Critical Review on Verification and Validation Frameworks for AI-based Diagnostics in Military CBM+

• Yun-Young Hwang,
• Hyeongsun Ka,
• Jinyoung Kim,
• Jiseong Son,
• Hyoungseop Shim,
• Jiwoo Choi,
• Heemin Kim,

Vol. 20, No. 3, March 31, 2026
10.3837/tiis.2026.03.005, Download Paper (Free):

• CBM+
• cost-sensitive metrics
• logistics and transport
• predictive maintenance
• Remaining useful life
• verification and validation

Abstract

The rapid adoption of Artificial Intelligence (AI) for Condition-Based Maintenance Plus (CBM+) has outpaced the development of rigorous Verification and Validation (V&V)frameworks. This discrepancy creates a substantial trust gap between promising remaininguseful life (RUL) prediction models and their reliable field performance in Logistics and Transport (L&T) operations. Distilling military-grade CBM+ V&V, this review proposes a cost-ensitive framework tailored to L&T operations, effectively addressing their asymmetric risks and offering a clear methodological distinction. This critical review distills military-grade CBM+ V&V practices and translates them into actionable requirements tailored to L&T contexts. We systematically searched Web of Science and Scopus for studies published between 2013 and 2025 using terms related to CBM+, predictive maintenance, V&V, and RUL, screened 427 records, and synthesized the findings from 123 studies core studies and guidelines that satisfied inclusion criteria on data fidelity, performance metrics, andoperational sustainment. The review identifies three pillars of trustworthy AI diagnostics inL&T: First, Verification must be explicitly cost-sensitive, asymmetrically penalizing false negatives and late RUL overestimation to minimize transport delay and safety risk. Second, Validation should rely on containerized, fully reproducible testbeds—such as Docker environments with pinned dependencies and complete data/version lineage—to ensure portability across distributed assets. Third, Sustainment requires continuous monitoring for data and model drift, supported by statistical tests and MLOps-enabled remodeling with auditable release management.

Statistics

Cite this article