As AMP8 raises the bar on reliability, emissions, and proactive maintenance, reactive approaches are becoming unsustainable. A recent Water Industry Journal article highlights that more than half of water-sector maintenance still happens after failure but there’s now a proven alternative: scientific machine learning (SciML). The feature showcased how Binnies UK, Williams Grand Prix Technologies, and JuliaHub are delivering that shift in practice using scientific machine learning (SciML).
Why SciML Instead of Traditional Machine Learning?
Traditional machine learning needs large, clean datasets. Water systems rarely have them. Instrumentation is sparse, telemetry is noisy, and models often fail when conditions change. SciML takes a different approach; it weaves physics, engineering design, and limited telemetry into explainable digital twins that predict faults and optimize operation.
With SciML, models start from what engineers already know (pump curves, process equations, site configuration). Telemetry then calibrates the model, and missing signals are inferred through physics, meaning actionable insights without massive sensor deployments.
This approach is already working. At Southern Water’s Matts Hill site, Binnies, Williams Grand Prix Technologies, and JuliaHub used SciML to build a pump digital twin using just four signals. The model delivered 90%+ fault prediction accuracy and identified energy-performance losses from control changes, findings that would have required expensive instrumentation under traditional AI.
The same physics-based approach applies to emissions reduction. In activated sludge plants, SciML digital twins simulate aeration and reaction kinetics to predict N₂O formation, helping operators balance treatment efficiency, energy use, and emissions, directly supporting Ofwat and IED compliance.
Ideal Use Cases
What makes JuliaHub essential is the ability to scale these models, calibrate them continuously, and deploy them securely without “black box” AI or costly pilots.
The article talks about the best use cases for SciML in predictive asset maintenance and optimization. SciML is most valuable for dynamic assets (pumps, rising mains, aeration, sludge treatment) and less suited for static structures (weirs, tanks).
Scientific machine learning has already shown results in aerospace, energy and elite motorsport. SciML offers an important breakthrough in how we understand, predict and optimize both maintenance and emissions performance across water operations.
For water utilities planning digital transformation, the core question is shifting from “How much data do we have?” to “How do we get physics-based predictive insight from what we already know?”
The new paradigm led by Binnies, Williams Grand Prix Technologies, and JuliaHub delivers that. Read the full article.
