21, rue d'Artois, F-75008 Paris http://www.cigre.

org

15-109

Session 2002
CIGR

CABLE DIAGNOSIS BASED ON DEFECT LOCATION AND CHARACTERIZATION THROUGH PARTIAL DISCHARGE MEASUREMENTS
M. de Nigris, G. Rizzi CESI F. Ombello Pirelli Cavi&Sistemi F. Puletti TechImp (Italy) A. Cavallini, G.C. Montanari, M. Conti University of Bologna

Summary A new approach to cable diagnostics is discussed in the present paper. The method and the system developed are based on the apparently well assessed measurement and location of partial discharges activity in cables, but are innovative and add substantial diagnostic value. This feature is achieved thanks to the use of most advanced statistical elaboration tools and fuzzy logic techniques which allow to obtain an efficient noise rejection and a high performance diagnosis in which the conventional information is complemented by an identification of the defect type, origin and position. The present paper describes extensively the working principle of this new diagnostic system and discusses its potential in the light of a laboratory activity carried out considering polymeric and paper-oil insulated cables, where artificial defects have been previously introduced. keywords: power cables, diagnostics, partial discharges, defect location, defect identification. 1 Introduction Diagnostic techniques constitute a fundamental tool for an optimized management of power networks. An efficient monitoring and diagnostic plan can reduce costs for maintenance and repair operations. It is consolidated knowledge that partial discharges (PD) measurements constitute one of the most promising tools for the evaluation of localized defects and damages. In the field of power cables, several condition assessment systems based on PD measurement are available on the market, each characterized by a specific supply voltage waveshape, data acquisition and elaboration philosophy. The performances of several advanced cable diagnostic systems have been recently compared in the frame of an extensive investigation

carried out in CESI laboratories and in representative field conditions; the relevant findings are reported in [1]. In all the systems considered, great attention is focused on defect location while little information is available about the origin of the defects and their consequent harmfulness. A new diagnostic system is presented here, which uses pulse shape and height-phase acquisition to correlate PD location and the recognition of the phenomena causing each PD activity. The system uses most advanced statistical elaboration tools and fuzzy logic techniques to obtain an efficient noise rejection and a high performance diagnostic value. At present, the diagnostic procedure adopted is based on an off-line approach to the condition assessment process: the cable being checked has to be de-energized and disconnected from any source or load from all terminals; a separate voltage generator is needed to energize the cable with a specific voltage waveshape. In this respect, the system has been successfully set up and tested with power frequency sources as well as with the very low frequency (0.1 Hz) and oscillating wave with low damping power sources. Most of the results reported in the present paper are relevant to the power frequency tests. The condition assessment of the power cable using the system object of the present paper consists in energizing the cable with a voltage stress having a value in the range of U0 to 2U0 and in performing the measurement of the PD activity. The PD signal acquired is subjected to a preliminary elaboration, called classification, in which the complex patterns acquired are classified in homogeneous subassemblies, having in common the same kind of PD source; successively the complex pattern is subjected to a separation process, allowing the original pattern to be split into different sub patterns; each one of them is relevant to one class characterized by homogeneous

pulses; the further step is the identification of the defect generating PD belonging to each class. An advanced tool for the location of the defects along the cable route is also integral part of the system, thus allowing to point out the presence and position of weak points and to give guidance to the cable repair working teams. The features and the characteristics of the diagnostic system based on advanced PD analysis will be analyzed and discussed in the following sections, as well as the application in laboratory to configurations simulating real situations, with artificial and controlled cable defects. Part of the activity described in the present paper has been carried out in the frame of the research projects foreseen in the Italian RD&D program on electrical system (Decree of the Italian Ministry for Industry, Trade and Handicraft dated 26 January 2000). 2 Measurement system and diagnostic tools In the following the features of the new diagnostic system based on PD detection, in terms of noise rejection, PD source location and diagnostic effectiveness, are described.

Fig. 1: Example of PRPD pattern of an entire acquisition. Fig. 2 shows the classification map that results from the feature extraction procedure applied to the data whose pattern is reported in Fig.1.

2.1

Classification Classification consists of separating the contribution of the different sources generating the recorded PD (generally summarized in a pattern reporting PD height vs. acquisition phase). Skilled operators can often classify successfully even complex patterns, but in the presence of multiple PD sources, possibly including significant background noise, neither skilled operators, nor artificial intelligence and statistical techniques can find univocal answers [2]. The solution employed in the new diagnostic system is to implement a fuzzy classifier in order to separate different PD phenomena according to the pulse shape of the discharges, under the assumption that discharge signals due to different sources are, generally, characterized by different pulse shape. Noise rejection, too, can be achieved thanks to this classification process since it occurs, in general, that signals generated by background noise or external disturbances are significantly different in shape from PD signals generated within the equipment under test. This result allows homogeneous classes of data to be treated separately through advanced statistical tools, so that an identification process can provide the recognition of background noise, disturbance or whatever PD phenomenon [3]. The following example will explain more in detail the classification procedure. Data are acquired and processed by a specific software that enables the visualization of the usual 3D Phase Related Partial Discharges (PRPD) pattern, as reported in Fig. 1. Each acquired pulse (corresponding to a single dot in the pattern) is processed in order to extract its most salient information. For this purpose, two conventional pulse parameters, that is, T equivalent time length and W equivalent bandwidth, are calculated and reported in a 2D map, the classification map [4].

Fig. 2: Classification map. Each point corresponds to a single pulse. The three clusters are relevant to pulses having different T,W features. The classification map is then processed by a fuzzy classification algorithm that generates different clusters of data. As a result of fuzzy classification, each pulse is considered as belonging to all classes, but with different membership degrees. Eventually, a given data point is assigned to the class to which it belongs with the highest membership. In this way, a crisp separation of the map in different classes can be obtained. This separation procedure allows the original PRPD pattern to be split into different sub patterns; each one of them is relevant to one class characterized by homogeneous pulses. An example of the results of the separation process is reported in Fig. 3 (referring, again, to the pulses collected in Fig. 1). As illustrated in Fig. 3, the classification map singles out the presence of different discharge distributions corresponding to groups of pulses characterized by different waveform shapes; therefore a single cloud in the classification map is associated to its characteristic pulse shape and identifies a sub-pattern composed by shape-homogeneous pulses. The final result is the separation of the original pattern into different pulsehomogeneous classes (sub-patterns), which can be treated separately for effective processing and diagnostic purposes [4].

Original PD pattern

Characteristic class pulse

Separated sub-PD patterns

0. 03

0. 02

0. 01

0. 00

- 01 0. 0 100 200 300 400 504

0. 04 0. 03 0. 02 0. 01 0. 00

- 01 0. 0 100 200 300 400 504

0. 04 0. 03 0. 02 0. 01 0. 00

Classification map

- 01 0. 0 100 200 300 400 504

Fig. 3: The separation process: starting from the original PD pattern, the classification map is derived and classified in different clusters corresponding to different sub patterns, each one characterized by homogeneous pulse shape discharges 2.2 Identification The further step is the identification of the defect generating PD belonging to each class, which is the basis for noise rejection and enhanced diagnostics. Statistical processing of phase and amplitude distributions is carried out in order to derive diagnostic markers and achieve identification. One of the statistical tools available derives from the application of the two parameter Weibull function to the whole set of PD height values, as well as to each single class provided by the fuzzy classifier. The two parameter Weibull distribution of PD pulse height is given by the following expression: that the shape parameter of the Weibull function, being associated to the dispersion of PD pulse height sequences, is characteristic of a given PD phenomenon, i.e. of the source generating PD pulse trains [5]. For example, values of <2 generally indicate surface discharges, while 2<<8 is characteristic of internal discharges. Corona discharges provide very large values. Other identification and diagnostic markers come from the evaluation of parameters such as: the discharges rate of occurrence, the phase distribution range, the maximum and minimum discharge amplitude, NQN (area under the curve of pulse repetition rate as a function of pulse magnitude) [6] and other indexes calculated on the basis of the histogram of the time elapsing between consecutive discharges [7]. Special routines have been arranged in order to identify background noise, as well as peculiar external disturbances. Noise recognition has not a unique solution, so that it must be approached by devising various techniques, each one tailored for a specific kind of noise. At present, experimental evidences suggest that random noise and noise due to AC/DC rectifiers are the most common disturbances that need to be rejected during on-field measurements (e.g., in substations, industrial environments, etc.). Background random

q F (q ) = 1 exp

(1)

where q is the pulse charge height, and are the Weibull function scale and shape parameters. The pulse phase distribution is basically treated through 2nd, 3rd and 4th order moment evaluation (used also for the height distribution). The estimation of (eq. (1)) is particularly important for PD identification purpose. In fact, it has been shown

noise is treated by investigating the lack of correlation of its phase distribution with the supply voltage wave, while the noise due to AC/DC converter units is recognized by means of special routines that identify the peculiar correlation of its phase distribution with the supply voltage waveform [8]. The identification procedure consists of several steps characterized by progressively more refined indications about the nature of the defect. The first step, after having excluded a noise nature of the recorded phenomenon, is based on a fuzzy classification system applied to a set of historical data of well-known PD phenomena. In particular, a restricted vector of parameters (such as shape parameter of the PD height Weibull distribution (), the minimum phase for the examined pattern (min), the phase range of discharge distributions (), the skewness of the height distribution) has been identified as the most significant set of markers for basic classification purposes. By analyzing this set of markers, three main PD categories can be separated and identified: corona discharges, internal discharges and surface discharges. This is the framework for an effective identification of each new recorded PD phenomenon. Afterward a further, more refined identification process can be carried out by means of routines that evaluate more in detail the ranges of variation of the same parameters and the correlation among them. In such a way other features of the PD source can be deeply Classify and split

investigated. For instance, an information about the proximity of the defect to one or the other electrode, or about the dimension and, even, the shape of the defect can be obtained [9]. A scheme of the complete identification procedure is shown in Fig. 4

2.3

Location Tool PD location tool has fundamental importance for cable diagnostic purposes. An adjustable timebase pulse acquisition of the PD measurement system allows data to be acquired also in a location mode, alternative to the usual pattern mode seen above, and the sources generating PD signals to be located by means of the reflectometric technique. The reflectometric technique can be summarized as follows. Referring to Fig. 5, when a discharge occurs in a cable section X, that is x meters far from termination A (the one where the measurement equipment is located) of a L meter long cable, the discharge pulse starts traveling along the cable, both in the direction of termination A and in the direction of termination B. By picking up each PD signal in correspondence of termination A, the detected waveform will appear as reported at the bottom of Fig. 5. The first peak is due to the pulse traveling in the direction of the termination A, the other two pulses are due to the reflection in correspondence of the termination B of both pulses generated in X (the reflections occur due to impedance mismatches). Identify Noise: N 1.0 - Y 0.0 Corona 0.05 Surface 0.50 Cavity 0.45 Multiple phenomenon 0.1 Internal interface 0.9

Noise: N 1.0 - Y 0.0 Corona 0.10 Surface 0.10 Cavity 0.80 Close to HV electrode 0.2 Embedded 0.7 Close to Ground electrod 0.1 e Noise: Y 0.98- N 0.02 Corona Surface Cavity Fig. 4: Example of the complete identification scheme. In bold the PD source identified as most likely by the diagnostic system. In italic an uncertain identification. The numbers from 0 to 1 correspond to the attribution rank given to a class by the fuzzy algorithm.

x meters

A
L meters

Signal Pickup Point

Fig. 5: The reflectometric technique. In correspondence of the signal pick-up point a pulse sequence will appear as a result of the different arrival time of the discharge pulse and its reflections Simple calculations based on the evaluation of the arrival times of a pulse sequence (that is a pulse and its reflections) allow the position (x) of the pulse-source to be derived, being known the length L of the cable or the propagation speed of pulses along the cable itself (and having carried out proper calibration) [10]. During PD location tests, a cable is energized until a PD activity is steadily detected; at this time a set of pulse sequences is acquired and processed. Noise can be rejected by comparing the attenuation of the calibration sequence with that of the acquired sequence. At the end of the processing phase a location map is derived, which reports the amplitude of each recorded PD versus its position of occurrence along the cable, as illustrated in Fig. 6.
0.7 0.6 0.5 0.4 0.3 -1.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0

Fig. 6: Example of a location map. Each point represents a discharge. The acquisition procedure (location mode) allows a PRPD pattern to be built, which is then processed, classified and identified. However there are some fundamental differences with respect to the pattern mode. The two modes differ for the timebase used to capture PD signals: large timebase in location mode, in order to capture the signal and its reflections, short timebase in pattern mode, in order to record the details of a single pulse waveform. This distinction permits different source identification procedures to be applied. In pattern mode different phenomena are separated by means of the fuzzy classification system, as described above, on the basis of the characteristic pulse shape of the discharges. In location mode each signal corresponds to a sequence consisting of a discharge with

its reflections. Thus, for each signal, a location can be derived for the discharge sources thanks to the reflectometric technique. In addition, or in alternative, to semi-automatic location procedure shown in Fig. 6, a fuzzy classification tool can be run on the acquired data and the separation of the original data set is provided on the basis of the characteristic signal shape of the acquired pulse sequences. Experimental results show that when pulses are acquired in sequences (due to a large time base), the fuzzy classification tool mainly distinguishes groups of sequences on the basis of the position of the intermediate pulse. This means that the classification tool is able to separate the data relevant to different discharge sites. As a result, homogeneous clusters of data are obtained, each one containing all the pulse sequences coming from the same discharge site. Once the first separation procedure has been carried out, a further analysis is required before proceeding to the identification phase. In fact, more than one PD phenomenon could be active in the same cable section. In order to distinguish among PD activities taking place in the same cable section, the shape of the first pulse of each sequence is taken into account for each location cluster; thus, data are reduced as if acquired in pattern mode (see Figs. 2,3) and a further classification procedure can be carried out on the basis of these pulse shapes. In such a way a complete separation of all PD phenomena active in the cable may be carried out, and for each one of them the identification is provided by appropriate routines of the system. A further correlation with the results of the pattern mode acquisition carried out on the same specimen can strengthen the indication provided by the location mode processing tool and allow a deeper inference on the nature and on the characteristics of the PD phenomenon to be achieved. In fact, the information related to the pulse shape of the PDs acquired in the pattern mode can provide a more refined separation of PD phenomena. Limitations to this approach are, of course, related to cable length. In fact, above a given length (varying as a function of type of insulation, cable and rated voltage)

defects are hardly inferred due to signal attenuation and dispersion, which may not allow to carry out separation within a located pattern. However, statistical tools can still be used, once separation as a function of PD source location has been achieved, even if differences among pulse shapes belonging to a series of pulse sequences coming from a certain location are hardly detectable [3]. 3 EXPERIMENTAL RESULTS Systematic laboratory tests were carried out in CESI laboratories. For the purpose of this paper we will discuss tests carried out on 2 medium voltage cables rated 12/20kV, insulated with different materials: i.e. paper-oil and XLPE. The cables were prepared with terminations having no defects and dimensioned in such a way as to allow dielectric tests with the lowest probability of flashover or puncture. Defects were created artificially in the cables or in the joints to check the capabilities of the diagnostic systems to reveal, locate and recognize the different types of possible defects. The first sample (C1) was a 130 m long XLPE insulated 1x120mm2 aluminum conductor cable, having a defect consisting of the removal of the outer semiconductor of the cable for a region 80x200mm, positioned 30 m away from the termination connected to the measuring equipment. The second sample (C2) was a 194 m long paper-oil insulated 1x240mm2 aluminum conductor cable with two joints positioned at 65 m and 130 m respectively from the termination connected to the measuring equipment; each joint was artificially damaged, by means of the introduction of a metallic sharp protrusion from the conductor penetrating the joint insulation for about 3 mm.

3.2

Measurement results

Cable C1: 130 m XLPE cable. Partial discharges incepted at 6 kV and were located at about 30 m from the measuring point, as can be seen in the map reported in Fig. 7.
Cabl l ne PD M appi e i ng
0. 1

0. 0 - 0 1. 20. 0 40. 0 60. 0 80. 0 100. 0 120. 0 134. 0

Fig. 7: Location on XLPE 130. All recorded discharges were located in a range of [2732] m, acquisition performed at 10 kV. Other acquisitions were carried out at higher voltage values, with the same location results. Thus, at the end of the first session (location mode), it was concluded that only one cable section was affected by partial discharges. The analysis of the first waveform of the pulse sequences did not single out the presence of multiple PD phenomena. During the second test session (pattern mode) further acquisitions were performed at increasing voltage levels (Fig. 8). In those acquisitions all the discharges had similar pulse shapes (see the single cloud in the classification map reported in Fig.8b). Thus, it was argued that just one PD phenomenon was active in the single discharge site located, as the measurements carried out in location mode had shown, too.

3.1

Measurement circuit and test setup The measurement circuit was the usual one for PD detection, with coupling capacitor, Ck, connected in series to the measurement impedance, Zm (indirect circuit). The cables were energized with sinusoidal voltage at 50 Hz and at 0.1 Hz. PD source location results were consistent both at 50 Hz and at 0.1 Hz. For the sake of brevity, only the results obtained at 50 Hz will be dealt with in this paper [11]. PD tests consisted of two measuring sessions, which correspond to the two acquisition modes that the measuring system is provided with. The first test session was carried out in location mode; in such a way pulse sequences were acquired and the location map was built (as well as the PRPD pattern). Thus, PD phenomena were separated firstly on the basis of their location and then on the basis of the pulse shape of the discharge waveform. Finally, an identification was attempted for each PD generating phenomenon. The second session was carried out in pattern mode, recording again each pulse, but in a shorter and more suitable time window. In particular, the latter mode was used for improved PD classification and recognition purposes. The comparison between pulses and PRPD patterns acquired in the two sessions provided a further help for understanding the recorded PD phenomena.

a)

b) Fig. 8: PD pattern acquired on cable C1, a), and relevant classification map, b).

Defect identification was carried out through the processing program. In particular, statistical analysis, performed separately for PD of different polarities, i.e. for discharges occurred during the first and the second semi-period of the applied voltage, gave interesting results. Figure 9 shows the two parameter Weibull PD pulse height distribution, relevant to negative discharges: a very good fitting is evident in the figure between the recorded data and the Weibull distribution. This fact, combined with the result of the fuzzy classification (all pulses are similar), strongly reinforce the evidence that there is only one discharge active site, thus only one defect.

Fig. 11. Let us note that both the location map and the pattern are built from the same base of data, i.e. form the pulses acquired in location mode.
0.7 0.6 0.5 0.4 0.3 -1.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0 194.0

Fig. 10: Location map on cable Paper-oil 194 (acquisition carried out at 24 kV).

Fig. 9: Two parameter Weibull PD pulse height distribution, relevant to negative discharges As illustrated in the previous paragraph, the value of (shape parameter of the Weibull distribution) is taken into account for identification, together with other parameters resulting from the phase analysis, particularly min and . The basic identification among the three main PD categories (i.e. cavity, corona, surface) was carried out, and the analyzed PD phenomenon was identified as a surface discharge. Then, a more refined indication was derived from the evaluation of several other markers. In particular, the value of and the trends of min and values with increasing voltage were considered, together with the discharge amplitude PD pattern asymmetry (evaluated by NQN and the ratio between the number of positive and negative discharges). On these basis, the diagnostic tool provided a final response and identified the defect as a superficial one, characterized by a large, opened (not embedded) discharging area located close to the grounded electrode. The diagnostic indication matched the real nature of the artificial defect, i.e. the removal of a 25 square centimeters area of the semi-conductive shield. Cable C2: 194 m paper-oil cable Partial discharges incepted at 7 kV; the location tool indicated the presence of a defect at 130 m from the measuring point. At higher voltage values (over 18 kV) other pulses were detected at 65 m from the measuring site. Hence, two different cable sections resulted as affected by PD, in correspondence to the artificial defects, positioned 65 and 130 meters away from the cable termination, respectively. The location result is shown in Fig. 10, while the PD pattern including discharges from both defective sections is reported in

Fig. 11: Pattern relevant to the acquisition performed at 24 kV on cable C2 The fuzzy classification tool applied to pulse sequences (location mode) singled out the presence of two phenomena occurring in different cable sections, that correspond to the two located discharge sites. Thus, the whole data set (Fig. 11) can be separated into two sub patterns, Figs. 12 and 13, relevant to the defect located at 65 m and to that at 130 m, respectively. The separated sub patterns can be processed one by one, in order to get the identification of the defect generating PD. Of course, indications could be also obtained by processing separately the patterns achieved between 7 and 18 kV, when only one PD source is active.

Fig. 12: Sub-pattern acquired at 24 kV and relevant to the PD phenomenon located at 65 m. The analysis of the pattern reported in Fig. 12 (regarding the defect located at 65 m), as well as other patterns relevant to the same phenomenon acquired at higher voltage values, indicates the presence of internal discharges. This indication matched the nature of the artificial defect inserted in the joint. However, it must be noticed that a reliable inference about the nature of

the source generating PD was very difficult because of the high inception voltage and the consequent small number of recorded discharges.

measurements is the lack of identification of the kind of source generating PD (not only of the source location). The approach here proposed constitutes a considerable step forward in making available reliable diagnostic indicators and rules to allow effective maintenance programs to be planned. 5 REFERENCES

Fig. 13: Sub-pattern acquired at 24 kV and relevant to the PD phenomenon located at 130 m. The analysis of the pattern reported in Fig. 13, as well as that of other patterns acquired at higher voltage values and relevant to the same phenomenon, showed that the PD height distribution did not always fit the Weibull distribution. This fact may indicated that more than one discharge phenomenon could be present at the same time. Also the evaluations based on the values of phase parameters gave indications that assigned an intermediate nature to the defect (between internal and surface). In fact, the fuzzy diagnostic routine evaluated the defect as internal with membership 58 % and surface-like with membership 42 %. This may support the presence of two different PD phenomena active in the same site, but it was not possible to achieve a good separation between the two phenomena on the basis of pulse shape, neither in location mode (considering the shape of the first pulse of each sequence) nor in pattern mode (where just the discharge pulse is acquired); this was likely due to attenuation and dispersion effects of traveling PD pulses, which smoothed the differences among pulse shapes. Nevertheless the system, by means of its statistical tools, singled out the presence of a multiple phenomenon. Indeed, the indication given by the system was consistent with the nature of the artificial defect, which consisted of the presence of a metallic nail inside the joint. It is possible that discharges incept both inside the cavity provided by the nail, having in this way an internal nature, and in the cable surface, starting from the nail body. 4 CONCLUSIONS The purpose of this paper is to show how an effective and complete diagnostic evaluation can be carried out on medium and high voltage cables by means of an innovative approach, consisting in the integration of reflectometric technique and advanced diagnostic procedures in the same system. In favorable conditions, i.e. when PD pulses are not excessively damaged and distorted, a double goal can be achieved: location from one side, separation and identification from another side. It is worthwhile to recall that one of the major concerns regarding the reliability of diagnosis made through PD

[1] V. Colloca, A Fara, M. de Nigris, G. Rizzi: Comparison among Different Diagnostic Systems for Medium Voltage Cable Lines, CIRED 2001 Amsterdam, the Netherlands. [2] H.G. Kranz, "Diagnosis of Partial Discharge Using Neural Networks and Minimum Distance Classification", IEEE Trans. on Electrical Insulation, Vol. 28, n. 6, Dec. 1993, pp. 1016-1024. [3] A. Cavallini, A. Contin, G. C. Montanari, G. Pasini, F. Puletti, Digital Detection and Fuzzy Classification of Partial Discharge Signals, to be published in IEEE Trans. on Dielectrics and Electrical Insulation, 2002. [4] A. Contin, A. Cavallini, G.C. Montanari, G. Pasini, F. Puletti, "Artificial Intelligence Methodology for Separation and Classification of Partial Discharge Signals", IEEE - Conference on Electrical Insulation and Dielectric Phenomena, Victoria, Canada, Oct. 2000, pp. 522-526. [5] A. Contin, G.C. Montanari, C. Ferraro, Partial Discharge Source Recognition by Weibull Processing of the Pulse Height Distribution, IEEE Trans. on Dielectrics and Electrical Insulation, Vol. 7, n. 1, pp. 48-58, Feb. 2000. [6] J. F. Lyles., G. C. Stone, M. Kurtz, Experience with PDA diagnostic testing on hydraulic generators, IEEE Trans. Energy Conversion, Vol. 3, n. 4, pp. 824-830, Dec. 1988. [7] M. Hoof, R. Patsch, Detection of Multiple Discharge Sites Using Pulse/Pulse Correlation, IEEE Trans. on Dielectrics and Electrical Insulation, Vol. 7, n. 1, pp. 12-20, Feb. 2000. [8] A. Cavallini, M. Conti, A. Contin, G.C. Montanari, Automatic Noise Rejection Tools in On-Field PD Measurements, 2001 Volta Colloquium, Como, Italy, Sept. 2001. [9] A. Cavallini, M. Conti, A. Contin, G.C. Montanari, Indexes for the Recognition of Insulation System Defects Derived from Partial Discharge Measurements, 2002 Int. Symp. on Electrical Insulation, Boston, USA, Apr. 2002. [10] M.S. Mashikian, F. Palmieri, R. Bansal, R.B. Northrop, Location of Partial Discharges in Shielded Cables in the Presence of High Noise. IEEE Trans. on Electrical Insulation, Vol. 27, n. 1, pp. 37-43, Feb. 1992. [11] M. De Nigris, U. Vercellotti, F. Zagliani, Condition Assessment for Medium Voltage Power Cables in industrial Environments by means of VLF Techniques, 5th International Conference on Insulated Power Cables, JICABLE, Versailles, France, Jun. 1999, pp. 921-926.