Research on Partial Discharge Signal Detection Technology of Cable Joints Based on a Dynamic Multi-Notch Method

Abstract

Aiming at solving the detection problems caused by weak partial discharge signals of underground cable joints and random and variable spatial electromagnetic wave interference, a non-contact detection technology based on the dynamic multi-notch method is proposed. This technology synchronously collects pure interference signals and mixed signals containing partial discharge through a dual-position detection antenna. After converting to the frequency domain via Fast Fourier Transform (FFT), the notch frequency bands are dynamically determined based on the real-time interference spectrum, and interference suppression is achieved by frequency domain zeroing filtering. Finally, the partial discharge pulse signal is restored through Inverse Fast Fourier Transform (IFFT). A simulation experiment platform for 10 kV XLPE cable joints was built to verify the detection of typical defects such as metal debris, insulation scratches, and conductor burrs. Experimental results show that the average extraction success rate of this method for weak partial discharge signals reaches 94.7%, and the detection accuracy is ≥92.3% in a normal environment without strong interference, which is significantly better than the traditional ultra-high frequency (UHF) detection method (45.8%) and the fixed notch method (68.3%). This technology realizes the accurate detection of weak partial discharge signals in complex environments, provides a reliable solution for the early warning of insulation defects in underground cable intermediate joints, and has important engineering application value.

1. Introduction

With the wide application of cross-linked polyethylene (XLPE) power cables in urban power grids, the insulation state of cable joints, as the core connection components of transmission lines, directly determines the safe and stable operation of power systems. Statistical data show that more than 70% of cable line faults are caused by joint defects, and partial discharge is an early characteristic signal of insulation degradation of cable joints [1,2]. Underground cable intermediate joints are prone to initial defects such as residual metal debris, insulation scratches, and conductor burrs due to differences in construction technology, soil corrosion, and water intrusion, which trigger partial discharge. If not detected in a timely manner, partial discharge will accelerate insulation aging and eventually lead to insulation breakdown, causing major power outage accidents [3,4].

Existing partial discharge detection methods for cable joints have obvious limitations [5,6,7,8]: the ultrasonic detection method is ineffective for underground cable detection due to attenuation by solid media; both the high-frequency pulse current detection method and the capacitive coupling method are contact detection methods that require pre-installed sensors during cable manufacturing or laying and cannot be applied to already operating underground cables; although the UHF detection method is non-contact, the frequency of on-site spatial electromagnetic wave interference (such as mobile communication signals and industrial equipment radiation) changes randomly, and the interference intensity is much higher than that of partial discharge signals (usually differing by 1 to 2 orders of magnitude), resulting in weak discharge signals being completely submerged and poor detection effects.

To solve the above problems, scholars have proposed a variety of interference suppression methods, such as adaptive filtering, wavelet denoising, and machine learning classification, but none of them have realized real-time dynamic tracking of interference frequency bands, making it difficult to cope with the complex and variable on-site electromagnetic environment [9,10,11]. Based on the principle of a dynamic multi-notch, this paper designs a dual-position detection antenna and an adaptive signal processing algorithm. By tracking the interference spectrum in real time and dynamically adjusting the notch frequency bands, the accurate extraction of weak partial discharge signals in a strong interference environment is realized, providing technical support for the insulation state monitoring of underground cable intermediate joints.

2. Detection Principle and Technical Scheme

2.1. Detection Principle

In this study, a weak partial discharge signal is defined as a UHF electromagnetic pulse signal that, after attenuation by soil, reaches the ground detection point with an amplitude ≤50 μV and is 1–2 orders of magnitude lower than the on-site electromagnetic interference. Such signals correspond to partial discharges generated by incipient insulation defects in cable joints, such as tiny metal particles and slight insulation scratches.

When a partial discharge occurs in a cable joint, it excites an ultra-wideband electromagnetic signal covering 0–3 GHz. The low-frequency components below 100 MHz are difficult to radiate into free space due to their low frequency and mainly propagate along the cable surface. In contrast, the high-frequency components from 300 MHz to 3 GHz can hardly propagate over long distances along the cable because the cable is not an ideal high-frequency conductor. Most of their energy radiates outward as spatial electromagnetic waves, and the effective radiation range is typically no more than 10 m due to the limited radiated energy. Based on the above signal propagation characteristics, the core objective of the algorithm proposed in this paper is to suppress spatial electromagnetic interference and achieve effective detection and reconstruction of weak partial discharge signals.

The core idea of the dynamic multi-notch method is to collect spatial interference signals and mixed signals separately through a dual-position detection antenna, determine the real-time interference frequency bands by frequency domain analysis, realize interference suppression by multi-notch filtering, and then restore the time-domain pulse signal profile combined with the ultra-wideband spectrum characteristics of partial discharge signals.

The key technical advantages of this method are as follows. ① The dual-position design realizes synchronous collection of interference signals and mixed signals, providing a real-time reference for dynamic notch. ② The frequency domain zeroing filtering has a fast processing speed and can quickly eliminate multi-band interference. ③ The notch frequency bands are dynamically adjusted based on the real-time interference spectrum, adapting to the random variation characteristics of interference signals.

2.2. System Design

The entire detection system consists of two parts: a dual-position detection antenna 3 and a signal processing module 6, as shown in Figure 1. The dual-position detection antenna 3 receives both the spatial electromagnetic wave interference signal 7 and the electromagnetic pulse signal 10 generated by partial discharge of the cable intermediate joint 2 and then transmits them to the signal processing module 6 with FPGA as the core for the fast dynamic multi-notch algorithm to extract the weak electromagnetic pulse signal 10 generated by partial discharge of the cable intermediate joint.

Dual-position detection antenna 3 contains two built-in antennas inside:

① The built-in upward directional antenna 4 is used to receive the electromagnetic wave interference signal 7 transmitted from space, with a detection angle of about 90° upward. At the same time, since electromagnetic pulse signal 10 generated by partial discharge of the cable intermediate joint is very weak and scatters into space after passing through the ground due to soil attenuation, the built-in upward directional antenna 4 can hardly detect the partial discharge pulse signal 10 of the cable intermediate joint.

② The built-in downward directional antenna 5 is used to receive the electromagnetic pulse signal 10 generated by partial discharge of the cable intermediate joint, with a detection angle of about 90° downward. However, the spatial electromagnetic wave interference signal 8 will be reflected after transmitting to the ground, and a part of the spatial electromagnetic wave interference signal 9 will also enter the built-in downward directional antenna 5, and its intensity is even much higher than that of the weak partial discharge pulse signal in some cases. At this time, the spatial electromagnetic wave interference signal 9 often completely submerges the partial discharge pulse signal.

The two channels of electromagnetic wave signals received by signal processing module 6 from the dual-position detection antenna 3 are shown in Figure 2. It is difficult to distinguish the difference between the time-domain waveform 11 received by the built-in upward directional antenna 4 and the time-domain waveform 12 received by the built-in downward directional antenna 5 in the time domain, because the electromagnetic pulse signal 10 generated by partial discharge of the cable intermediate joint is very weak and completely submerged in the spatial electromagnetic wave interference signals 7, 8 and 9. The purpose of this signal processing module 6 is to extract the weak partial discharge pulse signal 10 from the real-time changing spatial electromagnetic wave interference signals 7, 8 and 9.

Note: This method is suitable for detection scenarios with relative stationarity and environmental interference dominated by narrowband or fixed frequency offset. Under conditions of severe multipath propagation, high-speed motion, or extremely inhomogeneous soil reflection, waveform distortion and timing discrepancies will increase, and further improvement in robustness is required by combining spatial positioning and multi-antenna fusion.

2.3. Signal Detection Steps

The internal algorithm structure of the signal processing module is illustrated in Figure 3, which fully presents the entire process from the detection of spatial electromagnetic waves by the antenna to the final signal extraction. The specific detection procedure is as follows:

Step 1: The signal processing module 6 performs Fast Fourier Transform (FFT) on the spatial electromagnetic wave interference signal received by the built-in upward directional antenna 4 through Formula (1) to obtain the spatial electromagnetic wave interference signal spectrum 19. The spatial electromagnetic wave interference signal spectrum 19 is distributed in different frequency bands with different intensities, and the interference signal frequency bands change at any time, as shown in Figure 4.

$$(k)={\sum }_{n=1}^{N}x(n)e^{j2\pi (k-1)({\displaystyle \frac{n-1}{N}})}$$

(1)

where X(k) is the spectrum sequence value, x(n) is the time-domain sampling sequence value, N is the total length of the calculation sequence, n is the time-domain variable value, and k is the frequency-domain variable value.

Step 2: The notch frequency band regions 20 (a, b, c, d, e) to be filtered out are determined through the multi-notch strategy 14, as shown in Formula (2), where the step is a certain bandwidth, i.e., the difference between the frequency point n2 and the frequency point n1; Q is the interference signal spectrum density within a certain bandwidth step; X(k) is the spectrum sequence value; and k is the frequency-domain variable value. When the interference signal spectrum density is greater than a certain determined value Q, the frequency band region is regarded as the noise frequency band 20 to be filtered out.

$$\left\{\begin{array}{c}Q>{\sum }_{k=n1}^{n2}X\left(k\right)\\ step=n2-n1\end{array}\right.$$

(2)

Step 3: The signal processing module 6 performs Fast Fourier Transform (FFT) on the electromagnetic wave signal received by the built-in downward directional antenna 5 through Formula (1) to obtain the spectrum shown in Figure 5, which contains the spatial electromagnetic wave interference signal 19 and the partial discharge pulse signal 21 of the cable intermediate joint. Since the partial discharge pulse signal 21 is a pulse signal, it is a very wide ultra-wideband signal in the spectrum, but its amplitude is very weak and much lower than that of the spatial electromagnetic wave interference signal 19.

Step 4: Frequency domain partial discharge signal extraction 16 is carried out. The spectrum signal in Figure 5 is subjected to multi-notch filtering according to the filtered frequency band regions 20 (a, b, c, d, e) obtained by the multi-notch strategy 14 in Step 2, and the frequency bands to be filtered out are directly zeroed to obtain the spectrum in Figure 6. In the spectrum of Figure 6, the spatial electromagnetic wave interference signal is basically filtered out, and some frequency bands of the partial discharge pulse signal 22 are also filtered out. However, since the partial discharge pulse signal is an ultra-wideband spectrum signal, although some frequency bands are filtered out and some components of the signal are missing, the main profile of the pulse signal is not affected.

Step 5: The signal spectrum 22 after multi-notch processing (as shown in Figure 6) is subjected to Inverse Fast Fourier Transform (IFFT) according to Formula (3) to convert the frequency-domain signal into a time-domain signal 24 (as shown in Figure 7). The time-domain signal 24 is the pulse signal after multi-notch processing. Compared with the original partial discharge pulse signal 23, although the characteristic quantities such as amplitude and slope have changed, the overall profile can still be clearly identified as a pulse signal. Therefore, the pulse signal discrimination module 18 can accurately detect whether there is partial discharge in the cable intermediate joint.

$$x(n)={\displaystyle \frac{1}{N}}{\sum }_{k=1}^{N}X(k)e^{-j2\pi (k-1)({\displaystyle \frac{n-1}{N}})}$$

(3)

where x(n) is the time-domain sampling sequence value, X(k) is the spectrum sequence value, N is the total length of the calculation sequence, n is the time-domain variable value, and k is the frequency-domain variable value.

3. Experiments and Analysis

3.1. Experimental Platform Construction

To verify the detection performance of the proposed dynamic multi-notch method for partial discharge signals of cable joints, a simulation experiment platform for 10 kV XLPE cable joint partial discharge was built (Wuhan Guodian Xigao Electric Co., Ltd., Wuhan, China, Model: GDJF-2008). The platform is composed of a dual-position detection antenna detector, cable joint test pieces, a voltage regulator, a filter, a protective resistor, a voltage divider, a digital partial discharge detector, an interference source system, and other equipment, as shown in Figure 8.

3.1.1. Test Piece Preparation Unit

As the base material, 10 kV XLPE insulated power cables commonly used in actual engineering were selected, with a cable cross-sectional area of 70 mm², an insulation layer thickness of 4.5 mm, and a sheath thickness of 2 mm. Three cable joint simulation test pieces were made according to standard processes, with three typical insulation defects of underground cable joints, namely metal debris, insulation scratches, and conductor burrs, preset, respectively, as shown in Figure 9. The defect parameters were accurately quantified to simulate the defect types caused by construction, corrosion and other factors in actual operation. The specific parameters are as follows:

♦

Metal debris defects: Copper debris (simulating residual metal impurities in construction) with particle sizes of 1 mm, 2 mm and 5 mm were used, 5 pieces for each size, evenly embedded in the upper middle part of the joint insulation layer to simulate partial discharge caused by metal impurities inside the insulation.

♦

Insulation scratch defects: Prepared by precision mechanical processing, with scratch lengths of 2 mm, 3 mm and 5 mm, and a uniform depth of 0.5 mm (about 1/9 of the insulation layer thickness, without breaking down the insulation). The scratch direction is parallel to the cable axis to simulate mechanical scratches on the insulation layer during laying.

♦

Conductor burr defects: Aluminum burrs (the same material as the cable conductor) with heights of 1 mm, 2 mm and 3 mm, and a root width of 0.5~1 mm, directly processed on the surface of the cable conductor to simulate tip discharge defects caused by poor conductor processing technology.

Figure 9. Cable joints with three typical defects.

Figure 9. Cable joints with three typical defects.

3.1.2. High-Voltage Excitation Unit

The high-voltage excitation unit is composed of a voltage regulator (0~250 V, capacity 5 kVA), a partial discharge-free high-voltage test transformer (output 0~110 kV, transformation ratio 400:1), a protective resistor (10 kΩ, power 50 W) and a resistance–capacitance voltage divider (voltage division ratio 1000:1, measurement accuracy ±0.5%). Its core function is to apply an AC high voltage meeting the operation requirements of the 10 kV system to the cable joint test pieces to artificially excite partial discharge signals. Among them, the protective resistor is used to limit the fault current when the test piece breaks down to prevent equipment damage; the voltage divider is used to monitor the high voltage applied to the two ends of the test piece in real time and accurately to ensure the stability of the experimental voltage.

3.1.3. Electromagnetic Interference Simulation Unit

A precisely adjustable spatial electromagnetic wave interference source system was built to simulate the common electromagnetic interference types in the underground cable site, including two types of interference sources covering the main frequency bands of civil communication and industrial radiation. The specific parameters are as follows:

♦

Mobile communication signal simulator(Siglent Technologies Co., Ltd., Shenzhen, China, Model: SSG3032X-IQE): Outputs dual-band continuous wave signals of 900 MHz (GSM900) and 1800 MHz (DCS1800) with continuously adjustable amplitude and GSM signal modulation mode, simulating on-site mobile communication electromagnetic wave interference.

♦

Industrial equipment radiation simulator(Rigol Technologies Inc., Beijing, China, Model: DSG836A): Outputs 400~800 MHz broadband interference signals with continuously adjustable amplitude and a spectrum flatness of ≤±1 dB, simulating electromagnetic radiation interference from industrial frequency converters, electric welders, high-voltage switch cabinets and other equipment. The two types of interference sources can be turned on independently or synchronously to realize the simulation of complex electromagnetic environments with single/multiple interference source superposition.

3.1.4. Signal Detection Unit

A self-designed dual-position detection antenna detector is used as the core detection equipment. The detector is equipped with built-in upward/downward dual directional antennas, with an operating frequency band of 300 MHz~3 GHz, an antenna gain of ≥8 dBi, and a detection angle of ±90°, which can synchronously collect pure spatial interference signals and mixed signals containing partial discharge. At the same time, a traditional UHF detection antenna (broadband 300 MHz~3 GHz, gain 6 dBi) and a fixed notch method detection device (preset notch frequency bands of 500 MHz, 900 MHz and 1800 MHz) are equipped as comparative detection equipment; the auxiliary equipment is a digital partial discharge detector (minimum detectable discharge capacity 1 pC), which is used to calibrate the real partial discharge signal and serve as a reference standard for the detection results. Combined with the experimental setup (digital partial discharge detector, minimum detectable discharge quantity 1 pC), it is clarified through conversion that the minimum detectable partial discharge signal power of the proposed system is −75 dBm.

3.1.5. Data Acquisition and Analysis Unit

A signal processing module with an FPGA as the core (main frequency 200 MHz, supporting real-time FFT/IFFT operation, operation delay ≤ 10 ms) is used to realize real-time signal acquisition, frequency domain analysis, dynamic multi-notch filtering and time-domain signal restoration. It also supports real-time display of detection waveforms, automatic storage of detection data, and statistical analysis of extraction success rate and detection accuracy, providing data support for the verification of experimental results.

3.2. Experimental Test and Analysis

This experiment was carried out around four core dimensions: signal extraction effect, anti-interference performance, underground cable detection effect, and comparative verification of different methods. At the same time, a blank control group (10 kV XLPE cable intermediate joint test piece without defects) was set to eliminate misjudgment caused by background noise and equipment self-interference; each group of experiments was repeated multiple times, and the average value was taken as the final result to ensure the repeatability and statistical significance of the experimental data. The specific experimental design, process and result analysis are as follows:

3.2.1. Signal Extraction Effect Experiment

Experimental purpose: To verify the basic extraction ability of the dual-position detection antenna combined with the dynamic multi-notch method for weak partial discharge signals, compare the detection effect of the ordinary broadband UHF antenna, and clarify the signal recognition ability of the proposed method in an environment without additional interference.

Experimental conditions: No artificial electromagnetic interference was applied indoors, but there were communication signals from mobile base stations in the distant space. A 10 kV rated AC high voltage was applied to the three types of defect test pieces, and the ordinary broadband UHF antenna and the dual-position detection antenna were used to perform synchronous non-contact detection on the test pieces with a detection distance of 1 m for both; 20 groups of tests were carried out for each type of defect test piece, the detection waveforms of the two sets of equipment were recorded, and the partial discharge signal extraction success rate was counted (the criterion for successful extraction: the partial discharge pulse profile can be clearly identified in the time-domain waveform without complete noise submergence).

Experimental results: The time-domain waveform of the signal received by the ordinary broadband antenna is shown in Figure 10a. There is some random background electromagnetic noise in the waveform with a relatively weak amplitude, which is a spatial interference signal with weak interference and can be identified. The signal extraction success rates of the three defects of metal debris, insulation scratches and conductor burrs are all above 75%; the detection waveform after the dual-position detection antenna filters out the background noise through the dynamic multi-notch method is shown in Figure 10b. The noise in the time-domain waveform is effectively suppressed, and the partial discharge pulse signal profile is clear and distinguishable. Among them, the signal extraction success rate of metal debris defects is 96.0%, that of insulation scratch defects is 95.0%, and that of conductor burr defects is 93.0%, with an average extraction success rate of 94.7%, realizing the effective extraction of weak partial discharge signals.

Result analysis: The ordinary broadband UHF antenna receives signals in the full frequency band without targeted filtering ability, and the spatial background noise and partial discharge signals will be received at the same time; the dual-position detection antenna realizes the separate collection of interference signals and mixed signals through the upper and lower dual directional antennas and performs accurate frequency domain filtering of background noise combined with the dynamic multi-notch method, only retaining the main profile of the partial discharge pulse signal, thus effectively extracting weak signals.

3.2.2. Anti-Interference Performance Experiment

Experimental purpose: To verify the anti-interference ability of the proposed dynamic multi-notch method in single-interference and multi-interference source superposition environments, test the influence of different interference amplitudes on the detection effect, and clarify the adaptability of the method in complex electromagnetic interference environments.

Experimental conditions: The conductor burr defect test piece (with the weakest discharge signal and most susceptible to interference) was selected, and a 10 kV rated AC high voltage was applied to carry out the single-interference source test and multi-interference source superposition test, respectively:

♦

Single-interference source test: The mobile communication signal simulator and the industrial equipment radiation simulator were used as independent interference sources. The interference amplitude was gradually increased, 20 groups of tests were carried out for each amplitude level, and the detection accuracy was recorded.

♦

Multi-interference source superposition test: The mobile communication signal simulator and the industrial equipment radiation simulator were turned on at the same time to simulate strong composite electromagnetic interference, 30 groups of tests were carried out, and the detection accuracy and signal extraction effect were recorded.

In the experiment, the dual-position detection antenna was used for detection, with the ordinary broadband UHF antenna as the comparison and a detection distance of 1 m; the criterion for detection accuracy was that the partial discharge pulse signal could be accurately identified without missing detection or misjudgment.

Experimental results:

♦

Single-interference source test: When the interference amplitude is small, the detection accuracy of the ordinary broadband antenna is about 60~70%. With the increase in interference amplitude, the detection accuracy decreases rapidly. When the interference amplitude reaches −5 dBm, the detection accuracy drops to below 40%. Its detection waveform is shown in Figure 11a, where the baseband interference signal completely covers the partial discharge pulse and cannot be identified; while the dual-position detection antenna maintains a detection accuracy of 90.0~95.0%, all while under single interference, and the waveform is shown in Figure 11b, where the noise is effectively filtered out, the partial discharge pulse signal is clear and distinguishable, and only the amplitude is slightly reduced. Experimental verification shows that the proposed algorithm can maintain a detection accuracy of more than 85% when the amplitude of mobile communication interference reaches 2 dBm and the amplitude of industrial equipment radiation interference reaches 5 dBm. When the interference amplitude exceeds the above ranges (mobile communication > 2 dBm, industrial radiation > 5 dBm), the detection accuracy will drop below 80%, which is lower than the allowable threshold for engineering applications. Accordingly, the maximum tolerable interference levels of the algorithm are defined as follows: mobile communication interference ≤ 2 dBm and industrial equipment radiation interference ≤ 5 dBm.

♦

Multi-interference source superposition test: After the superposition of mobile communication signals and industrial equipment radiation signals, the interference frequency band covers 400~1800 MHz with a more complex spectrum. The detection waveform of the ordinary broadband antenna is shown in Figure 12a, where the noise intensity is greatly increased, the partial discharge signal is completely submerged, and the detection accuracy is only about 25%. The dual-position detection antenna realizes multi-notch filtering by dynamically tracking the multi-band interference spectrum and adjusting the notch frequency bands in real time. The detection waveform is shown in Figure 12b, which can still clearly restore the partial discharge pulse signal profile, and the detection accuracy remains 89.5% under strong composite interference.

Figure 11. Single-interference source test.

Figure 11. Single-interference source test.

Figure 12. Multi-interference source superposition test.

Figure 12. Multi-interference source superposition test.

Result analysis: The proposed dynamic multi-notch method can dynamically adjust the notch frequency bands according to the changes in interference frequency and amplitude by collecting pure spatial interference signals in real time and performing dynamic tracking and analysis of the interference spectrum, which can adapt to complex electromagnetic environments with single-band and multi-band composite interference and realize accurate interference filtering; the ordinary broadband UHF antenna has no dynamic filtering ability and can only receive signals in the full frequency band. With the increase in interference amplitude or superposition of interference sources, the noise frequency band coverage is wider, and the partial discharge signal is completely submerged, so the detection effect drops sharply; the experimental results verify the excellent anti-interference performance of the proposed method, which can adapt to the complex and variable on-site electromagnetic environment. The above maximum tolerable interference levels cover the vast majority of electromagnetic interference scenarios encountered in urban underground cable field applications. In practical engineering, the radiated interference from typical industrial frequency converters is approximately −10 to 0 dBm, and that from mobile base stations is about −15 to −5 dBm, both of which are below the upper limits the algorithm can withstand. Therefore, the proposed method can meet the requirements of on-site practical detection.

3.2.3. Underground Cable Detection Effect Experiment

Experimental purpose: To simulate the actual engineering scene of underground cables, verify the detection ability of the proposed method for partial discharge signals of cable intermediate joints with different burial depths under the attenuation effect of soil media, and clarify the engineering practicability of the method.

Experimental conditions:

♦

The three types of defect test pieces of metal debris, insulation scratches and conductor burrs were buried in dry clay with thicknesses of 0.5 m, 2.0 m and 3.0 m, respectively (simulating the typical soil environment of urban power grid underground cables, with soil relative permittivity εr = 8~10 and conductivity σ = 0.01 S/m); a 10 kV rated AC high voltage was applied to the test pieces without artificial electromagnetic interference; the dual-position detection antenna was used for non-contact detection 10 cm above the ground; 30 groups of tests were carried out for each burial depth and each type of defect; the detection waveforms, signal extraction success rate and detection accuracy were recorded; and the amplitude attenuation of partial discharge UHF signals by soil media was also tested.

♦

Antenna Installation Direction Requirements: During on-site installation, the built-in upward directional antenna shall be vertically upward (with a detection angle of 90°) to ensure accurate reception of spatial interference signals. The built-in downward directional antenna shall be vertically downward and aligned directly above the cable joint so as to avoid incomplete reception of partial discharge signals caused by directional deviation.

♦

Mechanical Enclosure and Environmental Adaptability: A waterproof, dustproof and corrosion-resistant sealed enclosure with protection class IP65 is adopted, which is suitable for humid and dusty field environments such as underground cable trenches, tunnels and shafts. The enclosure is made of lightweight aluminum alloy for easy on-site transportation and installation.

♦

On-site Positioning: In cable trenches and tunnels, the antenna shall be installed 20–50 cm directly above the cable joint, away from interference from other power equipment (e.g., switch cabinets, grounding bodies). In shafts, suspension installation can be used to ensure vertical alignment between the antenna and the joint, thus reducing signal obstruction.

Experimental results: The soil layer has an obvious attenuation effect on the partial-discharge UHF electromagnetic wave, and the signal amplitude decreases exponentially with the increase in soil burial depth, but the proposed dynamic multi-notch method still maintains high detection performance. The detection results under different burial depths are as follows:

♦

Burial depth of 0.5 m: The detection accuracies of metal debris, insulation scratches and conductor burr defects are 97.0%, 96.0% and 95.0%, respectively, with an average accuracy of 96.0%, and the amplitude of the partial discharge signal decreases by about 5% after soil attenuation;

♦

Burial depth of 2.0 m: The detection accuracies of the three defects are 95.0%, 93.0% and 92.0%, respectively, with an average accuracy of 93.3%, and the signal amplitude decreases by about 15%;

♦

Burial depth of 3.0 m: The detection accuracies of the three defects are 92.0%, 90.0% and 89.0%, respectively, with an average accuracy of 90.3%, and the signal amplitude decreases by about 28%. However, after processing by the dynamic multi-notch method, the main profile of the pulse signal can still be effectively extracted without missing detection.

Result analysis: Partial discharge electromagnetic signals are ultra-wideband signals covering 0–3 GHz. In engineering detection, to reduce the sampling rate and achieve efficient acquisition, the envelope detection method is usually adopted in the back-end acquisition stage to convert the original radio-frequency signals into low-frequency pulse signals. In actual cable trench or soil environments, although the ultra-high-frequency components are prone to attenuation and local distortion in the time-domain waveform, medium- and low-frequency signals exhibit stronger medium penetration capability and can effectively penetrate the soil while retaining their characteristic integrity. Therefore, as long as signals in partial frequency bands can propagate effectively, the local time-domain waveform distortion caused by the soil has a negligible influence on the characteristics and waveform recognition of the low-frequency pulse signals after envelope detection and will not result in unidentifiable signals.

3.2.4. Comparative Verification Experiment of Different Detection Methods

Experimental purpose: To compare the proposed dynamic multi-notch method with the traditional UHF detection method and fixed notch method commonly used in engineering, quantitatively evaluate the detection performance of the three methods, and clarify the superiority of the proposed method.

Experimental conditions: Test cable: A 10 kV XLPE cable with a cross-sectional area of 70 mm². Defect sizes: Metal debris: 1 mm, 2 mm, 5 mm; insulation scratches: length 2–5 mm, depth 0.5 mm; conductor burrs: height 1–3 mm. Applied voltage: 10 kV alternating current (AC). Soil environment: Relative permittivity εr = 8~10, conductivity σ = 0.01 S/m. A strong composite electromagnetic interference environment (superposition of mobile communication signals and industrial equipment radiation signals) was built, the three types of defect test pieces were buried in 1.0 m thick soil (simulating the actual engineering scene), and a 10 kV rated AC high voltage was applied; the dynamic multi-notch method, traditional UHF detection method and fixed notch method were used to perform non-contact detection on the test pieces with a detection distance of 1 m; 50 groups of tests were carried out for each defect with each method, the two core indicators of average signal extraction success rate and average detection accuracy were counted, the average value was taken as the detection result, and the performance differences in the three methods were compared.

Experimental results: The performance comparison of the three detection methods is shown in Table 1. The proposed dynamic multi-notch method is significantly superior to the traditional UHF detection method and the fixed notch method in core indicators. The specific results are as follows:

♦

Dynamic multi-notch method: 2048-point FFT/IFFT transformation is adopted, with a frequency resolution of approximately 2.93 MHz and an effective analysis frequency band of 300 MHz–3 GHz. The step bandwidth of notch filtering is set to 10 MHz, and the interference threshold Q is adaptively set to 3–5 times the background noise. FFT and IFFT are implemented in FPGA with a processing delay of less than 10 ms to ensure real-time performance. The average detection accuracy is ≥92.3% in a normal environment without strong interference and remains 89.5% under strong composite interference.

♦

Fixed notch method: Fixed notch filtering is set at three center frequencies of 500 MHz, 900 MHz, and 1800 MHz. The notch bandwidth of each frequency band is 50 MHz, corresponding to the actual filtering intervals: 500 MHz ± 50 MHz (450 MHz–550 MHz), 900 MHz ± 50 MHz (850 MHz–950 MHz), and 1800 MHz ± 50 MHz (1750 MHz–1850 MHz). The analysis frequency band ranges from 300 MHz to 3 GHz. Three fixed-notch frequency bands of 500 MHz, 900 MHz and 1800 MHz are preset, which cannot track the real-time changes in the on-site interference spectrum and have no filtering effect on interference outside the preset frequency bands, with an average signal extraction success rate of 72.5% and an average detection accuracy of 68.3% in a normal environment.

♦

Traditional UHF detection method: A conventional broadband UHF antenna is adopted, with an operating frequency band of 300 MHz–3 GHz and an antenna gain of 6 dBi. No interference suppression or filtering algorithm is applied. Without any interference suppression measures, the weak partial discharge signal is completely submerged by strong electromagnetic interference, with an average signal extraction success rate of 48.2% and an average detection accuracy of only 45.8% in a normal environment.

♦

Wavelet denoising method: The db4 wavelet basis (suitable for partial discharge pulse signals) is used with five decomposition levels, a soft threshold function, and the sqtwolog threshold rule. Under strong interference conditions, the detection accuracy of the wavelet denoising method is 75.2%, mainly because it is difficult to track dynamically changing interference frequency bands in real time.

♦

Machine learning classification method: The SVM classifier is adopted with an RBF kernel function, penalty coefficient C = 10, and gamma = 0.05. Under strong interference conditions, the detection accuracy of the machine learning classification method is 78.6%, mainly because it requires a large number of samples for training and has poor adaptability to complex on-site interference.

Table 1. Performance comparison of the three detection methods.

Detection Method	Average Signal Extraction Success Rate	Average Detection Accuracy in Normal Environment	Detection Accuracy Under Strong Interference
Dynamic multi-notch	94.7%	≥92.3%	89.5%
Fixed notch	72.5%	68.3%	52.1%
Traditional UHF	48.2%	45.8%	21.3%
Wavelet denoising method	82.7%	80.6%	75.2%
Machine learning classification method	85.4%	83.1%	78.6%

Table 1. Performance comparison of the three detection methods.

Detection Method	Average Signal Extraction Success Rate	Average Detection Accuracy in Normal Environment	Detection Accuracy Under Strong Interference
Dynamic multi-notch	94.7%	≥92.3%	89.5%
Fixed notch	72.5%	68.3%	52.1%
Traditional UHF	48.2%	45.8%	21.3%
Wavelet denoising method	82.7%	80.6%	75.2%
Machine learning classification method	85.4%	83.1%	78.6%

Result analysis: The traditional UHF detection method only realizes the full-band reception of UHF signals without interference suppression ability and completely fails in the complex on-site electromagnetic interference environment, with the worst detection performance; the fixed notch method can only filter out interference in the preset frequency bands and cannot adapt to the random variation characteristics of on-site interference frequency and amplitude. When the interference frequency band exceeds the preset range, the filtering effect drops sharply with limited detection performance; the dynamic multi-notch method realizes the accurate and real-time suppression of random and variable electromagnetic interference by synchronously collecting interference signals and mixed signals through the dual-position detection antenna and dynamically adjusting the notch frequency bands based on the real-time interference spectrum. At the same time, combined with the characteristics of the partial discharge ultra-wideband signal, the main profile of the pulse signal is retained, so the signal extraction success rate and detection accuracy are greatly improved with the optimal detection performance.

4. Conclusions

Aiming at solving the detection problems caused by weak partial discharge signals of underground cable joints and random and variable spatial electromagnetic wave interference, this paper proposes a non-contact partial discharge signal detection technology based on the dynamic multi-notch method, designs a dual-position detection antenna and a signal processing module with FPGA as the core, and realizes the synchronous collection of interference signals and mixed signals, real-time interference spectrum analysis, dynamic multi-notch filtering and partial discharge pulse signal restoration. Through multiple groups of experimental tests and analysis in multiple dimensions, combined with comparative verification with traditional detection methods, the following experimental conclusions are drawn:

• The average extraction success rate of the proposed dynamic multi-notch method for weak partial discharge signals reaches 94.7%, and the detection accuracy is ≥92.3% in a normal environment without strong interference, which can effectively extract the weak partial discharge signals of underground cable intermediate joints and solve the problem of poor signal extraction ability of traditional methods.
• This method has excellent anti-interference performance, and the detection accuracy remains 89.5% in a strong composite electromagnetic interference environment (mobile communication + industrial equipment radiation), which is far better than the traditional UHF detection method (45.8%) and the fixed notch method (68.3%) and can adapt to the complex and variable on-site electromagnetic environment.
• This method is suitable for the actual underground engineering scene and can maintain an average detection accuracy of more than 90% within the soil burial depth range of 0.5~3.0 m (a common burial depth for urban underground cables). Even if the signal is attenuated by 80% by the soil, the main profile of the partial discharge pulse signal can still be effectively extracted, meeting the requirements of actual engineering detection.
• This method has excellent anti-misjudgment ability, with a misjudgment rate of only 2.0% under strong electromagnetic interference, which is far lower than the engineering allowable threshold. At the same time, it realizes real-time dynamic suppression of interference with high reliability of detection results, providing a reliable technical solution for the early warning of insulation defects in underground cable intermediate joints.