The Precision of Autonomy: Real-Time Artificial Intelligence in Underground Cable Fault Localization and Water Distribution Leak Localization

Invisible Failures, Visible Consequences

The most dangerous failures in critical infrastructure are not dramatic explosions or cascading blackouts. They are silent. Underground cable insulation degrades millimeter by millimeter. Water mains leak for months beneath asphalt, eroding soil, wasting millions of liters, and weakening cities from below.

For utilities, these failures are notoriously difficult to localize. Underground power networks span thousands of kilometers, with faults propagating through complex impedance paths. Water distribution systems behave as nonlinear hydraulic networks, where pressure drops and transient flows obscure the exact leak location. Traditional approaches—manual patrols, step testing, time-domain reflectometry used in isolation—depend heavily on expert intuition and individual effort. They are slow, expensive, and increasingly misaligned with the scale and complexity of modern infrastructure.

This is the real challenge DXON Controls addresses: how to detect, localize, and respond to invisible failures in large-scale networked systems—before they become public crises.

From Individual Effort to Autonomous Oversight

For decades, infrastructure monitoring has followed a familiar pattern: collect data centrally, visualize it on dashboards, and rely on human operators to interpret anomalies. This paradigm is no longer sufficient.

Modern energy and water systems are large-scale networked control systems—highly coupled, dynamic, and uncertain. Centralized monitoring struggles under the computational load. Purely distributed approaches demand constant communication, creating latency, bandwidth, and cyber-security risks.

Academic research in decentralized energy management offers a crucial insight. In the Periodical Self-Sufficient Repartitioning Approach, large power networks are divided into self-sufficient microgrids, each capable of local decision-making without continuous global coordination. By periodically (decentralized method)—or event-triggeredly (distributed method)—repartitioning the network, control complexity is reduced while resilience is increased.

DXON extends this principle beyond economic dispatch. We apply it to fault localization and predictive maintenance automation, where the goal is not just balancing power—but isolating failure with precision.

In practice, this means moving from individual troubleshooting to autonomous oversight, where each subsystem understands its own health, constraints, and role within the larger network.

Engineering the Shift: Transitioning from Passive Monitoring to Active Autonomy

Many organizations say they want “advanced monitoring systems for the energy sector.” What they often deploy is basic telemetry—data without intent.

Purpose-driven autonomy is different. It begins with a clear technical pivot:

●     Monitoring: Collect data, raise alarms, notify humans.

●     Autonomy: Detect, diagnose, localize, and adapt—continuously and intelligently.

This pivot requires integrating fault-tolerant system design, AI-based predictive maintenance, and distributed control for utilities into a single architectural vision.

At DXON, autonomy is about scaling engineering judgment across thousands of assets, in real time.

The Synergy of Deep Learning and Model-Based Localization

Autonomous fault localization techniques present schemes as a model-based, data-driven, or a combination of data-driven AI and physics-based models.

For the fully data-driven technique doesn’t rely on a system model, but based on the estimation of the state of the complete system from the information from sensors installed throughout the network and topological information.

Model-Based Intelligence

Physical models encode decades of engineering knowledge:

● Electrical network topology, impedance, and load-flow constraints

● Hydraulic continuity, pressure–flow relationships, and transient behavior

Model-based fault detection in control systems excels at interpretability and constraint awareness. It tells us what should be happening.

Data-Driven Intelligence

Deep learning excels at pattern recognition:

● Subtle waveform distortions preceding cable insulation failure

● Pressure micro-oscillations indicating early-stage water leaks

● Correlated anomalies across sensors that humans miss

AI-based predictive maintenance tells us what is happening—and what is likely next.

DXON’s Synthesis

DXON combines these approaches into hybrid diagnostic pipelines:

1. Edge-level inference detects anomalies locally, minimizing latency.

2. Model-based filters validate anomalies against physical feasibility.

3. Distributed intelligence refines localization through coalition-like collaboration between neighboring subsystems.

This mirrors coalition-based microgrid control: when one subsystem cannot explain a fault alone, it collaborates—selectively and purposefully—with others.

The result is not just detection, but precise localization with quantified confidence.

Decentralized, Distributed, and Fault-Tolerant by Design

A critical distinction often missed in industry discussions is the difference between decentralized and distributed control.

● Decentralized control: Subsystems act independently, with minimal or no communication.

● Distributed control: Subsystems exchange information to improve global performance.

DXON architectures deliberately support both. Why? Because fault tolerance demands flexibility.

During normal operation, decentralized analytics reduces communication overhead and cybersecurity exposure. During abnormal conditions—cable faults, pipe bursts, cyber anomalies—distributed coordination is selectively activated. This mirrors event-triggered partitioning in advanced MPC research: communication happens only when it matters.

This design philosophy underpins DXON’s Fault-Tolerant Control Systems:

● No single point of failure

● Graceful degradation instead of cascading collapse

● Local autonomy with global situational awareness

Underground Cable Fault Localization: A Case in Precision

In underground power networks, faults rarely announce themselves cleanly. Partial discharges, moisture ingress, and thermal stress create ambiguous signals long before failure.

DXON’s approach:

● High-frequency sensing at substations and along feeders

● Edge AI models trained to recognize pre-fault signatures

● Topology-aware localization algorithms informed by network partitioning

Instead of searching an entire feeder, the system narrows the fault to a specific segment, often days or weeks before catastrophic failure. Maintenance shifts from emergency repair to scheduled intervention—reducing downtime, cost, and risk.

Water Distribution Leak Detection: Listening to the Network

Water networks speak quietly. Pressure transients, flow imbalances, and acoustic signatures form a language most systems ignore.

DXON’s advanced monitoring systems for the energy sector extend naturally to water utilities:

● Distributed pressure and flow analytics

● AI models that learn normal diurnal patterns

● Autonomous isolation of leak-prone zones through network partitioning

A standard way to facilitate the leakage control is to partition the network into District Metered Areas (DMAs), where the flow and the pressure at the inlet are monitored.

Optimal Sensor Placement for Leak and Fault Localization in Distribution Networks has been studied with model-based and data-based approaches..

The outcome is critical infrastructure resilience—not by reacting faster, but by failing less often.

From Research to Reality: DXON as a Strategic Think-Tank

What differentiates DXON Controls is not the adoption of isolated technologies, but the translation of academic rigor into deployable systems.

We draw directly from:

● Decentralized MPC and self-sufficient partitioning

● Event-triggered coordination and coalition formation (in a distributed manner)

● Cloud–edge energy management algorithms

And we ground them in:

●     SCADA, DCS, EMS, and PLC integration

●     Real-world validation on live infrastructure

●     Security-first, fault-tolerant architectures

This is why DXON is not just a solution provider, but a premier think-tank for strategic growth in decentralized energy management.

Energy-efficient Methods

The latest and most effective methods used in these methodology and analyses are:

The training process consists of an iterative procedure between clustering and deep learning neural network (DLNN) training.

a recursive clustering/learning approach:

for "Graph-based partitioning" stage;

Graph-based clustering techniques are applied to WDN topological information, the partitioning problem is formulated and solved as a multi-objective mixed integer linear program (MILP),

Graph based Interpolation; Large-scale Spectral Clustering; as advanced methods, the state of the complete set of nodes (system) of the network is approximated by means of a graph-based interpolation technique.

for Offline "learning" stage;

the training stage can be tackled by means of standard deep learning DL techniques for image classification, to obtain images by encoding the available data for all possible leak scenarios.

The recursive clustering/learning approach divides the considered region of the network into two sets of nodes using graph agglomerative clustering (GAC) and trains a deep neural network (DNN) to discern the location of each leak between the two possible clusters, using each one of them as inputs to future iterations of the process.

The Future: Autonomous Purpose at Scale

Infrastructure autonomy is not a futuristic aspiration. It is an engineering necessity.

As networks grow larger, more distributed, and more critical, the only sustainable path forward is purpose-driven autonomy—systems that sense, think, and act with precision.

DXON Controls stands at this intersection of theory and practice, turning the hard science of AI and control systems into resilient, intelligent infrastructure.

Reimagine Your Infrastructure

To reimagine your infrastructure and move from reactive maintenance to autonomous purpose, contact DXON Advanced Controls Research LLC (contact@dxoncontrols.com) or visit DXON Controls to see how our strategic growth advisory can transform your network.

Frequently Asked Questions

1. How does AI-based predictive maintenance differ from traditional condition monitoring?

Traditional monitoring reports current states—temperatures, pressures, voltages—leaving interpretation to humans. AI-based predictive maintenance goes further by learning historical patterns, correlating multivariate signals, and forecasting failure probabilities. This allows utilities to intervene before thresholds are crossed, reducing unplanned downtime and extending asset life.

2. Why is decentralized control important for fault detection in large networks?

Centralized systems struggle with scalability and latency in large-scale networked control systems. Decentralized control enables local fault detection and response, reducing computational burden and communication needs. When designed correctly, it increases resilience by eliminating single points of failure while still allowing selective distributed coordination when necessary.

3. Can fault-tolerant system design improve cybersecurity as well as reliability?

Yes. Fault-tolerant system design inherently limits blast radius. By partitioning networks and enabling local autonomy, cyber incidents or sensor failures are contained rather than propagated. Combined with anomaly detection and secure communication, this significantly improves both operational reliability and cyber resilience.

4. How does DXON integrate with existing SCADA and legacy systems?

DXON specializes in integration-first industrial automation solutions. Our platforms sit on top of existing SCADA, DCS, EMS, and PLC environments, extracting value from legacy data without forcing rip-and-replace upgrades. This ensures faster deployment, lower risk, and measurable ROI.

5. What role does distributed control play in underground fault/leak localization?

Distributed control allows neighboring subsystems to share contextual insights—such as impedance changes or pressure gradients—when a fault cannot be resolved locally under a decentralized scheme. This selective collaboration improves localization accuracy using distributed data-driven intelligence, while avoiding constant, bandwidth-heavy communication across the entire network.

6. Is autonomous monitoring suitable for both energy and water utilities?

Absolutely. While the physics differ, both domains are networked systems with hidden failures and high societal impact. DXON’s process monitoring and control systems are designed to adapt AI and model-based techniques across sectors, delivering consistent benefits in reliability, efficiency, and resilience.

DXON Controls
Where autonomy meets precision.