Abstract This paper explores gaps or potential gaps in systems engineering competencies shown in the INCOSE Systems Engineering Competency Framework (ISECF), derived by examining the competencies needed to perform 21st century systems engineering roles, and by reviewing real‐life examples of applying the INCOSE systems engineering competency framework. The authors recommend updates to the INCOSE framework to account for systems engineering roles (as opposed to other kinds of engineering roles) and how systems engineering is evolving in a global and ultra‐connected world.
Abstract Circular design is widely promoted as a strategy for reducing environmental impacts in high‐volume consumable product systems, yet firms remain uncertain about its economic implications. This paper develops a system dynamics model that integrates cost learning, material recovery, design improvement, customer satisfaction, demand elasticity, and profit‐optimizing price behavior. A linear business‐as‐usual (BAU) configuration is compared with circular design systems initiated under identical market and operational conditions. Results show that circularity does not yield a single characteristic outcome but instead exhibits varying performance regimes depending on the strength of the interactions among learning, recovery, customer value, and reinvestment feedback loops. Strong circular design significantly outperforms BAU across economic and environmental metrics, while moderate capability delivers meaningful, pronounced gains. Notably, even a weak circular design, though initially underperforming, eventually surpasses the BAU system as reinforcing mechanisms mature. These findings demonstrate that circular design is inherently advantageous and becomes increasingly superior when key system capabilities evolve sufficiently to activate reinforcing feedback. This study contributes to the scientific community by providing a generic system dynamics model for the circular economy that serves as a reusable blueprint adaptable to real‐world use cases across different industries.
Abstract In System of Systems (SoS) such as Manned‐Unmanned Teaming (MUM‐T), control structure is a critical design element that determines the allocation of decision‐making authority and interaction patterns among constituent systems. Previous research has adopted a static design approach that selects a single control structure at design time and fixes it throughout the system's lifecycle. However, in actual operational environments, the operational context—including communication availability, threat level, and time pressure—continuously changes, and the optimal control structure varies accordingly. This study proposes a context‐adaptive design methodology, applied at the early conceptual design stage, that enables control structure transitions in response to context changes defined by operational scenarios. The key conceptual shift is augmenting Operational Requirements (ORs) as specified in ISO/IEC/IEEE 29148 — retaining the standard OR as the base capability statement — with conditionally activated control modes. The methodology defines context as an operational environment state vector, reinterprets operational scenarios as sequences of context transitions, and specifies, for each control structure, the valid context range together with transition triggers, transition procedures — distinguishing planned (with handshake) and emergency (without handshake) classes — and safety constraints. These elements are integrated into Adaptive ORs. An illustrative application to a tank company‐drone MUM‐T system shows that the methodology can systematically specify control structure transition mechanisms at the OR level for twelve decision‐related functions, yielding eighteen transition triggers and two procedural classes (Planned and Emergency) with associated safety constraints. The application demonstrates feasibility and internal consistency; empirical validation through simulation, formal verification, and operational testing is identified as future work. This research extends prior work on static OR derivation to address dynamic design decision problems, providing a requirements‐layer foundation for operationally resilient SoS.
Abstract Requirements engineering (RE) plays an integral role in successful systems engineering (SE) as the vast majority of design rework and cost overruns are attributed to poor requirements. Assuring requirement quality remains a time‐consuming and expertise‐intensive activity. There is growing interest in leveraging artificial intelligence (AI) to support RE, especially with the broad availability of Large Language Models (LLMs) that work well with natural language, in which requirements have historically been defined. However, LLMs are stochastic tools, and the empirical evidence regarding the performance and reliability of LLMs in evaluating requirements quality is nascent. To address this gap, this study evaluates how well an off‐the‐shelf LLM evaluates requirement quality for a notional engineered system using INCOSE‐derived criteria. We assess LLM performance by comparing it against a human‐expert generated ground truth, characterizing its performance and variability of errors across independent runs. This study makes three important contributions. First, it provides the first quantitative evidence of the performance of LLM‐based requirement quality evaluation, documenting substantially higher performance in avoiding false alarms compared to reliably detecting quality issues. Second, it documents pronounced variability in LLM behavior, suggesting both its performance and manifestation of AI errors vary drastically. Third, it reveals differences in detection performance for different requirement quality categories. More specifically, LLMs perform poorly in detecting issues that require SE judgement, such as correctness, necessity, and boundedness, and exhibit more reliable performance in detecting linguistically explicit issues such as ambiguity and Verifiability. Overall, these findings indicate that off‐the‐shelf LLMs are not yet suitable for autonomous requirement quality evaluation; yet could be more suitable to assist human experts or serve specialized roles in agentic AI frameworks. Overall, this work provides empirical grounding for AI‐assisted RE quality assessment performance along with its errors and potential risks; highlighting the importance of trustworthy adoption of AI for SE.
Abstract Previous research has argued verification is a knowledge building activity that aims to reduce the uncertainty of the system in its ability to meet requirements. Information theory has been proposed as one method to quantify this reduction of uncertainty from individual verification activities, however, it has not yet been applied to analyzing entire verification strategies. Simultaneously, other researchers have proposed logical foundations to model valid verification strategies as Bayesian Networks, but have not derived a success metric to assess them. This paper proposes new metrics to analyze verification strategies when modeled as Bayesian Networks using information theory. The theory surrounding these metrics gives further insight into how verification reduces uncertainty and which design patterns do so more effectively.
