A method and system for on-line monitoring of insulation faults of an intrinsically safe cable for mining

Abstract

The application provides a kind of mine intrinsic safety type cable insulation fault on-line monitoring method and system, it is related to coal mine underground high-voltage cable insulation fault monitoring technical field, the method comprises: by mine intrinsic safety type high-frequency pulse signal sensor coupling cable ground wire on partial discharge current pulse signal, obtain high-frequency pulse signal;Analog conditioning is carried out to high-frequency pulse signal, and the analog signal after conditioning is obtained;The analog conditioning includes analog filter, preamplifier and impedance matching;Anti-aliasing filtering is carried out to the analog signal after conditioning, and the filtered signal is obtained;The analog signal in target band is extracted by selecting target band to the filtered signal;The analog signal in target band is converted into digital signal by A / D analog-digital conversion.The application identifies fault type, source and severity by cloud collaboration and multi-technology fusion, and real-time early warning visualization helps operation and maintenance to respond quickly, reduces accident risk.

Description

Technical Field

[0001] This invention relates to the field of insulation fault monitoring technology for underground high-voltage cables in coal mines, and in particular to an online monitoring method and system for insulation faults in intrinsically safe mining cables. Background Technology

[0002] High-voltage cables in coal mines are the lifeline of the power supply system. Insulation failures can lead to power outages (losses of hundreds of thousands of yuan per shutdown) and, in severe cases, can cause electrical sparks, which can trigger gas explosions. Therefore, insulation failure monitoring is a core aspect of mine power supply safety.

[0003] Current mainstream offline testing methods involving periodic power outages (such as megohmmeter testing and power frequency withstand voltage testing) have significant limitations: on the one hand, offline testing requires interrupting production, takes a long time per test, and has a high rate of human error, making it unsuitable for unmanned or minimally staffed mining operations; on the other hand, power frequency withstand voltage testing may accelerate insulation aging and break down minor defects, creating potential hazards.

[0004] As underground power supply systems become increasingly complex (with fully mechanized mining faces using over 2km of cables containing 8-12 joints), existing technologies exhibit significant shortcomings across the entire process of signal acquisition, signal processing, and fault diagnosis. Clamp-on sensors are susceptible to electromagnetic interference, leading to signal distortion; non-intrinsically safe types pose a risk of gas explosion (as demonstrated by a coal mine that damaged its roadways); intrinsically safe types suffer from 30% lower sensitivity due to current and voltage limitations, potentially making it difficult to acquire early, weak partial discharge signals. Current systems rely solely on analog filtering to remove fixed noise, which may not be sufficient to handle random interference, and lack time-frequency joint analysis capabilities, making it difficult to distinguish between suspended / internal / surface discharges. Current fault diagnosis heavily relies on manual comparison of phase-resolved partial discharge spectra (PRPD spectra), resulting in significant subjective errors and lengthy diagnostic times. More critically, traditional methods may fail to accurately pinpoint the source of discharge, making it difficult to differentiate between faults originating from cable joints, insulation internals, or terminals, necessitating a comprehensive inspection during maintenance, which is inefficient. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide an online monitoring method and system for insulation faults in intrinsically safe cables used in mining. By integrating cloud collaboration and multiple technologies, the system can identify the type, source and severity of faults, provide real-time early warning and visualization to help maintenance personnel respond quickly and reduce the risk of accidents.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] A first aspect is a method for online monitoring of insulation faults in intrinsically safe cables used in mining, the method comprising:

[0008] Step 1: Obtain the high-frequency pulse signal by coupling the partial discharge current pulse signal on the grounding wire of the cable using a mining intrinsically safe high-frequency pulse signal sensor;

[0009] Step 2: Perform analog conditioning on the high-frequency pulse signal to obtain a conditioned analog signal; the analog conditioning includes analog filtering, preamplification and impedance matching; perform anti-aliasing filtering on the conditioned analog signal to obtain a filtered signal; select the target frequency band of the filtered signal to extract the analog signal within the target frequency band; convert the analog signal within the target frequency band into a digital signal through A / D analog-to-digital conversion;

[0010] Step 3: Perform digital filtering and noise suppression on the digital signal, identify and eliminate narrowband interference and random noise, enhance the periodic partial discharge pulse signal, and extract signal features, including amplitude, phase, discharge quantity and occurrence frequency.

[0011] Step 4: Transmit the signal features to the cloud management system, perform signal denoising and time-frequency analysis through wavelet transform and short-time Fourier transform, and generate phase, amplitude, and frequency PRPD spectra.

[0012] Step 5: Compare the PRPD spectrum with the pre-stored typical defect partial discharge feature model library for mining cables, and use machine learning models to determine the type, source and severity of partial discharge, and obtain diagnostic results that include partial discharge attributes and severity.

[0013] Step 6: Compare the insulation severity in the diagnostic results with the preset warning threshold and alarm threshold to generate warning or alarm information, and send the warning or alarm information to the downhole monitoring host for real-time display and status indication.

[0014] Secondly, an online monitoring system for insulation faults in intrinsically safe cables used in mining includes:

[0015] The acquisition module is used to acquire high-frequency pulse signals by coupling the partial discharge current pulse signal on the grounding wire of the cable with a mining intrinsically safe high-frequency pulse signal sensor;

[0016] A digital signal module is used to perform analog conditioning on high-frequency pulse signals to obtain conditioned analog signals. The analog conditioning includes analog filtering, preamplification, and impedance matching. The conditioned analog signals are then subjected to anti-aliasing filtering to obtain filtered signals. The filtered signals are then subjected to target frequency band selection to extract analog signals within the target frequency band. The analog signals within the target frequency band are then converted into digital signals via an A / D converter.

[0017] The extraction module is used to perform digital filtering and noise suppression on digital signals, identify and eliminate narrowband interference and random noise, enhance periodic partial discharge pulse signals, and extract signal features, including amplitude, phase, discharge quantity and occurrence frequency.

[0018] The analysis module transmits signal features to the cloud management system, performs signal denoising and time-frequency analysis through wavelet transform and short-time Fourier transform, and generates phase, amplitude, and frequency PRPD spectra. The PRPD spectra are compared with a pre-stored library of typical defects in mining cables, and a machine learning model is used to determine the type, source, and severity of partial discharge, resulting in a diagnostic result that includes the attributes and severity of partial discharge.

[0019] The execution module compares the insulation severity in the diagnostic results with preset warning thresholds and alarm thresholds, generates warning or alarm information, and sends the warning or alarm information to the downhole monitoring host for real-time display and status indication.

[0020] Thirdly, a computing device includes:

[0021] One or more processors;

[0022] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0023] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0024] The above-described solution of the present invention has at least the following beneficial effects:

[0025] The sensor effectively overcomes the problems of magnetic saturation and output signal distortion of traditional open-close clamp sensors. The intrinsically safe protection circuit uses a series design of Zener diode and N-MOS field-effect transistor, with parameters that ensure the breakdown voltage is greater than the sum of the turn-on voltage and the transistor voltage drop. This achieves overvoltage protection to prevent explosion without compromising the integrity of the high-frequency pulse signal. Out-of-band noise is first filtered out and signal impedance is matched. Then, anti-aliasing filtering is used to prevent subsequent A / D sampling distortion. The frequency band of the mining partial discharge signal is specifically selected to reduce invalid signal interference. Furthermore, stable narrowband interference and random noise are eliminated. Periodic partial discharge pulses are enhanced through multi-cycle time-domain averaging, and dynamic thresholding is used to adapt to noise fluctuations, screening for effective partial discharge pulses and providing high signal-to-noise ratio data support for subsequent diagnostics. Leveraging cloud computing power, deep signal processing is achieved: wavelet transform accurately filters out broadband interference, short-time Fourier transform captures the dynamic frequency changes of partial discharge signals over time, and the generated PRPD spectrum intuitively presents fault characteristics; furthermore, through model library comparison and machine learning diagnosis, it can not only accurately identify fault types, but also locate the source of discharge (insulation, joints, etc.) and the severity of insulation, solving the problems of misjudgment, missed judgment, and knowing only the fault but not the source in traditional diagnosis; by comparing the quantified insulation severity with preset thresholds, early warning or alarm information is automatically generated; the downhole monitoring host displays details on an LCD screen and flashes LED lights (yellow warning, red alarm), ensuring that maintenance personnel can quickly obtain fault information and handle it in a timely manner in the complex downhole environment. Attached Figure Description

[0026] Figure 1 This is a schematic flowchart of an online monitoring method for insulation faults in intrinsically safe cables used in mining, provided by an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of an online monitoring system for insulation faults in intrinsically safe cables used in mining, provided by an embodiment of the present invention. Detailed Implementation

[0028] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0029] like Figure 1 As shown in the figure, an embodiment of the present invention proposes an online monitoring method for insulation faults in intrinsically safe cables used in mining, the method comprising the following steps:

[0030] A method for online monitoring of insulation faults in intrinsically safe mining cables, the method comprising:

[0031] Step 1: Obtain the high-frequency pulse signal by coupling the partial discharge current pulse signal on the grounding wire of the cable using a mining intrinsically safe high-frequency pulse signal sensor;

[0032] Step 2: Perform analog conditioning on the high-frequency pulse signal to obtain a conditioned analog signal; the analog conditioning includes analog filtering, preamplification and impedance matching; perform anti-aliasing filtering on the conditioned analog signal to obtain a filtered signal; select the target frequency band of the filtered signal to extract the analog signal within the target frequency band; convert the analog signal within the target frequency band into a digital signal through A / D analog-to-digital conversion;

[0033] Step 3: Perform digital filtering and noise suppression on the digital signal, identify and eliminate narrowband interference and random noise, enhance the periodic partial discharge pulse signal, and extract signal features, including amplitude, phase, discharge quantity and occurrence frequency.

[0034] Step 4: Transmit the signal features to the cloud management system, perform signal denoising and time-frequency analysis through wavelet transform and short-time Fourier transform, and generate phase, amplitude, and frequency PRPD spectra.

[0035] Step 5: Compare the PRPD spectrum with the pre-stored typical defect partial discharge feature model library for mining cables, and use machine learning models to determine the type, source and severity of partial discharge, and obtain diagnostic results that include partial discharge attributes and severity.

[0036] Step 6: Compare the insulation severity in the diagnostic results with the preset warning threshold and alarm threshold to generate warning or alarm information, and send the warning or alarm information to the downhole monitoring host for real-time display and status indication.

[0037] In this embodiment of the invention, the partial discharge current pulse coupled to the grounding wire of the cable by the intrinsically safe high-frequency pulse signal sensor for mining can adapt to the explosion-proof requirements of the explosive gas environment in coal mines, avoid the risk of equipment failure and ensure the safety of underground operations; analog conditioning can amplify weak partial discharge signals, filter out some low-frequency interference, reduce signal transmission loss and improve the signal-to-noise ratio; anti-aliasing filtering can avoid aliasing errors caused by high-frequency signal folding during subsequent A / D conversion, ensuring conversion accuracy; target frequency band selection can focus on the effective frequency range of the partial discharge signal, eliminate irrelevant frequency band interference, and further purify the signal; A / D analog-to-digital conversion converts analog signals into digital signals, providing a suitable format for subsequent flexible digital signal processing; digital filtering and noise suppression can specifically eliminate narrowband interference and random noise, effectively enhancing the identification of periodic partial discharge pulse signals; the extracted amplitude, phase, discharge quantity, and occurrence... Signal characteristics such as frequency are the core basis for subsequent partial discharge diagnosis; wavelet transform has a better denoising effect than traditional filtering, and can remove residual noise while preserving the details of the partial discharge pulse; short-time Fourier transform can clearly present the time-frequency distribution characteristics of the partial discharge signal, helping to analyze the signal's change pattern over time; the generated phase, amplitude, and frequency PRPD spectrum can intuitively reflect the statistical regularity of the partial discharge signal; comparison with the partial discharge characteristic model library of typical defects in mining cables can rely on historical defect data in coal mine scenarios to avoid the problem of insufficient adaptability of general models; machine learning models can utilize mature diagnostic experience and improve diagnostic accuracy through data iteration; by comparing with preset warning thresholds and alarm thresholds, early hidden danger warnings can be achieved, and warning or alarm information is sent to the underground monitoring host for real-time display and status indication, allowing on-site personnel underground to know the cable insulation status immediately and assisting in rapid response.

[0038] In a preferred embodiment of the present invention, step 1 above includes:

[0039] Step 11: Obtain the initial induced signal by the partial discharge current pulse on the grounding wire of the coupling cable. Specifically, this includes: First, conducting a survey of the operating conditions of the target monitoring cable in the coal mine to determine the rated voltage level of the cable, the rated grounding current range during normal operation, and the typical amplitude range of the partial discharge current pulse in historical partial discharge fault data (generally tens to hundreds of microamps), and the main frequency range of the partial discharge signal; at the same time, it is also necessary to consider the frequency of interference signals in the complex electromagnetic environment underground, such as the interference generated by frequency converters and motor starting, which is mostly concentrated in the range of tens to hundreds of kilohertz.

[0040] Secondly, based on the principle of electromagnetic induction and the above-mentioned operating parameters, the material selection and structural parameters are determined. In material selection, high permeability and low loss coefficient manganese-zinc ferrite materials are preferred. In calculating structural parameters, the amplitude of the partial discharge current pulse coupled to the target is first considered, combined with the expected number of turns of the magnetic core. Considering the convenience of downhole installation and coupling efficiency, the magnetic core is usually designed as a single-turn coupling structure. By comparing the magnetic saturation critical point of a closed-loop clamp-on magnetic core, it is found that closed-loop magnetic cores are prone to magnetic saturation when the partial discharge current exceeds 500μA, leading to signal distortion. The effective cross-sectional area of ​​the magnetic core is calculated: for example, when the maximum amplitude of the target partial discharge current pulse is 300μA, and the saturation magnetic flux density of the magnetic core material (using B... When the voltage (s) is 0.5T and the signal frequency is 1MHz, the effective cross-sectional area (S) of the magnetic core must meet the following condition: the magnetic flux density (B) generated by the partial discharge current must conform to the formula B = (the internal permeability μi of the magnetic core multiplied by the number of coil turns N, then multiplied by the partial discharge current I) divided by the magnetic path length l of the magnetic core. Furthermore, the magnetic flux density B calculated using this formula must be less than or equal to 0.8 times the saturation magnetic flux density Bs of the magnetic core. Here, μi represents the internal permeability of the core material, N represents the number of coil turns, I represents the partial discharge current, and l represents the magnetic path length of the magnetic core. Calculations show that under these conditions, the effective cross-sectional area of ​​the magnetic core needs to be no less than 0.8 cm². 2 The magnetic circuit length of the magnetic core needs to be controlled within 5cm. By meeting these two requirements, magnetic saturation of the magnetic core can be avoided. At the same time, the arc angle of the magnetic core is strictly controlled to 90°, that is, a quarter circle. Through machining precision control, such as the arc error being less than or equal to ±1°, the coupling gap between the magnetic core and the cable grounding wire is ensured to be uniform, with the gap being less than or equal to 2mm, thereby reducing the loss of coupling efficiency.

[0041] Finally, the coupling effect of the magnetic core was verified through prototype testing to determine the final initial induced signal. Specifically, the fabricated magnetic core was installed on the grounding wire of a normally operating underground cable, and the coupled signal was acquired using an oscilloscope to measure its amplitude and signal-to-noise ratio (SNR). Under a typical partial discharge current of 100μA, the initial induced signal amplitude should reach 0.5V to 2V to ensure effective identification by subsequent circuits, and the SNR should be no less than 20dB to distinguish between partial discharge and interference signals. If the measured amplitude is insufficient, the effective cross-sectional area of ​​the magnetic core needs to be adjusted, such as increasing it to 1.0cm². 2 Alternatively, the length of the magnetic circuit can be reduced, such as shortening it to 4cm, until the initial induction signal obtained by coupling meets the above-mentioned indicators. At this point, the signal is the initial induction signal.

[0042] Step 12: The initial induced signal is input to the intrinsically safe protection circuit. The intrinsically safe protection circuit adopts a dual-parallel structure, with each path consisting of a Zener diode and an N-MOS field-effect transistor connected in series. Based on the characteristics of the initial induced signal, the breakdown voltage of the Zener diode, the turn-on voltage of the N-MOS field-effect transistor, and the voltage drop parameter of the Zener diode in the intrinsically safe protection circuit are determined, ensuring that the breakdown voltage is greater than the sum of the turn-on voltage and the voltage drop. The intrinsically safe protection circuit configured with the voltage drop parameter is used to perform overvoltage protection and integrity maintenance processing on the initial induced signal to obtain the protected induced signal. Specifically, this includes: first analyzing the electrical characteristics of the initial induced signal. The initial induced signal obtained in step 11 is continuously monitored using an oscilloscope for at least 24 hours, covering different operating conditions such as cable load, no load, and light load. The normal peak voltage range, maximum overvoltage peak value, and signal frequency characteristics of the signal are recorded. The maximum overvoltage peak value is caused by power grid fluctuations or transient interference. For example, the measured normal peak voltage range of the initial induced signal is 1V to 5V. Under the condition of power grid voltage fluctuation of ±10%, the maximum overvoltage peak value is 10V. The signal frequency is still concentrated in the partial discharge signal frequency range of several hundred kHz to several MHz, and the signal is in the form of high-frequency AC pulse with no DC component.

[0043] Secondly, based on the above signal characteristics, the breakdown voltage VZ of the Zener diode, the turn-on voltage VGS of the N-MOS field-effect transistor, and the voltage drop VF of the Zener diode are determined sequentially. For the breakdown voltage VZ of the Zener diode, its value must be higher than the maximum overvoltage peak value of the initial induced signal to avoid normal signals triggering the Zener diode breakdown and causing signal attenuation; it must also be lower than the maximum withstand voltage of the subsequent high-frequency pulse signal conditioning board, which is usually 15V, to ensure effective clamping during overvoltage. Combined with the measured maximum overvoltage peak value of 10V, a Zener diode with VZ=12V is selected, such as the 1N4742 model, whose rated breakdown voltage is 12V with a deviation range of ±5%. It has been verified that 12V is higher than the maximum overvoltage peak value of 10V and lower than the conditioning board withstand voltage of 15V, which can simultaneously meet the requirements of overvoltage protection and normal signals not triggering breakdown.

[0044] For the turn-on voltage VGS of an N-MOS field-effect transistor, an enhancement-type N-MOS transistor that can reliably conduct within the normal peak voltage range of the initial induced signal (1V to 5V) and can quickly turn off when the signal is cut off should be selected. The IRLML2502 model N-MOS transistor is selected, with a typical turn-on voltage VGS(th) of 2.5V, ranging from 1.8V to 3.0V. According to calculations, although the lowest value of 1V in the normal peak voltage range of the initial induced signal (1V to 5V) is slightly lower than the minimum value of VGS(th) of 1.8V, considering that the signal is in pulse form, its instantaneous peak value can reach more than 1.8V. Moreover, the field-effect transistor has a slight conduction when VGS is close to VGS(th), which can ensure that the signal can pass smoothly. At the same time, the turn-off resistance of this model of field-effect transistor is greater than 100MΩ when VGS=0V, which can effectively avoid leakage current interference.

[0045] For the Zener diode's voltage drop VF, a silicon Zener diode matching the aforementioned Zener diode (1N4742) is selected. This model of Zener diode has a typical forward voltage drop VF of 0.7V at a rated operating current of 10mA, ranging from 0.6V to 0.8V. Next, the core condition is verified: the Zener diode's breakdown voltage VZ is greater than the sum of the N-MOS field-effect transistor's turn-on voltage VGS and the Zener diode's voltage drop VF. Combining the selected parameters: VZ = 12V, VGS = 2.5V, VF = With a voltage of 0.7V, the calculated VGS + VF = 2.5V + 0.7V = 3.2V. Since 12V is greater than 3.2V, this condition is met, ensuring that the Zener diode does not break down during the normal transmission of the initial induced signal, the N-MOS field-effect transistor remains on, and the signal passes through without amplitude reduction. Only when an overvoltage signal exceeding 12V occurs will the Zener diode break down. At this time, current flows through the N-MOS field-effect transistor, which uses its on-resistance (approximately 8Ω when VGS = 5V) to limit the current and protect the subsequent circuitry.

[0046] Finally, the intrinsically safe protection circuit with configured parameters is connected in parallel in two paths. Each path consists of a 1N4742 Zener diode and an IRLML2502N-MOS transistor connected in series and connected to the transmission path of the initial induced signal. The signal waveforms before and after processing are compared using an oscilloscope: After processing a normal signal such as a pulse with a peak value of 3V and a frequency of 1MHz, the amplitude remains between 2.9V and 3.0V, the waveform has no obvious distortion, and the rise and fall time deviations are less than or equal to 10ns; when an overvoltage signal of 15V is input, the circuit output voltage is clamped between 12V±0.5V, and the circuit can quickly return to normal after the overvoltage disappears, with a recovery time of less than or equal to 1μs. The signal obtained at this time is the protected induced signal.

[0047] Step 13, using the protected induced signal as a high-frequency pulse signal, specifically includes: First, determining the input parameter requirements of the high-frequency pulse signal conditioning board. These parameters are combined with the design goals of the conditioning board, such as analog filtering, preamplification, and other functions, as well as the relevant standards for intrinsically safe mining equipment. The input amplitude range of the conditioning board is 0.5V to 10V, the input signal frequency range is 100kHz to 100MHz, the input signal-to-noise ratio must be greater than or equal to 18dB, and the input impedance is 50Ω.

[0048] Secondly, the parameters of the protected induced signal were measured and compared. The protected induced signal was tested using equipment such as oscilloscope and spectrum analyzer. In terms of amplitude, when the cable has normal partial discharge, such as a partial discharge current of 100μA, the amplitude of the protected induced signal is 0.8V. When the cable has a slight partial discharge defect, such as a partial discharge current of 300μA, the amplitude is 2.5V, both of which are within the input amplitude range of 0.5V to 10V of the conditioning board.

[0049] Then, the attenuation of the protected induced signal after passing through the transmission link is calculated to ensure that the signal still meets the requirements when transmitted to the conditioning board. Based on the cable laying distance in the mine, the distance between the underground monitoring point and the conditioning board is typically 5 to 20 meters. A mining coaxial cable with a characteristic impedance of 50Ω, such as the MSYV-50-3 model, is selected. At a frequency of 1MHz, the attenuation coefficient is 0.3dB / 100m. Calculating with a maximum laying distance of 20 meters, the transmission attenuation = (0.3dB / 100m) × 20m = 0.06dB. Combining this with the initial amplitude of the protected induced signal at 1MHz being 2.5V, the amplitude after transmission = 2.5V × 10. (-0.06 / 20) =2.5V×0.993≈2.48V, which is still within the input amplitude range of the conditioning board, and the attenuation is minimal, so it does not affect the signal integrity.

[0050] Finally, the effectiveness of the protected induced signal was verified through on-site joint debugging. The protected induced signal was connected to the high-frequency pulse signal conditioning board through the aforementioned mining coaxial cable, and the output signal of the conditioning board was observed. After the conditioning board performed analog filtering on the signal, such as 100kHz to 100MHz bandpass filtering and pre-amplification, the signal amplitude was 24.8V. The waveform was consistent with the waveform of the protected induced signal, with no obvious distortion, and it could accurately reflect the partial discharge situation of the cable. When the partial discharge current increased, the signal amplitude increased synchronously. Thus, it can be confirmed that the protected induced signal fully meets the requirements for subsequent processing and can be officially used as a high-frequency pulse signal.

[0051] In this embodiment of the invention, compared to the completely closed magnetic core of a traditional open-close clamp sensor, the magnetic core can structurally break the completely closed state of the magnetic circuit, effectively avoiding the magnetic saturation problem caused by partial discharge current pulses during the operation of mining cables, and reducing the risk of distortion in the initial sensing signal. Simultaneously, the coupling method between the arc-shaped magnetic core and the cable grounding wire is more suitable for the narrow installation space underground, and can stably capture the original characteristics of the partial discharge current pulse, providing true, undistorted initial data for subsequent signal processing. From an explosion-proof safety perspective, the dual-parallel Zener diode and N-MOS field-effect transistor structure can specifically address the needs of explosive gas environments in coal mines: when a transient overvoltage signal occurs, the Zener diode breaks down and the N-MOS transistor limits the current, effectively clamping the overvoltage and preventing damage. The protection of the subsequent circuitry prevents explosion-proofing, meeting the explosion-proof requirements of intrinsically safe mining equipment. From a signal integrity perspective, by configuring the breakdown voltage to be greater than the sum of the turn-on voltage and voltage drop, it ensures that the Zener diode does not operate and the N-MOS transistor conducts stably when the initial induced signal is transmitted normally, solving the amplitude reduction problem of high-frequency pulse signals in traditional dual-Zener parallel circuits. By using the protected induced signal as a high-frequency pulse signal, it can directly output a signal that meets the input requirements of the subsequent analog conditioning stage, satisfying the intrinsically safe electrical safety standards for mining while avoiding invalid data problems caused by overvoltage or distortion. On the other hand, it can achieve seamless connection between signal acquisition, protection, and subsequent processing stages, ensuring that the signal entering the analog conditioning stage is both safe and effective.

[0052] In a preferred embodiment of the present invention, step 2 above includes:

[0053] Step 21 involves performing analog filtering on the high-frequency pulse signal to remove out-of-band noise and obtain a filtered analog signal. Specifically, this includes: firstly, through on-site electromagnetic environment surveys, statistically analyzing the frequency bands of typical underground interference sources. For example, power frequency interference is concentrated at 50Hz and its harmonics at 100Hz, interference generated by frequency converters and motor startup is mostly distributed between 50kHz and 200kHz, while the effective high-frequency pulse frequency band of the partial discharge signal from mining cables, verified by historical fault data and actual measurements, is concentrated between 300kHz and 60MHz. Based on this, it can be determined that the analog filtering needs to achieve the core objective of retaining the 300kHz to 100MHz frequency band and filtering out out-of-band noise below 300kHz and above 100MHz.

[0054] Next, a bandpass filter was selected as the analog filter type, and its key parameters were determined: the low cutoff frequency fL and the high cutoff frequency fH. The low cutoff frequency fL needs to be 200kHz higher than the highest interference frequency band, while being close to the lower limit of the effective frequency band of the partial discharge signal (300kHz) to avoid over-filtering of the effective signal; therefore, fL = 300kHz was chosen. The high cutoff frequency fH needs to be 100MHz lower than the upper limit of the effective frequency band of the partial discharge signal, while avoiding wireless clutter interference above 100MHz; therefore, fH = 100MHz was chosen. Regarding the selection of the filter order, to ensure... The attenuation effect of out-of-band noise needs to be calculated through attenuation measurement: the attenuation of a 200kHz signal (low interference upper limit) is required to be greater than or equal to 20dB, ensuring that the interference amplitude is reduced to less than 1 / 10 of the effective signal. The attenuation of a 100MHz signal (high interference lower limit) is required to be greater than or equal to 25dB. After selection and verification, a 4th-order RC bandpass filter can meet this attenuation requirement. The attenuation of the 4th-order filter can reach 22dB and 26dB respectively in frequency bands 10% away from the cutoff frequency, such as 270kHz and 100MHz, which completely covers the preset attenuation target.

[0055] Finally, the filtering effect was verified through prototype testing to determine the final filtered analog signal. The designed 4th-order bandpass filter was connected to the high-frequency pulse signal transmission link obtained in step 13, and the signals before and after filtering were acquired simultaneously using an oscilloscope and a spectrum analyzer. On the one hand, the time-domain waveform was observed to ensure that the partial discharge pulse waveform in the range of 300kHz to 100MHz had no obvious distortion and that the rise and fall time deviations were less than or equal to 15ns. On the other hand, the attenuation of out-of-band noise was measured using a spectrum analyzer. If the measured attenuation of the 200kHz interference signal was 23dB and the attenuation of the 101MHz interference signal was 27dB, and the signal amplitude in the range of 300kHz to 100MHz had no obvious loss and the attenuation was less than or equal to 1dB, then the output signal was the filtered analog signal that met the requirements. If the attenuation was insufficient, such as only 15dB attenuation of the 200kHz interference, then the filter order needed to be increased to 5th order, and the test needed to be repeated until the attenuation index was met.

[0056] Step 22 involves pre-amplifying and impedance matching the filtered analog signal to obtain a conditioned analog signal. Specifically, this includes: determining the target amplitude range for pre-amplification: based on the range of the subsequent A / D conversion, the weak signal from 0.1V to 0.5V needs to be amplified to 1V to 5V. Therefore, the target amplification factor range for pre-amplification can be calculated: minimum amplification factor A. min =Target minimum amplitude / Original minimum amplitude = 1V / 0.1V = 10 times, Maximum amplification factor A max=Target maximum amplitude / Original maximum amplitude = 5V / 0.5V = 10 times, meaning the amplification factor needs to be stable at 10 times. For amplifier selection, a low-noise operational amplifier such as OPA277 is chosen, with an input noise voltage less than or equal to 10nV / √Hz and a bandwidth greater than or equal to 110MHz, covering the highest effective frequency band of 100MHz for partial discharge signals, thus avoiding high-frequency signal amplification distortion. The OPA277 is configured as a non-inverting amplifier circuit, and its amplification factor formula is 1 + the ratio of the feedback resistor to the current-limiting resistor. To achieve 10 times amplification, a current-limiting resistor R1 = 1kΩ is selected. Therefore, the feedback resistor Rf needs to satisfy Rf / R1 = 9, hence Rf = 9kΩ.

[0057] The core objective of impedance matching is to ensure that the input impedance of the preamplifier circuit is consistent with the output impedance (50Ω) of the bandpass filter in step 21, thus avoiding signal reflection. A transmission line transformer matching network is used here, selecting a transmission line transformer with a 1:1 impedance transformation ratio, such as a BNC interface-specific matching transformer. Its input impedance is designed to be 50Ω to match the filter's output impedance, and its output impedance is designed to be 50Ω to match the preamplifier circuit's input impedance. The reflection coefficient of the matching network is measured using an impedance meter: the reflection coefficient is required to be less than or equal to 0.1, meaning the signal reflection loss is greater than or equal to 20dB. Testing shows that the 1:1 transmission line transformer has a reflection coefficient of 0.08 in the 300kHz to 100MHz frequency band, with a reflection loss of 22dB, meeting the impedance matching requirements.

[0058] The impedance matching network and the 10x preamplifier circuit were connected in series in the signal link. The measured output signals were as follows: the original 0.1V filtered analog signal was amplified to 1.02V, and the original 0.5V filtered analog signal was amplified to 5.05V. The amplitudes were all within the target range of 1V to 5V, and the waveforms were undistorted. The rise time changed from 20ns after filtering to 21ns, and the deviation was negligible. At the same time, the amplitude loss during signal transmission was reduced from 15% before matching to less than 2%. The output signal at this time is the conditioned analog signal that meets the requirements.

[0059] Step 23 involves applying anti-aliasing filtering to the conditioned analog signal. The cutoff frequency for the anti-aliasing filter is determined based on the sampling frequency of the subsequent A / D conversion, resulting in the filtered signal. Specifically, this includes determining the sampling frequency fs for the subsequent A / D conversion. According to the Nyquist sampling theorem, the sampling frequency must be greater than 1.2 times the highest frequency of the signal to avoid aliasing. Combining this with the highest effective frequency of the partial discharge signal (100MHz) determined in Step 21, fs must be greater than 1.2 × 100MHz = 120MHz. Simultaneously, to reserve some redundancy, fs = 120MHz is selected. 120MHz is 1.2 times 100MHz, satisfying both the theorem requirement and the conventional selection range for high-speed A / D chips, such as the ADS8364 chip, whose sampling frequency can be stably maintained at 120MHz.

[0060] Next, the cutoff frequency fc of the anti-aliasing filter is determined. In the industry, the cutoff frequency of anti-aliasing filters is usually taken as 1 / 2.5 to 1 / 2 of the sampling frequency. If 1 / 2 is taken, that is, fc=fs / 2=60MHz, then the effective signal of 100MHz is too close to the cutoff frequency of 120MHz, which can easily lead to the attenuation of the effective signal. If 1 / 2.5 is taken, fc=fs / 2.5=48MHz, then the upper limit of the effective signal of 48MHz and 50MHz is only 2MHz apart, which may filter out some effective partial discharge signals near 100MHz. After compromise calculation, fc=50MHz is selected. This value is equal to the highest effective frequency of the partial discharge signal and also meets the anti-aliasing requirement that fc is less than fs / 2=50MHz. It can ensure that effective signals of 100MHz and below can pass normally, while interference signals above 100MHz are filtered out.

[0061] In filter selection, a low-pass filter was used, and the order needed to be verified by attenuation: the attenuation of a 120MHz signal was required to be greater than or equal to 40dB. After testing, the attenuation of a 5th-order Butterworth low-pass filter in the 6MHz band reached 45dB, which fully met the requirements. Finally, the low-pass filter was connected to the conditioned analog signal link. The measured attenuation of the 100MHz effective signal was only 0.5dB, with no significant loss, while the attenuation of the 6MHz interference signal was 46dB. The output signal at this time is the filtered signal.

[0062] Step 24: Perform a Fast Fourier Transform on the filtered signal to convert the time-domain signal into a frequency-domain signal to obtain its spectrum. On the spectrum, extract features from the spectral envelope. Identify convex points in the signal energy distribution by constructing the convex hull shape of the spectral amplitude, and select the corresponding frequencies as candidate points for the frequency band boundaries. Based on the distribution of the candidate points and the prior frequency band characteristics of the partial discharge signal of the mining cable, finally determine the upper and lower boundary frequencies of the target frequency band, thereby setting the target frequency band range. According to the target frequency band range, use a bandpass filter to extract the analog signal within the target frequency band from the filtered signal. The system includes: First, receiving the filtered signal, and then performing a Fast Fourier Transform (FFT) to convert the time-domain signal to the frequency domain for frequency distribution analysis. Considering the highest effective frequency of 100MHz for underground partial discharge signals in mines and the real-time monitoring requirements, 1024 points are selected as the number of FFT operation points. Combined with a 12MHz A / D sampling frequency, this system can clearly distinguish between the effective partial discharge signal and the inverter interference signal from 50kHz to 200kHz, while keeping the operation time within 50μs to meet real-time requirements. At the same time, a Hanning window is used to suppress spectral leakage, resulting in a sidelobe attenuation of 31dB, preventing the sidelobe of the interference signal from masking the main lobe of the partial discharge signal.

[0063] After obtaining the frequency domain spectrum through FFT transformation, the amplitude values ​​of the spectrum obtained by FFT are first sampled at equal intervals from low to high frequency, with the sampling interval set to 1kHz. Based on a frequency resolution of 11.7kHz, the 1kHz sampling interval ensures that the sampling density contains 11 to 12 sampling points within each frequency resolution interval, without causing data redundancy due to too many sampling points. After sampling, several sets of frequency and amplitude data pairs are obtained. Subsequently, the Graham scan algorithm is used to construct the convex hull shape of these sampling points. This algorithm first selects the sampling point with the lowest frequency as the starting point, and then sorts the remaining sampling points according to the polar angle of the line connecting them to the starting point from small to large, gradually connecting each sampling point to form a convex polygon. Finally, the vertices of this convex polygon are the convex points of the spectrum amplitude distribution. These convex points correspond to the frequency positions where the signal energy is relatively concentrated, which is the key basis for judging the frequency band boundary. For example, after constructing the convex hull of a certain set of sampling points, the frequencies corresponding to the convex points are 280kHz power frequency interference band, 320kHz, 800kHz, 3.100MHz, 4.8MHz, and 5.2MHz wireless clutter band.

[0064] Furthermore, the prior frequency band characteristics are derived from historical fault data and field measurement summaries. The frequency range of the effective partial discharge signal of mining cables is stably concentrated between 300kHz and 100MHz. The frequency band below 300kHz is mostly power frequency and inverter interference, while the frequency band above 100MHz is mostly underground wireless communication noise. Based on this, the previously identified bumps are judged one by one. Bumps with frequencies between 280kHz and 300kHz and between 150MHz and 100MHz are judged to be interference signals and are eliminated. Only four bumps with frequencies in the range of 300kHz to 100MHz—320kHz, 800kHz, 3.30MHz, and 150MHz—are retained as candidate points for the frequency band boundary.

[0065] Subsequently, the selected candidate points were sorted by frequency from smallest to largest. The frequency of 320kHz corresponding to the first candidate point after sorting was taken as the lower boundary frequency of the target frequency band. This frequency is slightly higher than the lower limit of the effective frequency band of the partial discharge signal, which is 300kHz, so as to avoid residual interference near 300kHz from entering the target frequency band. The frequency of 90Hz corresponding to the last candidate point after sorting was taken as the upper boundary frequency of the target frequency band. This frequency is slightly lower than the upper limit of the effective frequency band of the partial discharge signal, which is 100MHz, so as to avoid wireless noise interference near 100MHz from entering the target frequency band. Finally, the target frequency band range was set to 320kHz to 100MHz.

[0066] An RC active bandpass filter was selected, with a low cutoff frequency of 320kHz and a high cutoff frequency of 100MHz. The fourth-order filter was chosen. Testing showed that the fourth-order RC active bandpass filter attenuated signals below 320kHz by 28dB, ensuring the amplitude of 280kHz interference signals was reduced to less than 1 / 25 of their original value. For signals above 90MHz, the attenuation reached 30dB, ensuring the amplitude of 5.2MHz clutter signals was reduced to less than 1 / 32 of their original value. Simultaneously, the attenuation of partial discharge signals in the 320kHz to 90MHz range was less than 1dB, with no significant signal loss. This bandpass filter was then connected to the filtered signal link, and the output signal was the analog signal within the target frequency band. Finally, the time-domain waveform of the output signal was measured using an oscilloscope. If the waveform was undistorted, the amplitude was stable within the 1V to 5V range, and the spectrum analyzer showed that the signal energy was concentrated in the 320kHz to 90MHz frequency band, then the target frequency band extraction was confirmed to be effective.

[0067] Step 25: The amplitude of the analog signal is equivalent to the height of a cylinder, and the duration of the analog signal within the target frequency band is equivalent to the base area of ​​the cylinder. The energy volume is obtained by multiplying the amplitude and duration. Based on the energy volume value, the corresponding A / D converter is determined. The corresponding A / D converter is used to convert the analog signal within the target frequency band into a digital signal. Specifically, this includes: First, receiving the analog signal within the target frequency band, to quantify its energy scale and determine the suitable A / D converter, energy volume is introduced as a core quantification indicator. The equivalent logic is: the amplitude of the analog signal is analogous to the height of the cylinder (reflecting instantaneous intensity), and the duration within the target frequency band is analogous to the base area of ​​the cylinder (reflecting time span). The product of the two is the energy volume, which can comprehensively reflect the cumulative scale of signal energy. Moreover, this indicator can be directly related to the range and resolution of the A / D converter, avoiding signal overflow or insufficient accuracy due to improper selection.

[0068] When calculating the energy volume, key parameters need to be acquired and calculations performed in two steps. First, the amplitude reference value of the signal is acquired. This involves capturing the amplitude changes of the analog signal within the target frequency band in real time using an oscilloscope, recording the maximum and minimum amplitudes within 10 consecutive 200ms cycles of 50Hz power frequency, and taking the arithmetic mean of these two values ​​as the amplitude reference value. Using only a single instantaneous amplitude can easily lead to calculation errors due to signal fluctuations. Second, the duration of the signal is acquired. This involves recording the continuous duration of the signal within the target frequency band using a timer, i.e., from the moment the signal first enters the 320kHz to 90MHz frequency band until… The total duration of the last time the signal leaves the frequency band is calculated. Considering that the partial discharge signal in mining is a periodic pulse signal, it is necessary to calculate the average duration of 5 consecutive pulses to avoid random errors. For example, the measured durations of the 5 pulses are 11μs, 12μs, 10μs, 13μs, and 12μs, respectively. The average value is the sum of 11μs, 12μs, 10μs, 13μs, and 12μs divided by 5, which equals 12μs. This value is the duration of the signal. The third step is to calculate the energy volume by multiplying the amplitude reference value by the duration to obtain the specific value of the energy volume, that is, 3V multiplied by 12μs equals 36V·μs.

[0069] Based on the calculated energy volume value, the corresponding A / D converter is determined, with the core being matching the A / D range and resolution. First, the relationship between energy volume and A / D range is analyzed. Energy volume is determined by the product of amplitude and duration, with amplitude being the core factor affecting A / D range. If the energy volume is in the range of 10V·μs to 50V·μs, corresponding to an amplitude of 1V to 5V and a duration of 10μs to 12μs, then the A / D converter's range needs to cover 1V to 5V to avoid signal amplitude exceeding the range and causing overflow distortion. If the energy volume is greater than 50V·μs, corresponding to an amplitude greater than 5V, then an A / D converter with a range of 0V to 10V must be selected. Next, the analysis... The relationship between energy volume and A / D resolution is as follows: the smaller the energy volume, the more subtle the amplitude fluctuation of the signal. For example, an energy volume of 10V·μs corresponds to an amplitude of 1V and a duration of 10μs. The amplitude fluctuation may only be tens of millivolts, requiring a higher resolution A / D converter to capture subtle fluctuations. Practical tests have verified that when the energy volume is in the range of 10V·μs to 50V·μs, a 12-bit resolution A / D converter calculated based on a 5V range, with a minimum quantization unit of approximately 1.22mV, can meet the accuracy requirements. Its quantization error does not exceed 2%, and it can accurately reproduce the amplitude changes of the partial discharge signal. If the energy volume is less than 10V·μs, a 16-bit resolution A / D converter should be selected to improve accuracy.

[0070] Based on the fact that the energy volume of the aforementioned 36V·μs falls within the range of 10V·μs to 50V·μs, a 12-bit resolution A / D converter with a range of 0V to 5V, such as the ADS8364 chip, was selected. This chip has a sampling rate of up to 1MSPS, meeting the high-frequency acquisition requirements of partial discharge signals in mining applications, and its operating temperature range is -40℃ to 85℃, making it suitable for the harsh temperature and humidity environment underground. To verify the rationality of this selection, an analog signal within the target frequency band was input into the selected A / D converter for analog-to-digital conversion, and then the output digital signal was restored. Analysis shows that the digital signal is converted into an analog signal by a digital signal processing algorithm. The amplitude difference between the restored signal and the original analog signal is compared. If the amplitude error does not exceed 2% and the phase information of the signal has no obvious shift and the phase difference does not exceed 5°, then the A / D converter is confirmed to be suitable for the current signal. If the error exceeds the allowable range, the selection method of the amplitude reference value is re-optimized, such as using a weighted average instead of an arithmetic average or replacing it with a higher resolution A / D converter, until the conversion accuracy requirements are met. Finally, the analog signal in the target frequency band is converted into a digital signal by the A / D converter.

[0071] In this embodiment of the invention, out-of-band noise is filtered out by analog filtering, which reduces interference from irrelevant frequency bands, initially purifies high-frequency pulse signals, and preserves the effective signal basis for subsequent processing; pre-amplification enhances weak partial discharge signals and prevents the signal from being submerged by noise; impedance matching reduces reflection and loss during signal transmission, improves signal transmission efficiency and integrity, and ensures the stability of the conditioned signal; anti-aliasing filtering sets the cutoff frequency according to the A / D sampling frequency, which can effectively prevent frequency aliasing during sampling, avoid digital signal distortion, and ensure the accuracy of subsequent analog-to-digital conversion; by analyzing the spectral characteristics through fast Fourier transform and combining the prior frequency band characteristics of the partial discharge signal of mining cables to determine the target frequency band, and then using a bandpass filter to extract the signal, the effective partial discharge signal frequency band can be accurately focused, other frequency band interference can be eliminated, and the signal-to-noise ratio can be improved; selecting a suitable A / D converter based on energy volume (the product of amplitude and duration) can avoid signal distortion caused by insufficient converter performance and prevent resource waste, ensuring the accuracy and efficiency of analog-to-digital signal conversion and adapting to the characteristics of partial discharge signals in mining scenarios.

[0072] In a preferred embodiment of the present invention, step 3 above includes:

[0073] Step 31 involves performing frequency domain filtering on the digital signal to identify and filter out stable narrowband interference, resulting in a denoised digital signal. Specifically, this includes: first, locating the specific frequency of the stable narrowband interference through spectrum analysis; inputting the digital signal obtained in step 25 into a digital signal processing unit, and using Fast Fourier Transform (FFT) to convert the time-domain signal into a frequency-domain spectrum. Considering that the effective frequency band of the partial discharge signal is 300kHz to 100MHz, the frequency resolution of the FFT needs to be set to 1kHz. Resolution = sampling frequency / number of FFT points. Taking the number of FFT points as 12000, 12MHz / 12000 = 1kHz ensures accurate capture of interference peaks within the range of 100kHz to 6MHz. Actual measurements show significant amplitude peaks at 150kHz and 350kHz in the spectrum, at 0.12V and 0.09V respectively, and no frequency drift was observed for 10 minutes. Therefore, these two frequencies are identified as the stable narrowband interference frequencies that need to be filtered out.

[0074] Secondly, design targeted frequency domain notch filter parameters. The core parameters of the notch filter are the center frequency (f0) and the 3dB bandwidth (B), which must filter out only the interference frequency without affecting adjacent effective frequency band signals: the center frequency f0 is directly set to the located interference frequency (150kHz, 350kHz); the 3dB bandwidth B needs to be determined according to the distance between the interference frequency and the effective frequency band. The distance between 150kHz and the lower limit of the effective frequency band 300kHz is 150kHz, and the distance between 350kHz and the lower limit of the effective frequency band 300kHz is 50kHz. To avoid affecting the floating discharge signal near 300kHz, from 300kHz to 1MHz, 1 The 3dB bandwidth of the 50kHz notch filter is set to 10kHz, covering 145kHz to 155kHz, and the 3dB bandwidth of the 350kHz notch filter is set to 8kHz, covering 346kHz to 354kHz. At the same time, it is necessary to ensure that the filter attenuates the interference frequency by more than or equal to 20dB, so that the interference amplitude is reduced to less than 1 / 10 of the effective signal. According to simulation calculation, a second-order infinite impulse response (IIR) notch filter can meet the requirements: the attenuation of the second-order IIR notch filter can reach 25dB at the center frequency, and the attenuation drops rapidly to less than 1dB at a distance of 2 times the bandwidth from the center frequency, without affecting the effective signal.

[0075] Finally, the filtering effect was verified through actual measurement to determine the denoised digital signal. The designed dual notch filters (150kHz, 350kHz) were connected in series to the digital signal link, and the spectrum and time domain signals before and after filtering were collected simultaneously. In the frequency domain, the interference amplitude at 150kHz decreased from 0.12V to 0.01V, and the interference amplitude at 350kHz decreased from 0.09V to 0.008V, and the signal amplitude in the effective frequency band from 300kHz to 100MHz did not change significantly. In the time domain, the periodic interference ripples that were originally superimposed on the partial discharge pulse disappeared, and the pulse waveform was clearer. The digital signal obtained at this time is the denoised digital signal that meets the requirements. If the measured attenuation is insufficient, the order of the notch filter needs to be increased to the third order, and the test needs to be repeated until the attenuation index is met.

[0076] Step 32 involves performing time-domain averaging on the denoised digital signal. This involves synchronously superimposing and averaging data from multiple power frequency cycles to enhance the periodically recurring partial discharge pulse signal and suppress random noise, resulting in an enhanced digital signal. Specifically, this includes: First, determining the core parameters for time-domain averaging: the number of power frequency cycles and the synchronization reference. Considering the real-time requirements for partial discharge monitoring in underground coal mines, the data processing delay must be less than or equal to 1 second. If the number of average cycles is too small, random noise cancellation will be insufficient; if it is too large, the delay will exceed the requirements. A compromise calculation selects 20 power frequency cycles as the averaging window, which ensures both noise suppression and delay control within 500ms (including data transmission time). The synchronization reference relies on the power frequency signal (50Hz sine wave) collected by the mine power supply sensor in Step 2. Digital phase-locked loop technology is used to use the starting point of each power frequency cycle as the time reference, ensuring precise alignment of the 20 cycles of digital signals on the time axis.

[0077] Twenty synchronously aligned power frequency cycles of digital signal, each cycle containing 12MHz × 20ms = 240,000 sampling points, are superimposed one-to-one, i.e., the first sampling point of the first cycle is added to the first sampling point of the second to 20th cycles, and so on, resulting in a total of 240,000 superimposed sums. Each superimposed sum is then divided by the number of cycles, 20, to obtain the average amplitude of each sampling point. In this process, the amplitude of periodic partial discharge pulses, because they appear at the same sampling point in each cycle, approaches 20 times the amplitude of a single-cycle pulse after superposition, and remains close to the amplitude of a single-cycle pulse after averaging. On the other hand, random noise, because it is randomly distributed at different sampling points in each cycle, approaches 4.47 times the amplitude of a single-cycle noise after superposition, and decreases to about 0.22 times the amplitude of a single-cycle noise after averaging, thus achieving noise suppression.

[0078] Finally, an oscilloscope was used to compare the signals before and after time-domain averaging to verify the enhancement effect: the signal-to-noise ratio of the partial discharge pulse before averaging was 15dB, corresponding to a pulse amplitude of 1V and a noise amplitude of 0.22V. After averaging, the signal-to-noise ratio increased to 26dB, the pulse amplitude became 0.98V, and the noise amplitude decreased to 0.04V. Moreover, the original 50ns rising edge and 60ns falling edge of the partial discharge pulse after averaging did not deviate by more than 5ns. At this time, the output signal was the enhanced digital signal. If the signal-to-noise ratio did not increase to 20dB, the number of averaging cycles was increased to 25, corresponding to 500ms, and the calculation was repeated until the signal-to-noise ratio met the requirements for subsequent pulse detection and was not lower than 25dB.

[0079] Step 33 involves pulse detection of the enhanced digital signal. Potential partial discharge pulses are identified based on signal amplitude and waveform characteristics, and grouped according to their power frequency phase to obtain a grouped pulse sequence. Specifically, this includes: First, based on the measured characteristics of partial discharge pulses in mining cables, two core judgment conditions are set: one is a rise time less than or equal to 60 ns; the other is a pulse width less than or equal to 120 ns. Simultaneously, to avoid false detection noise, an amplitude detection threshold is set: based on the average noise amplitude obtained in step 32, such as 0.04V, the threshold is set to three times the average noise amplitude, 0.12V. Only signals with an amplitude exceeding 0.12V and meeting the waveform characteristics are judged as potential partial discharge pulses.

[0080] Secondly, the enhanced digital signal is scanned point by point according to the sampling points. When the amplitude of a certain sampling point exceeds 0.12V, pulse capture is triggered: the amplitude changes of 100 sampling points before and after the point are recorded, and the rise time and pulse width are calculated. If the rise time is 45ns and the pulse width is 90ns, both of which meet the judgment conditions, it is marked as a potential partial discharge pulse, and its occurrence timestamp is recorded. If the rise time is 80ns, such as an electromagnetic impulse signal, it is judged as a non-partial discharge pulse and is excluded. According to actual measurement, the recognition rate of this detection method for effective partial discharge pulses is greater than or equal to 98%, and the false detection rate of non-partial discharge pulses is less than or equal to 2%.

[0081] Finally, based on the 50Hz, 20ms period of the power frequency signal collected by the mine power supply sensor, a mapping relationship between timestamps and power frequency phases is established: each power frequency cycle (20ms) is divided into 360 phase points, and the time corresponding to each phase point is approximately 55.56μs (20ms / 360). That is, phase 0° corresponds to the start of the cycle, phase 90° corresponds to the 5ms moment of the cycle, and so on. For the timestamp of each potential partial discharge pulse, the time position of its occurrence in the current power frequency cycle is calculated, and then converted to the corresponding phase = (6ms / 20ms) × 360° = 108°. According to the phase distribution characteristics of different fault types in mine cables, the phase is divided into 8 group intervals, each grouped at 45°. Each pulse is assigned to the corresponding interval according to its phase, ultimately forming a grouped pulse sequence classified by phase interval. If the number of pulses in a certain group interval (such as 45° to 90°) accounts for more than 60% of the total number of pulses, it is preliminarily judged that there may be an internal discharge fault, providing direction for subsequent diagnosis.

[0082] Step 34: Set a dynamic threshold for the grouped pulse sequence. Adjust the threshold size according to the noise floor level and signal statistical characteristics, remove noise pulses with amplitudes below the threshold, and screen out effective partial discharge pulses. Specifically, this includes: selecting 10 power frequency cycles (200ms) as the statistical window for the noise floor. If the window is too short (e.g., 2 cycles), the statistics will be inaccurate due to random noise fluctuations; if it is too long (e.g., 30 cycles), it will not be able to respond to noise mutations in time. Within each statistical window, extract sampling points where no potential pulses are detected, i.e., sampling points with amplitudes less than or equal to 0.12V, which are considered as pure noise points. Calculate the maximum amplitude (N). max ) and root mean square value (N) rm ), where N max N reflects the peak level of noise. rm Reflects the average fluctuation level of noise; based on actual measurements, N represents the noise level at a certain statistical window. max =0.05V, N rm =0.03V.

[0083] Secondly, to balance the need to avoid missing small pulses and the need to avoid false detection noise, a dynamic threshold (Th) = N is set based on multiple downhole measurement data. max ×1.5+N rm ×0.5, in this formula, N max ×1.5 ensures the threshold is higher than the noise peak, avoiding false detections; N rm ×0.5 represents a margin for small-amplitude pulses to avoid missed detections; substituting the measured values ​​above, Th = 0.05V × 1.5 + 0.03V × 0.5 = 0.075V + 0.015V = 0.09V; if the noise floor rises to N due to equipment start-up and shutdown within the next statistical window... max =0.1V, N rm =0.06V, then the dynamic threshold is synchronously adjusted to Th=0.1V×1.5+0.06V×0.5=0.15V+0.03V=0.18V, realizing the real-time adaptation of the threshold to the noise floor.

[0084] Finally, the amplitude of each pulse in the grouped pulse sequence is compared with the dynamic threshold of the current statistical window: if the pulse amplitude is 0.15V (higher than 0.09V), it is determined to be a valid partial discharge pulse; if the pulse amplitude is 0.08V (lower than 0.09V), it is determined to be a noise pulse and is removed.

[0085] Step 35: Extract the signal characteristics of the effective partial discharge pulses, including calculating the amplitude, phase, and discharge quantity of each pulse, and counting the frequency of pulse occurrence per unit time to form signal characteristics. Specifically, the amplitude of the effective partial discharge pulse is the peak amplitude of its time-domain waveform. For each effective pulse, extract the amplitude of the largest sampling point in its waveform and record this value as the pulse amplitude. To eliminate system errors, the signal link needs to be calibrated in advance: connect a standard signal source to the sensor. After processing, if the measured amplitude is 0.98V, then the calibration coefficient Kv = 1V / 0.98V ≈ 1.02. The final actual amplitude = measured amplitude × Kv, such as 1.2V × 1.02 ≈ 1.22V.

[0086] The pulse phase directly uses the converted power frequency phase value without additional calculation, but the phase accuracy needs to be verified: by synchronously acquiring the power frequency signal and the pulse timestamp, the phase conversion formula is rechecked. Phase = (time position within the period / 20ms) × 360°. If the time position within the period of a certain pulse is 6ms, the converted phase is 108°, which is completely consistent with the actual phase (108°) of the power frequency signal, then the phase data is valid.

[0087] The discharge quantity is a key parameter reflecting the partial discharge intensity and needs to be calculated in conjunction with the sensor's calibration curve (obtained in advance using a standard partial discharge calibrator): the standard calibrator outputs a partial discharge signal with a known discharge quantity Q0 (e.g., 100pC), which, after system processing, yields the measured amplitude V0 (e.g., 2V). Then, the discharge quantity calibration coefficient K... q =Q0 / V0=100pC / 2V=50pC / V; For the measured amplitude V of a certain effective pulse (e.g., 1.22V), its discharge quantity Q=V×Kq=1.22V×50pC / V=61pC; At the same time, the influence of cable length on the discharge quantity needs to be considered: If the cable length between the monitoring point and the sensor is 100m, and the discharge quantity attenuation coefficient per 100m of cable is 0.95 (based on actual measurement), then the actual discharge quantity =Q×0.95=61pC×0.95≈58pC.

[0088] Frequency is the number of effective partial discharge pulses per unit time. One minute (60 seconds) is selected as the statistical time unit. The total number of effective partial discharge pulses within 60 seconds is counted. If the total number is 300, then the frequency = 300 times / 60 seconds = 5 times / second. To avoid the influence of short-term fluctuations, a moving average is used: the frequency is updated every 10 seconds, and the total number of pulses in the previous 60 seconds is counted each time to ensure that the frequency data is stable.

[0089] Finally, the amplitude, phase, discharge quantity, and frequency per unit time of each effective pulse are integrated to form a signal feature set. If a certain feature parameter exceeds the system error range, the signal link needs to be recalibrated until the feature data is accurate.

[0090] In this embodiment of the invention, for fixed-frequency interference remaining from previous analog filtering, such as inverter and motor noise, frequency domain filtering is used to accurately locate and eliminate it, preventing interference from being misjudged as partial discharge pulses, solving the problem of incomplete analog filtering, and improving the purity of digital signals. Utilizing the power frequency periodicity of partial discharge signals and the randomness of noise, multi-cycle synchronous superposition and averaging are used to enhance the amplitude of partial discharge pulses and suppress random noise, making small-amplitude partial discharge pulses that were originally masked by noise clearly visible, thus improving the signal-to-noise ratio. Potential partial discharge pulses are screened based on waveform characteristics, namely short rise time and narrow width, eliminating non-partial discharge impulse signals. Grouping by power frequency phase, combined with the phase distribution differences of different faults, lays the foundation for accurate differentiation of partial discharge types. To address the problem of fluctuating noise floor in the well, thresholds are dynamically adjusted according to real-time noise levels to avoid missed or false detections with fixed thresholds, improving the accuracy of effective partial discharge pulse screening and reducing invalid data interference. Four core indicators—amplitude, phase, discharge quantity, and frequency—are extracted to transform digital signals into quantitative fault judgment criteria, connecting front-end signal processing and back-end fault diagnosis, providing support for accurately judging cable insulation status and assessing fault severity.

[0091] In a preferred embodiment of the present invention, step 4 above includes:

[0092] Step 41: The signal characteristics are transmitted to the cloud management system, which receives and parses the signal characteristics to obtain a pulse data sequence containing pulse amplitude, phase, and frequency of occurrence. Specifically, this includes: considering the network connection design between the cloud management system and the underground detection host in this invention, three methods can be used: network cable, single-mode dual-fiber, and WIFI. Given the characteristics of strong electromagnetic interference and long transmission distance in underground mining, single-mode dual-fiber optical cable is preferred for transmission. The transmission rate needs to match the amount of data in the signal characteristics: each signal characteristic contains pulse amplitude, phase, and frequency of occurrence. A single detection host generates approximately 100 sets of characteristic data per second, which translates to approximately 400 bytes / second of data. Therefore, the transmission rate is set to 10Mbps to ensure no data packet loss.

[0093] Due to differences in production batches, multiple underground detection hosts may output slightly different signal characteristic formats. Therefore, the cloud management system needs to first establish a parsing rule base: First, the amplitude unit should be unified, converting mV unit data to V (1mV=0.001V) and retaining 2 decimal places of precision. For example, if the detection host outputs 1200mV, it should be parsed as 1.20V. Second, the phase format should be unified. If it is a period percentage, such as 0.3 periods, it should be converted to an angle value, 0.3×360°=108°, retaining integer precision. Third, the frequency should be associated with the timestamp, binding the unit time frequency with the system timestamp of the detection host to form a structured data item of time, amplitude, phase, and frequency.

[0094] Every 100 sets of parsed data received by the cloud management system, a sampling comparison is performed with the original signal characteristics of the downhole detection host. The sampling ratio is 10%. If the deviation between the parsed amplitude and the original amplitude in the sampled data is less than or equal to 0.01V, the phase deviation is less than or equal to 1°, and the frequency is completely consistent, the parsing is deemed valid. These 100 sets of data are then arranged in chronological order by timestamp to form a pulse data sequence containing pulse amplitude, phase, and frequency of occurrence. If the deviation exceeds the standard, retransmission and parsing are triggered until three consecutive sampling verifications are passed, ensuring that the pulse data sequence can truly reflect the characteristics of the downhole partial discharge signal.

[0095] Step 42: Based on the pulse data sequence, analyze the energy distribution differences between noise and partial discharge signals at different scales, and set a denoising threshold according to the energy distribution differences; process the pulse data sequence using wavelet transform and according to the denoising threshold to filter out random noise and broadband interference, obtaining a denoised pulse data sequence, specifically including:

[0096] First, the partial discharge signal of mining cables is a high-frequency narrow pulse with a main frequency of 300kHz to 100MHz and a pulse width of less than or equal to 100ns. Wavelet basis functions with good time-domain tight support and high frequency-domain resolution are selected. After comparing the processing effects of the db series (db1 to db8) wavelets, the reconstruction error of the db4 wavelet for high-frequency pulses is less than or equal to 5%, which is better than db1 (error 12%) and db8 (high computational complexity). Therefore, the wavelet basis function is determined to be db4.

[0097] The number of decomposition layers needs to be determined based on the signal's dominant frequency and noise distribution: The sampling frequency of the pulse data sequence is 12MHz. According to the frequency coverage characteristics of wavelet decomposition (based on the Nyquist sampling theorem, the highest effective signal frequency is 1 / 2 of the sampling frequency, i.e., 6MHz), the frequency ranges corresponding to each decomposition layer are as follows: Decomposition layer 1 (d1) covers 3MHz to 6MHz, decomposition layer 2 (d2) covers 1.5MHz to 3MHz, decomposition layer 3 (d3) covers 0.75MHz to 1.5MHz, decomposition layer 4 (d4) covers 0.375MHz to 0.75MHz, decomposition layer 5 (d5) covers 0.1875MHz to 0.375MHz, and the low-frequency coefficient (a5) covers 0 to 0.1875MHz. Considering that the main frequency of the partial discharge signal is 300kHz (0.3MHz) to 100MHz, decomposing it into 5 layers can completely cover its effective sampling frequency band (0.3MHz to 6MHz, the part above 6MHz has been mapped to the low frequency band due to sampling aliasing), and can separate the interference of different frequencies to a5 (0-0.1875MHz) and the 0.1875-0.3MHz part of d5. Therefore, the number of decomposition layers is determined to be 5 layers.

[0098] Secondly, the energy values ​​of the five high-frequency coefficients (d1 to d5) and one low-frequency coefficient (a5) after decomposition were calculated: Since the partial discharge signal energy is concentrated in the range of 300kHz (0.3MHz) to 100MHz, the energy in the effective sampling frequency band is mainly distributed in layers d2 to d4 (corresponding to 0.375MHz to 3MHz). According to actual measurements, the total energy of layers d2 to d4 accounts for 82% of the total signal energy. Random noise and broadband interference energy are dispersed, with layer d1 (3-6MHz) accounting for only 5% of the energy (high-frequency interference), layer d5 (0.1875-0.375MHz) accounting for 8% of the energy (including some interference below 0.3MHz), and layer a5 (0-0.1875MHz) accounting for 5% of the energy (low-frequency interference). Based on this difference in energy distribution, layers d2 to d4 can be determined as the dominant layers of partial discharge signal, and layers d1, d5, and a5 are the dominant layers of interference.

[0099] Then, the threshold needs to suppress only the interference layer coefficients without damaging the partial discharge layer coefficients: First, calculate the maximum coefficient amplitude of the interference-dominant layers (d1, d5, a5). The maximum coefficient amplitude of layer d1 is 0.03V, layer d5 is 0.02V, and layer a5 is 0.01V. Take the maximum value of 0.03V as the basic threshold. Then consider the minimum coefficient amplitude of the partial discharge layers (d2 to d4) which is 0.08V. Finally, set the denoising threshold to 0.04V. Compare the coefficients of each decomposition layer with the threshold: set the coefficients with an absolute value less than 0.04V to 0, and retain the coefficients with an absolute value greater than 0.04V. The threshold processing is then completed.

[0100] Finally, the processed coefficients are reconstructed using a db4 wavelet to obtain the reconstructed signal. The signals before and after reconstruction are compared using an oscilloscope: the signal-to-noise ratio (SNR) before reconstruction is 26dB, corresponding to the enhanced signal in step 32; the SNR after reconstruction is increased to 34dB, and the amplitude deviation of the partial discharge pulse is less than or equal to 0.02V, with no phase deviation. The random noise amplitude decreases from 0.04V to 0.01V, and the broadband interference completely disappears. At this point, the reconstructed signal is the denoised pulse data sequence. If the SNR improvement is less than 30dB, the threshold needs to be adjusted, such as decreasing it to 0.035V, and the signal needs to be decomposed and reconstructed again until the requirement of an SNR greater than or equal to 32dB is met.

[0101] Step 43: Based on the time-frequency characteristics of the denoised pulse data sequence, determine the sliding time window length and overlap rate required for the short-time Fourier transform; segment the denoised pulse data sequence using a sliding time window with the specified window length and overlap rate; perform a short-time Fourier transform on the segmented signal segments to obtain the time-frequency distribution data of the signal. Specifically, this includes: First, the window length needs to balance time resolution and frequency resolution: the duration of a mining partial discharge pulse is approximately 50ns to 100ns, the time resolution needs to be less than or equal to 100ns, and the corresponding window time length needs to be less than or equal to 100ns. However, this correspondence has deviations and needs to be adjusted according to the actual application scenario. The STFT window length unit is usually in the millisecond range, because although the partial discharge pulse is short... However, it is necessary to analyze the time-frequency changes within the power frequency period (20ms). Therefore, the time resolution target is set to 2ms, and the frequency resolution target is set to 100kHz. According to the STFT resolution formula, the time resolution ≈ window length, and the frequency resolution ≈ 1 / window length. Therefore, the window length = 1 / frequency resolution = 1 / 100kHz = 10ms. However, the time resolution of 10ms window cannot meet the 2ms requirement. After compromise calculation, the window length is selected as 2ms. At this time, the frequency resolution = 1 / 2ms = 500kHz, the time resolution = 2ms, and the 2ms window can completely contain a single partial discharge pulse. The pulse duration is less than or equal to 100ns, which is much less than 2ms, thus avoiding the pulse being truncated by the window.

[0102] Secondly, the overlap rate ensures that there are no gaps between the signal segments of adjacent windows, while avoiding excessive overlap that would lead to a surge in computation: the window length is 2ms. If there is no overlap, the interval between adjacent windows will be 2ms, which may cause the partial discharge pulse within the 2ms interval to be missed. If the overlap rate is too high, such as 80%, the computation will increase by 5 times, affecting the efficiency of cloud processing. Considering the frequency of occurrence of partial discharge pulses in mining, the overlap rate is set to 50%. The overlap time between adjacent windows = 2ms × 50% = 1ms, and the window sliding step size = 2ms - 1ms = 1ms. This ensures that the pulse within the 1ms interval is not lost, while keeping the computation within an acceptable range.

[0103] Then, the denoised pulse data sequence is divided into 199 signal segments with a window length of 2ms and a step size of 1ms. Each signal segment is then weighted using a Hanning window and subjected to a Fourier transform to obtain the frequency spectrum of each segment, with a frequency range of 0 to 6MHz and a frequency point interval of 500kHz, matching the frequency resolution of the 2ms window length. The timestamp, frequency point, and corresponding amplitude of each signal segment are integrated to form time-frequency distribution data, for example, time 10ms, frequency 1MHz, amplitude 1.2V; time 10ms, frequency 2MHz, amplitude 0.3V.

[0104] Finally, a standard partial discharge signal with a known main frequency is input into the system, and its STFT time-frequency diagram is compared with the theoretical time-frequency characteristics: if the amplitude of the 1MHz frequency point in the time-frequency diagram at the corresponding timestamp is 1.2V, the deviation from the theoretical value is less than or equal to 5%, and there are no obvious side lobes, and the side lobe amplitude is less than or equal to 10% of the main lobe, then the time-frequency distribution data is determined to be valid; if the side lobe is too high, such as 20%, then a Hamming window needs to be replaced, and the data needs to be re-segmented and transformed until the time-frequency data can accurately reflect the time and frequency correlation characteristics of the partial discharge signal.

[0105] Step 44: Based on the time-frequency distribution data and pulse phase information, statistically map the partial discharge pulses detected within multiple power frequency cycles according to their phase and amplitude to generate a PRPD spectrum with phase and amplitude as coordinates and pulse occurrence frequency as intensity. Specifically, this includes: First, determining the two-dimensional coordinate range and scale value of the PRPD spectrum. The phase coordinate (horizontal axis) is based on a power frequency cycle of 20ms corresponding to 360°. To ensure the ability to distinguish the phase distribution differences of different faults, the phase range is set to 0° to 360°, and the scale value is set to 5. °, too small a division value, such as 1°, will result in too many grids and frequency dispersion; too large a division value, such as 10°, will mask phase details. A 5° division value can balance details and concentration. Amplitude coordinate (vertical axis): Based on the amplitude range of the denoised pulse data sequence, in order to cover all pulse amplitudes, the amplitude range is set to 0V to 5V, and the division value is set to 0.2V. Each 0.2V is a coordinate grid, for a total of 25 grids. The 0.2V division value can distinguish the small amplitude difference between 0.5V and 0.7V, and avoid the merging of pulses with similar amplitudes due to the division value being too large, which would affect pattern recognition.

[0106] Secondly, the statistical period needs to include a sufficient number of power frequency cycles to ensure the representativeness of the frequency statistics: if the period is too short, the number of pulses will be insufficient and the frequency distribution will be sparse; if it is too long, the calculation time will be too long. Considering the frequency of mine partial discharge pulses (less than or equal to 10 pulses / second), 20 power frequency cycles (400ms) are selected as the statistical period. At this time, the number of pulses is approximately 4 (10 pulses / second × 0.4 seconds = 4), which can form a stable frequency distribution. The frequency statistics rule is: for each partial discharge pulse, according to its phase... The position value (e.g., 108°) is assigned to the corresponding phase cell (108° ÷ 5° = 21.6, assigned to cell 22: 105° to 110°), and its amplitude value (e.g., 1.5V) is assigned to the corresponding amplitude cell (1.5V ÷ 0.2V = 7.5, assigned to cell 8: 1.4V to 1.6V). The count in the intersection cell of the phase cell and amplitude cell is incremented by 1. After counting all pulses within 20 cycles, the count in each intersection cell is the pulse frequency of that phase and amplitude combination.

[0107] Then, a mapping relationship between frequency and intensity is established to generate a PRPD spectrum. A color gradient is used to represent frequency and intensity: 0 is white (no pulse), 1 is light yellow, 2 is dark yellow, and 3 and above are red. This gradient setting can intuitively distinguish between no-pulse, low-frequency, and high-frequency regions, which conforms to the human eye's color recognition ability. The correspondence between phase coordinates, amplitude coordinates, and color intensity is imported into the spectrum generation module of the cloud management system to generate a two-dimensional PRPD spectrum. The statistical period, amplitude range, and phase range are marked below the spectrum to ensure the integrity of the spectrum information.

[0108] Finally, mining cables with known fault types (such as cables with floating discharge) were selected for testing. The generated PRPD spectrum was compared with the typical PRPD pattern of floating discharge: if the intersection area of ​​the 0° to 90° phase grid and the 1.0V to 2.0V amplitude grid in the spectrum is dark yellow (twice), and the intersection area of ​​the 180° to 270° phase grid and the 1.0V to 2.0V amplitude grid is light yellow (once), consistent with the typical pattern, then the spectrum is considered valid; if the high-frequency region deviates, such as being concentrated between 90° and 180°, then the accuracy of the phase calculation needs to be checked, such as whether the timestamp and phase mapping are incorrect, and the spectrum needs to be regenerated until it can accurately reflect the PRPD pattern of the fault, providing a visual basis for subsequent fault diagnosis.

[0109] In this embodiment of the invention, signal features are transmitted to the cloud, relying on the high-performance processor and large-capacity storage of the central server to avoid analysis delays or accuracy degradation caused by insufficient resources at the edge. Non-uniform format data from different detection hosts are parsed into standardized pulse data sequences, avoiding analysis errors caused by format confusion, while also supporting data aggregation from multiple devices. Thresholds are precisely set by analyzing energy distribution differences to filter out random noise and broadband interference, avoiding the drawbacks of traditional one-size-fits-all filtering, maximizing the preservation of partial discharge signal details, providing clean data for subsequent time-frequency analysis, and improving analysis accuracy. Short-time Fourier transform is used... The core of performing time-frequency analysis on partial discharge (PD) signals using transform is to overcome the shortcomings of traditional Fourier transform, which only displays frequency distribution but lacks time correlation. It accurately captures the frequency changes of PD signals over time, providing a basis for judging fault trends. After performing a short-time Fourier transform on a signal segment, the resulting time-frequency distribution data can intuitively present the frequency composition of the PD signal at a certain moment, allowing for early prediction of fault trends and saving maintenance personnel time for handling. By converting scattered pulse data into a visual spectrum of phase, amplitude, and frequency, the differences in spectrum patterns between different faults are intuitive, facilitating rapid differentiation of fault types and providing a standardized and comparable intuitive basis for subsequent fault diagnosis.

[0110] In a preferred embodiment of the present invention, step 5 above includes:

[0111] Step 51: Input the PRPD spectrum into a pre-stored model library of typical defects in mining cables. The model library contains PRPD spectrum feature vectors of various typical defects. Statistical feature extraction is performed on the PRPD spectrum to form a feature vector to be diagnosed. Specifically, this includes: First, determining the foundation and feature vector dimensions of the pre-stored model library. Considering the core characteristic that different defects have different PRPD patterns, the typical defect feature vectors in the model library must cover key statistical features that can distinguish various faults. After analyzing the PRPD spectrum patterns of partial discharge faults in mining cables, the feature vector dimension is determined to be 6 dimensions, corresponding to: the phase range of the high-frequency region, the amplitude range of the high-frequency region, the average frequency of the high-frequency region, the total number of pulses, the phase of the largest amplitude pulse, and the amplitude standard deviation. Simultaneously, the model library needs to pre-store standard feature vectors for four types of typical defects, and all vectors must maintain a consistent format and dimension.

[0112] Secondly, statistical feature extraction of the PRPD spectrum is carried out. This process requires calculation of each of the above 6-dimensional features. Taking a PRPD spectrum to be diagnosed generated in step 44, with a phase resolution of 5°, an amplitude resolution of 0.2V, and a statistical period of 20 power frequency cycles as an example: to extract the phase range of the high-frequency region, first traverse all phase and amplitude intersection cells in the spectrum, count the frequency of each cell, and mark cells with a frequency greater than or equal to 3 as high-frequency cells; then extract the phase values ​​of these cells, take the minimum phase value (e.g., 10°) and the maximum phase value (e.g., 85°), 185° and 275°, and merge them into 10° to 85° / 185° to 275° as the first dimension feature; similarly, extract the amplitude values ​​of the high-frequency cells, take the minimum amplitude value (e.g., 1.1V) and the maximum amplitude value (e.g., 1.9V), to obtain 1.1V to 1 0.9V is used as the second feature; the total frequency of all high-frequency cells is counted. For example, if there are 12 high-frequency cells in a spectrum and the total frequency is 48, then the average frequency = 48 ÷ 12 = 4.0, which is used as the third feature; all cells are traversed, and the frequency of each cell is accumulated. For example, if the total frequency of all cells is 17, it is used as the fourth feature; the cell with the largest amplitude in the spectrum is found, such as the cell corresponding to the amplitude of 2.0V, and its phase value is recorded as the fifth feature; the amplitude of all pulses is first counted, and the amplitude of each cell is multiplied by the frequency, such as 1.0V × 2 times, 1.2V × 3 times, etc., to obtain a list of amplitudes of all pulses; then the standard deviation of the list is calculated, such as 0.28V, which is used as the sixth feature; finally, the extracted 6-dimensional features are organized according to the vector format of the model library to form the feature vector to be diagnosed.

[0113] Step 52: Calculate the similarity between the feature vector to be diagnosed and the typical defect feature vector in the model library. Perform preliminary matching based on the principle of maximum similarity to obtain preliminary matching results of defect type and its confidence level. Specifically, this includes: First, both the feature vector to be diagnosed and the typical defect feature vector are multi-dimensional data of mixed numerical and interval types, and it is necessary to focus on reflecting the consistency of feature trends. After comparing Euclidean distance, Manhattan distance, and cosine similarity, the cosine similarity calculation method is preferred. This method reflects the similarity of vector directions by calculating the cosine value of the angle between two vectors, and is not affected by the absolute value of the features.

[0114] Secondly, the similarity between the feature vector to be diagnosed and each typical defect feature vector in the model library is calculated one by one. Before the calculation, the interval features are converted into numerical data: for example, the phase range is 10° to 85° / 185° to 275°, and the midpoint of the two intervals (47.5°, 230°) is taken; the amplitude range is 1.1V to 1.9V, and the midpoint 1.5V is taken; the maximum amplitude pulse phase is 40° / 220°, and the midpoint (40°, 220°) is taken. The cosine similarity between the vector to be diagnosed and typical vectors of floating discharge, internal discharge, etc., is calculated respectively.

[0115] Taking the calculation of a typical suspended discharge vector as an example, the numerical values ​​of each dimension of the two vectors are first standardized to eliminate the influence of dimensions; then the cosine value of the included angle is calculated using the vector dot product formula. According to actual measurements, the cosine similarity between the vector to be diagnosed and the typical suspended discharge vector is 0.92, the similarity with the typical internal discharge vector is 0.65, and the similarities with surface discharge and corona discharge are 0.58 and 0.43, respectively. The maximum similarity is 0.92, which initially points to suspended discharge.

[0116] After obtaining the similarity calculation results, the maximum similarity value is not directly used. Instead, a secondary verification is performed based on the fault feature verification criteria. These criteria are derived from historical monitoring data and engineering practice summaries of partial discharge faults in mining cables. For example, the high-frequency region of suspended discharge needs to include 0° to 90° and 180° to 270°, with amplitude ranges mostly between 1.0V and 2.0V. The high-frequency region of internal discharge needs to include 45° to 135° and 225° to 315°. For the above-mentioned vector to be diagnosed, its high-frequency region is 10° to 85° and 185° to 275°. These two intervals are completely within the typical phase range of suspended discharge, and its amplitude range is between 1.1V and 1.9V, which also conforms to the amplitude pattern of suspended discharge. The feature criterion verification passes. If the similarity of a vector to be diagnosed points to suspended discharge, but the high-frequency region falls between 90° and 180°, the feature criterion verification fails, and the defect type corresponding to the second largest similarity (such as internal discharge) needs to be reselected.

[0117] Finally, the confidence level of the preliminary matching result is calculated. The confidence level calculation needs to consider the degree of matching between the similarity value and the feature criteria: if the feature criteria match perfectly (as in the case above), then the confidence level = similarity value × 1.0; if the feature criteria match partially (as in the high-frequency region containing only one typical phase interval), then the confidence level = similarity value × 0.8; if the feature criteria do not match, then the confidence level = similarity value × 0.5. In the case above, the feature criteria match perfectly, and the similarity is 0.92, so the confidence level = 0.92 × 1.0 = 0.92. The final preliminary matching result is: defect type: floating discharge, confidence level: 0.92.

[0118] Step 53: Input the feature vector to be diagnosed and the preliminary matching result into a pre-trained machine learning classification model. The model calculates the probability of belonging to various defect types based on the input feature vector and, combined with the preliminary matching result, determines the partial discharge type, discharge source, and insulation severity level, generating a diagnostic result containing partial discharge attributes and severity. Specifically, this includes: First, determining the pre-training basis and model structure of the machine learning classification model; the model is trained based on historical measured data of partial discharge faults in mining cables. The training samples cover four typical defect types (suspended, internal, surface, and corona discharge), with 1000 samples for each defect type. Each sample set contains a 6-dimensional feature vector, preliminary matching results, actual discharge source, and actual severity label. The discharge source label includes internal insulation, cable termination, joint, and outer sheath. The severity label is divided into mild (Level I), moderate (Level II), and severe (Level III). The model structure adopts a fusion architecture of convolutional neural network (CNN) and long short-term memory network (LSTM): CNN is used to extract local correlation features in the feature vector, and LSTM is used to capture the temporal correlation between the preliminary matching results and the feature vector (such as the matching degree between confidence and feature dimension). The model output is the probability value of each type of defect, the probability value of each discharge source, and the probability value of each severity.

[0119] Secondly, the input features need to include two parts: one is the 6-dimensional feature vector to be diagnosed generated in step 51, which is converted into a numerical matrix with a dimension of 1×6; the other is the preliminary matching result in step 52, which is converted into numerical features. The defect type: floating discharge corresponds to the encoding [1, 0, 0, 0], and the confidence level: 0.92 corresponds to the numerical value 0.92. After integration, the dimension is 1×5. The two parts of features are concatenated into a 1×11 input matrix and input into the pre-trained model.

[0120] The model first performs convolution operations on the 6-dimensional feature vector through a CNN layer to obtain 6 feature maps; then, it performs temporal processing on the 1×5 preliminary matching features through an LSTM layer to capture the correlation between confidence and defect type encoding, and outputs probability values ​​through a fully connected layer; after obtaining the probability values ​​output by the model, it makes a comprehensive judgment based on the preliminary matching results to determine the final partial discharge type and discharge source; for the partial discharge type, the floating discharge probability (0.95) output by the model is highly consistent with the preliminary matching result (floating discharge, confidence 0.92), so the partial discharge type is confirmed to be floating discharge; if the maximum probability type output by the model is inconsistent with the preliminary matching type, it is necessary to re-check the feature extraction in step 51 and the similarity calculation in step 52, and after eliminating data errors, the type with the higher model probability shall be used.

[0121] Finally, the severity classification needs to refer to the pre-set threshold standards, which are based on historical failure data of insulation faults in mining cables: the standard for mild (Level I) is a total number of pulses less than or equal to 20 / 20 cycles, a maximum amplitude value less than or equal to 2.0V, and a high-frequency area less than or equal to 10 cells; the standard for moderate (Level II) is a total number of pulses greater than 20 and less than or equal to 30, a maximum amplitude value greater than 2.0V and less than or equal to 3.0V, and a high-frequency area greater than 10 and less than or equal to 20 cells; the standard for severe (Level III) is a total number of pulses greater than 30, a maximum amplitude value greater than 3.0V, and a high-frequency area greater than 20 cells; all judgment results are integrated to generate the final diagnostic result.

[0122] In this embodiment of the invention, the visualized PRPD spectrum is transformed into a structured feature vector to be diagnosed, achieving format unification between the spectrum data and the typical defect feature vectors in the model library, thus solving the problem that the visualized spectrum cannot be directly used for data comparison. By extracting key statistical features that can distinguish fault types, it is ensured that the vector to be diagnosed and the vector in the model library are fully compatible in feature dimensions. Through similarity calculation between the vector to be diagnosed and the vector in the model library, preliminary matching is performed, which can quickly narrow down the range of fault types, reduce the computational complexity of subsequent machine learning models, and improve diagnostic efficiency. The feature vector to be diagnosed and the preliminary matching results are integrated into the pre-trained model. The model calculates the probability values ​​of various defects and combines the directional guidance of the preliminary matching results to improve the accuracy of partial discharge type judgment. Compared with preliminary matching, which can only judge the type, this step can also accurately locate the discharge source and the severity of insulation.

[0123] In a preferred embodiment of the present invention, step 6 above includes:

[0124] Step 61: Read the insulation severity level from the diagnostic results and call the preset warning threshold and alarm threshold. Specifically, this includes: First, determining the level reading rules: The diagnostic results output in step 53 are structured data, containing partial discharge type, discharge source, and insulation severity. The cloud management system locates the insulation severity field through a preset interface. This field is formatted as Level I = Mild, Level II = Moderate, and Level III = Severe. The system needs to simultaneously acquire the encoding and description and temporarily cache them to avoid errors caused by missing single pieces of information; Second, determining the threshold calling basis: Warning and alarm thresholds... The thresholds are not subjectively set, but rather formulated based on the requirements of the "Coal Mine Safety Regulations" and long-term field fault data. For example, minor faults have a low probability of worsening, while severe faults are prone to causing accidents. Ultimately, the warning threshold is set to Level II (moderate) and the alarm threshold is set to Level III (severe). The thresholds must be verified by the coal mine safety supervision department for compliance. Finally, the threshold call verification is completed: after the system calls the threshold, it automatically checks whether its encoding format is consistent with the read level code. For example, if both are Level II and Level III, if there is a format discrepancy, such as the threshold being Level 2 and the level being Level II, an automatic conversion program is started until the two formats are unified.

[0125] Step 62: Compare the insulation severity level with the warning threshold and alarm threshold respectively, and determine whether the current insulation state meets the warning or alarm conditions based on the comparison results. Specifically, the comparison logic is set as follows: First, based on the progressive nature of the levels, there is a relationship between Level I and Level II, and Level II and Level III. On this basis, specific rules are formulated: When the level code is less than the warning threshold, such as the level code corresponding to Level I being less than the warning threshold corresponding to Level II, it is determined to be a normal state, and no warning or alarm operation is required; when the level code is greater than or equal to the warning threshold and less than the alarm threshold, such as the level code corresponding to Level II being greater than or equal to the warning threshold corresponding to Level II and less than the alarm threshold corresponding to Level III, it is determined to meet the warning conditions; when the level code is greater than or equal to the alarm threshold, such as the level code corresponding to Level III being greater than or equal to the alarm threshold corresponding to Level III, it is determined to meet the alarm conditions. In this way, the originally qualitative description of severity is transformed into a quantifiable relationship.

[0126] For the three levels of mild (Level I), moderate (Level II), and severe (Level III), the system compares each level with a threshold to verify the feasibility of the logic. For example, a mild level is considered normal because it is below the warning threshold, a moderate level meets the warning conditions, and a severe level meets the alarm conditions. Each level corresponds to only one result to avoid ambiguous judgments. For example, even if a Level III level meets both the warning and alarm thresholds, it is still preferentially judged as meeting the alarm conditions. At the same time, the system automatically records the comparison basis and generates logs to balance traceability and subsequent threshold optimization needs.

[0127] Step 63: If the warning conditions are met, generate a warning message; if the alarm conditions are met, generate an alarm message. Send the generated warning or alarm message to the underground mining intrinsically safe cable insulation fault monitoring host, driving its LCD screen to display the alarm status and the LED indicator to flash accordingly. Specifically, the information should include four elements: partial discharge type, discharge source, insulation severity, and maintenance instructions, based on the diagnostic and comparison results. Warning messages emphasize time-limited maintenance, such as a Level II fault requiring maintenance within 72 hours; alarm messages emphasize emergency handling, such as a Level III fault requiring immediate shutdown. After the information is generated, a type number (YJ=warning, BJ=alarm) and a timestamp are automatically added for easy storage and traceability. Relying on the underground and above-ground communication links, anti-interference single-mode dual-fiber optical cables are preferred. Set the transmission parameters: a rate of 10Mbps to ensure a delay of less than or equal to 1 second, and use CRC32 verification. The sending end calculates the verification value, and the receiving end verifies it. If the verification is inconsistent, a retransmission is requested to avoid data loss or tampering.

[0128] After confirming the completeness of the information, the intrinsically safe detection host for underground mining synchronously drives the hardware: the LCD screen displays the information type, core content, and maintenance instructions in separate lines. For example, the warning information displays the warning type as "Floating Discharge Level: Level II" and the instruction as "Maintenance within 72 hours," with a display duration of 24 hours or until manual confirmation. The LED indicator lights operate according to the type: the yellow light flashes once per second during a warning, and the red light flashes twice per second during an alarm, while the buzzer sounds intermittently, ensuring that underground maintenance personnel can detect the problem in a timely manner until the loop is closed by manual reset.

[0129] In this embodiment of the invention, by reading the insulation severity level from the diagnostic results and calling preset thresholds, a definite basis and unified standard are provided for subsequent status judgment. On the one hand, directly reading the quantified severity level avoids secondary interpretation errors of the diagnostic results. On the other hand, the preset thresholds are based on historical data of insulation faults in mining cables and customized according to coal mine safety regulations, rather than being subjectively set, ensuring that subsequent judgments meet the actual needs of underground safe production. By comparing the objective severity level with the preset thresholds one by one, it is possible to clearly define whether the current insulation status is normal, requires a warning, or requires an alarm, avoiding misjudging minor defects as requiring an alarm and failing to judge severe defects as only requiring a warning. This quantitative comparison logic improves the objectivity and consistency of the judgment. The generated warning / alarm information can be accurately sent to the underground monitoring host, avoiding information transmission delays. Through the dual methods of displaying details on the LCD screen and flashing LED indicators, it is ensured that underground maintenance personnel can quickly obtain information. The LCD screen can display specific details of the fault, while the LED indicator can quickly convey the status in the complex underground environment, allowing maintenance personnel to know the risks and locate the problems as soon as possible, buying time for timely handling.

[0130] like Figure 2As shown, embodiments of the present invention also provide an online monitoring system for insulation faults in intrinsically safe cables used in mining, comprising:

[0131] The acquisition module is used to acquire high-frequency pulse signals by coupling the partial discharge current pulse signal on the grounding wire of the cable with a mining intrinsically safe high-frequency pulse signal sensor;

[0132] A digital signal module is used to perform analog conditioning on high-frequency pulse signals to obtain conditioned analog signals. The analog conditioning includes analog filtering, preamplification, and impedance matching. The conditioned analog signals are then subjected to anti-aliasing filtering to obtain filtered signals. The filtered signals are then subjected to target frequency band selection to extract analog signals within the target frequency band. The analog signals within the target frequency band are then converted into digital signals via an A / D converter.

[0133] The extraction module is used to perform digital filtering and noise suppression on digital signals, identify and eliminate narrowband interference and random noise, enhance periodic partial discharge pulse signals, and extract signal features, including amplitude, phase, discharge quantity and occurrence frequency.

[0134] The analysis module transmits signal features to the cloud management system, performs signal denoising and time-frequency analysis through wavelet transform and short-time Fourier transform, and generates phase, amplitude, and frequency PRPD spectra. The PRPD spectra are compared with a pre-stored library of typical defects in mining cables, and a machine learning model is used to determine the type, source, and severity of partial discharge, resulting in a diagnostic result that includes the attributes and severity of partial discharge.

[0135] The execution module compares the insulation severity in the diagnostic results with preset warning thresholds and alarm thresholds, generates warning or alarm information, and sends the warning or alarm information to the downhole monitoring host for real-time display and status indication.

[0136] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for online monitoring of insulation faults in intrinsically safe mining cables, characterized in that, The method includes: Step 1: Obtain the high-frequency pulse signal by coupling the partial discharge current pulse signal on the grounding wire of the cable using a mining intrinsically safe high-frequency pulse signal sensor; Step 2: Perform analog conditioning on the high-frequency pulse signal to obtain a conditioned analog signal; the analog conditioning includes analog filtering, pre-amplification, and impedance matching; perform anti-aliasing filtering on the conditioned analog signal to obtain a filtered signal; select a target frequency band from the filtered signal to extract the analog signal within the target frequency band; convert the analog signal within the target frequency band into a digital signal through an A / D converter, including: The high-frequency pulse signal is subjected to analog filtering to remove out-of-band noise, resulting in a filtered analog signal. The filtered analog signal is then pre-amplified and impedance-matched to obtain a conditioned analog signal. The conditioned analog signal is then subjected to anti-aliasing filtering, with the cutoff frequency determined based on the sampling frequency of the subsequent A / D conversion, resulting in a filtered signal. The filtered signal is then subjected to a Fast Fourier Transform to convert the time-domain signal to a frequency-domain signal to obtain its spectrum. In the spectrum, the spectral envelope is feature-extracted, and the convex hull shape of the spectral amplitude is constructed to identify convex points in the signal energy distribution, selecting the corresponding frequencies as candidate points for the frequency band boundaries. Based on the distribution of the candidate points and the prior frequency band characteristics of the partial discharge signal of the mining cable, the upper and lower boundary frequencies of the target frequency band are finally determined, thus setting the target frequency band range. Based on the target frequency band range, a bandpass filter is used to extract the analog signal within the target frequency band from the filtered signal. The amplitude of the analog signal is equivalent to the height of a cylinder, and the duration of the analog signal within the target frequency band is equivalent to the base area of ​​the cylinder. The energy volume is obtained by multiplying the amplitude and the duration. Based on the energy volume value, the corresponding A / D converter is determined. The corresponding A / D converter is then used to convert the analog signal within the target frequency band into a digital signal. Step 3: Perform digital filtering and noise suppression on the digital signal, identify and eliminate narrowband interference and random noise, enhance the periodic partial discharge pulse signal, and extract signal features, including amplitude, phase, discharge quantity and occurrence frequency. Step 4: Transmit the signal features to the cloud management system, perform signal denoising and time-frequency analysis through wavelet transform and short-time Fourier transform, and generate phase, amplitude, and frequency PRPD spectra. Step 5: Compare the PRPD spectrum with the pre-stored typical defect partial discharge feature model library for mining cables, and use machine learning models to determine the type, source and severity of partial discharge, and obtain diagnostic results that include partial discharge attributes and severity. Step 6: Compare the insulation severity in the diagnostic results with the preset warning threshold and alarm threshold to generate warning or alarm information, and send the warning or alarm information to the downhole monitoring host for real-time display and status indication.

2. The online monitoring method for insulation faults in intrinsically safe mining cables according to claim 1, characterized in that, Step 1 includes: The initial induced signal is obtained by using the partial discharge current pulse on the grounding wire of the coupling cable; The initial induced signal is input to an intrinsically safe protection circuit, which adopts a dual-parallel structure, with each path consisting of a Zener diode and an N-MOS field-effect transistor connected in series. Based on the characteristics of the initial induced signal, the breakdown voltage of the Zener diode, the turn-on voltage of the N-MOS field-effect transistor, and the voltage drop parameter of the Zener diode in the intrinsically safe protection circuit are determined, ensuring that the breakdown voltage is greater than the sum of the turn-on voltage and the voltage drop. The intrinsically safe protection circuit configured with the voltage drop parameter performs overvoltage protection and integrity maintenance processing on the initial induced signal to obtain a protected induced signal. The protected sensing signal is used as a high-frequency pulse signal.

3. The online monitoring method for insulation faults in intrinsically safe mining cables according to claim 2, characterized in that, Step 3 includes: The digital signal is subjected to frequency domain filtering to identify and filter out stable narrowband interference, resulting in a denoised digital signal. The noise-reduced digital signal is subjected to time-domain averaging, and the data of multiple power frequency cycles are synchronously superimposed and averaged to enhance the periodically recurring partial discharge pulse signal and suppress random noise, thereby obtaining the enhanced digital signal. The enhanced digital signal is subjected to pulse detection. Potential partial discharge pulses are identified based on the signal amplitude and waveform characteristics, and grouped according to the power frequency phase in which they occur to obtain a grouped pulse sequence. A dynamic threshold is set for the grouped pulse sequence. The threshold value is adjusted according to the noise floor level and signal statistical characteristics to remove noise pulses with amplitudes lower than the threshold and select effective partial discharge pulses. The signal characteristics of the effective partial discharge pulses are extracted, including calculating the amplitude, phase, and discharge quantity of each pulse, and counting the frequency of pulse occurrence per unit time to form signal characteristics.

4. The online monitoring method for insulation faults in intrinsically safe mining cables according to claim 3, characterized in that, Step 4 includes: The signal characteristics are transmitted to the cloud management system, which receives and parses the signal characteristics to obtain a pulse data sequence containing pulse amplitude, phase and occurrence frequency. Based on the pulse data sequence, the energy distribution differences of noise and partial discharge signals at different scales are analyzed, and a denoising threshold is set according to the energy distribution differences; the pulse data sequence is processed using wavelet transform and according to the denoising threshold to filter out random noise and broadband interference, resulting in a denoised pulse data sequence. Based on the time-frequency characteristics of the denoised pulse data sequence, the sliding time window length and overlap rate required for the short-time Fourier transform are determined; the denoised pulse data sequence is segmented using a sliding time window with the specified window length and overlap rate; the segmented signal segments are subjected to a short-time Fourier transform to obtain the time-frequency distribution data of the signal. Based on the time-frequency distribution data and the phase information of the pulses, the partial discharge pulses detected in multiple power frequency cycles are statistically mapped according to their phase and amplitude to generate a PRPD spectrum with phase and amplitude as coordinates and pulse occurrence frequency as intensity.

5. The online monitoring method for insulation faults in intrinsically safe mining cables according to claim 4, characterized in that, Step 5 includes: The PRPD spectrum is input into a pre-stored partial discharge feature model library of typical defects in mining cables. The model library contains PRPD spectrum feature vectors of various typical defects. Statistical feature extraction is performed on the PRPD spectrum to form a feature vector to be diagnosed. The similarity between the feature vector to be diagnosed and the typical defect feature vector in the model library is calculated, and a preliminary matching is performed based on the principle of maximum similarity to obtain the preliminary matching results of defect type and its confidence level. The feature vector to be diagnosed and the preliminary matching result are input together into a pre-trained machine learning classification model. The model calculates the probability of belonging to various defect types based on the input feature vector, and combines the preliminary matching results to determine the partial discharge type, discharge source and insulation severity level, generating a diagnostic result that includes partial discharge attributes and severity.

6. The online monitoring method for insulation faults in intrinsically safe mining cables according to claim 5, characterized in that, Step 6 includes: Read the insulation severity level from the diagnostic results and call the preset warning threshold and alarm threshold; The insulation severity level is compared with the warning threshold and the alarm threshold respectively, and the current insulation status is determined based on the comparison results to see if the warning condition or the alarm condition has been met. If the warning conditions are met, a warning message is generated; if the alarm conditions are met, an alarm message is generated. The generated warning or alarm message is sent to the intrinsically safe cable insulation fault monitoring host for underground mining, which drives its LCD screen to display the alarm status and the LED indicator to flash accordingly.

7. An online monitoring system for insulation faults in intrinsically safe cables used in mining, wherein the system implements the method as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to acquire high-frequency pulse signals by coupling the partial discharge current pulse signal on the grounding wire of the cable with a mining intrinsically safe high-frequency pulse signal sensor; A digital signal module is used to perform analog conditioning on high-frequency pulse signals to obtain conditioned analog signals; the analog conditioning includes analog filtering, pre-amplification, and impedance matching; the conditioned analog signals are then subjected to anti-aliasing filtering to obtain filtered signals. The filtered signal is subjected to target frequency band selection to extract the analog signal within the target frequency band; the analog signal within the target frequency band is converted into a digital signal through A / D analog-to-digital conversion; The extraction module is used to perform digital filtering and noise suppression on digital signals, identify and eliminate narrowband interference and random noise, enhance periodic partial discharge pulse signals, and extract signal features, including amplitude, phase, discharge quantity and occurrence frequency. The analysis module transmits signal features to the cloud management system, performs signal denoising and time-frequency analysis through wavelet transform and short-time Fourier transform, and generates phase, amplitude, and frequency PRPD spectra. The PRPD spectra are compared with a pre-stored library of typical defects in mining cables, and a machine learning model is used to determine the type, source, and severity of partial discharge, resulting in a diagnostic result that includes the attributes and severity of partial discharge. The execution module compares the insulation severity in the diagnostic results with preset warning thresholds and alarm thresholds, generates warning or alarm information, and sends the warning or alarm information to the downhole monitoring host for real-time display and status indication.

8. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images