AI- and Ontology-Based Enhancements to FMEA for Advanced Systems Engineering: Current Developments and Future Directions

Technical Summary: AI- and Ontology-Based Enhancements to FMEA for Advanced Systems Engineering

Problem Statement
Traditional Failure Mode and Effects Analysis (FMEA) remains a cornerstone of reliability engineering but faces critical limitations in the context of modern, complex systems. Conventional FMEA is characterized as a manual, document-centric, and expert-dependent process, typically implemented via static spreadsheets. These methods struggle to keep pace with rapid design iterations, cross-disciplinary collaboration, and the integration of feedback across a product's lifecycle. Key issues include:

• Subjectivity and Bias: Reliance on unstructured brainstorming leads to inconsistent Risk Priority Number (RPN) calculations and analyst bias.
• Lack of Traceability and Interoperability: Failure knowledge is often trapped in natural language documents, lacking formal structure, semantic consistency, or linkage to digital system models (e.g., CAD, simulation).
• Inability to Handle Complexity: Traditional methods often fail to capture emergent interactions in highly coupled systems or "systems of systems," particularly when dealing with novel technologies lacking historical failure data.
• Static Nature: FMEA is frequently treated as a compliance exercise rather than a dynamic, living engineering tool, limiting its adaptability to design changes.

Methodology
This paper presents a state-of-the-art review and synthesis of recent advances aimed at transforming FMEA through the convergence of Artificial Intelligence (AI) and ontology-based knowledge engineering within the framework of Model-Based Systems Engineering (MBSE). The review is structured thematically:

1. Critical Analysis of Limitations: Examining the evolution of FMEA standards (e.g., IEC 60812, AIAG-VDA) and identifying gaps in managing system complexity and lifecycle integration.
2. AI-Driven Enhancements: Reviewing the application of Machine Learning (ML), Natural Language Processing (NLP), and Large Language Models (LLMs) to automate failure prediction, prioritization, and knowledge extraction from unstructured data (maintenance logs, repair reports).
3. MBSE and Function Modelling: Exploring the transition from document-centric to model-driven engineering, specifically focusing on the Function-Behaviour-Structure (FBS) paradigm as a foundation for linking system intent to reliability analysis.
4. Ontological Formalisation: Investigating the role of ontologies in formalizing system knowledge (functions, behaviours, failures) to enable semantic reasoning, interoperability, and explainability.
5. Hybrid Frameworks: Synthesizing approaches that combine AI (for data-driven prediction and generation) with ontologies (for semantic constraints and logical reasoning) to create explainable, adaptive FMEA workflows.

Key Contributions
The paper makes four primary contributions to the field of systems engineering and reliability:

1. Critical Analysis of Conventional FMEA: It systematically outlines the limitations of traditional FMEA in managing increasing system complexity and highlights the need for a shift from static documentation to dynamic, model-integrated processes.
2. Review of MBSE and Function Modelling: It details the transition toward MBSE and function modelling (specifically the FBS framework), emphasizing their role in improving traceability between design objectives, physical implementation, and risk management.
3. Synthesis of Ontological Roles: It articulates how ontologies serve as a formal mechanism to represent engineering knowledge, enabling machine-processable yet human-verifiable models that support semantic reasoning and cross-domain interoperability.
4. Evaluation of AI Techniques: It evaluates specific AI techniques—including probabilistic models (Bayesian Networks), supervised/unsupervised ML, NLP, and LLMs—demonstrating their capacity to augment FMEA in failure detection, RPN automation, and the extraction of insights from unstructured textual data.

Results and Findings
The review synthesizes evidence from numerous case studies and technical developments:

• AI Capabilities: AI techniques have shown significant potential in automating RPN calculation (reducing subjectivity), predicting failure modes using historical data (e.g., via ensemble learning and deep transfer learning), and extracting structured failure knowledge from unstructured maintenance logs using NLP and transformer models.
• Ontological Benefits: Ontologies provide a semantic layer that bridges the gap between disparate engineering artifacts. They enable the formalization of FMEA data, allowing for automated inference of failure propagation, consistency checking, and the creation of knowledge graphs that link failure modes to system functions and structures.
• Hybrid Synergy: The integration of AI and ontologies addresses the "black box" problem of pure AI. Ontologies constrain AI outputs (particularly LLMs) to ensure alignment with engineering logic and domain standards, thereby enhancing explainability and trust. Conversely, AI accelerates the population and maintenance of ontologies.
• Tooling and Standards: The paper identifies a growing ecosystem of tools (e.g., Protégé, MetaGraph, KARMA) and standards (e.g., SysML v2, Industrial Ontologies Foundry) that support the implementation of these integrated frameworks, though widespread industrial adoption remains a work in progress.

Significance and Future Directions
The paper posits that the convergence of AI, ontologies, and MBSE represents a paradigm shift toward "intelligent, explainable, and model-integrated" systems engineering. Its significance lies in offering a structured roadmap for embedding FMEA within knowledge-rich engineering environments, moving beyond static checklists to dynamic, adaptive risk management.

However, the authors maintain a modest and critical stance regarding current readiness. They identify several persistent challenges that must be addressed before these technologies can be fully scaled in industry:

• Data Quality and Availability: AI models require high-quality, granular datasets, which are often scarce for novel systems.
• Explainability and Trust: The "black box" nature of complex AI models remains a barrier to acceptance in safety-critical domains; ontologies are presented as a necessary component to ensure traceability and justification of AI decisions.
• Standardization and Interoperability: The lack of unified ontological patterns and semantic standards across different engineering tools and domains hinders scalability and knowledge reuse.
• Interdisciplinary Adoption: Successful implementation requires bridging the gap between ontology researchers, AI developers, and practicing systems engineers, necessitating new toolchains and cultural shifts toward formalized knowledge management.

Ultimately, the paper concludes that while AI and ontology-enhanced FMEA is not merely an evolution of existing practice but a cornerstone of next-generation intelligent systems engineering, realizing its full potential requires coordinated efforts in standardization, tool development, and interdisciplinary collaboration.