Biologically inspired survival analysis framework for failure monitoring of composite insulators using leakage current

• Multi-aspect feature engineering enables precise failure tracking. • A survival model using dynamic-weighted Cox-RSF provides accurate risk measurement. • Clinical interpretability allows reliable classification by risk scores. • Measurement correctness is validated using C-index and IBS benchmarks. Composite insulators are subject to degradation and failure due to harsh environmental stresses like pollution, moisture and discharge activities. Conventional condition-assessment techniques such as threshold-based leakage current (LC) analysis are unable to predict time-to-failure and present survival probability. A hybrid prognostic framework is proposed in this paper based on medical survival analysis that considers insulator failure as a time-to-event problem with purpose of uncertainty quantification. Electrical engineering diagnostics are integrated with bio-statistical survival models in this work for speed-precision compensation in real practices. LC waveforms from flashover tests under different pollution levels are analyzed to extract performance indexes including temporal quantities, frequency properties and nonlinear dynamics. These indexes are used to train two survival models: Cox proportional hazards model for pollution and LC entropy and random survival forest (RSF) for capturing nonlinear discharge paths with time-dependent weighting. Validation on test data from 5 insulators shows superiority of the presented approach to some similar methods in terms of measurement correctness as C-index, integrated brier score (IBS) and risk stratification. The presented monitoring framework can be used as a data-driven tool to reduce unplanned outages particularly in cases of poor-data environment and measurement uncertainties due to incomplete observations.