Abstract. This paper is devoted to the development and validation of a methodology for constructing a predictive maintenance model for aircraft systems based on statistical analysis of real operational data. The failure category “External Lights” was selected as the model object, as it is characterized by a high event frequency and a critical impact on flight schedule regularity. The study is based on a dataset comprising 13,204 maintenance and repair records for nine aircraft collected over a seven-year period. The methodological novelty of the work lies in the justification of the data preprocessing procedure. A 0.95-quantile-based filtering approach was applied to censor anomalous inter-event times associated with extended aircraft downtime. The key result is the statistical justification for selecting the exponential distribution to describe the failure process. A comparative analysis with the two-parameter Weibull distribution, using the Akaike Information Criterion and the Kolmogorov–Smirnov goodness-of-fit test, demonstrated no statistically significant improvement in accuracy when increasing model complexity (shape parameter β ≈ 1.04). The final outcome is a probabilistic model that enables quantitative assessment of failure risks. The practical significance of the study lies in the development of an interpretable decision-support tool for engineering and maintenance services which, unlike machine learning models, ensures transparency in decision-making and supports optimization of spare parts inventory management.

Keywords: predictive maintenance, reliability theory, failure analysis, exponential distribution, time series, predictive model.