RECDs for Data Centres: Integrating Retrofit Work with Redundancy Testing

Data centre teams do not change critical infrastructure on a whim. Every change goes through a written Method of Procedure (MOP), a change advisory review, a rollback plan and a set of acceptance tests. That discipline is exactly what an RECD retrofit needs, and it creates an opportunity many operators miss.

Fitting a Retrofit Emission Control Device takes each generator out of service for a period, then asks you to prove it still performs. Your facility already takes generators out of service and proves they perform, through maintenance windows, load bank tests and integrated systems tests. If you plan the retrofit as part of that testing calendar rather than as a separate project, you save fuel, run hours, customer notices and risk.

This article is laid out the way an operations team would lay out a MOP. Each section maps to a part of the change document, so you can lift the structure straight into your own template.

Short answer: Retrofit one generator at a time, only when the remaining sets still meet your design redundancy (or when a temporary set restores it). Take a baseline load bank reading before fitting, repeat it after fitting, and schedule the final integrated test only when every set in the block has been done.

Section 1: Purpose and Scope

Why data centres are in scope at all

Most data centre generators are large. In Delhi NCR, CAQM’s Direction No. 76 points sets above 800 kW (about 1,000 kVA at 0.8 power factor) towards an RECD, and gives sets between 125 kW and 800 kW a choice between an RECD and dual fuel. Under the Graded Response Action Plan, running an uncontrolled diesel set during Stage II and above becomes difficult; our GRAP guide for DG sets covers the stages. Outside NCR, your State Pollution Control Board consent sets the terms.

Data centres also have a reputational reason. Enterprise and hyperscale customers increasingly ask about on-site emissions in due diligence questionnaires. “Our generators rarely run” is a weaker answer than “every set carries a certified emission control device.”

What is changing

An RECD is installed in the exhaust line, usually after the silencer or in place of a straight section of pipe. The engine, alternator, controller and switchgear are not modified. That keeps the change narrow, which is good news for your change board.

What is out of scope

New sets bought to CPCB IV+ norms already meet stricter limits; see CPCB II vs III vs IV+ explained. If some of your fleet is due for replacement within two or three years anyway, run the numbers in our RECD vs new CPCB IV+ genset cost comparison before retrofitting sets you will soon retire. If you are building a new hall, see ordering an RECD before the facility is ready.

Section 2: Redundancy Position During the Change

This is the section your change board will read first. The retrofit reduces available generator capacity by one set for the duration of the work. What that means depends on your topology.

Concurrent maintainability is being tested for real

The Uptime Institute’s Tier Classification describes Tier III facilities as concurrently maintainable: any component can be taken out for maintenance without shutting down IT. An RECD retrofit is a real-world test of that claim. If taking one generator out makes your operations team nervous, the retrofit has already told you something useful.

Redundancy with one set in retrofit

Design	Example	During retrofit of one set	Recommended posture
N	2 sets, both needed at peak	Capacity below peak	Temporary set is mandatory, or schedule only when IT load is well below N-1 capacity with written risk acceptance
N+1	4 sets, 3 needed	Drops to N: no spare	Accept a short, controlled window with extra monitoring, or hire a temporary set to restore N+1
N+2	5 sets, 3 needed	Drops to N+1	Normal maintenance posture; proceed
2N	2 × 3 sets, each side carries full load	One side at reduced capacity, other side intact	Work on one side only; never both sides in the same window
Distributed or block redundant	Several blocks sharing a reserve	Depends on reserve block	Treat the reserve assignment as part of the MOP; confirm the reserve is not already committed

Check the rating, not just the count

Redundancy maths only works if you know what each set can deliver continuously. Data centre generators may carry standby, prime or data centre continuous ratings, and the difference matters when fewer sets share the load. The Uptime Institute’s technical paper on engine-generator ratings is a useful refresher. Our article on load factor and duty cycle explains the same ideas in simpler terms.

Also confirm with the vendor that the fitted device will not reduce the output you rely on. A correctly sized RECD should keep back pressure within the engine maker’s limit; our pieces on whether an RECD changes usable power output and how back pressure affects a DG set show how to check.

Section 3: The Testing Link Most Operators Miss

Your generators already get tested. Use those runs.

Many Indian data centres follow manufacturer schedules and borrow practice from the US standard NFPA 110. As summarised by Curtis Power Solutions, that standard calls for monthly exercise of at least 30 minutes at not less than 30 percent of nameplate kW or at the manufacturer’s minimum exhaust temperature, with supplementary load bank runs at 50 and 75 percent where monthly loading is too low. Whatever your exact regime, there are already scheduled occasions when each set runs under controlled load with engineers watching.

Plan the retrofit around three of them:

1. Baseline run (before fitting). Run each set on a load bank at 25, 50, 75 and 100 percent steps. Record exhaust back pressure, exhaust temperature, fuel rate, smoke and any alarms at each step. This is your “before” photograph.
2. Acceptance run (after fitting). Repeat the same steps on the same set. Compare back pressure and temperature at each step. The difference is the device’s real effect on your engine, measured, not assumed.
3. Integrated test (after the whole block is done). Once every set in a redundancy group is fitted, run your normal pull-the-plug or integrated systems test so that synchronising, load sharing and transfer timing are proved with all devices in place.

Combining these with tests you were going to do anyway avoids extra run hours and extra fuel, which matters during GRAP months when every run of a diesel set should be justified.

Why load bank testing matters more after an RECD

RECDs that use a catalysed particulate filter generally clean themselves by burning collected soot when the exhaust is hot enough for long enough. The Electrical Generating Systems Association’s handout on load banks with DPFs notes that passive filters typically need loads of roughly half of generator capacity or more to regenerate continuously, and that where the real emergency load is lower, a load bank should be used. It also advises stepping load up in small increments while watching back pressure, never forcing high load onto a heavily loaded filter.

This is where data centres differ from factories. A factory set works hard every day. A data centre set might run 30 minutes a month at partial load and a few hours a year on a load bank. Soot builds during the light runs and is cleared during the hot ones. So the retrofit should come with a written test regime:

• Monthly runs held long enough, and loaded enough, to reach sustained exhaust temperature
• A load bank session at least once or twice a year with gradual loading
• Back pressure logged at every run, with an alert threshold agreed with the vendor

What to record in the baseline and acceptance runs

Use the same sheet for both runs so the comparison is automatic. A simple layout:

Load step	Back pressure	Exhaust temperature	Fuel rate	Frequency and voltage stable?	Visible smoke	Alarms
25%
50%
75%
100%

Hold each step long enough for readings to settle, usually several minutes, and note ambient temperature on the sheet. A hot May afternoon and a cool February morning can give different exhaust temperatures for the same load, so like-for-like conditions make the comparison fairer.

Check fuel and engine readiness first

Data centre fuel often sits in tanks for months. Aged or contaminated diesel burns less cleanly, and a sooty engine sends more particulate matter into a new filter from its first hour. Before the baseline run:

• Sample the stored fuel for water and sediment; polish or replace it if your policy calls for it
• Replace air filters that are near their limit
• Check injectors on any set with a history of smoke on start
• Review oil consumption; an engine that burns oil adds ash that no self-cleaning process removes

Fixing these first protects the device and gives you a baseline that reflects a healthy engine rather than a tired one.

Background reading: what a diesel particulate filter is and how it helps, RECDs on generators that run only occasionally and the RECD maintenance guide. If a vendor calls the device “maintenance free”, read what that claim actually means before you accept it in your contract.

Section 4: Pre-Requisites

Complete all items before the change window is approved.

• Site survey done for every set, including exhaust route, stack height, support steel and crane access
• Device size and model confirmed per set; see typical sizing on the 2000 kVA and 2500 kVA RECD pages
• Engine maker’s back pressure limit obtained in writing
• Certification documents checked; our guide on verifying a CPCB-tested RECD lists what to ask for
• Baseline load bank data recorded for each set
• Alarm integration agreed: which device alarms (high back pressure, fault, communication loss) go to your BMS or EPMS, and in what format. Ask the vendor what outputs are available rather than assuming.
• Temporary generator arranged if Section 2 requires one, with its connection point tested
• Fuel stock confirmed above your policy minimum for the remaining sets
• Customer notification sent where contracts require notice of maintenance on critical systems
• Change approved by the change advisory board

Special cases worth checking in the survey

• Shared or common stacks. Some older halls merge exhausts into a common flue. An RECD is sized per engine; read whether two generators can share one RECD before anyone suggests a shortcut.
• Sets that already have aftertreatment. Newer engines may carry a DPF or SCR. See adding an RECD to a set with existing DPF or SCR.
• Containerised sets. Space inside the container or on its roof is often the constraint. Confirm whether the device goes inside, on the roof or on a separate frame, and check the container’s structural rating.

Section 5: Risk Register

Risk	Likelihood	Impact	Mitigation
Utility failure while a set is out of service	Low to medium	High	Remaining sets verified on load within 7 days; temporary set if below design redundancy; UPS autonomy checked
A second set fails during the window	Low	Very high	Pre-window test of all remaining sets; spares for common failure items on site; abort rule defined
Excessive back pressure after fitting	Low if sized correctly	Medium	Baseline vs acceptance comparison; vendor to correct before set returns to duty
Exhaust leak inside a generator room or container	Low	Medium	Leak check at idle and at load; CO monitoring during first runs
Fire risk from hot work	Low	High	Hot work permit; fuel lines protected; fire watch
Device alarm not reaching BMS	Medium	Medium	Point-to-point alarm test during acceptance
Soot loading from light monthly runs	Medium	Medium	Revised test regime; back pressure trending; planned load bank sessions

Section 6: Step Sequence With Hold Points

This is a typical sequence for one set. Adjust to your site.

Step	Action	Owner	Verification	Hold point?
1	Confirm remaining sets are available and on auto	Shift engineer	Panel status screenshot	Yes
2	Place the target set in manual, open its breaker and lock out	Electrical lead	Lock-out tag and photo	Yes
3	Isolate batteries and fuel supply to the target set as per site rules	DG technician	Checklist signed	No
4	Remove the exhaust section or fit the device at the agreed point	RECD vendor	Visual check by site engineer	No
5	Mount supports and connect device; route sensor and communication cables	RECD vendor	Photos; torque record	No
6	Restore fuel and batteries; run at idle; check for leaks	DG technician	Leak check record	Yes
7	Load bank acceptance run in steps	Test engineer	Back pressure and temperature vs baseline	Yes
8	Verify alarms at BMS or EPMS	Controls engineer	Alarm test log	Yes
9	Return set to auto; close breaker	Electrical lead	Panel status	Yes
10	Close change ticket after 24 hours of normal standby	Change manager	Ticket closed	No

A hold point means the next step does not begin until the named person signs off.

Section 7: Rollback Plan

Every MOP needs one. For an RECD retrofit, rollback means returning the set to its original exhaust configuration.

• Keep the removed pipe section, gaskets and fasteners labelled and stored next to the set until the change ticket closes.
• Agree in advance how long a reversal would take. If the answer is longer than your risk appetite for that set being out, schedule more crew or pre-fabricate a spool piece.
• If a utility failure happens mid-job, the set under work stays isolated. It is not started with a partially connected exhaust. The remaining sets carry the load and the incident procedure takes over.

Section 8: Acceptance Criteria

The set is accepted back into service only when all of these hold:

• Back pressure at each load step is within the engine maker’s limit and within the margin agreed with the vendor
• Exhaust temperature rise compared with baseline is within the expected range
• No visible smoke at steady load; no exhaust leaks at any joint
• Rated load reached and held for the agreed period
• Device display and remote monitoring show normal status
• Alarms confirmed at the BMS or EPMS
• Start and transfer time, where tested, matches the pre-retrofit record
• Documents filed: test certificate, device serial number, installation report and photographs

Section 9: Fleet Sequencing

Start with a pilot

Fit one set first, ideally the one with the easiest redundancy position. Run it through at least two scheduled tests before continuing. The pilot shows you how long the work really takes on your site and whether your test regime keeps back pressure stable.

Then wave through the fleet

Example: 2N design, eight sets (four per side). Retrofit one set per week, alternating sides: A1, B1, A2, B2 and so on. At no point are both sides reduced together. After the last set, run the integrated test.

Example: N+1 block with five sets, four needed. Each retrofit drops the block to N. The operator hires a temporary 2000 kVA set for the whole campaign and connects it to a pre-built tap-off, restoring N+1 throughout. The hire cost is set against avoided risk and against running the campaign faster, two sets per week instead of one.

Example: Edge site, two 500 kVA sets in N+1. IT load is 220 kW. With one set out, the other carries the load with margin. The operator proceeds without a rental but restricts work to a single day per set and places the remaining set’s technician on call. Typical sizing is on the 500 kVA RECD page.

Avoid these calendar clashes

• Your peak season for IT load or customer migrations
• Utility substation maintenance announced by the distribution company
• The October to January smog months in NCR, when running diesel sets for testing faces the most restriction
• Year-end change freezes that customers often request

What the pilot usually reveals

Operators who run a pilot set first tend to learn the same handful of things:

• Access takes longer than the fitting. Moving a device through a loading bay, up a service lift or onto a container roof is often the slowest part. The second set goes faster once the route is known.
• Alarm mapping needs a second pass. Point names in the BMS rarely match the vendor’s labels on the first attempt.
• Baseline data has gaps. Someone forgets to log ambient temperature or fuel rate. Fix the sheet before the next set.
• The test regime needs tuning. After a month of monthly exercises, the back pressure trend shows whether your runs are hot enough or whether a longer loaded run is needed.

Update the MOP after the pilot. The version you use for set number five should be better than the one you wrote for set number one.

Section 10: Communication and Sign-Off

Internal

• Change ticket raised with this MOP attached
• Shift handover notes updated each day of the campaign
• Operations dashboard shows which sets are fitted, pending or under work

External

• Customer notices where your service agreements require them. Keep the language factual: which system, what window, what redundancy remains.
• Pollution control board filings as your consent requires, with device documents kept ready for inspection

Sign-off

Approvals from the facility head, the critical environment manager, the electrical lead and the vendor’s project manager close the MOP. File the baseline and acceptance data together; they are your evidence the retrofit did not weaken resilience.

Frequently Asked Questions

Will an RECD affect generator start time or transfer?

The device sits in the exhaust and does not act on the starter, controller or switchgear. What can affect performance is back pressure beyond the engine maker’s limit, which is why the acceptance run compares against the baseline at every load step.

Can we fit the RECD during our annual load bank test?

You can schedule them together: the baseline run in the days before fitting and the acceptance run as part of the load bank session afterwards. Doing both on the same day is possible for some sites but leaves little room if anything needs adjusting.

Our sets run less than 20 hours a year. Will the device clog?

Low hours alone are not the problem. Low load is. If most runs are short and light, soot can build faster than it burns off. A test regime with longer, loaded runs and periodic load bank sessions keeps the device in its comfort zone.

Do we need a separate RECD for each generator?

Yes, in normal practice each engine gets its own device sized for its exhaust flow. Shared devices across paralleled engines raise flow, back pressure and backflow concerns.

How do we show customers the retrofit did not reduce resilience?

Share the before and after load bank data for each set, the completed MOP and the integrated test report. Numbers carry more weight than a vendor brochure.

Treat the Retrofit as a Change, Not a Purchase

Data centre operators already have the tools to do this well: MOPs, redundancy maps, load bank schedules and change boards. The RECD retrofit simply joins that system. Plan it around your tests, measure before and after, and never let the campaign reduce you below design redundancy without a written decision.

For sizing help across a fleet, or a per-set schedule that fits your test calendar, share your generator list via our RECD product page or contact the Aceget team.