Health prediction under limited degradation data for rocket engine bearings via conditional inference knowledge-enrichment approach

With the high reliability requirement of the liquid rocket engine (LRE), it is urgent to develop an accurate health condition prediction methodology for the key mechanical components in the engine. Owing to the high cost of rocket engine tests, degradation data of mechanical components such as bearings, are limited. Additionally, existing methods mostly rely on statistical indicators to monitor the condition of liquid rocket engines (LREs) and typically focus on a single operating condition, resulting in incompetent while dealing with prediction tasks under the inexperienced conditions and limited degradation knowledge. This paper addresses this challenge by proposing a health condition prediction framework that leverages the Conditional Inference Generative Adversarial Network (CIGAN) to enhance knowledge from limited degradation data. Firstly, a variational augmented degradation feature extraction model is employed to develop a generic health indicator (HI). Then, a condition inference restrained degradation data generation strategy is proposed based on the generative adversarial mechanism. Finally, the health indicator is predicted using the designed domain-adaptation method. The effectiveness of this knowledge-enrichment framework, based on the CIGAN approach, is validated through comparisons with benchmark algorithms using real monitoring data from a high-precision cryogenic rocket engine test platform.