Improving Utility Cables Diagnostics and Prognostics using Machine Learning

Abstract

Each year, millions of people and thousands of businesses are impacted by underground cable system failures. Underground cables are considered critical equipment within any power system, and typically one of the most expensive components of the system to repair. When they fail, the customer impact is immense and has the potential to cause severe collateral damage or worse, public safety concerns. Replacing underground power cables can be very expensive and time consuming and can take months or even years when associated with significant design, civil and construction work. Over 99% of solid dielectric (i.e.: XLPE or EPR) cable system failures are associated to Partial Discharge (PD). This paper characterizes the waveforms of Partial Discharge (PD) time domain signals utilizing a unique dataset of measured conditions of underground power cable systems. Machine Learning and Deep Learning models have been developed and evaluated for the purposes of providing the foundation for automated condition monitoring and predictive maintenance. The results demonstrate a step towards a predictive maintenance approach for underground cable systems.