A power transformer is an important piece of apparatus in a power system. Continuously exposed to mechanical, thermal and electrical stresses, a power transformer’s insulation system can deteriorate and this may eventually lead to faults in the transformer. Therefore, accurate insulation condition assessment for a power transformer has a significant importance for appropriately deciding on its maintenance and replacement schedule. To achieve this goal, various condition monitoring and diagnostic techniques have been developed and adopted by utilities. Amongst these techniques, partial discharge (PD) measurement has significant benefits since it provides a means of online monitoring and diagnosis of a transformer’s insulation system without interrupting the transformer’s normal operation.
PD is the localized breakdown in an insulation system, in which electric discharges occur and only partially bridge the insulation between conductors. If a transformer’s insulation system has a defect (e.g. cracks, holes, floating metal particles), repetitive discharge pulses can be generated. As the stresses accumulate on the defected insulation system, the amplitude and extensity of PD pulses will increase significantly, which can result in breakdown of the transformer insulation and even lead to the failure of the whole unit. This thesis aims to develop an online PD measurement system for condition monitoring and assessment of power transformer insulation system, including hardware implementation regarding an online sensor based PD signal acquisition as well as software development including algorithms for PD signals extraction, data representation and insulation integrity evaluation.
The characteristics of PD signals may differ according to insulation defects types, a transformer’s geometric configuration and its operation conditions, and the types of PD sensors and the configuration of PD measurement systems (i.e. the types and locations of sensors, measuring bandwidth, sampling method and sampling rate). Moreover, during online PD measurement of a transformer in a substation environment, acquired PD signals can be overwhelmed by extensive interference and noise. As such, it is necessary to apply signal processing techniques to extract PD signals from noise corrupted measurement signals. Furthermore, it needs to properly interpret measured PD signals and correlate PD signals to the transformer’s insulation integrity. The above considerations introduce significant challenges in developing online PD measurement system.
After conducting a comprehensive literature survey on various aspects of PD measurement of transformers, this thesis developed a high frequency current transformer (HFCT) and acoustic emission (AE) sensors based online PD measurement system targeting power transformer insulation condition monitoring and assessment. Important issues of developing such a system especially the proper selection of a sampling method and a sampling rate were investigated in this thesis. A novel probabilistic approach for improving the effectiveness of the wavelet transform based PD signals extraction and evaluation were proposed in this thesis. Another novel transient strength based PD extraction technique was also proposed as an alternative to the wavelet transform based PD signal extraction and evaluation in the thesis.
To tackle the limitations of the conventional wavelet transform for PD signal extraction, this thesis proposed a novel probabilistic wavelet transform approach. The proposed method computed the occurrence rate of PD events at each decomposition level with multi-scale thresholds. On the basis of this approach, automatic mother wavelet selection and decomposition level determination methods were developed. The proposed probabilistic approach was also used to provide a means for the evaluation of insulation integrity based on the characteristics of the extracted PD signal.
This thesis also proposed a differential PD signal extraction technique to facilitate transformer insulation assessment. This method can extract multiple waveforms of PD signals with a transient strength based threshold and can differentiate the PD signals caused by different types of PD sources. Moreover, it can maintain the signal polarity direction information, which can support PD source recognition.
This thesis also developed a spatial intersection method for PD source localization in a transformer. This method can indicate a suspected area for locating a PD source by finding the overlapped volumes of several AE sensors. A frequency window screening method was also implemented for extracting ultrasonic signals obtained by AE sensors, which can effectively remove noise signal and accurately identify the start point of the arriving sound wave signal obtained by AE sensors.
Extensive case studies using datasets obtained from PD measurements on experimental PD models and in-service transformers were conducted in this thesis to demonstrate the applicability of the above proposed approaches for online transformer insulation condition monitoring and assessment.