 DOI
 10.1139/facets20150005
Wavelet transform as a postprocessing diagnostic tool for fault identifications of highvoltage direct current transmission system
 Published Online
 12 April 2016
P. Sanjeevikumar
Research and Development, Ohm Technologies, Chennai 600 122, India
Contributions
 Conceived and designed the study
 Performed the experiments/collected the data
 Analyzed and interpreted the data
 Drafted or revised the manuscript
Frede Blaabjerg
Department of Energy Technology, Center for Reliable Power Electronics (CORPE), Aalborg University, Pontoppidanstraede 101, 9220 Aalborg, Denmark
Contributions
 Conceived and designed the study
 Performed the experiments/collected the data
 Analyzed and interpreted the data
 Drafted or revised the manuscript
Abstract
This short communication focuses on exploiting the inherent advantages of discrete wavelet transformation (DWT) as a diagnostic tool for postprocessing and for identifying the faults that occur in the standard highvoltage directcurrent (HVDC) transmission network. In particular, a set of investigations are developed and examined for singlelinetoground fault on the generation and on the load side converter, and DClink fault. For this purpose, a standard 12pulse linecommutated converter (LCC)HVDC transmission network along with the DWT algorithm is numerically modeled in the MATLAB/PLECS simulation software. Furthermore, in this paper, a set of designed faulty conditions are predicted using the output of DWT and the results of numerical simulation are presented. Results are in good agreement with expectations to prove that DWT is an effective tool for fault diagnostics.
Introduction
In recent years, a 12pulse voltagesource converter based on highvoltage direct current (HVDC) transmission network has been widely implemented in real time, because of its outstanding advantages, such as economical efficiency and flexibility. Moreover, HVDC transmission systems that utilize the linecommutated converter (LCC) are successful and have been running for many years (Wang and SaNguyen et al. 2014; Sanjeevikumar et al. 2014). However, on the point of safety, the operation of HVDC systems is essential for both monitoring and better protection. Many research articles have addressed various fault diagnostic postprocessing tools including artificial neural networks, fuzzy control, and other intelligent systems (Nanayakkara 2011, 2012; Mohagheghi et al. 2009). However, under highpower transmission conditions or produced complex strategies, they do not effectively find the fault that has actually occurred. Discrete wavelet transformation (DWT) has found wider applications recently for controlling the AC/DC drives and motors with major benefits and is discussed in detail by Saleh (2013), Hui et al. (2015), Febin Daya et al. (2013a, 2013b, 2015, 2016), and Sanjeevikumar et al. (2015a, 2015b, 2015c). Furthermore, wavelet transforms are applied in the field of HVDC transmission for diagnosis of fault location, analysis, and DClink failure, but a systematic and comprehensive investigation of types of fault identification has not been presented (De Kerf et al. 2011; Yeap and Ukil 2014; de Andrade and Ponce de Leao 2014) and moreover that is not robust to adaptations. DWT is the fast processing algorithm and its inherent capabilities are exploited as a fault diagnostic tool in this paper. A set of three drastic faulty conditions are developed for a 12pulse LCCHVDC transmission system as a preliminary investigation to verify the proposal—specifically, singlelinetoground, on generation and on load end converters, and DClink faults. The complete AC power system is numerically developed in the MATLAB/PLECS simulation software. The observed results presented in this paper show that DWT accurately predicts the types of developed fault conditions under all conditions, as expected.
Discrete wavelet transform for a 12pulse LCCHVDC power system
Figure 1 shows the schematic circuit of the standard 12pulse LCCHVDC transmission system in practice. The wellknown synchronous reference frame regulation scheme is employed to control the 12pulse rectifier and 12pulse inverter units to perform effectively in the available power transfer from generation to the load center (Sanjeevikumar et al. 2014). A set of three fault conditions are designed as a preliminary task to examine the most common and drastic issues in transmission networks—singlelinetoground, on generation and on the load side converters, and most important DClink faults. Furthermore, the research work is kept under study for investigation of types of fault identification and its detailed analysis as near future implementations.
In this analysis, the DWT plays the role of identifying about the set of faults under test and confirms the type of occurrences in the complete system. The input parameter is chosen as power, as it includes the characteristic variation in voltages and currents and adheres to be a simplified solution. The instantaneous power is calculated and fed as input to DWT, which begins processing when a discrete signal x[n] of length N is transmitted. DWT can easily distinguish between the instantaneous power into highpass signal resulting in an impulse response h[n] and that into a lowpass signal resulting in an impulse g[n], mathematically expressed as follows (Saleh 2013; Hui et al. 2015; Febin Daya et al. 2013a, 2013b, 2015, 2016; Sanjeevikumar et al. 2015a, 2015b, 2015c; Mallat 1989): $${d}^{1}[n]=\sum _{k=0}^{N1}x[k]h[nk];\text{\hspace{0.17em}}{a}^{1}[n]=\sum _{k=0}^{N1}x[k]g[nk]$$(1) where d^{1}[n] and a^{1}[n] are the outputs of the high and lowpass filters, respectively. After this processing, again the output from the lowpass filter is downsampled by two and passed through a low and a highpass filter, which resembles the ones in the first level, written as follows (second level of decomposition): $${d}^{2}[n]=\sum _{k=0}^{N/21}{a}^{1}[k]h[nk];\text{\hspace{0.17em}}{a}^{2}[n]=\sum _{k=0}^{N/21}{a}^{1}[k]g[nk]$$(2)
To be noted, several types of wavelet filters are available and have been investigated, but the minimum description length (MDL) is chosen as the best criterion and is mathematically represented as follows (Febin Daya et al. 2013a, 2013b, 2015, 2016; Sanjeevikumar et al. 2015a, 2015b, 2015c; Mallat 1989): $$\mathrm{MDL}(k,n)=\mathrm{min}\{\frac{3}{2}k\text{\hspace{0.17em}}\mathrm{log}\text{\hspace{0.17em}}N+\frac{N}{2}\text{\hspace{0.17em}}\mathrm{log}{\Vert {\tilde{\alpha}}_{n}\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\alpha}_{n}^{(k)}\Vert}^{2}\},\phantom{\rule{0ex}{0ex}}0<k<N;\text{\hspace{0.17em}}1\le n\le M$$(3) where ${\tilde{\alpha}}_{n}={W}_{n}f$ denotes a vector of the wavelet transformed coefficients of the signal f using wavelet filters (n). ${\alpha}_{n}^{\left(k\right)}={\varnothing}^{K}{\tilde{\alpha}}_{n}={\varnothing}^{K}\left({W}_{n}f\right)$ denotes a vector that contains k nonzero elements. The threshold parameter ${\varnothing}^{K}$ keeps k number of the largest elements of the vector ${\tilde{\alpha}}_{n}$ constant and sets all other elements to zero. N and M denote the length of the signal and the number of wavelet filters, respectively. The entropy H(x) of a signal x[n] of length N is defined by $$H(x)=\sum _{n=0}^{N1}x(n){}^{2}\mathrm{log}x(n){}^{2}$$(4)
For determining the optimal levels of decomposition, the entropy is evaluated at each level. For a new level j, the entropy works as the following constraint: $$H(x{)}_{j}\ge H(x{)}_{j1}$$(5)
A twolevel decomposition method is sufficient for the effective representation of HVDC transmission system identifying the type of fault. The components (low and highfrequency components) were scaled by their respective gains and then added together to generate the control signal u: $$u={k}_{{d}^{1}}{e}_{{d}^{1}}+{k}_{{d}^{2}}{e}_{{d}^{2}}+\cdots +{k}_{{d}^{N}}{e}_{{d}^{N}}+{k}_{{a}^{N}}{e}_{{a}^{N}}$$(6) where gains k_{d1}, k_{d2},…, k_{dN} are used to tune the high and mediumfrequency components of the fault (instantaneous power) signal (e_{d1}, e_{d2},…, e_{dN}). Gain k_{aN} is used to tune the lowfrequency components of the fault signal (e_{aN}) under test and N is the number of decomposition levels.
Numerical simulation results and discussion
For numerical verifications the following parameters are considered and fault occurrences are numerically designed with time intervals as described in Table 1. The DC transmission line length equals 300 km, and the AC generator was rated at 5000 MVA at 60 Hz, and load end utility of 500 kV. The transformers are rated, the rectifier and inverter ends at 1200 MVA, with grid 345 kV, 50 Hz, 10,000 MVA, and power transmission capacity of 1000 MW.
Type of fault  Fault time (s) 

SLG fault (generation end)  1.60–1.68 
DClink (HVDC transmission) fault  5.44–5.52 
SLG fault (load end)  6.72–6.80 
SLG=singlelinetoground.
In the first investigation test, Fig. 2 shows the simulation response when the 12pulse LCCHVDC transmission system is subjected to healthy and singlelinetoground fault on the generator side by DWT diagnostic tool behavior. Figure 2a–d depict the threephase generated voltage, generated currents, DClink voltage, and DClink current, respectively. From the results it is clear that healthy condition exists until the time interval 1.6 s, where the designed fault occurs in the transmission system, eventually leading voltages and currents to behave abruptly above the set ratings until 1.7 s, where again it returns back to healthy conditions as designed for investigation. Figure 2e illustrates the behavior response of fault identification by DWT; it is clearly observed until the healthy conditions output reaches a constant amplitude of 1.7, for all coefficients of twolevel decomposition according to eq. (6). Once, the designed fault occurs at 1.6 s, the DWT output value increases and reaches the peak amplitude of 6 and the contour area widens as expected.
From the amplitude of peak and contour areas, it could be easily distinguished that the fault that occurred is a singlelinetoground fault on the generator side. Furthermore, DWT returns back to its initial values once the fault is cleared at 1.7 s (eq. 6), and the contour area also reduces as it appears in healthy conditions. Hence, by DWT spectrum behavior, the healthy and faulty conditions are clearly distinguishable when the system is subjected to the disturbance.
The second investigation test is focused on singlelinetoground fault identification by DWT on the load end. Figure 3 describes the simulation response, as a 12pulse LCCHVDC transmission system subjected to healthy and singlelinetoground fault on the load side, and DWT diagnostic tool behavior. Figure 3a–d depict the threephase generated voltage, generated currents, DClink voltage, and DClink current, respectively. It was clear from the results that healthy conditions exist until the time 6.72 s, where the designed fault occurs in the transmission system, eventually leading the voltages and currents to behave abruptly as predicted above the set ratings until the time of 6.8 s, where again it returns back to healthy conditions as designed for investigation. Behavior of fault identification by DWT is described in Fig. 3e; it is notable until healthy conditions output reaches a constant amplitude of 1.7, all coefficient of twolevel decomposition according to eq. (6). Once, the fault occurs at 6.72 s, the DWT output value increases and reaches the peak of 5.8 in amplitude and the contour area widens as expected.
Furthermore, the contour area generated by DWT outputs is not similar to the first investigation test and it is clearly distinguishable. Hence, the amplitude of peak and contour areas shows that the fault occurred is singlelinetoground fault on load end. DWT returns back to its initial values once the fault is cleared at 6.8 s (eq. 6) and the contour area also reduces as it appears in healthy conditions. Hence, again by DWT spectrum behavior, the healthy and faulty conditions are clearly predictable when the system subjected to the disturbance.
In the third investigation test, Fig. 4 describes the simulation response, as a 12pulse LCCHVDC transmission system subjected to healthy and DClink fault, and DWT diagnostic tool behavior. Figure 4a–d depict the threephase generated voltage, generated currents, DClink voltage, and DClink current, respectively. It is noticed from the results that healthy conditions exist until the time interval 5.44 s, where the designed fault occurs in the transmission system, eventually leading the voltages and currents to behave abruptly as predicted above the set ratings until 5.52 s. Figure 4e elaborates the behavior of fault identification by DWT; it is observed until healthy conditions output reaches a constant amplitude of 1.7, all coefficient of twolevel decomposition according to eq. (6). Once, the fault occurs at 5.44 s, the DWT output value increases and reaches the peak of 5.8 in amplitude and the contour area widens as expected. To be noted, that the contour area generated by DWT outputs is not similar to the first and second investigation tests and it is clearly distinguishable.
Furthermore, it should be noted that once the DC fault occurs at time interval 5.52 s, drastic failure occurs and the complete system needs to be protected. So, the generator side is opencircuited from DC link similar to the load side; therefore, the AC generator, rectifier, and inverter on the load side are protected. However, voltage generated after the time 5.52 s is purely opencircuited (Fig. 4a), whereas threephase current on the generator side, DClink voltage, and DClink current are null, which confirms the protection and opencircuited conditions. Hence, the amplitude of peak and contour areas shown by DWT confirm that the fault occurred is DClink fault on the load end and also the opencircuit protection state. It is concluded from these three preliminary investigation tests that DWT is capable of predicting a diagnostic tool when the HVDC transmission system suffers with different faults.
Conclusion
This short communication exploited the capabilities of DWT as a postprocessing tool to identify the fault in a 12pulse standard LCCHVDC transmission system. A set of faulty conditions are designed and numerically implemented using the simulation software (MATLAB/PLECS). A set of observed results are presented and show good agreement with expectation under healthy and different fault conditions. Furthermore, the challenging task related to other types of fault and fault occurrences during transient conditions and the effect of leakage reactance on faults with sag/swell need to be investigated and will be addressed in future articles. Moreover, the DWT application can be extended to other transmission system types such as FACTS devices (STATCOM, SSSC, and UPQC) and active filters.
References

de Andrade L, and Ponce de Leao MT. 2014. Fault location for transmission lines using wavelet. IEEE Transaction on Latin America, 12(6): 1043–1048.

De Kerf K, Srivastava K, Reza M, and Bekaert D. 2011. Waveletbased protection strategy for DC faults in multiterminal VSC HVDC systems. IET Journal of Generation, Transmission and Distribution, 5(4): 496–503.

Febin Daya JL, Sanjeevikumar P, Blaabjerg F, Wheeler PW, and Ojo JO. 2015. Implementation of wavelet based robust differential control for electric vehicle application. IEEE Transaction on Power Electronics, 30(12): 6510–6513.

Febin Daya JL, Sanjeevikumar P, Blaabjerg F, Wheeler PW, Ojo JO, and Ahmet HE. 2016. Analysis of wavelet controller for robustness in electronic differential of electric vehicles—An investigation and numerical implementation. Journal of Electric Power Components and Systems.

Febin Daya JL, Sanjeevikumar P, Subbiah V, and Iqbal A. 2013a. A novel waveletfuzzy based indirect field oriented control of induction drives. Journal of Power Electronic, 13(4): 656–668.

Febin Daya JL, Subbiah V, and Sanjeevikumar P. 2013b. Robust speed control of an induction motor drive using waveletfuzzy based selftuning multiresolution controller. International Journal of Computational Intelligence System, 6(4): 724–738.

Hui L, Poh Chiang L, and Blaabjerg F. 2015. Submodule short circuit fault diagnosis in modular multilevel converter based on wavelet transform and adaptive neuro fuzzy inference system. Journal of Electric Power Components and Systems, 43(8): 1080–1088.

Mallat SG. 1989. A theory for multiresolution signal decomposition: The wavelet representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 11(7): 674–693.
 DOI: 10.1109/34.192463

Mohagheghi S, Harley RG, Habetler TG, and Divan D. 2009. Condition monitoring of power electronic circuits using artificial neural networks. IEEE Transaction on Power Electronics, 24(10): 2363–2367.

Nanayakkara OMKK. 2011. Location of DC line faults in conventional HVDC systems with segments of cables and overhead lines using terminal measurements. IEEE Transaction on Power Delivery, 27(1): 279–288.

Nanayakkara OMKK. 2012. Travelingwavebased line fault location in starconnected multiterminal HVDC systems. IEEE Transaction on Power Delivery, 27(4): 2286–2294.

Saleh SA. 2013. The implementation and performance evaluation of 3Ф VS wavelet modulated ACDC converters. IEEE Transaction on Power Electronics, 28(3): 1096–1106.

Sanjeevikumar P, Febin Daya JL, Blaabjerg F, MirNasiri N, and Ahmet HE. 2015a. Numerical implementation of wavelet and fuzzy transform IFOC for threephase induction motor. The Engineering Science And Technology: An International Journal, 19(1): 96–100.

Sanjeevikumar P, Febin Daya JL, Blaabjerg F, Wheeler PW, Szcześniak P, Oleschuk V, and Ahmet HE. 2015b. Waveletfuzzy speed indirect field oriented controller for threephase AC motor driveinvestigation and implementation. Engineering Science and Technology: An International Journal.

Sanjeevikumar P, Febin Daya JL, Wheeler PW, Blaabjerg F, Fedák V, and Ojo JO. 2015c. Wavelet transform with fuzzy tuning based indirect field oriented speed control of threephase induction motor drive. In Proceedings of the 18th IEEE International Conference on Electrical Drives and Power Electronics, IEEEEDPE’15, Tatranska Lomnica, Slovakia, 21–23 September 2015, pp. 111–116.

Sanjeevikumar P, Paily B, Basu M, and Conlon M. 2014. Classification of fault analysis of HVDC systems using artificial neural network. In Proceedings of the 49th IEEE Universities Power Engineering Conference, IEEE–UPEC’14, ClujNapoca, Romania, 2–5 September 2014, pp. 1–5.

Wang L, and SaNguyen TM. 2014. Stability enhancement of largescale integration of wind, solar, and marinecurrent power generation fed to an SGbased power system through an LCCHVDC link. IEEE Transaction on Sustainable Energy, 5(1): 160–170.

Yeap YM, and Ukil A. 2014. Wavelet based fault analysis in HVDC system. In Proceedings of the 40th Annual Conference of IEEE Industiral Electronics Society, IECON’14, Dallax, TX, USA, 29 October–1 November 2014, pp. 2472–2478.