How to Compare RECD Test Results from Different Laboratories

Sooner or later, most organisations with a Retrofit Emission Control Device end up holding two test reports from two different laboratories that seem to disagree. The commissioning lab said 76%. The lab hired for the annual consent renewal measured a downstream concentration that, set against the old baseline, implies 64%. A buyer’s consultant arrives with a third figure.

The natural reaction is to ask which lab is right. Usually that’s the wrong first question. More often both labs measured correctly, but they measured different things, under different conditions, and reported them on different bases.

This guide sets out a step-by-step way to compare RECD test results from different laboratories: what to check first, how to put the numbers on a common footing, how to account for measurement uncertainty, and when to stop comparing and order a joint test instead.

Step 0: Collect Everything, Not Just the Summary Page

A fair comparison is impossible from two summary tables. Before analysing anything, gather from each lab:

• The full test report, including annexes
• The sampling method and any deviations from it
• Raw data sheets: filter masses, sample volumes, gas readings, temperatures, pressures
• Engine operating data during each sample: load in kW, power factor, exhaust temperature
• The calibration annex (see which instrument calibration records should accompany RECD tests)
• The lab’s accreditation scope, showing that stack emission monitoring and each parameter were covered

If one lab can’t provide raw data, note that. Its result can still be considered, but it can’t be recalculated on a different basis, which limits how far the comparison can go.

Step 1: Are Both Reports Answering the Same Question?

Labs are hired for different purposes. Three common ones are:

Purpose	Typical result	What it tells you
Type approval of the RECD model	% PM reduction over a weighted multi-mode cycle on a test engine	Whether the design meets the requirement
Site acceptance of the installed device	% reduction before vs after, or upstream vs downstream, at site load points	Whether the device works on your engine
Routine stack monitoring for consent	Concentration at the stack, compared with a consent limit	Whether the stack currently complies

Comparing a type-approval percentage with a site-acceptance percentage, or a routine stack concentration with an acceptance baseline, mixes different questions. That doesn’t make the comparison worthless, but it changes what you can conclude. Our guides on how to read an RECD test certificate and factory acceptance vs site acceptance tests explain the differences in more detail.

Stop here if: one report is a type-approval cycle result and the other is a single site load point. Treat them as separate pieces of evidence, not as two measurements of the same thing.

Step 2: Did They Measure the Same Thing in the Same Way?

Next, line up the methods. Differences in any of these can explain a gap on their own:

Item	Lab A	Lab B	Why it matters
Particulate method	Gravimetric, isokinetic?	Gravimetric, isokinetic?	Opacity or optical methods aren’t comparable with mass
Raw or diluted sampling			Dilution ratio errors and condensation differ
Probe and filter temperature			Changes how much condensable material is collected
Sampling port location			Representativeness and leaks; see exhaust sampling location
Sample duration and number of runs			Short single runs carry more scatter
Comparison design	Sequential / bypass / upstream-downstream		Same-day designs remove many variables

Fill in a table like this for your two reports. Differences in method aren’t failures, but each one is a possible explanation for the gap.

Step 3: Were the Engine and Device in the Same State?

Even with identical methods, the engine and the RECD may not have been in comparable conditions. Check:

• Load in kW at each sample, not kVA or “approximate percentage.” See why generator load must be matched in RECD comparison tests.
• Thermal stability before sampling. See how to account for warm-up time when testing an RECD.
• Regeneration status and differential pressure across the device. See RECD testing during regeneration.
• Engine condition and fuel at each test. See how fuel and engine condition can confound RECD performance tests.
• Device age and operating hours since installation.

A test at commissioning and a test eighteen months later are measuring a device at different points in its life. Some change is expected, and that is useful information. It only becomes a problem when it is mistaken for a disagreement between labs.

Step 4: Put Both Results on the Same Basis

This is where most apparent disagreements disappear. Concentrations can be reported in many ways, and two labs often choose differently. Before comparing numbers, convert both to a single agreed basis.

The four basis questions

1. Wet or dry? Diesel exhaust contains several percent water vapour. Dry-basis concentrations are higher than wet-basis ones for the same gas.
2. Reference oxygen? Is the concentration at actual O₂, or corrected to a reference such as 15%? At what reference?
3. Reference temperature and pressure? “Normal” can mean 0 °C or 25 °C, depending on the lab and method.
4. Concentration or mass rate? mg/Nm³ depends on dilution. g/h and g/kWh don’t, provided flow is measured properly. See RECD emissions data in mg/Nm³ vs g/kWh.

The normalising formulas

To convert a wet concentration to dry:

C_dry = C_wet ÷ (1 − moisture fraction)

To correct to a reference oxygen level:

C_ref = C_measured × (20.9 − O₂_ref) ÷ (20.9 − O₂_measured)

To convert between 0 °C and 25 °C reference temperature (at the same pressure):

C_at_0°C = C_at_25°C × (298.15 ÷ 273.15)

That last conversion changes a concentration by about 9%, enough on its own to make two correct reports look inconsistent.

Worked Example: Two Labs, One RECD

The figures below are illustrative.

Lab A (commissioning, January) ran a sequential test at 60% load and reported:

• Baseline PM: 85 mg/Nm³, dry, corrected to 15% O₂, 0 °C
• Post-RECD PM: 21 mg/Nm³, same basis
• Reduction: (85 − 21) ÷ 85 = 75.3%
• Stated uncertainty: ±15% on each result

Lab B (consent monitoring, July) sampled the downstream stack only, at whatever load the building needed, and reported:

• PM: 18 mg/Nm³, wet basis, at actual O₂ of 13.0%, 0 °C
• Moisture: 8%
• Load: about 40%
• Stated uncertainty: ±20%

The facility manager compares Lab B’s 18 against Lab A’s baseline of 85 and gets a “reduction” of 78.8%. A consultant compares it with Lab A’s post-RECD figure of 21 and concludes the device has improved. Both comparisons are invalid, because the bases differ.

Normalise Lab B to Lab A’s basis:

1. Wet to dry: 18 ÷ (1 − 0.08) = 18 ÷ 0.92 = 19.57 mg/Nm³ dry
2. Correct to 15% O₂: 19.57 × (20.9 − 15) ÷ (20.9 − 13.0) = 19.57 × 5.9 ÷ 7.9 = 19.57 × 0.747 = 14.6 mg/Nm³

On a common basis, Lab B’s downstream result is 14.6 mg/Nm³, against Lab A’s post-RECD 21 mg/Nm³.

Is the difference real? One way to judge is the Eₙ number, used widely in proficiency testing between laboratories. It compares the difference between two results with their combined expanded uncertainty:

Eₙ = (x_A − x_B) ÷ √(U_A² + U_B²)

• U_A = 15% of 21 = 3.15 mg/Nm³
• U_B = 20% of 14.6 = 2.92 mg/Nm³
• Eₙ = (21 − 14.6) ÷ √(3.15² + 2.92²) = 6.4 ÷ √(9.92 + 8.53) = 6.4 ÷ 4.30 = 1.49

A common rule of thumb is that |Eₙ| ≤ 1 means two results agree within their stated uncertainties, and |Eₙ| > 1 means the difference is larger than the uncertainties explain. Here it is 1.49, so the difference is probably real.

But is it a lab disagreement? Step 3 says no. Lab B tested at about 40% load, and Lab A at 60%. Engine-out particulate, exhaust temperature and device efficiency all change with load. Lab B also didn’t measure a baseline, so it can’t produce a reduction percentage at all. The most that can be said is that downstream concentration at 40% load in July was lower than at 60% load in January. That is plausible for many engines and says nothing bad about either lab.

Conclusion: Neither report contradicts the other. Lab B’s report can’t be used to recalculate the RECD’s reduction. If a current reduction figure is needed, a new test must measure upstream and downstream at matched load.

Step 5: Use Uncertainty Properly

Every measurement carries uncertainty, and a competent lab should report it. For gravimetric particulate at stack level, combined uncertainties of 10–20% or more are not unusual, especially at low concentrations.

Some practical rules:

• Don’t compare single numbers as if they were exact. Compare ranges.
• Uncertainty grows for the reduction percentage. Because the reduction combines two uncertain measurements, its own uncertainty is wider than either alone. Our guide on calculating RECD particulate reduction shows a worst-case approach.
• Ask how the uncertainty was estimated. Was it a full uncertainty budget following the approach in the internationally used Guide to the Expression of Uncertainty in Measurement (GUM), or a generic figure?
• Use Eₙ or overlapping ranges to decide whether two results disagree, instead of looking at the raw difference.

If a report gives no uncertainty at all, ask for it. Without it, there is no objective way to say whether a difference between labs is meaningful.

Step 6: Look for Systematic Differences Between the Labs

If differences persist after normalising and allowing for uncertainty, look for systematic causes. These are the most common:

Possible cause	How to check
Different calibration states	Compare calibration annexes; look for expired or out-of-range certificates
Different filter media or weighing practice	Ask about filter type, conditioning and blanks
Different probe or filter temperature	Check method settings; affects condensable PM
Leaks or dilution at one lab’s port	Compare O₂/CO₂ at the sampling point with engine-out values
Different isokinetic performance	Check reported isokinetic ratios for each run
Single run vs multiple runs	Check run-to-run spread; a single run can be an outlier
Different averaging (time or mode weighting)	Check whether results are per mode or weighted

A lab that participates in proficiency testing schemes, where several labs measure the same sample and compare results, has evidence about its own systematic bias. Ask each lab whether it takes part in such schemes for stack emissions, and how it performed.

Compare Reductions, Not Just Concentrations

If both labs ran their own complete before-and-after (or upstream/downstream) test, compare the reduction percentages they each calculated, as well as the individual concentrations.

A reduction is a ratio of two measurements made by the same lab, with the same instruments, on the same basis. Many systematic biases that differ between labs, such as a slightly different filter temperature, a balance offset or a different reference condition, affect both of a lab’s own measurements in a similar way and largely cancel in the ratio.

For example, with illustrative figures:

	Before (mg/Nm³)	After (mg/Nm³)	Reduction
Lab A	80	20	75.0%
Lab B	92	25	72.8%

Lab B’s concentrations are consistently about 15–25% higher than Lab A’s, perhaps because of a different basis or a systematic method difference. Comparing Lab B’s “after” with Lab A’s “before” would give a misleading 68.8%. But each lab’s own reduction is within about two points of the other, which is well within normal measurement scatter. The device performance is consistent. Only the absolute levels differ.

This is why the most useful comparison between two complete tests is usually reduction against reduction, with load and conditions matched, and why mixing one lab’s baseline with another lab’s post-RECD figure should be avoided.

Red Flags That One Report Is Less Reliable

Sometimes the comparison shows that one report simply carries less weight. Common warning signs:

• No raw data or filter masses, only final results
• Particulate measured by opacity or an optical instrument and presented as mass
• No O₂ or CO₂ at the sampling point, so dilution can’t be checked
• No load record, or load only in kVA
• No calibration annex, or expired certificates
• A single short run at each load point, with no repeat
• No stated uncertainty
• Accreditation scope that doesn’t cover stack emission monitoring or the parameters reported

A report with several of these isn’t necessarily wrong, but it can’t carry equal weight with a fully documented one.

Step 7: Decide What to Do Next

After working through the steps, most comparisons end in one of four places:

1. Results agree once normalised. Record the normalised comparison and file it with both reports.
2. Results differ, but for a known operational reason (load, device age, regeneration). Note the reason. If a current performance figure is needed, plan a properly controlled retest.
3. Results differ, and the cause looks like method or instruments. Ask both labs for an explanation, share the raw data, and consider a joint test.
4. Results can’t be compared because one report lacks the information needed. Treat it as a limited piece of evidence and don’t base decisions on the comparison.

When to Commission a Joint or Split Test

If the stakes are high, for example a disputed acceptance payment, a regulatory question or a supplier warranty claim, the cleanest way forward is a joint test:

• Both labs sample on the same day, at the same load points, using ports that are as close as possible to equivalent
• Or, where port access allows, a split arrangement in which each lab samples at its own port alternately while the engine is held steady
• An agreed protocol fixes load, stability criteria, regeneration status, basis and reporting format beforehand
• Both parties witness the test (see how to set up a witnessed RECD performance demonstration)

If two competent labs, sampling the same exhaust under the same protocol, still disagree beyond their uncertainties, the cause will almost certainly be in method or instruments. That is far easier to find when everything else is fixed.

A Comparison Template You Can Reuse

Field	Lab A	Lab B
Test purpose
Date and device operating hours
Comparison design
Particulate method and sampling train temperature
Load at each sample (kW, PF)
Thermal stability evidence
Regeneration status / ΔP
Fuel and engine condition notes
Reported value, unit and basis
Normalised value (agreed basis)
Expanded uncertainty
Eₙ (or range overlap)
Calibration annex complete?
Accreditation scope covers parameter?
Conclusion

Keep the completed template with both reports. It makes later reviews, for example during a Pollution Control Board inspection or a facility sale, much quicker.

Frequently Asked Questions

Which result should I trust when two labs disagree?

Neither, until you’ve checked that they measured the same thing on the same basis under the same conditions. Most apparent disagreements come from differences in load, basis or test purpose. If a genuine disagreement remains, the lab with fuller raw data, complete calibration records and stated uncertainty gives you more to rely on.

Can I compare a lab’s downstream-only result against another lab’s baseline?

Only with great caution. Unless load, engine state, fuel and basis all match, the “reduction” you calculate mixes the device’s effect with everything else that changed. A proper reduction needs a baseline and a post-RECD measurement taken under matched conditions.

What does it mean if a lab doesn’t report uncertainty?

It means you can’t judge objectively whether a difference from another result is significant. Ask the lab for an uncertainty estimate. Accredited labs working to ISO/IEC 17025 are expected to be able to estimate uncertainty for their tests.

Is a 5% difference between two labs a problem?

Usually not. Stack particulate measurements typically carry uncertainties well above 5%, so a 5% difference is generally within normal scatter. Use Eₙ or overlapping uncertainty ranges to judge.

Can I compare the RECD’s type-approval result with my own site result?

Only loosely. The type-approval figure is a weighted result over several load modes on a test engine in a laboratory. A site result is usually measured at one or two load points on your own engine, which may be older and dirtier. A site result a few points below the type-approval figure is common and doesn’t by itself mean anything is wrong. A large gap is worth investigating, starting with load, engine condition and the reporting basis. The type-approval figure tells you the design is capable. The site result tells you how it performs on your installation.

Should I always use the same lab to avoid this problem?

Using one lab for related tests reduces some sources of difference, but it doesn’t remove changes in load, device state or fuel. It also concentrates trust in a single lab. Specifying a consistent method and reporting basis matters more than which lab you choose.

Make Your Reports Comparable From the Start

The easiest comparison is one that was planned for. Specify the method, load points, reporting basis and uncertainty requirements in every test scope, whichever lab you use. Aceget supplies CPCB type-approved Retrofit Emission Control Devices for DG sets from 25 kVA to 2500 kVA. We help clients structure acceptance tests so later reports can be compared with them. Choose labs accredited by NABL for stack emission monitoring, review our certifications, and contact our team if you need help interpreting two reports.