A parametric survival regression framework using additive Weibull for transformer reliability

To safely allow connections beyond conventional capacity limits, system operators require reliability models that capture the full lifecycle of their assets. Although condition-monitoring-based approaches are well established for high-voltage components, they require intensive data collection which is unavailable in the medium-voltage grid. Previous research on statistical models assume monotonic failure rates and therefore generally ignore infant mortality failures, leading to overestimated lifetimes. This paper introduces a parametric survival regression framework using the additive Weibull distribution, which expresses the failure rate as superposition of Weibull components for the infant mortality, useful life, and aging phases. The model parameters are estimated through maximum likelihood while accounting for censoring and truncation, and regression of the component’s characteristics (voltage and rating) enables interpretable, asset-specific lifetime estimation. Validation with synthetic data confirms accurate recovery of known parameters, whereas application to a real dataset of distribution transformers shows that early-life failures depend mainly on rating, while aging-related failures depend on both rating and voltage.