With the advancement of deep-space missions such as crewed lunar exploration and Mars exploration, artificial intelligence (AI) technologies are being increasingly embedded into aerospace engineering management processes, and managers' AI competency has become a critical factor affecting mission delivery performance and safety compliance. However, traditional project management competency models have been primarily constructed around process control and redundancy design within deterministic systems, failing to explicitly address governance challenges introduced by AI integration—including black-box opacity, accountability allocation, traceable review of AI outputs, and accountability mechanisms for assisted decision-making. Although existing NASA and ESA frameworks provide fundamental support for aerospace project delivery, AI-related competencies in aerospace management exhibit inherent interdependencies. Managers are required to continuously balance delivery schedules and cost pressures against mission assurance obligations such as verification and validation, configuration control, and compliance, making it impossible to simply represent AI competency as a linear list of attributes. Therefore, how to construct an AI competency assessment framework that captures governance priorities and multi-attribute interdependencies has become a critical issue urgently to be addressed in aerospace engineering management in the AI era.
In a recent study published in Space: Science & Technology, a joint team comprising researchers from Tsinghua University, Nanjing University of Aeronautics and Astronautics, and other institutions proposed a closed-loop modeling and assessment framework for aerospace engineering managers' AI competency based on multi-attribute decision-making. The study first identified core competency characteristics through RepGrid interviews with 30 aerospace engineering managers. Through principal component analysis (PCA) on the exploratory sample and confirmatory factor analysis (CFA) on the validation sample, a competency model was constructed and validated, comprising five dimensions—AI risk control and management, lifecycle AI coordination, AI cognitive readiness, AI compliance and safety assurance, and aerospace AI scenario enablement—with a total of twenty attributes. Subsequently, 14 experts were invited to construct an influence network relationship diagram using the DEMATEL method, identifying AI risk control and management as the upstream driving dimension and lifecycle AI coordination as the cross-dimensional hub, with global weights derived via the DANP method. Finally, using five real candidates as subjects, the improved VIKOR method was employed for candidate ranking and bottleneck diagnosis, achieving an intraclass correlation coefficient (ICC) of 0.912 for inter-rater reliability, with rankings remaining stable under different risk preferences. This framework provides a traceable decision-support tool for the selection, performance management, training planning, and career development of aerospace engineering managers in AI-enabled environments, offering significant engineering management value for ensuring the delivery quality, safety, and compliance of deep-space exploration missions in the AI era.
First, this study focuses on the competency challenges faced by aerospace engineering managers in the AI era, and constructs a five-dimensional AI competency model through a systematic combination of qualitative and quantitative methods. With the advancement of deep-space missions such as crewed lunar exploration and Mars exploration, AI technologies are being increasingly embedded into aerospace engineering management processes. Managers are not only required to meet delivery schedule and cost targets but also to critically evaluate AI recommendations, organize the verification and validation of AI outputs, and maintain human accountability in key decision-making. Traditional competency models have not explicitly addressed the governance challenges introduced by AI integration—including black-box opacity, accountability allocation, and traceable review—and AI-related competencies in aerospace management exhibit inherent interdependencies, making it impossible to simply represent them as a linear list of attributes. To address this, the study first extracted core competency characteristics through RepGrid interviews with 30 aerospace engineering managers. After screening by mention frequency, importance scoring, and expert review, 20 core attributes were finalized. Subsequently, principal component analysis (PCA) was performed on the exploratory sample, extracting five dimensions as shown in Fig. 1: AI risk control and management, AI compliance and safety assurance, AI cognitive readiness, lifecycle AI coordination, and aerospace AI scenario enablement. Confirmatory factor analysis (CFA) on the validation sample was then conducted to test the model's reliability and validity. Both the first-order five-factor model and the second-order hierarchical model achieved satisfactory fit indices; the composite reliability of each dimension exceeded 0.8 and the average variance extracted (AVE) exceeded 0.5, indicating that the model exhibits good reliability and validity.
Second, the paper constructs an influence network relationship diagram among the competency dimensions and attributes using the DEMATEL (Decision Making Trial and Evaluation Laboratory) method, and derives the global weights of each dimension and attribute via the DANP (DEMATEL-based Analytic Network Process) method. Fourteen experts with extensive experience in aerospace engineering management conducted pairwise evaluations of the direct influence relationships among the dimensions and attributes. The average relative error rate of rating consistency was 2.35% (<5%), indicating a high degree of consensus among the experts. Fig. 2 presents the influence network relationship diagram at the dimensional level, where the AI risk control and management dimension exhibits the highest positive driving degree, indicating its upstream driving position; the lifecycle AI coordination dimension possesses the highest centrality, serving as the hub connecting all dimensions; and the aerospace AI scenario enablement dimension exhibits a negative driving degree, belonging to the outcome-oriented downstream dimension. This "safety-first" governance logic is highly consistent with the high-reliability culture and "zero-failure" requirements of aerospace engineering, as risk boundaries and assurance mechanisms must be defined prior to the large-scale application of AI scenarios—a concept aligned with the "safety cage" architecture proposed by the European Cooperation for Space Standardization (ECSS). Fig. 3 presents the influence networks among attributes within each dimension, where attributes such as confidentiality red-line enforcement and AI fault response and emergency handling demonstrate relatively strong driving effects. Based on the converged limit supermatrix, the DANP method derives the global weights of each attribute. AI use-case identification across the lifecycle, critical evaluation of AI outputs, human–AI decision boundary management, and AI project integration management rank as the top four, with the weight allocation emphasizing traceable definition of AI use cases, clear decision boundaries, and rigorous review of AI-assisted outputs.
Finally, the paper conducts competency assessment, ranking, and bottleneck diagnosis on five real candidates using the improved VIKOR (VlseKriterijumska Optimizacija I Kompromisno Resenje) method, and validates the robustness of the results through sensitivity analysis. Four independent experts evaluated the twenty attributes of the five candidates. Inter-rater reliability analysis indicated an average intraclass correlation coefficient (ICC) of 0.912, demonstrating good consistency in the ratings. Fig. 4 presents the proximity radar charts of the five candidates across the twenty attributes, where a proximity value closer to 1 indicates that the candidate's radar line approaches the outer ring more closely, signifying that the candidate is closer to the ideal state for that attribute, thereby enabling rapid identification of each candidate's relative strengths and weaknesses. Based on the global weights derived from the DANP method, the weighted total gap S and the maximum weighted regret L for each candidate were calculated. The results show that candidate H5 performs optimally on both S and L metrics. Fig. 5 presents the sensitivity analysis, demonstrating that candidate rankings remain stable under different risk preference coefficients V, indicating that the recommended results are insensitive to decision preference settings. However, the gap between candidates H5 and H3 does not reach the VIKOR acceptable advantage threshold, suggesting that both candidates should proceed together to the final review stage, with targeted verification conducted based on the bottleneck characteristics identified by the Lj terms. This framework provides a traceable and interpretable decision-support tool for the selection, performance management, training planning, and career development of aerospace engineering managers in AI-enabled environments.
