Martin Hoof

PD source identification with novel discharge parameters using counterpropagation neural networks

Computer aided partial discharge (PD) source identification using different multidimensional discharge patterns is widely regarded as an important tool for insulation diagnosis. In this paper, a neural network (NN) approach to PD pattern classification is presented. The approach is based on applying variants of the counterpropagation NN architecture to the classification of PD patterns. These patterns are derived from physically related discharge parameters, different from those commonly used. It is shown that considerable improvements of the classification quality can be obtained when an extended counterpropagation network with a dynamically changing network topology is applied to patterns that employ the voltage difference between consecutive pulses instead of the phase of occurrence as the main discharge parameter. Furthermore, using a particular parameter vector that takes the correlation between consecutive discharges into account also allows to solve the rejection problem with this type of NN.

Analyzing partial discharge pulse sequences-a new approach to investigate degradation phenomena

This paper presents a new method of partial discharge data analysis for the investigation of space charge and degradation phenomena in high voltage insulation systems. The basic principles of the pulse-sequence-analysis are illustrated and the characteristic differences with regard to standard procedures that have been established during the past few years are outlined. A description is given of an electronic partial discharge acquisition system, which is designed on a plug-in interface board for a standard personal computer to perform long term measurements of all the discharge parameters required for the succeeding PSA. Electrical treeing in polyethylene is used to demonstrate the usefulness of the PSA method. It is shown that information on space charge phenomena and the development of the local degradation can be obtained by considering correlations between consecutive discharge pulses.<<ETX>>

A physical model, describing the nature of partial discharge pulse sequences

A description is given of a straightforward physical model, which is based on only a few physically meaningful parameters and that allows to describe the PD behaviour resulting from specific insulation defects. It is demonstrated that computer simulations of PD pulse sequences on the basis of this model correspond closely to experimental results. The comparison of simulated sequences and experimental results suggests that the discharge activity is not mainly a stochastic process but consists of periods, with different prevailing processes, each showing a nearly deterministic behaviour.

Mining Diagnostic Text Reports by Learning to Annotate Knowledge Roles

Several tasks approached by using text mining techniques, like text categorization, document clustering, or information retrieval, operate on the document level, making use of the so-called bag-of-words model. Other tasks, like document summarization, information extraction, or question answering, have to operate on the sentence level, in order to fulfill their specific requirements. While both groups of text mining tasks are typically affected by the problem of data sparsity, this is more accentuated for the latter group of tasks. Thus, while the tasks of the first group can be tackled by statistical and machine learning methods based on a bag-of-words approach alone, the tasks of the second group need natural language processing (NLP) at the sentence or paragraph level in order to produce more informative features.

Physical modeling of partial discharge patterns

Partial discharge phenomena in which solid dielectrics are involved can be simulated by a mainly deterministic physical model. The comparison of computer simulations with measured partial discharge data shows that in each case only one or two of the model parameters are responsible for the different behaviour of different defects.

The influence of space charge and gas pressure during electrical tree initiation and growth

The analysis of the sequence of partial discharges that occur during the growth of electrical trees shows that local space charges play an important role, especially at the early stage. In the space charge build-up, that occurs within the polymer as a consequence of the partial discharges a marked polarity dependence can be found. At the late stage of the tree growth this difference diminishes and the gas pressure in the tree channels is the decisive parameter.<<ETX>>

Electrical treeing-physical details obtained by pulse-sequence-analysis

The paper demonstrates that pulse-sequence-analysis allows a better insight into the phenomena which occur during tree initiation and growth in polyethylene. Tree growth in polyethylene is not a continuous monotonous process, but a phenomenon which incorporates different processes. Even with constant voltage load, one or another effect is predominant. After a period of very low discharge activity, a higher discharge activity (but still including discharge free intervals) may occur for some ten seconds. This may be followed by a longer period of very high discharge activity without any discharge free time intervals.

Using counterpropagation neural networks for partial discharge diagnosis

In high voltage engineering, various methods of non-destructive fault diagnosis are applied for investigating the quality of insulating materials and systems. The methods are aimed at classifying patterns derived from the measured characteristics of the electrical signals typically resulting from insulation defects. In this paper, variants of the counterpropagation neural network architecture are used to classify patterns representing various properties of partial discharges. It is shown that the classification quality can be improved considerably when an extended counterpropagation network with a dynamically changing network topology, and an additional vigilance unit for monitoring the behaviour of the network during the learning phase is applied. The extended network has significant advantages over the standard counterpropagation network in cases where outliers in the training data seriously degrade the approximation quality of the standard network. When using the proposed network in conjunction with physically motivated discharge data, input patterns from defect categories not considered during training can be rejected more reliably. This rejection problem is particularly important for practical applications where misclassifications cannot be tolerated.

Investigation of PD resulting from transformers using the PSA method

An analysis of partial discharges that result from transformers is presented. The PSA-method (Pulse-Sequence-Analysis) is applied to investigate the discharge behaviour. It is shown that a PD data base, that comprises typical model defects, can be efficiently used to identify the types of PD defects within high voltage (HV) equipment. When applying physically meaningful discharge parameters a reliable diagnostic decision can be obtained and the risk of misclassification can be reduced considerably.

Knowledge Extraction and Summarization for an Application of Textual Case-Based Interpretation

This paper presents KES (Knowledge Extraction and Summarization), a new knowledge-enhanced approach that builds a case memory out of episodic textual narratives. These narratives are considered as generated probabilistically by the structure of the task they describe. The task elements are then used to construct the structure of the case memory. The KES approach is illustrated with examples and an empirical evaluation of a real-world scenario of textual case-based interpretation for a technical domain.