Liquid Cooling System Cleanliness Testing

Where contamination comes from: particles, ions and suspended solids

Contamination is not simply a matter of "dirty, so flush it". There are three independent failure paths: particle blockage, ionic corrosion, and deposition of suspended solids and biofilm. Collapsing them into one vague "water quality" number is how sites end up with a turbidity reading that looks fine while the cold plate is clogging.

Particles: size and channel geometry decide the outcome

GB/T 48023-2026 sets two filtration tiers: secondary-side filters should be no coarser than 50 µm, primary-side no coarser than 300 µm, and filters should be serviceable while the system is running. T/CAEE 103-2026 goes further for cold-plate manifolds with countable limits: after internal flushing, particles above 100 µm must not exceed 10 per litre, and particles above 50 µm must not exceed 100 per litre. Both point at the same threshold: above 50 µm you are already in the range that needs control.

Bar lengths are an order-of-magnitude guide, not a linear scale. µm(c) denotes the equivalent particle size under the ISO 4406 calibration convention.

Ions: conductivity and pH are a pair, not alternatives

Conductivity reflects total dissolved ions in the coolant; pH reflects acid-base balance. Table A.1 of GB/T 48023-2026 specifies incoming secondary coolant pH 8.0–10.0 for copper cold plates or 8.0–9.0 for aluminium; operating pH within ±1 of the incoming baseline and inside 7.0–10.0 (copper) or 7.0–9.0 (aluminium). For conductivity the standard gives a reported value, to be agreed between supplier and buyer.

That is why "the national standard caps conductivity at some number" is a common misreading. The enforceable limit has to be written into the technical agreement, and the report should state "limit per agreement" rather than a blanket "meets the national standard". Flushing has its own hard figures: conductivity no higher than 10 µS/cm at 25 °C, copper and iron each no higher than 0.5 mg/L.

Low pH costs you metal corrosion and a degraded passive film; high pH costs you inhibitor effectiveness, faster scaling and aged seals. Both appear first in conductivity and in dissolved copper and iron, so read these three as a single trend.

Suspended solids and microbes: turbidity measures quantity, not form

Turbidity is an optical quantity, the scattering and absorption of light by suspended particles, and it is not the same observable as a particle count. The standard sets incoming secondary coolant turbidity at ≤5 NTU and operating at ≤20 NTU, with flushing water ≤5 NTU. The same table carries an indicator that is easy to skip: total colony count, ≤100 CFU/mL incoming and ≤1000 CFU/mL operating. Once biofilm forms inside a cold plate channel it raises both pressure drop and turbidity, and cleaning it costs far more than monitoring it.

Secondary coolant (ethylene glycol or propylene glycol aqueous solution): incoming turbidity ≤5 NTU, operating ≤20 NTU. Conductivity is to be agreed between supplier and buyer. After system flushing the water should meet: conductivity ≤10 µS/cm at 25 °C, pH 5–7.5, turbidity ≤5 NTU, copper ≤0.5 mg/L, iron ≤0.5 mg/L.

GB/T 48023-2026, Data center cold-plate liquid cooling system technical specification, Annex A Table A.1 and Annex B flushing requirements (summary; acceptance per the standard text)

How the four parameters divide the work

Site disputes are rarely about instrument accuracy; they are usually about using the wrong parameter to answer the question. Put all four in one table, define each boundary, then talk about selection.

"Incoming / operating" values are the secondary coolant frame of GB/T 48023-2026 Annex A Table A.1, covering ethylene glycol or propylene glycol aqueous solutions and other formulated fluids. Formulations differ by project, so check the standard text and the agreement before applying them.

Five stages of liquid cooling cleanliness testing

Cleanliness cannot be secured by one final inspection. Put measurement at five positions, each with its own criteria and instruments, and the data forms a traceable chain.

How to cite the 15 liquid cooling standards

In this system, national standards govern the system and the fluid, industry standards govern planning and test methods, and association standards govern component delivery and specialised methods. Mixing the layers is the most common reason a cleanliness report gets rejected: the data is right, but the basis is filed at the wrong level.

"Scope" means the standard is cited only for its field of application, to define the object and stage, with no figures extracted. "Clause" means specific figures are cited and traceable to a clause. GB/T 48023-2026 was published on 30 July 2026 and takes effect on 1 February 2027; YD/T 7105-2026 took effect on 1 November 2026. Status of the remaining standards should be checked on the National Standards Information Public Service Platform and the national association standards platform.

When standards conflict, which one wins

Two sets of limits in one report is normal: a general association standard gives an ISO 4406 class, a dedicated one gives particles per litre; a national standard gives an operating range, a customer agreement gives a tighter internal value. Resolve it in this order: contract or technical agreement first; without one, the dedicated standard before the general one, and the stricter of equals; where the standard is silent (secondary conductivity, for example), fall back to the agreement and label it "reported value". Whichever way it resolves, put one source in the verdict field and the other in the comparison column.

One point is routinely overlooked: GB/T 48023-2026 is a recommended national standard (GB/T) and is not mandatory in itself, but once a tender, purchase contract or technical agreement cites it, it binds that project. Before 1 February 2027 you can use it for technical benchmarking and pre-acceptance, but you should not publicly claim "certified to this national standard". Clause-by-clause notes are in our GB/T 48023-2026 clause review.

Measurement points and integration: placement beats specification

A sensor in the wrong place makes precision irrelevant. Three rules govern the secondary side: representative sampling, serviceable in operation, and no disturbance to the main loop.

All four instrument families offer RS485 Modbus RTU or 4–20 mA output and can tie into the CDU PLC or the data center BMS. Four rules for integration:

• Unified timestamps. Water quality and particle data must share one clock, otherwise a correlation such as "pH was drifting while counts spiked" cannot be established.
• Tiered alarms. A warning near the incoming baseline, an alarm at the operating limit, and a response window between them. With a single threshold the site can only ignore it or shut down.
• Trend retention. Keep raw curves for at least one full O&M cycle for filter life assessment and post-incident review; daily summaries alone will not reconstruct an event.
• Serviceable side stream. The particle sampling branch must be removable, cleanable and calibratable while the system runs, otherwise every service visit stops the whole loop.

Instrument selection: range, installation and sampling

Four parameters, four instrument families. Select in this order: range from the fluid and the criterion, then installation and sampling from the piping, and finally communications and maintenance.

BS-PHS276 pH sensor for liquid cooling

Low-conductivity fluids

Designed for the low ion concentration of deionised water and ethylene or propylene glycol solutions. A low-impedance sensing glass stays stable at low conductivity, PT1000 automatic temperature compensation removes the effect of temperature swings, and internal signal isolation copes with the electrical noise of a data hall.

Measuring range0–14 pH

Accuracy±0.02

ResponseT90 ≤20 s

Temperature0–80 °C

Pressure0–0.6 MPa

Process connection3/4 NPT

Power / output9–36 VDC / RS485

MediaPure water, DI water, PG25, EG25

BS-PHS276 Liquid-Cooled pH Sensor 0-14 pH ±0.02

BS-EC350 conductivity sensor for liquid cooling

Four ranges

Anti-polarisation and signal isolation reject external interference, and the sensor is compatible with propylene glycol, ethylene glycol and deionised water. Ranges A, B, C and D span deionised water through concentrated glycol solutions; choose by the actual fluid and operating window rather than defaulting to the widest range.

Range A / B0–20 / 0–200 µS/cm

Range C / D0–1000 / 0–2000 µS/cm

Cell constantK=0.01 / 0.1 / 1 / 1

Accuracy±1.5 %FS

ResponseT90 ≤60 s

Temperature0–100 °C, PT1000 compensation

Power / output18–27 VDC / 4–20 mA or RS485

Process connection3/4 NPT, stainless steel wetted parts

BS-EC350 Liquid-Cooled Conductivity Sensor ±1.5%FS

BS-TURB113 turbidity sensor for liquid cooling

Through-beam optics

The emitter sends light at a fixed wavelength and the receiver measures transmission through the coolant; suspended solids scatter and absorb it, and the attenuation maps to turbidity. A through-beam layout with a sapphire window is unaffected by reflections from stainless piping, and the built-in wiper brush limits bubbles and fouling for long-term online use.

Range0–10 NTU / 0–100 NTU

Resolution0.001 NTU

Repeatability≤1 %

Response<30 s (10 s measurement cycle)

Working pressure<6 bar

Ambient0–50 °C

OutputRS485

DimensionsΦ30 × 267 mm

BS-TURB113 Liquid-Cooled Turbidity Sensor 0.001 NTU

Online and portable liquid particle counting

Light extinction · 8 channels

Light-extinction counting: a particle passing the cell blocks a laser, pulse amplitude maps to size and pulse count to number. BT8500 covers long-term side-stream monitoring and production cleaning stations, BT8200 suits multi-point line integration, and the portable BTP8300 handles incoming inspection and flushing acceptance.

Size range1–200 / 2–400 / 2–750 µm

Channels8 standard, 16 / 24 / 32 optional

PrincipleLight extinction

Display7-inch colour touchscreen (BT8500)

OutputRS485 Modbus RTU, 4–20 mA

SamplingContinuous online, optional degassing unit

Portable checksBTP8300, under 1 min per test

StandardsISO 4406 and NAS 1638 class output

See the Liquid Particle Counter in the English catalogue

Three selection rules

• Size the conductivity range to the fluid. Deionised water uses 0–20 µS/cm (range A); PG25 / EG25 service usually falls in 0–200 or 0–1000 (B / C); wide or concentrated duty uses D. An oversized range wastes resolution where the readings matter.
• Size the turbidity range to the criterion. A 0–10 NTU range decides the ≤5 NTU incoming limit; 0–100 NTU covers the ≤20 NTU operating limit and the high turbidity of flushing. A model that switches ranges online suits a shared flushing-and-operation instrument.
• Set particle channels by target size. ISO 4406 needs the >4, >6 and >14 µm(c) bands; for the 50 / 100 µm component limits, confirm the lower range limit, channel setup and calibration can resolve the target size - never let "widest range" stand in for "effective resolution".

These parameters were checked against Bebur product pages on 2026-11-10. Configurations and options can change between batches; confirm against the current datasheet and order agreement before ordering.

Field notes and five common myths

The same instruments, installed differently, vary by an order of magnitude in data credibility. The six points below are where sites most often go wrong, and what is worth confirming first when a method goes live.

Five common myths

• "Turbidity is normal, so cleanliness passes." Turbidity is a total and cannot resolve size. ISO 4406 counts the 4, 6 and 14 µm(c) bands, and microchannel risk comes mainly from large particles; a low turbidity reading does not prove they are absent.
• "The national standard sets a conductivity ceiling." GB/T 48023-2026 gives secondary conductivity as a reported value to be agreed between the parties. The enforceable limit comes from the technical agreement, and the report should state "limit per agreement".
• "A 50 µm filter means no particle exceeds 50 µm." A filter rating describes the element's retention class, not the maximum particle size in the system. Bypass, element failure and start-stop surges all let particles larger than 50 µm reach the cold plate.
• "Twenty-four hours of flushing is a pass." 24 h is the minimum duration the standard gives; pH, conductivity, turbidity and copper/iron still have to be judged afterwards. Treating the clock as the conclusion mistakes a process requirement for an acceptance criterion.
• "With online instruments, lab sampling is unnecessary." Online instruments carry trend and alarms; arbitration and disputes still require sampling and retesting by the standard method. They are complementary, not substitutes.

One integration that is already in service

Component cleaning is where all four parameters are easiest to use together: you need to know when the count has reached its endpoint, and whether the bath's ion load has already saturated.

The method is simple: integrate an online particle counter, a pH sensor and a conductivity sensor on a side stream of the cleaning machine's circulation line, without cutting the main pipe or changing the existing flow, and watch the count-decay curve alongside the bath's pH and conductivity. The endpoint is decided from data, not from an operator's stopwatch.

According to Bebur's published technical material, this integration is in service in the liquid cooling component cleaning line at Boyd's Dongguan plant, monitoring solid particles, ionic impurities and pH online and feeding the cleaning result back in real time. The same material notes an improvement in the defect rate after integration.

Frequently asked questions

These are the questions that come up most often in selection discussions and site commissioning. Where a clause exists, the answer cites it rather than saying "it depends".

Criteria first, then points, then instruments

Cleanliness programmes are often run in reverse: buy the instrument, find somewhere to mount it, then open the standard. The right order is to fix the basis and the stage first, then decide points and sampling; the instrument only executes. Get the criterion wrong and greater accuracy makes the report less credible, not more.

If you are configuring water quality and particle monitoring for a cold-plate project, check the range against the fluid and the criterion, confirm sampling and installation conditions, then look at communications and maintenance. Full specifications and measuring principles for the four instruments are in the links below.

Related reading

• BS-TURB113 Liquid-Cooled Turbidity Sensor
• BS-EC350 Liquid-Cooled Conductivity Sensor
• BS-PHS276 Liquid-Cooled pH Sensor
• Liquid Particle Counter for Water, Oil and Coolant Monitoring
• Streaming Current Detector for Coagulation Control
• Peracetic Acid Analyzer for Water PAA Monitoring
• Water Hardness Analyzer BC300
• Hydrogen Peroxide Analyzer for Water H2O2 Monitoring