Applications of artificial intelligence technology in space nuclear reactor systems

Space nuclear power systems have attracted increasing attention as enabling technologies for future deep-space exploration because of their high energy density, long-duration power generation capability, and independence from terrestrial environmental conditions. Artificial intelligence (AI) has shown considerable potential for addressing key challenges in space nuclear systems, including intelligent design optimization, state prediction, fault diagnosis, and autonomous operation. This review provides a systematic overview of AI applications in space nuclear power systems, with particular emphasis on current capabilities, limitations, and future development directions. The classifications, system architectures, and representative technologies of space nuclear reactors are first introduced to establish the engineering context and identify the challenges that motivate AI integration. Machine learning (ML), intelligent optimization algorithm (IOA), deep learning (DL), reduced-order model (ROM), and data assimilation (DA) are reviewed with emphasis on their applicability under the specific constraints of space nuclear systems, including limited sensing capability, stringent size, weight, and power (SWaP) requirements, radiation environments, and scarce operational data. Representative AI applications are discussed in terms of system design optimization, state prediction and digital twins (DTs), fault diagnosis, health management, and prospective autonomous operation. The current maturity of AI-enabled applications in space nuclear systems is also assessed. Existing studies rely primarily on high-fidelity simulations, ground-based demonstrations, and transferable methodologies from terrestrial nuclear engineering, while operational data from space nuclear reactors remain unavailable. Key challenges associated with onboard computational limitations, radiation-induced hardware degradation, model reliability, and verification and validation (V&V) requirements are also discussed. Future research directions toward trustworthy AI, physics-informed learning, adaptive digital twins, and intelligent autonomous operation frameworks are then summarized. This review provides a reference for understanding the current status and future potential of AI technologies in space nuclear power systems.