The innovation hybrid architecture behind
cognitive operations
The architecture combines IoT data collection with deterministic processing to calculate real-time indicators for performance, sustainability, resilience, and operational state across complex systems of systems. Advanced Large Language Models (LLMs) and AI agents in Cognitive Digital Twin interpret these quantitative outputs, translating complex numerical data into clear, natural-language insights, predictive analytics, and interactive decision support.
How It Works
Project Definition
Establishing a Cognitive Digital Twin project is a structured, joint initialization process that harmonizes operational objectives, domain ontologies, application constraints, and performance indicators. By avoiding rigid, commercial implementation patterns, the architecture adapts dynamically to the unique operational realities of the target system.
What we connect to:
- — IoT sensors and smart meters of any type
- — Industrial control and automation systems (SCADA, PLC, DCS)
- — Building automation systems via BACnet, Modbus, KNX gateways
Building and Customizing Cognitive Digital Twins without code
Based on the client's requirements, we use the Cognitive Digital Twin Builder to model the target system as interconnected subsystems and processes, then configure its data processing and analysis using deterministic methods and machine learning. AI agents assess the change trends, impacts, while an LLM in a specialized harness supports user interactions and interpretation of results. The deployed solution includes an Operational Twin and a no-code Adapter, enabling operational teams to extend the model's scope and adapt its logic as operating conditions change.
System hierarchy
The complex system under study is modeled as a hierarchy of objects—including facilities, production lines, environmental zones, and infrastructure assets—each with defined properties, operational states, and relationships.
Digital Twin Builder
The Twin Builder enables domain experts to construct system models and embed them within the domain knowledge base. It facilitates defining properties for each object—including target metrics, operational statuses, assessment rules, and composite KPI models tailored to capture the key drivers of status transitions.
Operational targets
Operational targets, thresholds, and sustainability rules are configured for each process and object status using indicators whose mathematical formulas link directly to live data streams.
Standards compliance
Relevant international standards—such as ISO 37120 (urban sustainability), ISO 50001 (energy), ISO 14001 (environment), or ISO 9001 (quality)—can be embedded into the model as domain-specific compliance targets.
Data source linking
Data integration has two stages: prototype validation, using representative time-series data to map sources and test the twin; and live integration, connecting the deployed system to continuous data feeds that are checked, normalized, timestamped, and prepared for processing.
Access permissions
Role-based access permissions are configured across user groups—including management, operators, public stakeholders, and external partners—ensuring each persona accesses only authorized system views and data.
The no-code development environment allows domain experts—such as engineers, operations managers, and city planners—to adapt the system model independently, without requiring software development. Organizations retain full ownership of their Cognitive Digital Twins, including all underlying data, intellectual property, and end-user rights.
Each process in the target system is monitored in real time automatically
Once deployed and connected to data sources across the target system, the Cognitive Digital Twin operates continuously. It processes incoming data streams, evaluates key indicator composition, and applies machine learning models to detect status transitions. These assessment results are then interpreted by an LLM engine, providing operators and stakeholders with holistic decision support and actionable operational insights.
What the analytical engine does:
Real-time KPI processing
Dedicated indicator processors transform incoming data streams into indicator time-series representing specific processes within the target system. Values are calculated continuously and aggregated across the system hierarchy, enabling status and change metrics to propagate from individual sensor telemetry up to top-level system health, resilience, and sustainability scores.
Sustainability scoring
Each object and process within the Cognitive Digital Twin is assigned a real-time sustainability status, calculated automatically against defined rules and performance targets. The platform maintains a unified composite health score for the overall system at any moment, while machine learning models continuously evaluate predictive status trends.
Anomaly detection
The Twin platform monitors live operational patterns against established targets to detect statistical deviations and early-stage anomalies before escalation occurs. Diagnostic insights are communicated alongside prescriptive recommendations and severity ratings to guide rapid decision-making.
Predictive analytics
Using machine learning models — including time-series forecasting, supervised classification, and deep learning pattern recognition — the platform projects the future trajectory of each process. Emerging risks are identified hours or days before they become operational problems.
Root cause analysis
When an anomaly or non-compliance is detected, the system analyses contributing factors across connected processes — identifying correlations, cascading effects, and likely root causes automatically. The results are communicated to relevant end users.
Compliance monitoring
The model continuously checks every relevant KPI against embedded international and national standards. Non-compliance events are detected in real time, logged as auditable evidence, and flagged immediately.
Autonomous Oversight
Moving beyond manual screen monitoring, the Cognitive Digital Twin continuously analyzes, interprets, and communicates the operational states and evolving trends of complex real-world systems to relevant stakeholders in real time. It enables AI-assisted, comprehensive assessment of every critical detail and generates actionable recommendations for corrective actions and decisions, supporting both supervisory oversight and closed-loop control to enhance sustainability, performance, and resilience while preventing system failures.
This solution transitions complex system management from passive, manual monitoring to proactive, automated operations, unlocking new levels of improvement in resource efficiency and environmental performance while advancing economic, business, ecological, and societal development goals.
Proven through operational prototype deployments within several Horizon Europe consortia, the framework introduces a mature foundation for upcoming R&D initiatives and strategic technology transfers.
Architectural Alignment
Strong technical fit with legacy and modern infrastructure.
Domain Problem Alignment
Target addressability for complex operational bottlenecks.
Data Feasibility
Robust ingestion of measurements from heterogeneous data sources.
Socio-Ecological & Economic Return
Verifiable impact on performance and sustainability goals.
Targeted Demonstrators
Fast-track validation through small-scale operational pilots.
Low-Friction Integration
Modular, adaptable deployment without vendor lock-in.
Replicability
High transferability across multi-sectoral applications.
Collaborating in Innovation
Developed within the framework of the Horizon Europe program, our technology prototypes represent advancing stages of research and innovation. We invite public sector partners, industrial operators, and research institutions to explore potential applications and discuss tailored deployment scenarios.
Exploring the Live Prototype
Visit pharosnavigator.com to explore a live demonstration of our core baseline architecture.
Please note: This portal represents a previous operational generation. Next-generation features—such as continuous autonomous oversight and advanced LLM-driven decision support—are currently being integrated into the updated framework.