A Multi-Paradigm Change Impact Analysis for System Model Histories

Zusammenfassung

Model-Based Systems Engineering (MBSE) develops complex systems through system models that engineers collaboratively evolve. Local changes propagate through structural, historical, and behavioral dependencies in non-obvious ways across viewpoints and hierarchy levels. Change Impact Analysis (CIA) is the established response, but existing MBSE approaches are mostly static, single-paradigm (rules, traceability, or graph propagation), and answer only whether an element is impacted, not how severely or in what specific way. We present CHIMERA, a CIA approach for Eclipse Modeling Framework (EMF)-based system model histories. CHIMERA lifts version-to-version differences into semantic actions, discovers four relationship types (family, co-difference, co-metric, dependency), and combines rule-, pattern-, graph-, and ML-based signals in a hybrid predictor for three practitioner questions: binary impact (Q1), magnitude (Q2), and property-change type (Q3). On two independently developed system model histories (769,598 and 768,550 element pairs across 102 and 104 versions), CHIMERA raises Q1 MCC to 0.90/0.88 over the strongest baseline’s 0.56/0.66 and also leads on Q2 and Q3. CIA paradigms provide complementary signals, and combining them yields project-context-specific CIA support for evolving system models.