Executive Summary
AI augments engineering — it never replaces the safety barrier. SAFE-C's decision path is deterministic, not probabilistic.
The Principle
AI and analytics operate strictly as an advisory and optimisation layer within TransO2. Deterministic, PLC-based safety logic remains the actual safety barrier for every safety-relevant decision.
Why This Matters
Probabilistic inference is well suited to trend analysis, anomaly detection and predictive maintenance. It is not an appropriate basis for a safety-critical isolation or suppression decision, which must be verifiable, repeatable and certifiable. SAFE-C's Detect → Verify → Isolate → Suppress/Control → Alarm → Record sequence runs on deterministic control logic; AI-derived signals may inform monitoring dashboards but do not trigger suppression directly.
This publication describes conceptual frameworks and proposed architectures. Where quantitative figures appear, they are engineering projections intended for planning and are subject to experimental validation.
Human Override
Crew and maintenance staff retain override authority over automated responses at every stage of this sequence.
Glossary
References
References point to established engineering disciplines and public frameworks, not fabricated citations.
- General fail-safe / deterministic-control engineering principles
- TransO2 SAFE-C architecture documentation
Suggested Citation
TransO2 Engineering Team. Deterministic Safety Logic: Why AI Does Not Control Suppression. Knowledge Centre Publication TB-103. TransO2. v1.0. 2026.
About TransO2
TransO2 is developing a modular Environmental Operating System for railway rolling stock — integrating air, thermal, energy, safety, water and digital intelligence. Engineering concepts and performance targets are subject to detailed design, testing, validation and certification.