Failure Prediction and Prescriptive Maintenance for Power Systems

Business challenge

  • Large electricity utility has many transformers at the end of their lifetimes
  • Expensive to replace all; need to replace only the ones that are likely to fail soon


  • Utilized multiple data sources including weather, usage patterns and age to develop an automatic prediction system for impending failure
  • Predictions are used to identify transformers that should be replaced first

Business benefit

  • Costly downtime avoided by replacing before failure, and cost reduced by delaying replacement when possible
  • Strong predictors of failure identified and used to increase lifetime