- November 22, 2026
- Posted by: editor
- Category: RECD & Emission Control Devices
Pressure alarms get attention. A fall in pressure rarely does. When the differential pressure across a retrofit emission control device (RECD) climbs, everyone knows what to do: investigate, clean, regenerate. When it suddenly falls, the instinct is to relax. The trend line has dropped, the alarm has cleared and the filter looks healthier than it has in months. Someone writes “regeneration successful” in the log and moves on.
Sometimes that is exactly right. A good regeneration or a cleaning does lower the pressure drop, often dramatically. But a sudden fall in ΔP has other possible causes, and some of them are the opposite of good news. A filter that has cracked or lost its seal lets exhaust go around the material that is supposed to catch the soot, and its pressure drop falls precisely because it is no longer restricting the flow. A sensing tube that has come off or split reads low. A load that has dropped by a third gives a lower reading with nothing changed in the filter at all.
So the right response to a sudden fall is not delight and not alarm, but curiosity. This guide treats the fall as a puzzle with four candidate explanations. For each, we say what else must be true if it is the right one, which gives you a set of predictions to check against the data. We add a simple rule about the clean-filter baseline, a shape test using ΔP against load, a signature table for how the fall happened in time and a 48-hour containment plan for the cases where you cannot yet tell. The advice is general. Your supplier’s data and procedures take priority.
First, what makes pressure drop at all
The differential pressure across a filter is the pressure the gas loses in passing through it. In general it depends on how much gas flows, how hot and thick that gas is, how much material the filter holds and the geometry of the filter itself. When the number falls, one of four things has happened.
- Less gas is flowing through the filter. The load fell, or some of the gas is taking another route.
- The filter is holding less material. Soot has been burned off, or the filter has been cleaned.
- The route through the filter has become easier, because the filter media has been breached, a plug has been lost or a channel has opened.
- The measurement is wrong. A tap, tube, transmitter or setting has changed.
Numbers 1 and 2 can be benign. Number 3 is damage or bypass. Number 4 is a monitoring problem that may be hiding another. The task is to work out which you have, using the cheapest tests first. Because a damaged filter can keep running with no alarm at all, it pays to take a sudden fall as seriously as a rise.
The four hypotheses and what each predicts
For each hypothesis, ask: “If this were true, what else would I expect to see?” Then look. This is far more reliable than looking at the ΔP trace alone.
Hypothesis A: less flow (load or duty changed)
Story. The set is running at lower load or lower flow than before, so ΔP is lower at the same filter condition. This is the most common explanation of all, and the cheapest to test.
What else must be true.
- Load (kW) was lower when the lower reading was taken.
- ΔP at the same load as the earlier reading is unchanged.
- Exhaust temperature and engine data fit the lower load.
How to test. Compare readings at the same load band. Our guide on comparing RECD pressure readings at different generator loads explains how, and the article on using DG load logs shows how to find the load history. If the “fall” disappears when you compare like with like, the explanation is A.
Hypothesis B: the soot has gone (a real improvement)
Story. A regeneration (heater-assisted, passive or loaded run) or a cleaning has removed soot, so the filter is less restrictive.
What else must be true.
- There is a regeneration or cleaning event in the log just before or during the fall, and the event completed normally. See interpreting an RECD regeneration completion message.
- Temperatures rose during the event to the expected range.
- The fall is gradual or steps down during the cycle, as soot burns, and then holds.
- The new ΔP at a reference load is at or above the clean-filter baseline, not below it (see the rule below).
- Smoke and outlet appearance are unchanged or better.
How to test. Read the regeneration log and the temperature trace. Compare the new settled ΔP at your reference load with your record of ΔP after previous good regenerations and with the commissioning figure.
Hypothesis C: the gas is going around or through a breach (damage or bypass)
Story. Exhaust is finding an easier path than the filter walls. The substrate has cracked, melted or lost plugs, the mat that seals the core in the canister has failed, a seal or door has opened, a bypass damper is partly open, or a leak upstream of the inlet pressure tap is venting gas before it reaches the filter.
What else must be true.
- ΔP is lower than the clean-filter baseline at the same load, or lower than it should be for the soot you expect to be present.
- Soot, ash or ceramic dust appears on the outlet side: at the outlet face, in the downstream pipe, at the silencer or at the stack. A fleet-maintenance inspection article states that soot accumulation on the outlet side is a telltale sign that a filter is internally cracked (Fleet Maintenance). A workshop guide on DPF diagnosis says zero or abnormally low pressure should make you suspect a collapsed core or a bypass crack, to be confirmed by checking for excessive soot at the tailpipe (Snap-on). A diagnostic code reference for a low differential pressure fault likewise describes bypassing of the filter through a cracked or melted substrate, or through disconnected sensor lines, as the usual meaning (O Trucking code reference).
- Smoke may have increased or changed character. See black smoke after RECD installation.
- There may be a recent thermal event: a failed or interrupted regeneration, a very hot cycle, a load step during regeneration, an oil or fuel leak into the exhaust. See repeated failed regeneration and filter damage.
- There may be a recent mechanical event: lifting, impact, a service visit. See rattling inside an RECD.
- The ΔP-versus-load curve is flatter or lower in a way that does not fit soot loss (see the shape test below).
How to test. Inspect the outlet side when cold and safe, check damper positions and seals, compare ΔP with the baseline, arrange an opacity or particulate check and ask the supplier for an inspection. Do not open the canister yourself.
Hypothesis D: the measurement is wrong
Story. The reading fell because the instrument chain changed. A sensing tube has come off, split, been crushed or filled with water, a transmitter has failed (often toward zero), a tap has been reversed or a setting, scale or unit has been changed.
What else must be true.
- An independent gauge at the same taps disagrees with the controller.
- The controller trace may be unnaturally smooth, flat or stuck, or have lost its response to load changes.
- The change coincides with a service visit, a firmware update, a controller replacement or a cable or tube disturbance.
- There is no corresponding change in temperature, smoke, power or fuel consumption.
How to test. Use a calibrated portable gauge. Inspect the tubes. Compare the controller’s settings with an earlier export. See separating a sensor fault from an RECD performance problem, pressure sensors versus pressure switches and detecting missing data in RECD monitoring records.
The four at a glance
| Check | A: less flow | B: soot gone | C: breach or bypass | D: wrong reading |
|---|---|---|---|---|
| Load lower when it fell? | Yes | Not necessarily | Not necessarily | Not necessarily |
| ΔP lower at the same load? | No | Yes | Yes | Controller yes, gauge no |
| Regeneration or cleaning just before? | Not needed | Yes | Maybe (if thermal event) | Maybe (service visit) |
| Below clean baseline? | No | No | Often yes | Can be yes |
| Soot or dust at outlet side? | No | No | Often yes | No |
| Independent gauge agrees? | Yes | Yes | Yes | No |
| Smoke change? | No | No | Possible | No |
| Settles and holds near the baseline? | Follows load | Yes | Stays low, may keep falling | Unstable or flat |
The clean-baseline rule
Here is the most useful single test in this article. A filter cannot be less restrictive than a clean filter.
If a filter has been completely cleared of soot, its ΔP at a given flow should approach the value for a clean filter, which is higher in an aged filter than in a new one because ash and residual deposit remain. Your commissioning data gives the clean-filter figure at a reference load, and your history of good post-regeneration readings gives the typical “floor” for the aged filter.
- A post-regeneration ΔP near the floor is a good result.
- A ΔP well above the floor means the regeneration only partly worked.
- A ΔP below the floor cannot be explained by soot removal. Something else is going on: less flow (A), a breach (C) or a reading error (D).
The rule is not perfect. Temperature, instrument tolerance and the gas properties give some spread, and a first cleaning of a very ashy filter can lower the floor. But a drop that takes the reading clearly below the commissioning clean figure at a matching load is a red flag that deserves a test, not a celebration. Record the floor in your asset file, and see our guide to why an RECD keeps clogging for the rebuild rate and floor method.
The shape test: ΔP against load
A single reading at a single load says little. A curve says much more. ΔP across a filter generally rises with load, because flow rises with load. Plot ΔP against load, for the period before the fall and after it, using steady readings.
The following patterns are illustrative and based on general reasoning, so ask your supplier how your device should behave.
| What the new curve looks like against the old one | Suggests |
|---|---|
| Same shape, shifted down by a constant amount at every load | An instrument offset (zero shift) or a changed tap position (D) |
| Same shape, smaller in proportion (scaled down) | Less restriction, a real change such as soot burn-off (B) or a changed scale setting (D) |
| Much flatter, ΔP barely rises with load | A big share of the flow is going around the filter (C), or the sensor is stuck (D) |
| Matches the old curve once load is taken into account | Less flow (A) |
| Lower at high load but similar at low load | A bypass that opens with flow, or a partly open damper (C) |
| Noisy, stepping, unrelated to load | Sensing-line or electrical problem (D) |
To build the curve you need readings at several load steps, taken when steady. A load bank or a staged test makes this easy. See RECD commissioning with load bank testing and the staged-step method in our article on alarms during generator load steps. If you have a commissioning curve, compare to that.
The timing signature
How the fall happened in time adds information.
| How it fell | What it suggests |
|---|---|
| A step within seconds or minutes, at a known event | A tube coming off, a damper moving, a crack from a thermal event, a settings change |
| A smooth fall over the length of a regeneration cycle | Soot burning off (B) |
| A slow fall over days or weeks, flattening the curve | A developing leak or mat failure, or gradual passive regeneration |
| A fall that tracks load or weather | Flow (A) or moisture effects on the instrument |
| A fall followed by a rise to a higher level than before | A partial breach that has now been plugged by debris, or a heater or sensor issue |
| A fall that appears with another change (smoke, noise, smell) | Combine with the smoke, rattle and smell articles |
A step that coincides with a service visit deserves the first check, since disturbed tubes, taps and joints are a common source of sudden change. See auditing an RECD service report for missing information.
A fast checklist, cheapest first
- Same-load comparison. Is the fall still there at the same load? (A)
- Event check. Was there a regeneration, cleaning, service visit, setting change or thermal event? (B, C, D)
- Independent gauge. Does it agree with the controller? (D)
- Floor check. Is the new ΔP below the clean-filter baseline? (C or D)
- Outlet-side look. With the system cold, isolated and safe, inspect the outlet flange, downstream pipe and silencer inlet for soot, ceramic dust or fibres. (C)
- Smoke and smell. Any change at the stack? (C)
- Shape test. Build the ΔP-versus-load curve and compare it with the old. (A, B, C, D)
- Temperature trace. Did the cycle reach the right temperatures? Did a spike occur? (B, C)
- Damper and bypass. Position, actuator and limit switches. (C)
- Supplier review. Send the dataset and photographs. Ask for an inspection if any step points to C.
Checks 1 to 4 take under an hour, and they settle a surprising number of cases.
Forty-eight hours of containment
Sometimes you cannot tell yet. In that case, a short containment plan reduces risk without hiding the evidence.
Do
- Photograph the screen, the outlet side and the stack.
- Export or photograph the controller history and note the exact time of the fall.
- Keep running only as the supplier and your compliance position allow. If you suspect a breach, ask whether to continue loaded running, since a damaged filter may shed material downstream.
- Take a baseline of readings at three loads with an independent gauge.
- Retain an outlet-side deposit sample if one can be taken safely.
- Tell the supplier early.
Do not
- Run further regeneration cycles “to be sure”. A damaged filter can degrade further.
- Reset history or learned values. See RECD alarms that return after a reset.
- Bypass the filter unless the supplier’s procedure and your compliance position allow it.
- Strike, shake or open the canister.
- Assume the fall is a success because the alarm has cleared.
If the fall may affect compliance, see what to do during an RECD fault: business continuity.
Two worked illustrations
The following are illustrative scenarios, not real customers.
Scenario 1: the fall that was good news
A cold-storage facility’s 320 kVA set shows ΔP at 55% load falling from 47 mbar to 29 mbar over a 40-minute period. The regeneration log shows a heater-assisted cycle that ran at the same time, reached the expected bed temperature and ended with a normal completion message. Load was steady at 54 to 56% throughout. An independent gauge agrees with the controller within 1 mbar. The commissioning clean-filter figure at 55% load was 24 mbar, and the typical post-regeneration floor over the last year has been 28 to 30 mbar. The new value sits at 29 mbar, at the floor and above the clean baseline. The curve over loads has the same shape as earlier post-regeneration curves. Smoke is unchanged and the outlet is clean. Reading: hypothesis B. The log notes the cycle, the new floor and the verification, and the team continues normal monitoring.
Scenario 2: the fall that was not
A hotel’s 400 kVA set shows ΔP at 60% load drifting down over three weeks from 36 mbar to 16 mbar, with no regeneration logged and no maintenance. Load is steady. The independent gauge agrees with the controller, so the reading is real. The commissioning clean figure at 60% load was 22 mbar, so 16 mbar is below the clean baseline. The curve against load has become flatter. An operator notices a fine grey film on the downstream flange and a slight haze at the stack at higher load. A cold inspection finds soot on the outlet face of the canister when the downstream pipe is opened by the supplier. The supplier’s inspection finds a failing mat and a cracked section of core. Reading: hypothesis C. The team had been told for three weeks that the filter “looked healthy”. The supplier replaces the core, and the operator asks for the regeneration and thermal history to be reviewed for the cause. The key signals were the below-baseline pressure, the flatter curve and the outlet-side soot.
Preventing a silent failure
A falling ΔP is silent compared with a rising one. Three habits close that gap.
- Set a low-pressure watch, not just a high one. Ask the supplier whether your controller can alarm on ΔP that falls below an expected floor for a given load, and make sure any such alarm is not silently disabled. See RECD controller settings and access control.
- Trend ΔP at a reference load, and chart it with the floor line and the alarm line. See what an RECD operating dashboard should show.
- Do periodic outlet-side and smoke checks, and a particulate or opacity check at a sensible interval. See smoke opacity versus particulate mass for an RECD and RECD test results versus day-to-day site readings.
Questions people ask
Is a lower differential pressure always better?
No. Low ΔP can mean a clean filter, but it can also mean the gas is not passing through the filter material. A filter that is no longer restrictive is also no longer filtering.
Can a regeneration cause the filter to crack?
Violent or uncontrolled regenerations, particularly of a heavily loaded filter or one contaminated with oil or fuel, can crack or melt a core. That is why unexpected pressure behaviour after a cycle should be checked and not just welcomed. See RECD regeneration temperature.
How much of a drop is too much?
There is no universal figure. Compare with your own history of cycles and with the clean-filter baseline at a matching load. A drop into the normal post-regeneration range is expected. A drop below the floor is not.
Can a sensor fault make the pressure look low while the filter is blocked?
Yes. A tube that is disconnected or a transmitter that has failed can read near zero while the filter is heavily loaded. That is why an independent gauge matters.
Should I stop the set if I suspect damage?
Ask the supplier. The right answer depends on your compliance position, the engine limits and the risk. Do not decide in the dark: keep the evidence and call.
Closing thought
A falling pressure drop is a result to be explained, not a reward to be collected. Check the load, look for a regeneration or service event, compare with an independent gauge, apply the clean-baseline rule, build the curve against load and look at the outlet side for soot. Most falls are good news or instrument news. The few that are damage are the ones where acting early saves the filter, the silencer and the downstream equipment. If you would like help reviewing a pressure trend or building a ΔP-versus-load baseline, browse our retrofit emission control devices, read the FAQ or contact the Aceget team.
Related Reading
- Comparing RECD Pressure Readings at Different Generator Loads
- Interpreting an RECD Regeneration Completion Message
- Repeated Failed Regeneration and RECD Filter Damage
- Separating a Sensor Fault from an RECD Performance Problem
- Smoke Opacity vs Particulate Mass for an RECD