Eskom is using sophisticated analytical intelligence software to predict equipment failure. This plant maintenance optimisation project is currently in pilot phase. When fully rolled out, it is expected to save the power giant millions of rand by accurately predicting when its multimillion-rand transformers will require maintenance.
Eskom Distribution is using Enterprise Miner, a data mining solution from SAS, leaders in business and analytical intelligence. It is envisaged that the utility will use the tool for everything from cable theft prediction to prepaid metering fraud detection. Right now, however, the focus is on predicting the failure of its transformers.
The power giant's capital investment in transformers is substantial and intelligent predictive maintenance is needed to ensure proactive intervention to protect these assets.
"The pilot project focuses on monitoring the oil within the transformers," explains independent consultant, Francois Horton, Business Analyst, Distribution Information Strategy Management at Eskom. "Oil serves as an isolator within the transformers, so is a good indicator of their condition.
"If Eskom does not detect transformer problems in time, massive losses occur. Taking a transformer out of service for unplanned maintenance typically costs R1.5 million per maintenance activity."
By analysing the results of laboratory tests on the oil, proper planned maintenance can be carried out, and action taken as soon as a fault is detected.
"The potential cost savings of properly managing this environment using data mining run into millions of rand," says Horton.
Basically, the laboratory analyses the oil for gases which indicate what kind of fault is starting to occur. For example, a high level of corona in the oil indicates leakage currents, meaning that the insulation is breaking down. Arcing in the oil is an indicator of an imminent fault which, if left to continue, would cause the transformer to blow up.
The data mining pilot project aims to predict what gases would occur at levels which would indicate that maintenance was required, given the measurement of only one input, preferably oil temperature.
Results were gathered from the laboratory, and then data mined. "In the past, employees decided on the tests they would request from the laboratory based on experience and gut-feel," says Horton. "If they noticed x, then they would ask only for y test, believing this to be a good enough indicator.
"Data mining gave them an understanding of the importance of different variables to the outcome. Even slight variations in the presence of gases could change the causes dramatically."
Following the success of the pilot project, Eskom will embark on SAS Enterprise Miner training for its staff.
"Eskom is on the verge of taking huge strides into this whole area of data mining," says Horton. "It will assist management to improve availability of the plant, increase ROI and reduce loss of income."
It is envisaged that SAS Enterprise Miner will be used in a number of other areas in future, including customer knowledge, prepaid metering fraud detection, cable theft prediction, root cause of failure analysis, as well as plant maintenance optimisation in other areas of the network, for example circuit breakers, poles and lines.
Share