Functional Safety (FuSa) Analysis

Updated 7 February 2026

Functional Safety (FuSa) Analysis is a systematic process that identifies, assesses, and mitigates hazards in electrical/electronic systems by adhering to standards like ISO 26262 and IEC 61508.
It integrates comprehensive risk assessment methods including FMEA, FTA, and model-based approaches to ensure lifecycle safety from concept to runtime supervision.
Recent advances in FuSa incorporate machine learning and formal methods to manage complex, software-intensive systems and address emerging challenges in automotive safety.

Functional Safety (FuSa) Analysis encompasses the systematic identification, assessment, and mitigation of hazards resulting from failures or dysfunctional behaviors in electrical/electronic (E/E) systems, as mandated by domain standards such as ISO 26262 and IEC 61508. FuSa is distinguished by its lifecycle-wide integration, bridging concept definition, advanced engineering, verification/validation, and runtime supervision—aimed at reducing risk due to both random faults and systematic errors, especially in complex, software-intensive automotive systems. Recent advances incorporate model-based, formal, and machine learning–augmented methodologies to address the rising complexity and open operational domains of modern mobility platforms.

1. Core Principles and Standards of Functional Safety

Functional safety is predicated upon the rigorous management of failure-induced risk, requiring explicit theoretical models, formalized processes, and compliance with established regulatory frameworks:

ISO 26262: The foundational automotive FuSa standard, prescribing process requirements for hazard identification, risk assessment, and the definition of safety goals—then decomposing these into functional and technical safety requirements, and allocating Automotive Safety Integrity Levels (ASIL A–D, with D the most stringent) based on Severity (S), Exposure (E), and Controllability (C) triplets (Stolte et al., 2017, Shi et al., 2024).
IEC 61508: The cross-domain framework emphasizing probabilistic safety integrity metrics such as the average Probability of Dangerous Failure on Demand (PFDavg) and the architectural constraints of safety-instrumented systems under both demand and continuous operation (Brissaud et al., 2015).

Safety concepts are supported by modular decomposition, redundancy, coverage assurance, detection and mitigation mechanisms, and, increasingly, by formal methods and advanced assurance cases to address emergent, software-induced hazards (Abdulkhaleq et al., 2017).

2. Hazard Analysis and Risk Assessment (HARA)

Hazard Analysis and Risk Assessment underpins the initial and ongoing phases of FuSa:

Hazard Identification: Exhaustive enumeration of potential malfunctions and their operational consequences using systematic guidewords (e.g., HAZOP variants, skill-graphs, functional decomposition) (Bagschik et al., 2018).
Risk Classification: Each hazardous event is characterized by S (Severity: harm extent), E (Exposure: frequency of occurrence), and C (Controllability: ability of humans/automation to prevent harm), yielding an ASIL via ISO 26262 risk matrices (Stolte et al., 2017, Shefa et al., 4 Nov 2025).
Emergent/Systemic Risks: Functional Failure Identification and Propagation (FFIP), Systems-Theoretic Process Analysis (STPA), and other model-based methods are increasingly employed to detect complex causal chains, system-level control errors, and emergent risks not covered by traditional FMEA/FTA (Shefa et al., 4 Nov 2025, Abdulkhaleq et al., 2017).

HARA Workflow Example

Step	Input/Procedure	Output
Item Def.	Functional range, architecture	List of functions, operating modes
Hazard ID	Systematic guidewords (HAZOP)	Full list of hazards
Risk Score	Rate S, E, C	(S,E,C) tuples; assign ASIL
Safety Goal	Based on ASIL ≥ threshold	Imposed requirements

The HARA is iteratively refined via scenario enumeration, field feedback, and simulation, especially for Automated Driving functions with vast scenario spaces (Bagschik et al., 2018, Stolte et al., 2017).

3. System and Design-Level Analysis

Rigorous system analysis is embedded at each design level and variant, with techniques adapted to system abstraction and complexity:

Early-Stage Methods: FMEA (bottom-up, failure-centric), FHA (top-down, hazard-centric), and FFIP (dynamic, propagative) are widely used for early-phase risk quantification and control design (Shefa et al., 4 Nov 2025).
Variant Management: Representative variant selection is essential for scalable FuSa in highly configurable products, ensuring that the most stringent (“worst-case”) configuration guides all downstream safety requirements, per ISO 26262 (Schranner et al., 2020).
Redundant Architectures: Safety Shells (multi-channel arbitration), monitoring, and fail-operational paradigms are adopted to guarantee both safety and availability in automated and cooperative driving platforms (Hanselaar et al., 2023, Kochanthara et al., 2021).

Component- and subsystem-level artefacts (e.g., diagnostic monitor tables, architecture diagrams) support verification, traceability, and reuse across legacy and new designs, promoting maintainability and auditability (Mohan et al., 2019).

4. Quantitative Safety Metrics and Modeling Techniques

Quantitative metrics and modeling are applied to demonstrate, estimate, and assure functional safety integrity:

Failure Probability Metrics: Metrics such as PFDavg and Probability of Failure per Hour (PFH) are derived via logical fault trees, continuous-time Markov models, stochastic Petri nets, and compact ML-based regression models for hierarchical abstraction (Brissaud et al., 2015, Alexandrescu et al., 2021).
Analytical vs. Simulation-Based Methods:
- Fault-Tree Analysis (FTA): Boolean-algebraic, minimal cut sets, closed-form for static architectures.
- Markov Models: Explicit modeling of repair, test, and time-dependent transitions.
- Stochastic Petri Nets: Simulation of asynchronous/dynamic and cross-hierarchical scenarios.
- ML-Based Compact Models: Provide black-box, IP-protected mapping from parameter vectors to failure/reliability metrics, facilitating cross-layer integration (Alexandrescu et al., 2021, Arunachalam et al., 2024).

Method selection is guided by system complexity, architecture dynamism, required precision, and computational resources (Brissaud et al., 2015).

5. Advanced and Model-Based Approaches

Recent developments advance FuSa by embedding analysis and supervision directly in model-driven or data-centric workflows:

MBSE Integration: SysML-based approaches offer model-to-model transformations, built-in profiles for FMEA, and simulation-driven FFIP within unified digital threads, albeit with varying degrees of tool support for FHA/FFIP (Shefa et al., 4 Nov 2025).
Behavioral Supervisors: Behavior Tree (BT)–based supervisors render the static safety artifacts of ISO 26262—safety goals, FTAs—into dynamic, run-time monitors and orchestrators. BT nodes directly encode hazard detection (from FTA cut-sets) and recovery actions (safe states), ensuring coverage and traceability from hazard analysis to reaction logic (Conejo et al., 2024).
Machine Learning–Driven Anomaly Detection: Unsupervised learning frameworks (feature extraction, clustering, time-series analysis) deliver low-latency, 100% accuracy anomaly detection for analog/mixed-signal circuits, mapping early warnings to safety actions and DTCs (Arunachalam et al., 2024).

Model-based and AI-enabled methods are increasingly crucial for handling the scale and intricacy of software-defined vehicles, realizing rapid iteration, improved traceability, and enhanced coverage of non-obvious hazards (Shi et al., 2024, Petrovic et al., 5 Jan 2026).

6. Practical Implementation, Runtime Supervision, and Lifecycle Integration

Successful FuSa analysis mandates alignment between design-time assurance and runtime operational monitoring:

Runtime Supervisors: ISO 26262-aligned BT-based supervisors guarantee that every detected fault combination, as encoded from FTAs/HARAs, triggers the required safe-state transition within specified fault-reaction latencies (e.g., ≤1.5 s at ASIL D in an industrial Level 3 AD implementation) (Conejo et al., 2024).
Multi-Agent and AI-Enhanced Processes: LLM–empowered multi-agent systems, augmented by Retrieval-Augmented Generation and reflective critique, automate hazard analysis, requirements authoring, and test-case derivation, reducing error rates and development latency by an order of magnitude (Shi et al., 2024).
Scenario-Based Validation: Co-simulation platforms enable exhaustive scenario sampling, requirement refinement, and coverage analysis early in development, bridging the gap between abstract safety goals and concrete implementation constraints (Mohan et al., 2019).

Continuous monitoring, field-data feedback, and safety argumentation link lifecycle stages, facilitating adaptive safety management as systems are updated or retrained (notably in AI-enabled architectures) (Abbaspour et al., 5 Feb 2026).

7. Open Challenges and Future Directions

Key challenges and research directions include:

Emerging Complexity and Emergent Behavior: The high interconnectedness and variability of modern automotive systems—and the use of learning-enabled components—require the adoption of methods capable of modeling systemic, non-local failure propagation (STPA, FFIP, model checking) (Abdulkhaleq et al., 2017, Shefa et al., 4 Nov 2025).
Holistic Safety across Standards and Domains: SOTIF (ISO 21448) analysis must be integrated with classical FuSa for comprehensive coverage of performance insufficiencies, particularly in AI-driven perception, with continuous risk re-classification of QM components based on operational evidence (Abbaspour et al., 5 Feb 2026).
Security-Safety Integration: Addressing intentionally induced faults and cyber-physical threats is now mandatory, with unified threat/hazard logs and risk frameworks (Hänninen et al., 2018).
Toolchain and Methodology Gaps: Further unification of MBSE, formal verification, AI model assurance, and compositional safety arguments—supported by extensible profiles and parametric risk models—is needed for scale and regulatory acceptance (Shefa et al., 4 Nov 2025).

Overall, Functional Safety Analysis has evolved into a multidimensional discipline: deeply model-driven, highly quantitative, and lifecycle-first, with increasing reliance on formal, simulation, and AI-based methods to assure robust operation across expanding operational envelopes and system boundaries.