A model learning strategy adapted to health assessment of multi-component systems

For the issue of multi-component system health assessment, an approach of steady algebraic system model learning from the condition data collected over the system serving period was discussed. Faced with the problem of measurement deficiency and inter-component coupling in system working process modeling and drift evaluation, an inclusive data-driven framework supported with corresponding domain knowledge was proposed, aimed at building up appropriate component-level surrogate models and having them organized in agreement with conservation laws and other real-in-world physical constraints. The comparison between component performance curves before and after degradation was achieved by feeding the learning model with corresponding condition data. The efficacy and interpretability of the proposed methodology was validated by implementation on health assessment process of a type of commercial aircraft turbofan-engine, and provided detailed degradation status which is in accordance with real degradation modes of engine components.