Empowering All-in-Loop Health Management of Spacecraft Power System in the Mega-Constellation Era via Human-AI Collaboration

It is foreseeable that the number of spacecraft will increase exponentially, ushering in an era dominated by satellite mega-constellations (SMC). This necessitates a focus on energy in space: spacecraft power systems (SPS), especially their health management (HM), given their role in power supply and high failure rates. Providing health management for dozens of SPS and for thousands of SPS represents two fundamentally different paradigms. Therefore, to adapt the health management in the SMC era, this work proposes a principle of aligning underlying capabilities (AUC principle) and develops SpaceHMchat, an open-source Human-AI collaboration (HAIC) framework for all-in-loop health management (AIL HM). SpaceHMchat serves across the entire loop of work condition recognition, anomaly detection, fault localization, and maintenance decision making, achieving goals such as conversational task completion, adaptive human-in-the-loop learning, personnel structure optimization, knowledge sharing, efficiency enhancement, as well as transparent reasoning and improved interpretability. Meanwhile, to validate this exploration, a hardware-realistic fault injection experimental platform is established, and its simulation model is built and open-sourced, both fully replicating the real SPS. The corresponding experimental results demonstrate that SpaceHMchat achieves excellent performance across 23 quantitative metrics, such as 100% conclusion accuracy in logical reasoning of work condition recognition, over 99% success rate in anomaly detection tool invocation, over 90% precision in fault localization, and knowledge base search time under 3 minutes in maintenance decision-making. Another contribution of this work is the release of the first-ever AIL HM dataset of SPS. This dataset contains four sub-datasets, involving 4 types of AIL HM sub-tasks, 17 types of faults, and over 700,000 timestamps.

💡 Research Summary

The paper addresses the looming challenge of spacecraft power‑system (SPS) health management in the upcoming era of satellite mega‑constellations (SMC), where thousands of satellites will operate simultaneously and traditional health‑management processes become unsustainable. Statistics show that more than 85 % of on‑orbit anomalies involve SPS, leading to a high failure rate and substantial mission risk. Conventional All‑in‑Loop Health Management (AIL HM) follows a four‑step pipeline—work‑condition recognition, anomaly detection, fault localization, and maintenance decision‑making—largely performed by human experts using rule‑based logic, statistical or machine‑learning detectors, and extensive documentation. In the SMC context, three fundamental problems arise: (1) manpower cost explosion, as maintaining 24/7 expert teams for thousands of satellites is infeasible; (2) textual information deluge, because design manuals, maintenance handbooks, and logs become too voluminous for manual review; and (3) operational complexity, as the number of logical rules and fault signatures exceeds human cognitive limits.

To overcome these issues, the authors propose the Aligning Underlying Capabilities (AUC) principle and develop SpaceHMchat, an open‑source Human‑AI Collaboration (HAIC) framework that integrates large language models (LLMs) with domain‑specific tools across the entire AIL HM loop. The AUC principle first classifies each sub‑task by its intrinsic nature (logical reasoning, tool‑dependent, learning‑and‑approximation, knowledge‑intensive) and then matches the human capabilities required (judgment, tool operation, experience accumulation, knowledge retrieval) with corresponding LLM abilities (chain‑of‑thought reasoning, function calling, fine‑tuning, retrieval‑augmented generation). This systematic alignment guides the design of SpaceHMchat.

Key technical contributions:

1. Work‑Condition Recognition – Implemented via prompt engineering, step‑wise prompting, and chain‑of‑thought (CoT) to emulate expert logical reasoning. The system achieves 100 % conclusion accuracy on a test set of rule‑based scenarios.

2. Anomaly Detection – Leveraged LLM function‑calling and a Model Context Protocol (MCP) to invoke advanced statistical, machine‑learning, or deep‑learning detectors as external tools. The framework attains >99 % success in tool invocation and maintains real‑time detection performance.

3. Fault Localization – Trained a specialist LLM using LoRA, Supervised Fine‑Tuning (SFT), and Group Relative Policy Optimization (GRPO) on a curated historical fault database. The model reaches >90 % precision in identifying fault types from telemetry streams, effectively replicating expert pattern‑recognition.

4. Maintenance Decision‑Making – Integrated Retrieval‑Augmented Generation (RAG) with a domain knowledge base containing design documents, maintenance manuals, and logs. Knowledge‑base search and synthesis are completed in under three minutes, providing transparent reasoning traces and decision justifications.

To validate the framework, the authors built a hardware‑realistic fault‑injection experimental platform that reproduces the electrical, thermal, and power dynamics of a real SPS. They also released a high‑fidelity simulation model mirroring the hardware platform. Both resources, together with the XJTU‑SPS dataset, are open‑source. The dataset comprises four sub‑datasets aligned with the four AIL HM stages, includes 17 fault categories, and contains more than 700 k timestamps—making it the largest publicly available SPS health‑management dataset to date.

SpaceHMchat also introduces a three‑tier human‑AI staffing model: junior assistants interact with the LLM via a conversational interface to handle routine monitoring and diagnostics; senior subsystem experts intervene only for high‑complexity tasks such as root‑cause analysis and strategic maintenance planning; the LLM continuously learns from these interactions, reducing the need for large expert teams. Experiments demonstrate that this model can cut required personnel by an order of magnitude while doubling overall maintenance efficiency.

Transparency and interpretability are emphasized by logging the LLM’s reasoning steps, tool calls, and knowledge‑base citations, which are presented to operators in real time. This satisfies the stringent reliability and accountability requirements of space missions, where ultimate authority must remain with human controllers.

Overall impact: The paper delivers a comprehensive, reproducible solution for scaling SPS health management to the mega‑constellation scale. By systematically aligning human capabilities with LLM functions, providing a realistic experimental testbed, and releasing extensive data and code, the work sets a new benchmark for Human‑AI collaborative fault management in aerospace. Future research can extend SpaceHMchat to other spacecraft subsystems, incorporate multimodal inputs (e.g., imagery, acoustic data), and explore tighter integration with autonomous on‑board agents while preserving human oversight.