Abstract:

To address the issue that the failure rate series per 10⁴ flight hours of general aviation aircraft components is limited in length due to constraints from operational cycles and sample acquisition conditions， thereby degrading the performance of time-series modeling， a Bayesian prediction method for the failure rate per 10⁴ flight hours of general aviation aircraft components enhanced by neighborhood features is proposed. The large language model is utilized to extract component entities from unstructured historical failure texts to construct a component-level failure time-series dataset. The key failure components are identified by combining rank-Copula co-occurrence analysis and weighted principal component analysis， and the sets of neighbor components with significant failure co-occurrence relationships are mined. The failure sequences of these neighbor components are then incorporated as covariates into Bayesian time-series modeling to realize the short-term prediction of the failure rate per 10 000 flight hours of key components under small-sample conditions. Comparative experiments showed that the proposed method outperforms other comparison methods across multiple error metrics. Furthermore， cross-aircraft-type validation and rolling time-window experiments are conducted to evaluate the generalization ability and stability of the method. The results demonstrate that the proposed method achieves favorable prediction performance under different aircraft types and training sample sizes， and can more accurately characterize the short-term variation of the failure rate per 10⁴ flight hours of aircraft components under small-sample and non-stationary conditions. This research is expected to assist maintenance managers in the rational allocation of maintenance resources and improve fleet operational management efficiency.

Key words: general aviation, failure rate, small-sample, reliability, large language model