How Ontology Can Turn the Nuclear Industry into a Computable System

The presentation explains that expanding a nuclear‑industry project from 16 centrifuges to 11,520 creates a massive state‑space where traditional spreadsheets and manual reconciliation become systemic risk, especially when project‑control data can lag up to eight weeks.

01<br/>Expanding System State Space

When only a few devices operate, teams can rely on experience, meetings, and tables to maintain project status. Once the number of centrifuges reaches the ten‑thousand level, every object links to materials, processes, quality records, personnel shifts, suppliers, and critical paths, causing relationships and exception combinations to grow faster than the device count and turning the management problem into continuous computation of system dependencies.

02<br/>Unified Operational Model Makes Equipment Living Objects

The solution starts with defining a sustainably updatable object model that includes sites, centrifuges, parts, tasks, suppliers, defects, inventory, personnel, and milestones. Each object carries attributes and connects to BOMs, process routes, quality batches, schedules, and plans, so a material defect propagates through a graph to compute its impact on installation, staffing, cost, and delivery dates. Unlike simple ETL that merely joins tables, an industrial ontology preserves object identity, business state, semantic relationships, and action permissions, turning each centrifuge into a "living object" with a digital identity across supply, engineering, manufacturing, quality, and operation stages.

03<br/>Agent Workflow: Impact Graph and Solution Optimization

On this architecture, an Agent follows a typical chain: it listens for defect, delay, or resource‑change events; retrieves the relevant objects and their dependencies; generates root‑cause hypotheses; computes impacts on critical path, inventory, and personnel; compares alternative actions such as switching suppliers, expediting shipping, or re‑assigning staff; and finally submits the recommendation and its justification to an approver. The presentation cites a bearing‑quality incident where the system evaluated cost versus schedule, suggested a supplier switch and expedited order, and reduced the transportation time from three weeks to one week. The key point is that the recommendation references quality batch, supplier capability, schedule nodes, and budget impact, avoiding decisions based on isolated data silos.

04<br/>Human‑in‑the‑Loop Transaction Chain

In highly regulated domains, an Agent’s ability to invoke tools does not equal autonomous control. The architecture mandates that every human or AI action be recorded and traceable. The recommended execution flow is: the Agent proposes a solution; experts review and modify it; an authorized person approves; a orchestration layer writes back to scheduling, payroll, procurement, and site‑notification systems via controlled interfaces. This chain requires identity authentication, least‑privilege access, approval policies, idempotent writes, failure rollback, and non‑repudiable audit logs. Model output is only a candidate decision; actual production actions must be deterministic transactions. Maturity for nuclear, aerospace, or pharma is measured by explainability, clear approval, and traceable execution rather than autonomous rate.

05<br/>Metrics Focus on Decision Latency

When project control expands across the full value chain, each defect handling, supplier replacement, and personnel adjustment leaves a structured trace that becomes training data for risk models and optimizers. However, “the system will learn” must not be a vague slogan; enterprises need continuous monitoring of data freshness, anomaly detection latency, recommendation‑to‑approval cycle time, write‑back success rate, critical‑path deviation, and audit coverage. Scaling from 16 to 11,520 devices is really about scaling the organization’s ability to handle complexity. The foundation of industrial AI is not a large model but a closed loop of real‑time events, operational ontology, constraint optimization, human approval, and transaction execution. Only when this loop is stable can Agents move from demo interfaces to high‑regulation production systems.