An Uncertainty-Aware Provenance Framework for Enhanced Traceability in Engineering Systems

Abstract

In the design and analysis of complex aerospace systems, maintaining rigorous traceability of design decisions, data transformations, and modeling assumptions is essential for ensuring system integrity, enabling certification, and supporting life cycle assurance. Data provenance— defined as structured documentation of data sources, transformation processes, responsible agents, and applied tools—has emerged as a key enabler or transparency and auditability across a wide range of engineering workflows. Formal standards such as the W3C PROV-O ontology model provides the foundation for representing relationships in a consistent format. However, despite their utility, most existing provenance frameworks fall short in capturing and managing uncertainties- particularly when it rises from simulations, sensor measurements, or engineering assumptions characterized by incompleteness, variability, or evolving probabilistic definitions. This limitation presents a significant barrier to robust traceability in high-stakes contexts, where confidence in the model-based reasoning is critical. Although broadly relevant across engineering domains, this challenge is especially pronounced in Model based Systems Engineering (MBSE), where digital models are treated as authoritative throughout the system lifecycle and traceability under uncertainty is paramount. To address this gap, we prose a unified, extensible framework that integrates uncertainty quantification (UQ) with structured provenance modelling. The framework systematically combines quantitative analysis with expert judgment and validate observational data, enabling the rigorous tracking, classification, and reduction of uncertainty across engineering workflows. Each source of uncertainty—whether originating in data, models, or assumptions—is traced to its origin and ranked by its impact on system-level outcomes via sensitivity analysis. Parameters with negligible influence are pruned to simplify the provenance model and prioritize critical drivers. The refined set of uncertainty is then propagated through the system, with each stage—from identification and transformation to reduction and verification—captured within an extended and machine-interpretable provenance graph. This graph not only links the inputs to outputs but also maps assumptions to decisions, quantifies their influence, and embedding the metadata to support interpretability. In particular, it provides a transparent and auditable record of the decision-making process, allowing engineers to understand not only what was done, but also why, by whom, under what confidence level, and with what implications—thereby advancing traceability, accountability, and systems assurance throughout the lifecycle. As a demonstration of applicability and effectiveness of the proposed framework, the electric motor of an Electromechanical Actuator (EMA) is presented as use case. This illustrates how the framework enhances end-to-end traceability, reduces modeling overhead, and promotes more transparent, auditable, and risk-informed design practices. The current simulations yield robust design with less computational time. While the EMA is situated within aerospace domain, the framework could be generalized to other safety-critical and interdisciplinary contexts where uncertainty is prevalent. The proposed provenance-aware workflows expand the current traceability paradigm. It equips decision-makers with a verifiable and actionable account of how uncertainties are identified, propagated, and mitigated—supporting robust model validation, regulatory compliance, and long-term accountability across the system lifecycle.