L. Hao

Partial discharge source discrimination using a support vector machine

Partial discharge (PD) measurements are an important tool for assessing the health of power equipment. Different sources of PD have different effects on the insulation performance of power apparatus. Therefore, discrimination between PD sources is of great interest to both system utilities and equipment manufacturers. This paper investigates the use of a wide bandwidth PD on-line measurement system consisting of a radio frequency current transducer (RFCT) sensor, a digital storage oscilloscope and a high performance personal computer to facilitate automatic PD source identification. Three artificial PD models were used to simulate typical PD sources which may exist within power system apparatus. Wavelet analysis was applied to pre-process measurement data obtained from the wide bandwidth PD sensor. This data was then processed using correlation analysis to cluster the discharges into different groups. A machine learning technique, namely the support vector machine (SVM) was then used to identify between the different PD sources. The SVM is trained to differentiate between the inherent features of each discharge source signal. Laboratory experiments where the trained SVM was tested using measurement data from the RFCT as opposed to conventional measurement data indicate that this approach has a robust performance and has great potential for use with field measurement data.

Discrimination of multiple PD sources using wavelet decomposition and principal component analysis

Partial discharge (PD) signals generated within electrical power equipment can be used to assess the condition of the insulation. In practice, testing often results in multiple PD sources. In order to assess the impact of individual PD sources, signals must first be discriminated from one another. This paper presents a procedure for the identification of PD signals generated by multiple sources. Starting with the assumption that different PD sources will display unique signal profiles this will be manifested in the distribution of energies with respect to frequency and time. Therefore the technique presented is based on the comparison of signal energies associated with particular wavelet-decomposition levels. Principal component analysis is adopted to reduce the dimensionality of the data, whilst minimizing lost information in the data concentration step. Physical parameters are extracted from individual PD pulses and projected into 3-dimensional space to allow clustering of data from specific PD sources. The density-based spatial clustering of applications with noise (DBSCAN) clustering algorithm is chosen for its ability to discover clusters of arbitrary shape in n-dimension space. PD data from individual clusters can then be further analyzed by projecting the clustered data with respect to the original phase relationship. Results and analysis of the technique are compared for experimentally measured PD data from a range of sources commonly found in three different types of high voltage (HV) equipment; ac synchronous generators, induction motors and distribution cables. These experiments collect data using varied test arrangements including sensors with different bandwidths to demonstrate the robustness and indicate the potential for wide applicability of the technique to PD analysis for a range of insulation systems.

Comparison of support vector machine based partial discharge identification parameters

Partial discharge (PD) may have a significant effect on the insulation performance of power apparatus. Therefore, identification of PD sources is of interest to both power equipment manufacturers and utilities. With the development of PD measurement techniques, data analysis, signal processing and pattern recognition are gaining more interest. Research to date has considered varieties of different identification parameters such as phase resolved information, statistical operators, pulse shape analysis, pulse sequence analysis, frequency spectrum and wavelet analysis, which are also combined with so-called classifiers such as fuzzy logic, neural networks (NN) and learning machines. This paper investigates the performances of PD source identification using a support vector machine (SYM) based on different feature parameters. Due to the unique characteristics of SVM, some feature parameters that are not suitable for other classifiers are applicable. In this paper, comparisons of recognition rate and generalization capability between different feature parameters are discussed. Investigation reveals that recognition rate and generalization capability are influenced by the input PD parameters. Initial results indicate that, by using appropriate kernels and feature parameters, the automatic identification results obtained by using the support vector machine technique are very encouraging

Improving detection sensitivity for partial discharge monitoring of high voltage equipment

Partial discharge (PD) measurements are an important technique for assessing the health of power apparatus. Previous published research by the authors has shown that an electro-optic system can be used for PD measurement of oil-filled power transformers. A PD signal generated within an oil-filled power transformer may reach a winding and then travel along the winding to the bushing core bar. The bushing, acting like a capacitor, can transfer the high frequency components of the partial discharge signal to its earthed tap point. Therefore, an effective PD current measurement can be implemented at the bushing tap by using a radio frequency current transducer around the bushing-tap earth connection. In addition, the use of an optical transmission technique not only improves the electrical noise immunity and provides the possibility of remote measurement but also realizes electrical isolation and enhances safety for operators. However, the bushing core bar can act as an aerial and in addition noise induced by the electro-optic modulation system may influence overall measurement sensitivity. This paper reports on a machine learning technique, namely the use of a support vector machine (SVM), to improve the detection sensitivity of the system. Comparison between the signal extraction performances of a passive hardware filter and the SVM technique has been assessed. The results obtained from the laboratory-based experiment have been analysed and indicate that the SVM approach provides better performance than the passive hardware filter and it can reliably detect discharge signals with apparent charge greater than 30 pC

Partial discharge identification using a support vector machine

Partial discharge (PD) on-line monitoring and diagnosis is an important tool to assess the condition of power equipment. Different PD sources have different effects on the insulation performance of power apparatus. Therefore, identification of PD sources is of interest to both the power equipment manufacturers and utilities. A method based on machine learning theory, namely the support vector machine (SVM) was used for PD identification. Obtained experimental results from different partial discharge sources were pre-processed by using phase based information and wavelet analysis. Pre-processed data were also used as the SVMs input vectors, which was initially trained by known discharge source data, and then applied to identify different types of discharge sources. Initial results indicate that, by using appropriate kernels and parameters, the automatic identification results obtained using the SVM technique is encouraging.

Comparison of two partial discharge classification methods

Two signal classification methods have been examined to discover their suitability for the task of partial discharge (PD) identification. An experiment has been designed to artificially mimic signals produced by a range of PD sources that are known to occur within high voltage (HV) items of plant. The bushing tap point of a large Auto-transformer has been highlighted as a possible point on which to attach PD sensing equipment and is utilized in this experiment. Artificial PD signals are injected into the HV electrode of the bushing itself and a high frequency current transformer (HFCT) is used to monitor the current between the tap-point and earth. The experimentally produced data was analyzed using two different signal processing algorithms and their classification performance compared. The signals produced by four different artificial PD sources (surface discharge in air, corona discharge in air, floating discharge in oil and internal discharge in oil) have been processed, then classified using two machine learning techniques, namely the support vector machine (SVM) and probabilistic neural network (PNN). The feature extraction algorithms involve performing wavelet packet analysis on the PD signals recorded over a single power cycle. The dimensionality of the data has been reduced by finding the first four moments of the probability density function (Mean, Standard deviation, Skew and Kurtosis) of the wavelet packet coefficients to produce a suitable feature vector. Initial results indicate that very high identification rates are possible with the SVM able to classify PD signals with a slightly higher accuracy than a PNN.

Analysis of Partial Discharge Measurement Data Using a Support Vector Machine

This paper investigates the recognition of partial discharge sources by using a statistical learning theory, support vector machine (SVM). SVM provides a new approach to pattern classification and has been proven to be successful in fields such as image identification and face recognition. To apply SVM learning in partial discharge classification, data input is very important. The input should be able to fully represent different patterns in an effective way. The determination of features that describe the characteristics of partial discharge signals and the extraction of reliable information from the raw data are the key to acquiring valuable patterns of partial discharge signals. In this paper, data obtained from experiment is carried out in both time and frequency domain. By using appropriate combination of kernel functions and parameters, it is concluded that the frequency domain approach gives a better classification rate.

Partial discharge in medium voltage three-phase cables

Cable distribution networks are inherently complex and inaccessible systems; many of them are coming to the end of their original design life. As assets, they represent a dynamic and challenging issue with regard to the tasks of maintenance and management. Partial discharge (PD) has long been recognized as both a cause and symptom of the degradation of dielectric materials that protect high voltage plant. Utilities use the analysis of PD activity to make pre-fault decisions in areas such as maintenance, supply continuity and asset management. On-line PD monitoring systems are still in their relative infancy. An EDF Energy Networks funded research is investigating and identifying trends in PD activity associated with specific faults that commonly occur in distribution networks. In this paper an experiment to mimic the conditions experienced by on-line cable sections in the field is described. PD measurement has been obtained using conventional techniques covered in IEC 60270 in parallel with a commercially available substation monitor that is employed in distribution networks worldwide. Later work will involve using this experiment to PD test cable samples that contain a range of defects. It is hoped that each defect mechanism will produce an unique trend in PD activity as it degrades towards failure.

Identification of multiple partial discharge sources

Partial discharge (PD) measurements are an important tool for assessing the health of power equipment. Different PD may have different effects on the insulation performance of power apparatus. Therefore, identification of PD sources is of great interest to both system utilities and equipment manufacturers. This paper investigates the use of a wide bandwidth PD on-line measurement system which consists of a wide bandwidth sensor, a sophisticated digital signal oscilloscope and a high performance personal computer to facilitate automatic PD source identification. Wavelet analysis was applied to the obtained raw measurement data. The pre-processed data was then processed using correlation analysis. The obtained results have also been processed by accepted approaches, such as phase resolved information. A machine learning technique, namely the support vector machine (SVM) has been used to identify between the different PD sources.

Partial discharge diagnostics of defective medium voltage three-phase PILC cables

There are a number of medium voltage (MV) power distribution cable networks worldwide that are constructed predominantly of mass impregnated paper cables - London being one of these. Paper insulated lead covered (PILC) cables were extensively laid in the 50s and 60s before the introduction of cheaper polymeric alternatives that were sufficiently reliable. The current operational state of these networks has shown a gradual increase in failure rates of the previously reliable paper cables that are drawing to the end of their expected design life. Utilities are faced with the prospect of the impending failure of large sections of their prized asset and are keen to develop tools to better understand the health of their hardware. The analysis of partial discharge (PD) signals produced by the cables has been identified as a economically viable option to provide continuous condition monitoring of PILC cable circuits. Clearly, a comprehensive understanding of how PD activity relates to the various failure mechanisms exhibited by cable circuits in the field is required. In order to generate representative and repeatable PD signals from a number of PILC cable samples under rated conditions, an experiment was designed and commissioned. The scope of this paper is to introduce two cable degradation mechanisms that were applied to test cable and show the associated PD activity that was produced. A recently published technique for PD source discrimination was coupled with an understanding of the system and applied to the experiment data to isolate the signals specific to each degradation mechanism.