A Covariate‐Dependent Survival‐Copula Framework for Two‐Dimensional Automotive Warranty Analysis and Residual‐Warranty Preventive‐Maintenance Decisions

Warranty claims constitute a significant cost component for automotive manufacturers; preventive‐maintenance (PM) can reduce them by intercepting failures under warranty. Warranty coverage depends on service age and mileage, but analyses often reduce it to one dimension or assume independence, biasing warranty‐corner risk. Coarse segment claim rates can under‐ or over‐cover vehicles. We propose a framework for estimating joint age–mileage warranty risk under censoring and heterogeneous usage and translating residual risk into PM decisions. A censored‐likelihood copula estimates joint failure probability at the warranty limits and PM decision date. A covariate‐dependent survival copula estimates eligible vehicles' conditional failure probabilities over their remaining warranty, informing an economic threshold rule. Machine learning completes missing mileage, provides a discrimination diagnostic, and supplies a rank‐based PM‐allocation benchmark against which the copula‐based vehicle‐level policy is evaluated. In a controlled simulation with age–mileage dependence and an injected defect segment, the AIC‐selected Gumbel–Weibull copula estimates a 9.57% cumulative warranty‐corner failure probability, with a further 3.41% accruing after the PM decision date. Vehicle‐level targeting yields higher expected savings than segment‐level and ML‐allocation campaigns across PM‐cost scenarios when evaluated against data‐generating‐process risk, although realized savings depend on the evaluation scale. These gains require field validation of vehicle‐level risk calibration.