Huashan space AI model completes full in-orbit verification, enabling satellites’ autonomous health management

The Huashan vertical aerospace AI health management module, independently developed by Xi'an Zhongke Tianta Technology Co Ltd, has completed full-process in-orbit verification aboard the Chenguang-1 technical experimental satellite, marking the transition of satellite intelligent operation and maintenance from ground simulation to practical application in space.

The module entered orbit with Chenguang-1 on July 24. Designed for in-orbit satellite health management, the Huashan model can analyze telemetry data, detect anomalies, predict trends and make autonomous decisions. Supported by an on-board eight-card high-performance computing cluster, it has completed more than 300 model training tasks covering multiple time-series neural networks and envelope prediction algorithms.

As of Aug 12, the system had completed training, evaluation and prediction for multiple algorithms. Multiple prediction models maintained steady convergence, with comprehensive relative deviation ranging from 8.21 percent to 8.86 percent. Upper-envelope early warning accuracy reached 94.67 percent to 96 percent, enabling early prediction of potential faults in on-board equipment.

Unlike the traditional model of "ground training and in-orbit inference", Huashan establishes a closed loop covering data collection, in-orbit training, risk early warning and health assessment. By enabling autonomous data verification and model training on board, it reduces the need to downlink raw telemetry data and eases space-to-ground communication bandwidth pressure. The approach provides a potential path for the large-scale operation and maintenance of low-Earth orbit mega-constellations.

Zhongke Tianta has also advanced its hardware development. In April, it released a new-generation on-board laser communication terminal with a transmission rate at the 100-gigabit-per-second level. Its specialized intelligent production line officially entered operation in July, with a phase-one annual capacity of 600 units and plans to exceed 1,000 units annually by 2027.

The company will further integrate laser communication hardware with on-board intelligent algorithms, build high-speed space-ground data channels and contribute to the development of China's national space computing network.