Jump to content, Georgia Institute of Technology, College of Engineering, School of Electrical and Computer Engineering (ECE), ECE Research, ECE Research Centers, NEETRAC
College of Engineering
Search | Contact ECE | Feedback | BuzzPort
GT Home > COE Home > ECE Home > Research > Centers > NEETRAC
National Electric Energy Testing, Research & Applications Center

Predictive Failure Analysis for Distribution Cables and Power Apparatus


Asset management is increasingly recognized as a critical issue for electric utilities. Knowing how much investment is needed to provide for a reliable transmission and distribution system is very important, but difficult to establish. If utilities could replace “best guess” estimates on when to repair or replace system components with an optimized decision making processes, substantial savings could be realized. One of the more acute infrastructure replacement issues involves aging underground cable circuits. This problem is particularly difficult because utilities often keep very few records regarding cable failure history.

This page describes a methodology for projecting future cable system failure rates using limited historical failure data. The methodology, developed at Georgia Tech NEETRAC, can be applied on cable populations or other apparatus for which only incomplete past failure data is available. The only historical failure data needed to successfully forecast future failures are listed below:

  • number of installed components per year
  • number of replaced components per year
  • number of retired components per year
  • number of component failures per year.
It is not necessary to know the age of the failing set of components in order to project future failures. This is a tremendous advantage because information on the age of a failed component is typically not known. However, if it is available, it can be used to increase the accuracy of the forecast - but it is not required information!


When a population of components is installed, the aging process brings about increasing rates of failures over time (see Figure below). A partial replacement of the population resets the aging process and reduces the failure rates.
Example of Failure Analysis
The problem becomes more complex when multiple populations of various ages and multiple replacement events need to be taken into account. Our procedure enables us to not only estimate the failure rates in the near future, but also their confidence ranges. It is also possible to plan for replacement events in order to maintain a desired failure rate in the future.
Example of Failure Analysis
The only restriction is that the population must be homogeneous. That is, they must consist of the the same material type installed in the same general environment, such as directly buried, unjacketed crosslinked polyethylene insulated cable. The entire procedure, which involves a determination of the statistical parameters of the aging component and Monte Carlo simulations for assessment of the confidence ranges, is symbolically depicted in the diagram below.

The novel methodology used is depicted in the adjacent flowchart. Such an approach provides the ability to estimate distributions of failures rather than just the most likely number of failures. The figure below shows the 75 th percentile of failures of the component population (shown as "+" symbols). When extrapolated into the future (a short time horizon of 1-5 years), the procedure provides a failure forecast, enhanced with additional information on the confidence ranges.
Observed failures over time (+). The 75 percent confidence ranges (solid lines) are also shown.
From this information, the amount of cable needed to maintain a constant failure rate over the next year can be established as a function of the desired level of confidence as shown below.
Replacement rates needed to maintain a constant yearly failure rate vs. confidence range.
The benefits derived from this failure forecasting model include:
  • An ability to effectively plan for cable
    (or other component) replacements
  • An ability to project future failures based on optimal replacement rates
  • An increase in system reliability

Last revised on Spril 11, 2014