Quick Fault Severity Determination using Dissolved Gas Analysis with different gas ratio Fault Identification Techniques

When a fault is suspected to have occurred in an oil-filled electrical equipment, it is vital not only to identify it but also prudent to find the extent of its severity. Suppose one wants to get a rough idea on this after fault identification, then simple methods using established gases ratio techniques may be applied to quickly estimate roughly the extent of fault severity. This work seeks to find the variabilities and applicability in fault severity approximation using four known fault identification methods. These are Rogers, Doernenburg, IEC and CIGRE gas ratio techniques. The fault severity is evaluated based on the extent of how far the ratio is lying from the maximum or minimum value which is considered as signifying a fault presence. To facilitate in this rapid faulty severity estimation, faults are basically divided into three categories: partial discharges, electrical discharges, and thermal faults. Statistical analysis parameters: mean, standard deviation and range are applied for the variability study. One hundred and thirty-nine fault cases of physically examined equipment mainly found in IEC TC 10 databases and other published literatures are used for verification. The aggregate level of severity is calculated as the mean of fault severities emanating from various combinations of gas ratios applicable in each technique. The results show that the four techniques may be applicable but somewhat depends on the fault category as some gas ratios from these techniques might produce unrealistic extremely high ratio values. Partial discharges and thermal faults severity are best determined by CIGRE, IEC or Rogers techniques, as Doernenburg gives severity extremely high which may not be dependable. Electrical discharge severity can be estimated by all methods. This is based partly by the number of data outliers given by each ratio for each technique for the various faults classes which may highly skew the end statistic.