WFI distribution loop

The loop must stay hot, fast and clean. Pump wear shows up first as lost velocity.

Water for injection circulates hot and fast so nothing can grow. If return velocity or temperature drops, the loop is out of its qualified state. The twin checks the pump against its curve and the loop against its heat loss, so wear is a planned repair, not a deviation.

Hot WFI loop, 80 °C, 6 m³/h, DN40 316L, return ≥ 1.0 m/s, 5.5 kW sanitary pump on a drive

Built and run to USP <645> (conductivity), USP <643> (TOC), ASME BPE

Start with one utility See how we prove results

Reynolds at 1 m/s: 95,256
Supply velocity: 1.75 m/s
Return at peak draw: 1.17 m/s
Pump input power: 1.79 kW

Anatomy and nameplate

1Supply and return flow. Gives velocity in the supply run and at the end of the loop.
2Return temperature. The coldest point in the loop. It must stay at temperature.
3Conductivity. Checked against the USP <645> limit for the water temperature: 2.7 µS/cm at 80 °C.
4TOC. Total organic carbon against the 500 ppb limit.
5Pump speed and power. From the drive. Speed and power put the pump on its curve.
6Pump discharge pressure. With flow, tells pump wear from a valve change.
7Storage tank level. Draw from the points of use and make-up from the still.
8Heat exchanger inlet and outlet temperatures. Heat duty and the exchanger’s UA, which falls as it fouls.

WFI distribution loop nameplate
Loop
Service	Hot water for injection, 80 °C, continuous circulation
Design flow	6 m³/h
Tube	ASME BPE 1½ in (DN40), 38.1 mm OD × 1.65 mm wall, 34.8 mm bore, 316L electropolished
Loop length	300 m, 30 mm insulation
Supply velocity	1.75 m/s at 6 m³/h
Return velocity at peak draw	1.17 m/s with 2 m³/h drawn, minimum 1.0 m/s
Reynolds number at 1 m/s	95,256 (ρ 971.7 kg/m³, μ 0.355 mPa·s at 80 °C)
Loop head at design flow	46.1 m, including 15 m of return back-pressure
Pump and drive
Type	Sanitary end-suction centrifugal, open impeller, 316L, on a variable frequency drive
Motor	5.5 kW, 2-pole, 2,950 rpm rated
Best efficiency point	12 m³/h at 75 m, 55% at rated speed
Shutoff head	90 m at rated speed
Duty	6 m³/h at 46.1 m, 2,196 rpm (74% speed)
Pump efficiency at duty	49.1%
Shaft and input power at duty	1.49 kW shaft, 1.79 kW from the supply
Seal	Single mechanical seal with WFI flush
Heat and water quality
Heat exchanger	Double-tube-sheet shell and tube, plant steam on the shell side
Heat lost by the loop	4.2 kW, 0.61 K supply to return
Insulated tube heat loss	14 W per metre at 80 °C in a 22 °C room
Conductivity limit	2.7 µS/cm at 80 °C (USP <645> Stage 1, not temperature compensated)
TOC limit	500 ppb (USP <643>)

The clean utilities behind a batch

Three paths run side by side. Water goes from the still to the hot loop. Steam sterilizes parts, and product moves on to drying and cold storage. Air goes through HEPA filters into the graded rooms.

Water
Air
Steam
Product
Has its own twin page

The physics

The equations behind every curve on this page.

Velocity

v = Q / A

Return velocity must stay above 1.0 m/s, even with every point of use open.

Reynolds number

Re = ρ·v·D / μ

About 95,000 at 80 °C and 1 m/s in DN40. Well turbulent, so the wall stays scoured.

Loop curve

H = H_bp + (f·L/D + ΣK)·v²/2g + H_hx

Back-pressure at the return plus friction around the loop. Friction grows with velocity squared.

Pump curve and affinity laws

H = (1 − w)·H0·s² − k·Q² Q ∝ N P ∝ N³

Wear lowers the curve. Speed moves it. The crossing with the loop curve is the flow you get.

Loop heat loss

Q_loss = U·A·(T_loop − T_room)

Heat leaks through the insulation in step with the temperature difference.

Temperature around the loop

T(x) = T_room + (T_0 − T_room)·e^(−U′x/ṁc_p)

The water cools a little along every metre. Fast flow keeps the drop small.

The pump and the velocity

The drive holds a fixed speed. As the pump wears, its curve drops and the loop gets less flow. The margin above 1 m/s at the far end is what goes first.

The pump wears and the operating point slides down the loop curve

Drive held at 2,196 rpm

Pump curve, new
Pump curve, month 3 of wear
Pump curve, month 6 of wear
Loop curve

New pump: 6.00 m³/h at 46.1 m, 1.79 kW
Month 3: 5.63 m³/h at 42.7 m, 1.75 kW
Month 6: 5.25 m³/h at 39.2 m, 1.70 kW

Wear lowers the pump curve. Flow and discharge pressure fall together along the loop curve. A valve change would raise pressure as flow falls, so the twin can tell the two apart.

H = (1 − w)·H0·s² − k·Q² | H_sys = H_bp + (f·L/D + ΣK)·v²/2g + H_hx

Return velocity at peak draw against pump speed

All points of use open (2 m³/h drawn)

New pump
After 6 months of wear

Speed for 1.0 m/s, new: 69.8%
Speed for 1.0 m/s, worn: 76.0%
Return at drive setting, new: 1.17 m/s
Return at drive setting, worn: 0.95 m/s

A worn pump needs more speed for the same velocity. At a fixed drive setting, the margin above 1 m/s is what wear eats first.

v_return = (Q_supply − Q_draw) / A

Keeping the loop hot

A hot loop loses heat all day. The heat exchanger puts it back. The supply-to-return drop shows how much is going, and where.

Water temperature around the loop

300 m loop, 30 mm insulation, 22 °C room

Insulation whole
40 m of insulation missing

Return, insulated: 79.39 °C (4.2 kW lost)
Return, bare section: 79.11 °C (6.1 kW lost)
Heat loss per metre at 80 °C: 14 W insulated, 62 W bare

Fast hot water loses little heat per metre. A bare section shows as a bigger drop from supply to return, and the heat exchanger works harder to make it up.

T(x) = T_room + (T_0 − T_room)·exp(−U′·x / (ṁ·c_p))

Failure modes

What breaks, what it looks like, what the twin does about it.

Failure mode	The physics	Signals that show it	How the twin catches it
Pump wear	Impeller and casing wear lower the head the pump makes at a given speed. The loop curve does not change, so flow falls.	Return velocity falls at the same pump speed, about 2% a month here. Discharge pressure falls with it.	Solves for the head lost to wear and flags it in week 8, 13 weeks before return velocity at peak draw would drop under 1.0 m/s.
Seal wear	Worn seal faces open and leak. Heat from friction rises before the leak shows.	Seal flush flow and temperature. Visible leakage. A small rise in pump power at the same duty.	Tracks seal flush against pump speed and loop temperature, and asks for a check when it drifts.
Heat exchanger leak	A tube or tube-sheet leak lets plant-side fluid near the WFI. The double tube sheet gives the leak a path to drain, away from the water.	Flow at the tube-sheet tell-tale drain. Conductivity rise. Pressure across the tube sheet.	Watches conductivity against temperature and the exchanger’s UA together. A step in both points at the exchanger.
Sanitization temperature not reached	Heat lost around the loop is more than the exchanger can put back, so the coldest point falls short.	Return temperature below set point. Supply-to-return drop above the heat-loss model.	Compares the drop with the model, 0.61 K insulated. A larger drop points at missing insulation or a weak exchanger.
Conductivity drift	A rising reading is either the water or the sensor. Source water and still performance change the water. Fouling and temperature error change the sensor.	Conductivity creeping up at steady temperature. Disagreement with the still outlet sensor.	Checks the reading against the USP <645> limit for the measured temperature and against the still outlet, so a sensor fault does not look like bad water.

What the twin does not do

The loop controller and your QA alarms stay primary. The twin warns earlier. It does not release water.
We do not change drive speed or a qualified setting. We recommend. Your change control and QA decide.

How we catch it early

A physics twin and a PINN, working on one wfi distribution loop.

The twin is a set of equations for this machine, tuned to its nameplate. Give it the flow, speed and conditions and it returns what every reading should be. It always gives the same answer for the same inputs.

A physics-informed neural network, or PINN, sits on top. It is trained on the plant’s own data and it is penalized whenever it breaks a physics rule. So it can only learn what the equations leave out: wear, fouling, friction. It cannot invent an answer that the machine could never produce.

Step 1
Measure
Flow, pressure, current, temperature, vibration. Whatever the asset already has.
Step 2
Physics twin
The equations of this machine, set to its nameplate. It says what the readings should be right now.
Step 3
Residual
Measured minus expected. Demand, weather and speed are already taken out.
Step 4
PINN
A small neural network that learns the cause of the gap, under the physics rules.
Step 5
Finding
A named cause, a date, and the kWh or hours at stake. An engineer reviews it.

loss = (measured − predicted)² + λ · (physics residual)²

The second term is the rulebook. It is zero when the network obeys the equations.

Why this beats black-box AI

A black-box model learns patterns from history. It needs examples of each failure, and it goes quiet when the pump runs at a point it has not seen. The twin starts from physics, so it works on day one. The PINN adds what physics cannot know about this particular machine.

That is also what makes the savings countable. The baseline is a calculation, not a guess, so the gap between baseline and meter is a number a customer can check.

What the twin checks on this asset

Return velocity at peak draw: 1.17 m/s against 1.0 m/s, a 17% margin
Pump on its curve: 6.00 m³/h at 46.1 m and 2,196 rpm. After six months of wear, 5.25 m³/h at the same speed
Heat exchanger UA against its clean value, from inlet and outlet temperatures and steam flow
Conductivity against the USP <645> limit for the measured temperature: 2.7 µS/cm at 80 °C
Supply-to-return temperature drop against the heat-loss model: 0.61 K

	Fixed alarm	Black-box AI	Twin + PINN
Needs failure history to learn	No	Yes, lots	No
Says what is wrong	No	Rarely	Yes, by name
Works at a flow it has never seen	No	Often fails	Yes, the equations still hold
Puts a number on the saving	No	No	Yes, in kWh and dollars
Lead time on slow wear	Late	Varies	Weeks to months
Every number can be traced	Yes	No	Yes, to an equation

What the twin puts a number on

Energy and cost, from the equations.

Warning before velocity is lost

13 weeks

Flagged in week 8. Return at peak draw would fall under 1.0 m/s in week 21.

Pump energy per year

15,706 kWh

1.79 kW, around the clock, $1,885 at $0.12 per kWh

Holding flow with a worn pump

+0.40 kW

Speed up from 74% to 81.0% after six months of wear, 3,530 kWh a year

40 m of missing insulation

1.9 kW of heat

16,797 kWh of steam a year, $840 at $0.05 per kWh

What we do for you

We watch it. We catch it. We prove it.

Our reliability engineers run the twin on your wfi distribution loop, review every finding, and help your crew fix it. After the repair we compare the meter to the baseline and write the result in the ledger.

Start with one utility See the 90-day pilot

More assets

More from pharmaceutical, then the lead asset of each other industry.