costume.
Lie four — the "steady" window was not. A firing-rate swing or a one-point flue sample from a stratified duct puts the analyser's O₂ somewhere the average never was. The fifteen-minute log is the defence: if the numbers wander, the window moves or widens — it does not get averaged into respectability.
The indirect method as cross-examination
Same day, same window, B4's loss ledger from the flue log: O₂ 8% (excess air ~60%), stack 210 °C → dry flue ≈ 12.5%, moisture & hydrogen (wet husk) ≈ 9%, radiation ≈ 1.5%, unburnt ≈ 1%, CO ≈ 0.5% — efficiency ≈ 75.5%.
The two methods measure different truths — output over input versus one-hundred- minus-losses — so agreement within a point or two is the credibility check, and the residual gap is diagnostic, not embarrassing: here it pointed straight back at the blowdown valve. The plant's folklore number of 75 survives, barely; what changed is that it now has a fence, a window and a repair list attached.
Evaporation ratio — the every-shift proxy
Between tests, track the number that needs no boundary drawn: kg of steam per kg of fuel. This window gave 10,000 ÷ 2,600 ≈ 3.85, against the plant's dry-season ≈ 4.2 (5,000 kg/h on ~1,200 kg/h of husk). Typical bands, each at its own GCV and feed temperature: husk 4–4.5, Indian coal 5–6, furnace oil 13–14, natural gas 12–13. The ratio moves with fuel moisture and feed temperature as much as with the boiler — which is precisely its value: a falling ratio is a question, and a quarterly test is how the question gets answered. Log it every shift; it costs two readings you already take.
- The evening before: zero and span the flue analyser, stage weighed fuel in a marked bunker, brief the shift — no blowdown, no soot-blowing, no fuel-pile scooping inside the window (or log each event with its time).
- Bag a sealed fuel sample every 30 minutes; send the composite for moisture and GCV. The lab result moves the answer more than any instrument on the test.
- Bookmark drum level and pressure at start and end — the drum is either a measured line item or a liar; there is no third option.
- Log the six numbers every 15 minutes: steam integrator, feed °C, stack °C, flue O₂, drum pressure, drum level. A wandering log means move the window, not average it.
- Report a band with its basis stapled on: GCV as-fired, feed temperature, load, window. A decimal without a basis is an opinion in fancy dress.
- Efficiency belongs to a boundary and a window — two steady hours, representative load, one settling hour discarded.
- The four lies: wet fuel charged at dry GCV · the drum lending or hiding steam · blowdown passing · a window that only looked steady.
- Wet husk at dry-season GCV cost this test six phantom points — sample the fuel, always.
- Direct = output over input; indirect = 100 − losses. A 1–3 point gap is diagnostic; here it re-arrested the blowdown valve.
- Evaporation ratio (steam ÷ fuel, ≈ 4.2 for the practice plant) is the free every-shift trend; the quarterly test is its examiner.
Boiler testing was born competitive. In 1811, Cornish mine engineers began publishing each pumping engine's "duty" — water lifted per bushel of coal — in a monthly circular, league table and all. Duties climbed for decades, and so did the gamesmanship: trials run on hand-picked coal, stokers firing with unnatural care while the observer watched, engines nursed for the test and flogged the rest of the month. Two centuries later the tricks have not changed — only the units. Every rule in this lesson, from sealed fuel samples to the discarded first hour, exists because some Cornish engineer's grandfather already found the loophole.
FAQ
How long should a boiler efficiency test run?
Two steady hours of logged data is the honest minimum for a shell boiler, after about an hour of stabilisation you discard — storage effects and a twitching integrator dominate anything shorter. Run it at the load the boiler actually carries, and if production swings, pick the representative window rather than the flattering one. Longer is better arithmetic; steadier is better still.
My direct and indirect results disagree by three points — which one is right?
Possibly both — they measure different boundaries. First check the usual suspects on the direct side: fuel moisture versus the GCV you used, drum-level movement, blowdown or soot-blowing inside the window, meter calibration. The indirect side is blind to blowdown and leaks but sees the flue honestly. When the gap survives that audit, report the band between them with the basis stated — and chase the gap, because it usually names a specific valve or meter (B4 holds the ledger).
Can I run the test at low load if production is slow?
You can, but label it. Radiation and convection loss is roughly constant in kilocalories, so at 30% fire its percentage share triples, and a husk bed at low fire burns unevenly besides — the result is a true number for a condition the boiler should not be living in. Test at the load the plant actually runs; if that load is chronically 30%, the finding is right-sizing, not tuning (B4's point).
What is the minimum instrument kit for a credible test?
A platform scale for fuel, the steam integrator or the feedwater meter (with blowdown shut — often the more honest route, D8), a thermometer in the feed line, a flue-gas analyser with a stack thermocouple, and a wristwatch. Add sealed sample bags for the fuel and you can defend the result; skip the samples and the monsoon owns your answer.
The vendor guaranteed 82% but my best test says 75. Was the guarantee false?
Probably not — it was true somewhere else: a new boiler, dry reference fuel, full steady load, and very often NCV arithmetic, which scores several points higher than GCV with nothing changed (B4's trap). Ask for the guarantee's basis sheet and restate it on yours. The gap that remains after the bases match is real, and the indirect method's ledger tells you which knob it lives behind.
Check yourself
Five quick questions on this lesson. No marks, no records — the score is for you.