Water Utilities / Indian Ocean Territory / 2021
Resolving flow-pressure instability across a regional water network
Pump activation cycles, sensor lag and pressure regulation were interacting in ways the utility's SCADA data could not explain. We built an augmentable variable model with virtual sensors to close the gap between what the network did and what its instrumentation reported.
Results
- 63%
- Reduction in flow-pressure anomaly error rate
- 41%
- Reduction in reactive maintenance cost
Disciplines
- Process control
- Sensor modelling
- Industrial data
Context
The utility was carrying a persistent anomaly rate in flow and pressure readings, and a maintenance function that spent most of its time responding to events rather than preventing them. The instrumentation was adequate on paper. The difficulty was that pump activation cycles, sensor lag and pressure regulation interacted on timescales the historian could not resolve, so the recorded data described a network that did not quite exist.
Approach
We ran a discovery programme to map the interdependencies directly — establishing, for each anomaly class, whether it originated in the physical network or in the measurement of it. That distinction had not previously been drawn, and it accounted for a substantial share of the reported faults.
We then implemented an augmentable variable model: virtual sensors that estimate the quantities the physical instrumentation cannot observe at the required rate, feeding algorithmic process controls that act on the estimate rather than the lagged reading. The model is extensible by design, so that new instrumentation improves it rather than requiring it to be rebuilt.
Outcome
Within four months the error rate in flow-pressure anomalies fell by 63% and reactive maintenance costs by 41%, returning a net-positive position inside six months of deployment. The larger change was to the maintenance function, which could now distinguish a failing asset from a failing measurement.