Partial Discharge Noise Filtering - Essay Example

Partial Discharge Noise Filtering Name Institution Course Instructor Date Abstract This proposal presents an advanced technique of partial discharge (PD) signal noise filtering. The proposed technique uses the support vector machine (SVM) in conjunction with wavelet transform (WT). It is different from other linear WT-based techniques and other conventional denoising techniques in such a way that it is capable of separating noise coefficients from data coefficients in high voltage (HV) electric power appliances. The methodology will characterize the spatial correlations via the decomposition of PD wavelet. SVM classifier will play a major role in the identification and separation of maxima lines associated with PD and noise. The justification reveals that this method is viable in denoising of PD signals based on the substantive methodology and literature. Partial Discharge Noise Filtering Introduction The technique of insulating electrical wiring plays a crucial role in the protection and performance of high voltage (HV) electrical appliances. If HV happens to be transmitted via damaged or uninsulated equipment, there are high chances that the equipment will malfunction or produce partial discharges (PDs). Research shows that the failure of electric appliances results from a cumulated damage of the insulation caused by a prolonged discharge of PD signals for a period of over one year (Chang, et al., 2005). That is why several methods have been devised to detect and filter (denoise) the occurrence of PD signals. This paper seeks to present a proposal of the most appropriate and recent approach to partial discharge noise filtering (denoising). It will achieve its objective by highlighting the types of noise filtering in partial discharge and identifying the most appropriate system used recently. Research problem definition and research outcomes specification Depending on the power transmission conditions, there are different types of HVPD signals that cause different degrees of damages on particular electrical equipments. Some of these signals include the power-line system carrier communication noise, rotating machine commutator noise, corona discharge noise, colored noise and electric appliances switching circuit noise, among others. Advances in technology have transformed the analogue converters to digital format (Mota, et al., 2011). Whereby, noise suppression has become easier and possible by measuring digital PD via Digital Signal Processing (DSP) device. Scientists have proposed and implemented various conventional PD denoising algorithms by the use of noise filters such as FFT thresholding, finite impulse response, adaptive filtering, moving average, FIR, IIR, digital filtering and infinite impulse response, among other algorithms. The conventional PD denoising algorithms have shown success in detection and removal of colored noise, stationary signal and DSI. However, they have failed to detect and filter White Gaussian Noise (WGN). From this evidence, it is clear that the operation of the algorithms is linear (limited). Moreover, the linearity nature of the algorithms is ineffective in the detection and suppression of non-stationary PD signals characterized with sharp edges, limited time and transient impulses. In this context, the linear algorithms work in such a way that they filter both the significant signal element and the noise because they appear similar in the spectrum. In a nutshell, as long as the filtering occurs linearly, the transformation will also occur linearly. This clearly shows that there is need to come up with a new algorithm or improve the existing one to detect and eliminate all types of PD signals without affecting the significant signals. Recent advances show that wavelet transform (WT) is a powerful model than the conventional PD filtering algorithms because it is able to detect and filter both time-limited and inhomogeneous signals (Dou, et al., 2012). Based on the advances of WT, this paper seeks to propose a significant PD filtering model which is superior to the conventional ones discussed above. The PD filtering model being proposed is based on the principle that decompositions of WT partial discharges are distinct, whereby; it is characterized by clustering of coefficients in similar positions at every scale of decomposition. Since it is possible to separate one position from another, then technicians can separate coefficients that are associated with noise from those associated with PD signals. By doing this it is possible to filter noise easily. The method being proposed relies on the translation-invariant discrete wavelet transform (the Discrete Wavelet Transform (DWT) variant) that prevents decimation-oriented data loss. The DWT local modulus maxima are used coupled with scales to characterize spatial correlations. In the methodological context, noise filtering is performed by utilizing the vector machine-supported classifier to detect maxima lines associated with noise (Prinza, et al., 2014). Based on preliminary tests, it is prospected that the technique can offer improved noise filtering results for the denoising of both harmonic and white noises in comparison to the conventional WT-associated denoising algorithms. Additionally, this algorithm is advantageous in a way that it filters a localized and pulsating noise which is completely undetectable to conventional strategies. Literature review As already described PDs are electric power discharges that are occur due to presence of insulation cavities caused by contaminants or insulation damages in high voltage appliance. The PDs are associated with the malfunctioning of the electrical equipment over time (Ma, Zhou and Kemp, 2002b). It is because; they occur repetitively at a particular point generating constant damage. It is upon manufacturing companies to closely monitor the transmission of HV in electrical appliances to prevent unnoticed breakdown of machines (Chang, et al., 2005). Many PD test are performed easily in the laboratories equipped with necessary equipments such as shield walls to decrease interference. Contrary to PD test, the PDs produced by HV machineries are hard to monitor because of the rampant interference from communication signals, power systems signals and switching mechanism signals, among other sources. That is why measurement and filtration of PDs on-site requires a much more sophisticated mechanism that is devoid of external interference. Research shows that several analogue and digital literature techniques have been proposed to curb the on-going problem. However, there is substantive approach yet to be commissioned to overcome the interference and diversity of the signal during on-site denoising. Recent findings suggest that WT is a promising tool in PD processing because in several preliminary attempts it has outshined the conventional algorithms. Such tests include the effectiveness of WT in processing the Gaussian White noise corruption from PD signal power cable (Ma, Zhou and Kemp, 2001). The approach was enhanced further by incorporating techniques fitted with PD to assess the levels of threshold and also to reveal the outcome of measuring electric power cable. Furthermore, another test utilized the multi-resolution decomposition approach in PD noise filtering which outperformed the conventional linear algorithms on a point-plane. It is done via the analytical comparison with other PD denoising algorithms such as time-frequency, wavelets, statistically-associated filters, linear filters and adaptive filters. In recent research findings, techniques associated with WT yielded superior results and proved to be efficient in processing signals on a point-plane arrangement. On the same note, researchers revealed that the effectiveness of the disintegration of Wavelet Packets in the noise filtering of PD signals emanating from HV generators and switch gears insulated by gas (Ma, Zhou and Kemp, 2002a). In this regard, an experimental technique was proposed to assess the threshold magnitude and record the outcomes which are used in the PD processing of electric power cables. That is why a lot of scientific prospects revealed that the Second Generation Wavelet Transform was the only imperative technique to complement the use Wavelet in PDs and remove the decorrelation of Wavelet from noise. It is due to the fact that Wavelet demonstrated its efficacy in the signal processing of HV electric power cables. Alternatively, another approach used decomposition of Wavelet Packets to denoise PDs with respect to the phase-resolved distributions images. This means that they managed to demonstrate the efficacy in processing of PD signals in generators and motors. Research methodology and planning First stage The PD signals are usually produced on the external side of the circuitry as low-amplitude current flows fast in the circuit in form of pulses. The shape of the pulses is determined by several factors which include the type of the defect, the test equipment size, measurement methodology, the discharge site-sensor distance and the location of the disruption. The amplitudes of the pulses are hard to detect in a noisy environment because they usually measured in miro-volts to milli-volts (Zhang, et al., 2007). Therefore, in the initial stage of the methodology the appropriate way of characterizing spatial correlations will involve the use of the procedural concepts of the WT local modulus maxima propagation theoretical model. This process will be effective based on the fact that the largest coefficients (which other WT-based noise filtering strategies seek) are found in the PD influence cone. But unlike the other methods that analyze the whole time-scale plane or magnitude-by-magnitude, this technique will assess the features of propagating coefficients between adjacent levels. By doing so, it will differentiate the PD- associated coefficients from the noise-associated coefficients. Second stage On the other hand, a wavelet transform is characterized by its ability to detect function singularities. Due to this character, the wavelet transform is significant in the resolution of singularity location problems. Based on this significance, it acts as data compression software and as a detector of image border. In this regard, the second stage of the methodology will involve assessment of the spatial correlations PD in two stages. The first stage will involve the determination of the decomposition on-start level. The level will be linked to the frequency spectrum of PD with the uppermost energetic band which will be determined in various ways. For instance, the decomposition of discrete wavelet transform (DWT) will appear in form of logarithmic partitions in the domain of frequency where the coefficients of the first stage will correspond to the band with high frequency varying from π and π/2. The second stage will vary from π/2 and π/4 and so forth. It is important to note that coefficients of DWT always group in the high energetic frequency band levels of the signal. Therefore, they directly proportionate with the measurement system pass-band frequency. Based on this information, the first approach will involve the determination of the initial analysis level via the measurement system characteristics knowledge. Third stage Previous research revealed that the important feature of the maxima lines of PD is that they tend to reduce immensely when proliferating from superior to inferior levels of decomposition. It is a characteristic behavior that associates directly with the PD frequency spectrum shape, especially due to the concentration of energy in small band numbers (Zhou, Zhou and Kemp, 2005). That is why when the corruption of the signal by noise occurs, the modulus maxima proliferation tends to rely on the frequency spectrum and type of the noise. Therefore, the separation of the maxima lines will be conducted by a support vector machine (SVM) based pattern classifier. The maxima lines will be connected directly to the classifier because the space of input will be defined as the maxima which will be located at each stage. Thereafter, the task of the machine will be to show if the maxima line is associated with the noise or PDs. In this regard, the lines associated with noise will be identified, followed by the separation of the coefficients from the combined set. Finally, the reconstruction will be based entirely on PD coefficients. However, the SVM classifiers will be chosen based on two main factors. The first factor will rely on the previous research that will indicate whether the SVM classifier matches or outperforms other classification techniques. The second factor will be based on the experimental and imperative procedures of the process because SVM classifiers are not flexible enough when it comes to the adjustment of the parameters. After SVM selection, the training of SVM will be performed in a procedural manner. Whereby, artificial data will be used only in the initial development stage, followed by decomposition and manual classification of PD signal that is free from noise from one form of noise sample. Thereafter, the data will be used in of SVM and adjust it to appropriate parameters. Finally, the machine will be used to differentiate other signals based on the set type of noise and PD followed by recording and evaluation of the results. Justification/evaluation of the results Adherence to the procedure is tantamount to the success of the technique. In this regard, after the methodology, the evaluation process reveals if the process is viable and if it is advantageous over the linear WT-based algorithms. For instance, based on the described methodology the PD signals are obtained via the use of calibration pulses. But before that, the calibration pulses are infused into the advanced and improved WT-based algorithm system before occurrence of energization (Satish and Nazneen, 2003). Alternatively, the PD pulses are obtained when the whole system is in operation (turned on). After obtaining the prospected PD pulses, the nose sample is retrieved immediately before the PD initiation level (stage). Both the PD pulses and noise sample data are important in the training process. However, the data is considered raw and unfit for training the machines until it undergoes manual classification. The training of the machines is important in defining the initial parameters. Recent research establishes that parameter definition allows the machines to classify and differentiate new signals that contain both noise and PD. Therefore, drawing from this evidence, it is clear that the experiment does not only rely on executable procedure and contextual methodology, but also it is also based on the previously successful experiments (Tang, et al., 2013). It is important to consider that at this moment there is no reliable procedure suitable for denoising. But since this approach seems the only remedy, it is vital to put into consideration the significance of its success. Conclusion In conclusion, the paper presented a proposal on the most appropriate and recent approach to partial discharge noise filtering (denoising). It achieved its objective by highlighting the types of noise filtering in partial discharge and identifying the most appropriate system used recently. In this regard, it proposed an effective and new PD noise filtering technique associated with spatial correlations of WT coefficients with respect to the stages of decomposition. The proposals revealed the high efficiency of the approach compared to the linear WT-based strategies and other conventional techniques because it exploits the localized features of the PDs to separate noise coefficients from viable data coefficients. Therefore, it is imperative to perform this technique practical to justify the theoretical concepts outlined in the proposal. Bibliography Chang, C.S., Jin, J., Chang, C., Hoshino, T., Hanai, M. and Kobayashi, N. 2005. Separation of corona using wavelet packet transform and neural network for detection of partial discharge in gas-insulated substations. Power Delivery, IEEE Transactions on, 20(2), pp.1363-1369. Dou, D., Yang, J., Liu, J. and Zhao, Y. 2012. A rule-based intelligent method for fault diagnosis of rotating machinery. Knowledge-Based Systems, 36, pp.1-8. Ma, X., Zhou, C. and Kemp, I.J. 2001. Wavelets for the analysis and compression of partial discharge data. In Electrical Insulation and Dielectric Phenomena, 2001 Annual Report. Conference on, IEEE. pp.329-334. Ma, X., Zhou, C. and Kemp, I.J. 2002a. Automated wavelet selection and thresholding for PD detection. Electrical Insulation Magazine, IEEE, 18(2), pp.37-45. Ma, X., Zhou, C. and Kemp, I.J. 2002b. Interpretation of wavelet analysis and its application in partial discharge detection. Dielectrics and Electrical Insulation, IEEE Transactions on, 9(3), pp.446-457. Mota, H.D.O., Rocha, L.C.D.D., Salles, T.C.D.M. and Vasconcelos, F.H. 2011. Partial discharge signal denoising with spatially adaptive wavelet thresholding and support vector machines. Electric Power Systems Research, 81(2), pp.644-659. Prinza, L., Mohanalin, J., Beenamol, M. and Joshy, P.V. 2014. Denoising Performance of Complex Wavelet Transform with Shannon Entropy and its Impact on Alzheimer Disease EEG Classification Using Neural Network. Journal of Medical Imaging and Health Informatics, 4(2), pp.186-196. Satish, L. and Nazneen, B. 2003. Wavelet-based denoising of partial discharge signals buried in excessive noise and interference. Dielectrics and Electrical Insulation, IEEE Transactions on, 10(2), pp.354-367. Tang, Y., Wu, D., Chen, S., Zhang, F., Jia, J. and Feng, X. 2013. Highly reversible and ultra-fast lithium storage in mesoporous graphene-based TiO 2/SnO 2 hybrid nanosheets. Energy & Environmental Science, 6(8), pp.2447-2451. Zhang, H., Blackburn, T.R., Phung, B.T. and Sen, D. 2007. A novel wavelet transform technique for on-line partial discharge measurements. 1. WT de-noising algorithm. Dielectrics and Electrical Insulation, IEEE Transactions on, 14(1), pp.3-14. Zhou, X., Zhou, C. and Kemp, I.J. 2005. An improved methodology for application of wavelet transform to partial discharge measurement denoising. Dielectrics and Electrical Insulation, IEEE Transactions on, 12(3), pp.586-594. Read More

Literature review As already described PDs are electric power discharges that are occur due to presence of insulation cavities caused by contaminants or insulation damages in high voltage appliance. The PDs are associated with the malfunctioning of the electrical equipment over time (Ma, Zhou and Kemp, 2002b). It is because; they occur repetitively at a particular point generating constant damage. It is upon manufacturing companies to closely monitor the transmission of HV in electrical appliances to prevent unnoticed breakdown of machines (Chang, et al., 2005). Many PD test are performed easily in the laboratories equipped with necessary equipments such as shield walls to decrease interference.

Contrary to PD test, the PDs produced by HV machineries are hard to monitor because of the rampant interference from communication signals, power systems signals and switching mechanism signals, among other sources. That is why measurement and filtration of PDs on-site requires a much more sophisticated mechanism that is devoid of external interference. Research shows that several analogue and digital literature techniques have been proposed to curb the on-going problem. However, there is substantive approach yet to be commissioned to overcome the interference and diversity of the signal during on-site denoising.

Recent findings suggest that WT is a promising tool in PD processing because in several preliminary attempts it has outshined the conventional algorithms. Such tests include the effectiveness of WT in processing the Gaussian White noise corruption from PD signal power cable (Ma, Zhou and Kemp, 2001). The approach was enhanced further by incorporating techniques fitted with PD to assess the levels of threshold and also to reveal the outcome of measuring electric power cable. Furthermore, another test utilized the multi-resolution decomposition approach in PD noise filtering which outperformed the conventional linear algorithms on a point-plane.

It is done via the analytical comparison with other PD denoising algorithms such as time-frequency, wavelets, statistically-associated filters, linear filters and adaptive filters. In recent research findings, techniques associated with WT yielded superior results and proved to be efficient in processing signals on a point-plane arrangement. On the same note, researchers revealed that the effectiveness of the disintegration of Wavelet Packets in the noise filtering of PD signals emanating from HV generators and switch gears insulated by gas (Ma, Zhou and Kemp, 2002a).

In this regard, an experimental technique was proposed to assess the threshold magnitude and record the outcomes which are used in the PD processing of electric power cables. That is why a lot of scientific prospects revealed that the Second Generation Wavelet Transform was the only imperative technique to complement the use Wavelet in PDs and remove the decorrelation of Wavelet from noise. It is due to the fact that Wavelet demonstrated its efficacy in the signal processing of HV electric power cables.

Alternatively, another approach used decomposition of Wavelet Packets to denoise PDs with respect to the phase-resolved distributions images. This means that they managed to demonstrate the efficacy in processing of PD signals in generators and motors. Research methodology and planning First stage The PD signals are usually produced on the external side of the circuitry as low-amplitude current flows fast in the circuit in form of pulses. The shape of the pulses is determined by several factors which include the type of the defect, the test equipment size, measurement methodology, the discharge site-sensor distance and the location of the disruption.

The amplitudes of the pulses are hard to detect in a noisy environment because they usually measured in miro-volts to milli-volts (Zhang, et al., 2007). Therefore, in the initial stage of the methodology the appropriate way of characterizing spatial correlations will involve the use of the procedural concepts of the WT local modulus maxima propagation theoretical model.

Read More