Real-time hierarchical risk assessment for UAVs based on recurrent fusion autoencoder and dynamic FCE: A hybrid framework

Abstract

Abstract Effective risk assessment is critical for unmanned aerial vehicles (UAVs) to ensure their safety and reliability. Up to now, the researchers have proposed quite a few methods for the above target. However, these methods are mainly based on path planning and collision theory, the risk caused by the abnormal status of UAVs themselves is generally ignored, which limits the further improvement on their performance. In practice , due to factors such as complicated compositions, variable condition monitoring (CM) data, and scarce failure records, etc., it is always a great challenge to implement the complete information fusion and accurate risk assessment for UAVs based on their real-time status. In this regard, a novel hybrid framework is proposed in this paper, which integrates the qualitative knowledge and the quantitative CM data, to evaluate the real-time hierarchical risk of UAVs. Specifically, the complicated UAV is firstly abstracted as a multi-level evaluating index system considering its qualitative logic compositions. Then, for each low-level index, given its multivariate CM data of several time instants, recurrent fusion autoencoder (RFA), a novel unsupervised neural network architecture, is proposed to extract their robust and complete feature embeddings automatically, where not only the information of variate dimension but also the information of time dimension can be fully fused. Furthermore, the risk of each low-level index is quantified by the adaptive Gaussian mixture model in a probabilistic way, which is truly data-driven with the help of the Bayesian hyperparameter optimization. Finally, the dynamic fuzzy comprehensive evaluation is utilized to evaluate the hierarchical risk of UAVs level by level, it should be noticed that our method can dynamically adjust the weights of each index employing the variable weight coefficients, which can capture the preliminary risk of UAVs more timely compared with the traditional methods. The proposed framework is validated on two typical datasets: the turbofan engine datasets (simulation) and the UAV flight datasets (real). The experimental results demonstrate the effectiveness and superiority of the hybrid framework on robust information fusion and accurate hierarchical risk assessment.