Cyber-physical production systems increasingly rely on artificial intelligence to coordinate sensing, control, and decision making across tightly coupled physical and digital layers. As learning models become embedded within production workflows, conventional automation architectures struggle to maintain consistent quality assurance and lifecycle governance. Model behavior evolves over time, data distributions shift, and decision logic becomes less transparent, particularly in safety and quality sensitive environments. This work introduces an AI-native decision support framework that integrates quality assurance mechanisms and lifecycle controls directly into cyber- physical production pipelines. The framework combines model- centric orchestration, continuous validation, explainability-aware monitoring, and governance feedback loops to support reliable operation across deployment stages. Evaluation across representative production scenarios demonstrates improved defect detection stability, reduced decision volatility, and enhanced operational transparency without compromising system scalability.
Hrytsenko et al. (2022) studied this question.