A novel analytical redundancy method based on decision-level fusion for aero-engine sensors

This paper addresses sensors’ analytical redundancy by a decision-level data fusion approach to improve the aero-engine control system reliability in the cases of impulsive disturbance and performance degradation. To bridge this gap, we propose a data-driven and model-based decision-level fusion framework to provide analytical redundancies of aero-engine sensors in dynamic behavior. On the one hand, a meta-learning model of the online sequential extreme learning machine (OSELM) algorithm is developed to support data-driven redundancy, whose architecture consists of several ELM units and one supervisor OSELM. The flight data are used to train the ELMs offline and then adjust the OSELM weights online, so the data information can be fully captured to increase data-driven regression accuracy. On the other hand, the decision integration strategy is designed and formulated as a convex optimization problem rather than a multivariate linear regression based on individual decisions from data-driven and model-based calculations. Although the model’s inputs are intentionally perturbing, the model-based decision can improve the fused signal’s dependability. For the data-driven method, several ELM units in meta-OSELM can alleviate the impact of impulsive noise and reflect sensitive signal changes without a physical model. We aim to provide an efficient analytical redundancy of the aero-engine sensors in the flight envelope. Simulations on benchmark databases and aircraft engine flight data show that the meta-OSELM yields better generalization performance and stability than the OSELM and the ensemble of OSELM. Furthermore, the analytical redundancy of engine sensors provided by the proposed decision-level fusion strategy performs superior in prediction accuracy compared to data-driven or model-based at typical operation points in the flight envelope.