Jordan faces severe water scarcity, with 50% of water produced lost to leakage, theft, and metering issues — a problem known as non-revenue water (NRW). According to a recent paper on arXiv, traditional reactive approaches have proven insufficient for sustained NRW reduction. The authors propose an intelligent framework that integrates EPANET hydraulic modeling, digital twin technology, SCADA systems, and large language model (LLM)-based AI agents for continuous network monitoring and adaptive decision-making.
Framework Architecture and Technology Stack
The system combines real-time data streams with physics-based simulation to detect anomalies. It employs retrieval-augmented generation (RAG) for policy interpretation and function calling for network control. The proof-of-concept implementation uses EPYT (a Python toolkit for EPANET) with offline LLMs — specifically llama3.1:8b via Ollama — on a 1,164-junction district network in Amman, Jordan. This setup enables automated hydraulic simulation, flow-based anomaly detection aligned with water distribution zone (DZ) practice, and AI-generated health reports with response times under 2 minutes and zero API costs.
Burst Detection and Validation
The paper reports a specific validation: a simulated leak of 30.1 L/s produced measurable flow redistribution in 15 pipes, flagging a 15-junction cluster that localizes the burst. This confirms alignment with water distribution zone monitoring practice. The framework accommodates Jordan's intermittent supply patterns and limited automation through a phased implementation strategy, offering a scalable pathway for water-scarce regions to leverage intelligent automation for NRW reduction and operational efficiency.
Implications for Enterprise Infrastructure
For technology leaders managing critical infrastructure, this framework demonstrates how combining physics-based digital twins with LLM agents can move beyond reactive maintenance to adaptive, real-time control. Key metrics from the proof-of-concept include:
| Metric | Value |
|---|---|
| Water loss in Jordan | 50% of produced water |
| Simulated leak flow | 30.1 L/s |
| Pipes showing redistribution | 15 |
| Junctions in flagged cluster | 15 |
| Response time | Under 2 minutes |
| API cost | Zero (offline LLMs) |
By using offline LLMs like llama3.1, the system avoids recurring API costs and can operate in environments with limited connectivity — a common challenge in developing regions. The integration of RAG allows the AI to interpret local policies and regulations, while function calling enables direct control over network valves and pumps.
The framework's reliance on EPANET, an open-source hydraulic modeling standard, and SCADA integration means it can be adopted by utilities worldwide. For enterprise buyers, this represents a proven, cost-effective approach to reducing NRW, which in Jordan accounts for half of all water production. The paper does not disclose specific commercial partners or funding, but the technical validation provides a blueprint for similar initiatives in other water-stressed regions.