Analysis of the Use of Field Data Under Variable Conditions to Develop Lifetime Models for Electrical Distribution Devices

Lifetime models have been predominantly developed using constant but accelerated conditions to assess their base lifetime and the acceleration factor under different conditions. This approach is expensive and time-consuming, especially for highly reliable devices, as found in electrical distribution systems. On the other hand, online monitoring provides a large amount of data on the conditions and failures of the fleet of devices. However, constant conditions are not generally present. Therefore, developing efficient methods to estimate parameters from field data is of interest.

Proportional hazard (PH) and accelerated failure time (AFT) models are commonly used to describe the failure of devices under time-varying stress factors. This work analyses how these can be used efficiently to estimate reliability models' parameters, focusing on real-world electrical distribution devices.

The reliability function of a highly reliable device is challenging to acquire, as failure will generally only happen after a long time, and most of the time, devices are not run until failure. In addition, the dependency of the failure rate on environmental conditions the device is operating in requires to make a series of experiments to infer the acceleration factors in the classical setting. Therefore for such devices, accurate reliability curves or hazard rates are often not known, which limits the application of lifetime models, e.g., for maintenance or service planning. Up to now, mainly the "average" reliability of a type of device was used, meaning that the environmental conditions were often unknown. For this, most often, field or fleet data was already used. Where even this was not possible, the reliabilities of whole classes of devices were studied. Overall the effect of an aggregation of failure data over a diverse population will lead to a spread of the reliability curve compared to the one using a specific device type or specific environmental conditions, hindering a precise prediction of its failure. It is therefore of interest to find ways to make use of all available information to improve this. We explore this in this study for two different models and using simulated failure data coming from real environmental conditions.

Keywords: Proportional hazard model, Accelerated failure time model, Field data, Variable conditions, Electrical distribution, Capacitors, Breakers.

Download PDF