Wavepin
Pipeline intelligence

Find the leak.
Know where.
Know what to do.

Your pipeline already has gauges. Wavepin reads them, works out what is really happening, and gives the control room one clear answer.

See it workShow me the numbers
97 of 100leaks found
13 secondsto raise the alarm
190 metrestypical distance from the real leak
6 sensorson a 20 km line

Measured on 2,500 simulated cases the system had never seen (1,164 leaks). It also raised 7 false alarms per 100 quiet cases. Not yet tested on a real pipeline.

How it works

1. Listen

Reads the pressure and flow gauges you already have. Every second.

2. Think

Tests every explanation: leak, valve, bad sensor, weak pump. Keeps the one that fits.

3. Advise

Simulates each response first. Recommends the cheapest safe one. A person approves.

Hands on

Try it. Three steps.

2

Watch the system work it out

Now showingSudden leak A hole opens. Product escapes.
Simulated pipeline
PUMPSTERMINAL0 km5 km10 km15 km20 kmV1V2V3V4PFPPFPSYSTEM SAYS KM 18.2
The system says

Leak near km 18.2

About 15.5 litres per second. Somewhere between km 17.4 and km 18.7.

Over 99% sure
What else could it be?
Leak
>99%
Faulty sensor
<1%
Does not add up
<1%
Valve problem
<1%
All normal
<1%
Weak pump
<1%
See the pressure along the line34474211390 km5 km10 km15 km20 kmkPa

Green: the system’s estimate, with its margin of doubt. White dots: real gauges. Press “Show what really happened” to compare with the truth.

3

Get one clear recommendation

Recommended now

Stop the pumps. Close the valves around the leak.

Safety check passedNeeds shift supervisor approvalNothing runs on its own
Expected cost over the next hour, per option
Keep running
$279k
Slow the pumps
$264k
Throttle a valve
$277k
Stop and isolate
$24k
Full shutdown
$24k

Cost = spilled product and clean-up + lost delivery + pump energy. Each option is simulated first, against every explanation still in play. Prices are demo assumptions.

Proof

Better than the methods in use today

We built the standard methods ourselves and ran everything on the same cases. Then we broke the model on purpose.

How far off is the location?

Typical distance from the real leak. Shorter is better.

Basic flow-balance alarm
355 m
State observer + classifier
670 m
Moving-horizon estimator
206 m
Neural network alone
254 m
Wavepin engine
190 m
Wavepin engine + learning
191 m

How many leaks does it catch?

Out of 100 real leaks. Longer is better.

Basic flow-balance alarm
80%
State observer + classifier
97%
Moving-horizon estimator
76%
Neural network alone
99%
Wavepin engine
97%
Wavepin engine + learning
99%

All alarms tuned to the same false-alarm rate (5 in 100) on separate data. The standard methods are our own implementations.

4.6×

closer to the leak than a moving-horizon estimator when the model is wrong.

95%

of the time it names the right cause: leak, valve, sensor or pump.

43%

of wrong-model cases where it says “do not trust my model”. It says so in 0.2% of normal cases.

Money

Less product on the ground

Average spill per leak, over 15 minutes. 100 simulated leaks.

11.6 m³

No response

5.5 m³

Today: alarm, operator shuts down after 2 minutes

3.3 m³

Wavepin: checked recommendation, acted on in 15 seconds

Fair warning: most of this gain comes from reacting in 15 seconds instead of 120. With the same 2-minute delay, a better alarm alone saved little (5.5 vs 5.2 m³). No pressure limit was broken in any case.

Optional hardware

Want it sharper? Add small nodes.

The software works on your SCADA alone. The node is optional. It listens faster than SCADA, with an exact clock, so it can time the pressure wave a leak sends down the pipe.

km 0km 4km 8km 12km 16km 20EDGE BOXWAVEPIN SOFTWAREYOUR SCADAread onlyCONTROL ROOMadvice

Six nodes at places you can already reach: both ends and the four valve stations. They send out by mobile network. Nothing can send commands back.

Sudden leak

157 → 20 m

Typical location error, SCADA alone → with six nodes.

Slow, growing leak

No gain

130 m before, 137 m after. A slow leak sends no sharp wave to time.

People and hazard

Planned

The same node can watch the station: someone present, a cabinet opened, vapour in the air. In design. Not tested yet.

Simulation, 447 leaks. Node sampling 5 times a second, clock good to 5 ms. The 20 m figure is at the limit of what the simulator can resolve; expect worse on real pipe. The nodes are in design; none is installed yet.

Getting started

Three steps. No risk to your operations.

1. Send one month of data

An export from your historian, for one line. Nothing is installed. Nothing touches your control system.

2. We replay it

You see what Wavepin would have caught, where, and how fast. On your own events.

3. Run it alongside

A pilot on one line. It reads your SCADA, read-only. Your operators stay in charge.

Straight talk

What is proven. What is not.

Proven, in simulation

  • Finds 97% of leaks. Typical location error 190 m.
  • Tells a leak from a valve fault, a bad sensor and a weak pump.
  • Keeps working when the pipeline model is wrong.
  • Every recommendation passes a fixed safety rulebook first.
  • 12,800 simulated cases. Every number is reproducible.

Not proven yet

  • No real pipeline data has been used. None.
  • Tiny leaks (under 1% of flow): it catches 68%. A neural network catches more.
  • Pump trips still cause about 1 false alarm in 5.
  • Two faults at once are often read as one.
  • One straight 20 km line, one product. No branches, no batches.
  • The optional nodes are in design. None is installed yet.