Lifetime models are primarily developed using constant but accelerated conditions to assess lifetimes and acceleration factors under various scenarios. This is costly and time-consuming for reliable devices like those in electrical distribution systems. Alternatively, online monitoring yields ample data on device conditions and failures, though constant conditions aren't always present. Thus, efficient methods for parameter estimation from field data are valuable.

Proportional hazard (PH) and accelerated failure time (AFT) models describe device failure under time-varying stress. This work examines their efficient use in estimating reliability parameters, especially for real-world electrical distribution devices.

Acquiring the reliability function for highly reliable devices is challenging, given rare failures and limited runtime until failure. The failure rate's dependence on environmental conditions necessitates experiments to infer acceleration factors conventionally. Consequently, accurate reliability curves or hazard rates are often unknown, hindering model application in maintenance or service planning. Typically, "average" device reliability was used, assuming unknown environmental conditions and utilizing field or fleet data. When this isn't feasible, whole device classes' reliabilities were studied. Aggregating failure data across diverse devices spreads the curve, limiting precise predictions. Thus, leveraging available data to enhance predictions is vital. This study explores this using two models and simulated failure data from real environments.