Fault Diagnosis Framework for Space TWTAs by Combining the Knowledge Graph and Large Language Model

The statistical data of space traveling-wave tube amplifier (TWTA) fault events record fault symptoms, causes, and solutions in the form of text. Effectively mining such text data can potentially enhance the reliability of devices. In this study, a fault diagnosis framework for space TWTAs is designed by combining knowledge graph and large language model. Specifically, the requirement analysis, graph retrieval, information judgment, and answer generation of the framework are all accomplished by a single large language model to enhance its stability. The knowledge graph serves as the retrieval knowledge database of the framework to enhance the professionalism. The performance of the framework is tested in practical scenarios. The results demonstrate that 92 responses are provided for 100 fault description by the framework, among which 87 responses are adopted, with an adoption rate of 94.56%. Moreover, the framework exhibits stable response consistency for the same questions and shows stability against data with noise. The performance on a larger-scale test set has slightly decreased. However, it still maintains an accuracy of 88.13%, which implies that the framework can provide effective auxiliary decision-making for fault diagnosis of devices.