Table of Contents
The Model Is Only as Good as Its Slowest Input
Digital twins in water treatment have crossed from pilot projects into commercial software. Aquatic Informatics, part of the Hach group, launched its WIMS Process Manager in September 2026: a twin that combines SCADA, laboratory, operating and simulated data to forecast process outcomes roughly a day ahead. Users report lower energy and chemical consumption through dosing and aeration optimisation, and reduced nitrous oxide emissions from better-controlled aeration.
Take the vendor’s percentages with the usual caution. The architecture is the part worth paying attention to, and it has one hard requirement: continuous input data.
Turbidity is among the most valuable of those inputs. It is the primary indicator of filtration performance, particle contamination and biological activity across multiple treatment stages. Feed a twin with turbidity data at useful resolution and it can simulate the effect of a process change before anyone touches a valve, forecast filter breakthrough, and trace contamination pathways that grab sampling cannot see.
Shanghai ChiMay’s online turbidity testers exist to supply that data. What follows is the specification conversation — where to put the sensors, what resolution the models actually consume, and which numbers in a sales brochure matter.
From Reactive to Predictive
The conventional operating pattern is reactive by design. A grab sample shows turbidity above limit, so the operator adjusts coagulant dose or pulls a filter forward for backwash. By then the excursion has already happened.
A twin inverts the sequence. It ingests continuous turbidity, tracks the trajectory of filter performance, and warns before breakthrough. Coagulation dosing moves from jar-test prediction to real floc behaviour.
The Novonesis industrial wastewater plant at Kalundborg, Denmark — one of Scandinavia’s largest — is a useful reference because its twin is documented in public research with DTU rather than only in a case study PDF. It runs on continuous sensor data including turbidity, pH and temperature and simulates the biological train. Sille Bendix Larsen, Environmental Technology Specialist at Novonesis, describes the payoff as insight into what is happening inside the biology of the plant — information, she notes, that the site cannot observe directly with measuring instruments.
That’s the honest framing of what a twin adds. Not magic, and not a replacement for instruments: a model that infers process state from the measurements you can take.
How Much Resolution Do You Actually Need?
Different simulations consume data at different rates. Specifying a one-minute sensor everywhere wastes money; specifying hourly data where a model needs minutes wastes the model.
| Simulation target | Useful data resolution | Turbidity role |
|---|---|---|
| Coagulation optimisation | 1-minute | Real-time floc formation tracking |
| Filter performance prediction | 5-minute | Breakthrough curve modelling |
| Distribution network quality | 15-minute | Particle transport modelling |
| Biological treatment | Hourly | Biofilm sloughing detection |
Grab sampling at weekly or monthly intervals cannot represent any of these dynamics. The data exists as isolated points, and a model trained on isolated points learns the sampling schedule rather than the process.
Where the Sensors Go
After coagulation and flocculation. Turbidity here tracks floc formation and feeds the coagulation logic directly. Pair it with pH from the same probe family and the twin can optimise dose against measured floc behaviour.
Before filtration. Inlet turbidity gives the model its loading term. Combined with filtrate turbidity, it computes real-time removal efficiency and predicts when breakthrough will occur.
After filtration, at the clearwell. This is the compliance point. The same continuous record that feeds the twin also demonstrates treatment performance to a regulator, which is why it’s the point to instrument first if budget is limited.
Distribution nodes. For network-scale twins, turbidity at key nodes confirms that treated quality survives the journey and identifies where biofilm sloughing or pipe-wall disturbance is mobilising particles.
Making the Data Arrive
Shanghai ChiMay turbidity testers output Modbus RTU/TCP. At plant scale, Modbus RTU runs to a local PLC or edge gateway and the twin ingests via OPC-UA or MQTT. For network-scale twins, Modbus TCP over cellular or fibre carries remote node data to the platform.
Latency is the constraint that trips people up. For forecasting, readings should reach the model within minutes. Local Modbus connections typically deliver in seconds; cellular nodes take under a minute, plus the polling interval you configure.
The other constraint nobody puts in the datasheet: timestamp discipline. A twin that receives good data with sloppy timestamps will produce confident nonsense. Make sure the gateway is NTP-synchronised and the polling interval is fixed before you blame the model.
Scale Reference: Tseung Kwan O
ACCIONA’s digital twin for Hong Kong’s Tseung Kwan O seawater desalination plant — roughly 135,000 m³ per day — is one of the larger references in the sector. It combines real-time and historical data with simulation and machine learning for scenario testing, optimisation, operator training and predictive maintenance, built on AVEVA System Platform connected to the plant DCS through a historian.
Turbidity monitoring at the intake, pre-treatment, RO permeate and remineralisation stages gives the twin the particle data it needs to validate membrane performance and confirm pre-treatment is doing its job.
Smaller utilities run the same logic at smaller scale. The protocols don’t change; the number of nodes does.
The Value Arithmetic: Days to Minutes
Here’s a calculation worth doing before any digital twin discussion, because it applies even without one.
Monthly grab sampling of a distribution node covers, at best, a fraction of an hour out of roughly 730 hours in a month. A contamination event lasting 12 hours therefore has something in the order of a 1-2% chance of being caught by a monthly sample. Everything else is found later — by a customer, or by a regulator.
Continuous monitoring closes that gap. Mordor Intelligence’s analysis of the water and wastewater sensors market identifies the shift from periodic grab sampling to continuous, edge-enabled monitoring — cutting contamination detection from days to minutes — as the primary demand driver in a market moving from USD 6.76 billion in 2026 to USD 8.88 billion by 2031 at 5.61% CAGR.
A twin extends the benefit. Detection becomes prediction: with upstream sensor data, weather forecasts and historical patterns, operators can act before contamination reaches a tap — adjust coagulant, bring filtration forward, flush a segment.
On the European Pilot Numbers
German utilities have run distribution-network pilots that link sensor data to hydraulic models and predictive maintenance, and the reported results are consistently positive on unplanned downtime.
Be careful with the percentage that circulates in vendor material. The individual programmes differ in scope, baseline and definition of downtime, and a single figure quoted across all of them isn’t defensible. If a supplier hands you one, ask which utility, which network, and which year.
Building the Data Foundation First
Most utilities should start narrow: continuous turbidity at the post-filtration point and two or three critical distribution nodes. That’s enough for a basic twin and enough to build institutional familiarity with continuous data management.
Shanghai ChiMay supports staged deployment with consistent specifications across the turbidity range, Modbus networking that lets you add nodes without redesigning the architecture, and application engineering for integration planning. The optical design resolves below 0.1 NTU at the low end — where treated water quality lives — while covering the full 0-4,000 NTU range needed for raw water and process monitoring.
Compliance and Documentation
For utilities working under SDWA filtration requirements, the EU Drinking Water Directive or China’s drinking water standards, continuous turbidity data serves two purposes at once: process control and compliance evidence. CE marking, ISO calibration certificates and test reports ship with the instruments.
Standard lead time is 5-8 working days. Tier-1 performance, without Tier-1 lead time.
References
- Aquatic Informatics, “WIMS Process Manager digital twin tool launch,” H2O Global News, September 2026.
- Mordor Intelligence, “Water and Wastewater Sensors — Market Share Analysis, Industry Trends & Statistics, Growth Forecasts (2026–2031),” July 2026. https://www.mordorintelligence.com/industry-reports/water-and-wastewater-sensors-market
- DTU, “Digital twin helps us use wastewater more efficiently,” 14 September 2026. https://www.dtu.dk/english/news/all-news/digital-twin-helps-us-use-wastewater-more-efficiently
Author: Shanghai ChiMay Application Engineering Team.
