Taking the guesswork out of asset failure

Overhead powerlines (People's Panel)

Systems, engineering and technology company Frazer-Nash Consultancy is delivering a ground-breaking project that has the potential to help SA Power Networks better predict asset failure for its high voltage network using machine learning techniques.

The ability to better predict infrastructure asset failures, to direct maintenance funds to the right areas and to help prevent power outages is a significant development for power networks across the nation.

Frazer-Nash’s unique methodology is based upon machine learning techniques, linking key features of influence to the probability of failure of SAPN’s overhead conductors, as well as the underground Paper Insulated Lead Covered (PILC) cables in the Adelaide CBD. Some of these cables are more than 70 years old and difficult to access.

Related article: Space-based solar: baseload power from the sun 24/7

The Frazer-Nash system identifies assets at the greatest risk of failure, which could enable SAPN to direct maintenance resources at the right time, and better ensure reliability of service for its 900,000 home and business customers across the state. 

Frazer-Nash senior engineer Dr Simon Inverarity said it was the first time a machine learning methodology had been developed in Australia to predict distribution network asset health. The method relies only on existing failure records, which eliminates the need for time-consuming and costly sample testing and inspections often required by other techniques. 

“In the past, asset management has been based mainly on asset age, known defects and known reliability issues. However, these factors alone are inadequate for quantifying the probability of failure,” he said.

“The probability of failure is, in general, a result of many contributing factors. Our system uses a health index, which factors in variables such as corrosion levels, rainfall, wind speed, the proximity to large bodies of brackish water or salt, the material and configuration of the asset, as well as the age of the asset, including the operational environment, the conductor or cable attributes and conditions, and its operational history.”

Senior engineer Andrejs Jaudzems, who developed the machine learning algorithms, said, “Machine learning presents an excellent tool for determining the weightings of these features, due to its ability to categorise large volumes of data based upon continuous inputs. 

“These techniques make use of powerful algorithms to identify patterns in large, complex datasets.

Related article: SA Power Networks drone helps re-string powerlines

“Our team recently completed the first phase of the development for overhead conductors. The encouraging results have demonstrated the algorithms’ ability to not only correlate asset failure with relevant contributing factors, but also predict an asset failure rate that agrees with the actual number of failures. 

“Work is ongoing, for both overhead conductors and underground cables, to consolidate our findings for implementation into SA Power Networks’ asset management system.

“We are excited to be a part of this significant project, and this work is indicative of Frazer-Nash’s strategic commitment to the energy sector.” 

Previous articleIf the opposition wants a mature discussion about nuclear power, start with a carbon price. Without that, nuclear is wildly uncompetitive
Next articleWA to retire state-owned coal plants by 2030