Automated identification of insulation faults using Electro Magnetic Interference methods

On-line condition monitoring of substation electrical equipment depends on reliable, non-invasive surveillance techniques. Early detection of faults helps to mitigate the need for reactive maintenance and unplanned system downtime, thus ensuring continuity of supply. The Electro Magnetic Interference (EMI) method is a surveillance technique that can assist in identifying insulation degradation and conductor faults; such as Partial Discharge (PD) and Arcing. EMI frequency scans are used to identify the frequencies that are characteristic of fault conditions. Time-resolved analysis at these frequencies provides crucial data necessary for the classification of these faults. With the emergence of continuous on-line monitoring, there is an increasing need to embed more intelligence within monitoring devices to automatically recognise developing fault conditions. The main challenges faced with this method is that there is too much emphasis put on engineers in the field being able to identify these key frequencies by eye or knowledge alone, which limits the ability to automate the process. This paper presents a novel diagnostic assistant that will automatically identify the spot frequencies the engineer would manually capture for further, time-resolved analysis. The resultant time-resolved scans are then analysed to perform feature extraction and dimensionality reduction to automatically classify the data to a known fault category. Validation of the proposed techniques has been performed on real world data captured and labelled by engineers in the field. The accuracy of this method is established through direct comparison between the choices made by the engineers in the field to the classification of fault conditions and the decisions of the automated diagnostic assistant. The consistent accuracy of the results obtained paves the way for a fully automated expert system that can identify and classify possible emerging fault conditions utilising EMI diagnostics.