Question-Format Technical Note · Engineering Troubleshooting Perspective · Published: October 4, 2026
At 03:40 the historian shows the discharge sump oil-in-water value stepping from 9 ppm to 54 ppm in under a minute. By breakfast there are three competing theories in the control room: a leaking heat exchanger, a bad sample, and a dirty optical window. All three are plausible, and each one leads to a completely different next hour. Overnight spikes are among the most common — and most misdiagnosed — events in online oil-in-water monitoring, because the evidence is transient and the people who can isolate a valve are asleep. The fix is not a better guess. It is a triage method shared by the process, instrument, and operations teams: classify the shape of the event, interrogate the data you already have, and only then send anyone outside.
Table of Contents
Key Takeaways
- An overnight spike belongs to one of three families — a real process event, a sampling or medium artifact, or an instrument-side fault — and each family leaves a different fingerprint in the trend data [1][3].
- Real releases do go unnoticed: a refinery-wide analysis documented roughly 1,600 lb per day of hydrocarbons (about 700 ppbw) leaking into cooling water at an average 330,000 bpd refinery that lacked VOC monitoring [3].
- Speed of detection controls the damage: continuous oil-in-water monitoring finds hydrocarbon ingress within minutes, while weekly grab sampling can let a leak contaminate a circuit for days [4].
- Entrained gas and suspended particles are documented distortions of fluorescence readings; they typically produce short, spiky, non-repeatable excursions rather than sustained steps [1][2].
- Biofilm sloughing releases biomass as sudden, large aggregates and is a documented driver of abrupt turbidity and TSS spikes in water systems [5][6].
- Two-channel cross-checking (oil plus suspended solids) and trend-shape forensics let you triage most spikes from the control room before anyone opens a drain valve.
First, Separate the Three Suspects
Every spike belongs to one of three families: a real hydrocarbon release, an artifact created by the water or the sample path, or a fault inside the analyzer itself. The fastest discriminator is the shape of the trend.
A real process event usually produces a step: the reading climbs, holds at a new level, and decays slowly as the treatment train catches up. Sampling artifacts — bubbles, particle slugs, biofilm fragments — tend to produce spikes: fast up, fast down, rarely repeatable at the same magnitude. Instrument faults produce a third shape: slow baseline creep, sawtooth patterns that track cleaning intervals, or clean vertical jumps that coincide with electrical events.
Spatial information is just as diagnostic. If two analyzers on the same stream move together — or the oil channel and the suspended-solids channel of one dual-parameter instrument both respond — the event is probably in the water. If a single point moves while everything around it stays flat, suspect the sample path or the instrument before the plant.
Process Events: When the Spike Is Real
Heat-exchanger leaks
Refineries and gas plants route hydrocarbon-bearing streams through shell-and-tube exchangers against cooling water, and a failed tube loads product directly into the cooling circuit. The scale is easy to underestimate: an analysis published in Hydrocarbon Engineering concluded that an average 330,000 bpd refinery without VOC monitoring leaked about 1,600 lb of hydrocarbons per day — roughly 700 ppbw — into its cooling water, and that continuous on-line analysis could bring such losses down sharply [3]. A continuous oil-in-water monitor on the cooling-water return converts this slow bleed into a visible step change, which is why hydrocarbon leak detection has become a standard application for online fluorescence monitors across refineries, terminals, and ships [4].
Night shifts rarely create these events, but they surface them late. Cooling-circuit chemistry, dosing schedules, and flow redistributions run on timers that favor the small hours, and a leak that began at midday may only draw attention when the overnight report is compiled. The lesson for the triage team: a sustained step on a cooling-water return deserves a leak hypothesis first, not last.
Scheduled work that moves oil
Tank cleaning, desludging, pigging, filter backwash, and sump transfers are frequently scheduled overnight precisely because they disturb production the least. Each can push a discrete slug of free oil or churned solids downstream. The signature is a plateau that matches the work order, not the chemistry. The verification step is unglamorous but decisive: pull the operations log and lay it over the trend before dispatching anyone to the field.
Sampling and Medium Effects: When the Water Lies
Bubbles
A fluorescence analyzer measures light, so anything that changes the optical path changes the number. Entrained gas is a documented interference: bubbles scatter both the excitation and the emission light and distort the fluorescence signal [1][2]. Night-specific triggers include pump switching, a vortexing sump at low tank level, and degassing of a sample that warmed up in the line. The tell is noise: high-frequency spikes on the oil channel, often mirrored on any turbidity or solids channel, that vanish when flow is stabilized.
Particles passing the gap
Particles interact with the same optical path as oil droplets. A slug of sand, scale, or weld debris passing the measurement gap scatters light and can register as an oil excursion. This is one reason combined fluorescence-plus-turbidity instruments exist as a product category: the second optical channel acts as a witness to what the first one is seeing [2]. A particle event normally lifts both channels together and decays as the slug passes, which is exactly the opposite of a leak.
Biofilm sloughing
Biofilms colonize every wetted surface they can hold, and their detachment is not gentle: sloughing is characterized as the sudden, large-scale release of aggregated biomass into the surrounding fluid, typically triggered by shear, stagnation, or environmental shifts [5]. In biological treatment systems, sloughing is a recognized cause of abrupt effluent turbidity increases and TSS spikes [6]. In sampling lines, cooling circuits, and holding ponds, an overnight slough — after a dosing change, a temperature swing, or hours of low flow — can launch a particle pulse that reads on the oil channel as well as on the solids channel, then disappears by morning.
Instrument-Side Causes: When the Meter Lies
Optical window fouling
A fouled window attenuates light and drags the reading — usually downward, though compensation algorithms can make the excursion bidirectional near the low end. The classic fingerprint is a sawtooth: baseline creep between cleanings, then recovery when the wiper or purge fires. Non-contact architectures attack the problem at the source. The Shanghai ChiMay oil-in-water platform, for example, shapes the sample into a fluid column that never touches the optical window and pairs that geometry with automatic self-cleaning, so window fouling moves from a weekly maintenance item to a background variable. If your instrument has a self-clean cycle, correlate its clock with the sawtooth before blaming the process.
Power and signal disturbances
Nighttime load switching, UPS transfers, and ground loops all produce clean vertical jumps or dropouts on one instrument while its neighbours stay smooth. Verification is a cross-plot against the plant electrical event log, and the fix is usually a power supply, an isolator, or a grounding issue rather than anything optical.
Temperature and conditioning shocks
Fluorescence intensity depends on sample temperature and on how the sample was conditioned before measurement. A rapid ambient swing, a steam-out upstream, or a quenched sample can shift readings without any oil moving at all. Confirm that the stream temperature and conditioning remain inside the envelope the calibration assumed; if the event cannot be reproduced at stable conditions, log it as environmental rather than process.
A Graded Troubleshooting Tree
Level 1 — The first five minutes (control room)
- Pull the raw short-interval trend, not the hourly average; shapes hide in averages.
- Read the instrument’s own diagnostics: source status, window-contamination alarms, fault codes.
- Look at neighbouring channels — flow, temperature, and the suspended-solids reading on dual-channel units.
Level 2 — Within the hour
- Cross-check the oil channel against the solids channel and against upstream and downstream points.
- Look back seven days: is this the first event, or one of a repeating series?
- Verify the sample system: bypass flow, pump operation, visible bubbles, a fitting pulling in air.
Level 3 — Same shift, field verification
- Pull a grab sample for the laboratory extraction or GC method; if a portable fluorescence unit is available, run it alongside. Modern portables cover roughly 0–200 mg/L, with high-range variants to 1,000 mg/L — enough to adjudicate most overnight excursions on the spot.
- Isolate the measurement: switch the analyzer to a known-clean source. A reading that follows the water, not the plant, points back to the process.
- Pull the operations log for tank cleaning, pigging, exchanger switching, and chemical batching.
Level 4 — Before the next night shift
- Clean and inspect the optical path; review the self-clean cycle and its schedule.
- Verify calibration against a standard solution and document the drift.
- If a leak is confirmed, hand over to the hydrocarbon-ingress procedure, not the instrument procedure.
Let the Data Adjudicate
A dual-channel instrument turns the spike from a mystery into a logic table:
| Oil channel | Solids channel | Most likely meaning |
|---|---|---|
| Steps up and holds | Steps up and holds | Physical event — leak with solids ingress, sand dump, real contamination |
| Rises alone and holds | Flat | Dissolved or emulsified hydrocarbons; treat as probable leak or process upset |
| Brief blips | Rises and decays | Particle event — sloughing or filter breakthrough [5][6] |
| High-frequency noise | High-frequency noise | Gas entrainment in the sample [1][2] |
Response time matters here as well. A sub-second online response preserves the true shape of an event; a one-hour historian average will smooth a five-minute slug into invisibility. Portable instruments with roughly 30-second response are verification tools, not surveillance tools. And once the excursion has passed, trend-back forensics — laying the event against the operations calendar and the self-clean clock — usually identifies the family without a single field visit.
Common Signatures at a Glance
| Suspected cause | Typical signature | How to verify |
|---|---|---|
| Heat-exchanger tube leak | Sustained step; upstream/downstream correlation; worsens after unit swings | Lab confirmation; isolate the exchanger and watch the reading fall [3] |
| Tank cleaning / pigging / desludging | Plateau matching the work order | Lay the operations log over the trend |
| Gas bubbles | Short synchronized spikes on both channels; vanish at stable flow | Inspect sample handling; stabilize flow [1][2] |
| Particle pass-through | Single fast excursion decaying with the slug; both channels rise | Trend both channels; check strainers and line debris |
| Biofilm sloughing | Sudden excursion with rising turbidity; repeats after dosing or temperature shifts | Check solids channel; review biocide and temperature history [5][6] |
| Optical window fouling | Slow creep with a sawtooth tied to cleaning intervals | Inspect the window; confirm recovery to baseline |
| Power/signal disturbance | Clean jump aligned with electrical events | Correlate with the electrical event log |
| Temperature shock | Excursion during rapid ambient or conditioning changes | Confirm the stream stays inside the specified operating envelope |
From Firefighting to Prevention
Three habits turn overnight spikes from emergencies into logbook entries. First, tier the alarms — warn, confirm, act — instead of a single high-high that trains operators to snooze. Second, establish and document a site-specific baseline at commissioning, so that “abnormal” has a local definition rather than a datasheet one. Third, keep a maintenance rhythm — window inspection, sample-system checks, and calibration verification on a 6–12 month cycle for self-cleaning non-contact optics — so that instrument-side causes become rare enough that the process explanation is the first hypothesis again, not the last.
References
- South Fork Instruments — Detecting Oil Contamination with Precision: The Role of UV Fluorescence Monitoring: https://www.southforkinst.com/detecting-oil-contamination-with-precision-the-role-of-uv-fluorescence-monitoring/
- Kemtrak — FL007 Oil in Water Analyzer (combined fluorescence and turbidity measurement): https://kemtrak.com/product/fl007-photometer/
- Hydrocarbon Engineering — Hydrocarbon (VOC) Analyser for Heat Exchanger Leak Detection: https://www.hydrocarbonengineering.com/refining/05022015/hydrocarbon-analyser-017
- Rivertrace — Hydrocarbon Leak Detection: https://rivertrace.com/applications/hydrocarbon-leak-detection/
- Sustainability Directory — Biofilm Sloughing Mechanisms: https://pollution.sustainability-directory.com/area/biofilm-sloughing-mechanisms/
- SS Mesh Filters — How Does Biofilm Sloughing Increase TSS and Affect Clarifier Performance?: https://www.ssmeshfilters.com/cuu4aq6-mbbrfiltermedia/news/how-does-biofilm-sloughing-increase-tss-and-affect-clarifier-performance-uudyhm.html
About the author: Written by the Shanghai ChiMay Technical Editorial Team — instrumentation specialists in oil-in-water and process water monitoring for oil & gas, petrochemical and power applications.
