Electrical equipment fault analysis method and system based on big data

Abstract

The invention relates to the field of fault analysis, and discloses an electrical equipment fault analysis method and system based on big data. The method comprises the following steps: extracting a starting current spectrum and a steady-state current spectrum in a non-salt-fog environment from a historical library, and training corresponding fault diagnosis models by using a one-dimensional convolutional neural network and a two-dimensional convolutional neural network respectively; the method comprises the following steps: acquiring a real-time starting current frequency spectrum, environmental parameters, equipment operation duration, salt mist corrosion characteristic parameters and conductive ion conductivity parameters, compensating the real-time starting current frequency spectrum, eliminating pseudo harmonic peaks and baseline offset distortion caused by corrosion of copper ions by salt mist and corrosion products thereof, and obtaining a compensated starting current frequency spectrum; in the starting stage, the fault probability weight is identified through full-frequency scanning, and FPGA resources are dynamically allocated; and acquiring a steady-state frequency spectrum for fault detection in an operation stage. According to the method, the influence caused by the salt mist environment is eliminated, the fault misjudgment rate is reduced, and the resource occupancy rate is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault analysis, in particular to an electrical equipment fault analysis method and system based on big data. BACKGROUND

[0002] Electrical equipment is one of the important components of a ship, and its main function is to provide power for machinery operation. During the operation of electrical equipment, it is inevitably affected by both internal and external factors, and faults may occur. The occurrence of ship electrical faults directly affects the safety of navigation.

[0003] Under certain corrosive conditions, such as severe salt spray corrosion environment, salt spray particles will deposit on the insulating surface of electrical equipment, forming a thin electrolyte film with high conductivity. At the same time, copper ions in the corrosion products also have certain conductivity, further deteriorating the insulation performance. The decrease of insulation resistance will result in that the damping formed by the resistance of the winding itself and the insulation resistance is not enough to effectively suppress the resonance loop, and then a pseudo-resonance peak of non-fundamental frequency is generated. The traditional electrical equipment fault analysis method compensates or analyzes the fundamental wave, which cannot effectively eliminate the pseudo-resonance peak induced by the corrosion environment, resulting in an increase in the misdiagnosis rate of fault diagnosis.

[0004] Therefore, it is urgent to develop an electrical equipment fault analysis method and system based on big data with environmental compensation capability to meet the severe requirements of the salt spray scene. SUMMARY

[0005] (1) Technical problem to be solved

[0006] The purpose of the present application is to provide an electrical equipment fault analysis method and system based on big data to eliminate the non-fundamental pseudo-resonance peak in the current signal caused by the decrease of insulation resistance under salt spray environment.

[0007] (2) Technical scheme

[0008] To achieve the above purpose, on the one hand, the present application provides an electrical equipment fault analysis method based on big data, characterized in that the method comprises:

[0009] Step S1: extracting the starting current spectrum and its corresponding fault type under non-salt spray environment from the historical fault library to form the first current fault data, and extracting the steady-state current spectrum and its corresponding fault type to form the second current fault data; after preprocessing the first current fault data, using one-dimensional convolutional neural network to train the starting current fault diagnosis model to output the current frequency band and the fault type association probability; after preprocessing the second current fault data, using two-dimensional convolutional neural network to train the steady-state current fault diagnosis model to output the corresponding association probability.

[0010] Step S2: Obtain the equipment operating time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment; obtain the conductivity correction value based on the copper ion concentration of the corrosion product, the environmental parameters, and the salt spray corrosion characteristic parameters; and compensate the real-time starting current spectrum to eliminate the pseudo-resonance peak and baseline offset distortion caused by the salt spray and its corrosion product copper ions to obtain a compensated starting current spectrum.

[0011] Step S3: Perform a full-band scan to obtain a real-time starting current spectrum, input the compensated starting current spectrum into the starting current fault diagnosis model, and identify the fault type probability weight corresponding to the fault frequency band; dynamically allocate FPGA computing resources based on the fault type probability weight, collect the real-time steady-state current spectrum, and input it into the steady-state current fault diagnosis model for fault detection.

[0012] Furthermore, the method of obtaining the equipment operating time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment includes:

[0013] Get the device running time and environmental parameters, including chloride ion concentration and ambient temperature ;Concentration of copper ions as corrosion products in salt spray environment for: .

[0014] Furthermore, the method for obtaining the conductivity correction value according to the copper ion concentration of the corrosion product, the environmental parameters and the conductive ion conductivity parameters includes: obtaining the conductive ion conductivity parameters, including the chloride ion conductivity coefficient and the conductivity coefficient of copper ions of corrosion products The conductivity correction amount for: .

[0015] Furthermore, the method of obtaining the equipment operation time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment also includes: real-time monitoring of the insulation resistance change rate: When the monitored insulation resistance change rate exceeds the preset change rate threshold, the copper ion concentration correction mechanism is automatically activated; the salt spray corrosion characteristic parameters are obtained, and the corrected corrosion product copper ion concentration is generated based on the coupling relationship between humidity change and resistance change: ; The corrected copper ion concentration of the corrosion product Replace the original corrosion product copper ion concentration Conductivity correction amount Calculate; where, is the concentration correction coefficient in the salt spray corrosion characteristic parameters; is the attenuation ratio of the copper ion concentration of the corrosion product caused by the sudden change of humidity.

[0016] Further, the real-time starting current spectrum is compensated to eliminate the pseudo-resonance peak and baseline offset distortion caused by salt spray and its corrosion product copper ions, and the method for obtaining the compensated starting current spectrum includes: obtaining the real-time starting current spectrum and salt spray corrosion characteristic parameters, and correcting the real-time starting current spectrum according to the conductivity correction amount

[0017] is a frequency variable, representing a frequency domain coordinate axis of the current spectrum analysis; is a chloride ion influence factor in the salt spray corrosion characteristic parameters; is a resonance bandwidth in the salt spray corrosion characteristic parameters; is a resonance center frequency in the salt spray corrosion characteristic parameters; and α is a corrosion product copper ion conductivity correction coefficient in the salt spray corrosion characteristic parameters; is an initial insulation resistance in the salt spray corrosion characteristic parameters; is an amplitude-frequency response exponentially decaying around is a pseudo-resonance peak amplitude; is an insulation resistance change rate; is a baseline offset compensation term.

[0018] Further, the real-time starting current spectrum is compensated to eliminate the pseudo-resonance peak and baseline offset distortion caused by salt spray and its corrosion product copper ions, and the method for obtaining the compensated starting current spectrum includes: obtaining the real-time starting current spectrum and salt spray corrosion characteristic parameters, and correcting the real-time starting current spectrum according to the conductivity correction amount Further, the real-time starting current spectrum is compensated to eliminate the pseudo-resonance peak and baseline offset distortion caused by salt spray and its corrosion product copper ions, and the method for obtaining the compensated starting current spectrum includes: obtaining the real-time starting current spectrum and salt spray corrosion characteristic parameters, and correcting the real-time starting current spectrum according to the conductivity correction amount When ;

[0019] The updated initial insulation resistance is calculated as follows: ; wherein is a calibration coefficient in the salt spray corrosion characteristic parameters; and is replaced by for subsequent compensation calculation, and when , the calibration is ended.

[0020] ​​​​​​​​​Further, after the first current fault data is preprocessed, a one-dimensional convolutional neural network is used to train a starting current fault diagnosis model capable of outputting a current frequency band and fault type associated probability; after the second current fault data is preprocessed, a two-dimensional convolutional neural network is used to train a steady-state current fault diagnosis model capable of outputting a corresponding associated probability, including: the starting current spectrum is divided into different frequency bands according to different fault types j Segments; the starting current spectrum is directly normalized after segmentation to obtain one-dimensional input data; the frequency band feature map output by the one-dimensional convolutional neural network is globally averaged and pooled along the frequency domain dimension to generate a fault sensitive feature vector for each frequency band ; and the starting associated probability matrix is mapped through a fully connected layer .

[0021] The steady-state current spectrum is divided into different frequency bands according to different fault types j Segments; the steady-state current spectrum is converted into a time-frequency matrix through a short-time Fourier transform, and the time domain and frequency domain are independently normalized to obtain two-dimensional input data; the frequency band feature map output by the two-dimensional convolutional neural network is globally averaged and pooled along the time domain and frequency domain dimensions to generate a fault sensitive feature vector for each frequency band ; and the steady-state associated probability matrix is mapped through a fully connected layer .

[0022] Further, the dynamic allocation of FPGA computing resources includes: according to the fault type probability weight obtained in the starting stage, the FPGA computing resources are first allocated according to the fault type probability weight : The fault type probability weight is compared with a preset first threshold value and a second threshold value to divide the FPGA computing cores; subsequently, the steady-state associated probability matrix is updated according to the real-time diagnosis result of the steady-state current fault diagnosis to generate a frequency band priority weight : The frequency band priority weight is compared with a preset third threshold value and a fourth threshold value to divide the FPGA computing cores;

[0023] When the frequency band priority weight change rate exceeds the preset priority threshold value within 5 consecutive cycles , the FPGA resource reallocation is triggered.

[0024] Based on the same inventive concept, in another aspect, the application also provides a big data-based electrical equipment fault analysis system, which comprises a fault diagnosis model module, a data acquisition module, a corrosion compensation module and a phased diagnosis module, and the modules are sequentially and communicatively connected.

[0025] The fault diagnosis model module extracts starting current spectrum and corresponding fault types in a non-salt spray environment from a historical fault library to form first current fault data, and extracts steady-state current spectrum and corresponding fault types to form second current fault data; after preprocessing the first current fault data, a one-dimensional convolutional neural network is used to train a starting current fault diagnosis model that outputs current frequency band and fault type association probability; after preprocessing the second current fault data, a two-dimensional convolutional neural network is used to train a steady-state current fault diagnosis model that outputs corresponding association probability.

[0026] The corrosion compensation module obtains the device runtime and environmental parameters to obtain the corrosion product copper ion concentration in a salt spray environment, obtains the conductivity correction amount according to the corrosion product copper ion concentration, environmental parameters and salt spray corrosion characteristic parameters, compensates the real-time starting current spectrum, eliminates the pseudo-resonance peak and baseline offset distortion caused by the salt spray and its corrosion product copper ion, and obtains the compensated starting current spectrum.

[0027] The phased diagnosis module is used to perform full-band scanning to obtain a real-time starting current spectrum, input the compensated starting current spectrum into the starting current fault diagnosis model, identify the fault type probability weight corresponding to the fault frequency band, dynamically allocate FPGA computing resources according to the fault type probability weight, collect real-time steady-state current spectrum, input the steady-state current fault diagnosis model for fault detection.

[0028] Further, the corrosion compensation module further comprises an insulation resistance correction unit configured to, when and , the degradation compensation formula is: ; calculate the updated initial insulation resistance: ; wherein, is a calibration coefficient in the salt spray corrosion characteristic parameters; replace with for subsequent compensation calculation, and when is detected, the calibration is ended.

[0029] The humidity impact correction unit is configured to monitor the insulation resistance change rate in real time: When the insulation resistance change rate is monitored to exceed the preset change rate threshold, the copper ion concentration correction mechanism is automatically activated; the salt spray corrosion characteristic parameters are obtained, and the corrected corrosion product copper ion concentration is generated according to the coupling relationship between humidity change and resistance change: ; the corrected corrosion product copper ion concentration Substitute original corrosion product copper ion concentration Participate in conductivity correction amount Calculate; wherein, It is the concentration correction coefficient in the salt spray corrosion characteristic parameter; It is the decay ratio of the corrosion product copper ion concentration caused by humidity mutation.

[0030] (3) beneficial effects

[0031] Compared with the prior art, the beneficial effects of the present application are: the present application quantifies the interference of the spectrum according to the salt spray corrosion mechanism, compensates the real-time starting current spectrum, eliminates the pseudo-resonance peak and baseline offset distortion caused by the salt spray and its corrosion product copper ion, and combines with phased diagnosis, more efficiently and accurately analyzes the electrical equipment fault in the offshore salt spray environment. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 It is an electrical equipment fault analysis method based on big data according to the embodiment 1 of the present application; Figure 2 It is an electrical equipment fault analysis system based on big data according to the embodiment 2 of the present application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0034] Before examples are given, the application scenarios of the present application concept need to be described. The present application mainly faces the electrical equipment fault analysis in high salt spray corrosion environment such as ships, offshore platforms, etc. In the salt spray environment, an electrolyte film will be formed on the surface of the insulation resistance, and after pitting the insulation surface, corrosion product copper ions with conductivity will be generated, which will interact and affect the insulation resistance, and then the resistance will not be able to damp the generation of harmonics, so that the collected current spectrum will cover the real fault characteristics.

[0035] Embodiment 1: as Figure 1As shown, the embodiment provides a big data-based electrical equipment fault analysis method, which comprises the following steps: S1: extracting starting current spectrum and corresponding fault type in a non-salt spray environment from a historical fault library to form first current fault data, and extracting steady-state current spectrum and corresponding fault type to form second current fault data; after preprocessing the first current fault data, a one-dimensional convolutional neural network is used to train a starting current fault diagnosis model that outputs current frequency band and fault type association probability; after preprocessing the second current fault data, a two-dimensional convolutional neural network is used to train a steady-state current fault diagnosis model that outputs corresponding association probability; S2: obtaining equipment running time and environmental parameters to obtain the concentration of corrosion product copper ions in a salt spray environment; obtaining the conductivity correction amount according to the concentration of corrosion product copper ions, environmental parameters and salt spray corrosion characteristic parameters; compensating the real-time starting current spectrum to eliminate the pseudo-resonance peak and baseline offset distortion caused by salt spray and its corrosion product copper ions, and obtaining the compensated starting current spectrum; S3: performing full-band scanning to obtain a real-time starting current spectrum, inputting the compensated starting current spectrum into the starting current fault diagnosis model to identify the fault type probability weight corresponding to the fault frequency band; dynamically allocating FPGA computing resources according to the fault type probability weight, collecting real-time steady-state current spectrum and inputting it into the steady-state current fault diagnosis model for fault detection.

[0036] For example, taking a certain ship propulsion motor as an example: the equipment running time is 10 years, the environmental chloride ion concentration is 0.5 mol / L, and the environmental temperature is 25℃.

[0037] The chloride ion concentration is 0.5 mol / L, which is converted to molar concentration as needed for calculation: , where 35.45 is the molar mass of chloride ion , in particular, the unit is mol / L in the corrosion rate calculation , and the unit is mol / L in the conductivity calculation .

[0038] The copper ion concentration is 0.5 mol / L, which is converted to mol / L for conductivity calculation: . .

[0039] The salt spray corrosion characteristic parameters are calibrated through an acceleration experiment: the experimental equipment is a salt spray chamber with specifications conforming to ASTM B117-19 standard; the temperature range is set to 25±2℃ ​​​​​​50 ± 2 °C; salt spray deposition rate set to 1.5 ± 0.5 mL / 80 cm2-h; humidity controlled at 30% to 98% RH, ± 3% fluctuation; copper conductor with mica insulation stator winding of marine motor was prepared before the experiment, the surface was sandblasted to ISO 8501-1 Sa2.5 level, and each group of repeated samples was 6 times in the experiment.

[0040] Chloride ion influencing factor , resonance bandwidth , resonance center frequency Obtained from experiments, and the experimental data are shown in Table 1:

[0041]

[0042] The copper ion conductivity coefficient a is obtained from electrochemical impedance spectroscopy experiments, and the equipment uses a PARSTAT 4000 electrochemical workstation; the equivalent circuit parameters are measured in a 0.01-0.05 mol / L copper ion solution, and then the formula is used to derive: .

[0043] The humidity resistance coupling coefficient η is obtained from experiments, and the experimental data are shown in Table 2:

[0044]

[0045] Chloride ion conductivity coefficient and corrosion product copper ion conductivity coefficient Obtained from experiments, and the experimental data are shown in Table 3:

[0046]

[0047] Further, the method for obtaining the device running time and environmental parameters to obtain the corrosion product copper ion concentration in a salt spray environment includes: obtaining the device running time and environmental parameters, wherein the environmental parameters include the chloride ion concentration and the environmental temperature ; the corrosion product copper ion concentration in a salt spray environment is:

[0048] .

[0049] Exemplary, wherein the conversion chloride ion concentration unit , .

[0050] In particular, the coefficient 0.18 is obtained by ASTM B117 salt spray cabin weight loss experiment, the salt spray deposition rate is set to 1.5 mL / 80 cm2-h, and the value range is 0.17 0.19, the error is less than 2.1%; the exponential term coefficient -0.015 is obtained through the temperature-varying corrosion current experiment, and the temperature is set at 25 The value range at 50℃ is -0.014 -0.016, error less than 1.8%. Data based on six sets of repeated samples, salt spray concentration 3.0 5.5 mg / m 3 , in compliance with IEC 60068-2-52 standard.

[0051] Furthermore, the method for obtaining the conductivity correction value according to the copper ion concentration of the corrosion product, the environmental parameters and the conductive ion conductivity parameters includes: obtaining the conductive ion conductivity parameters, including the chloride ion conductivity coefficient and the conductivity coefficient of copper ions of corrosion products The conductivity correction amount for:

[0052] .

[0053] For example, the chloride ion conductivity coefficient is , conductivity coefficient of copper ions of corrosion products , , corresponding conductivity correction amount:

[0054] .

[0055] Furthermore, the method of obtaining the equipment operation time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment also includes: real-time monitoring of the insulation resistance change rate: When the monitored insulation resistance change rate exceeds the preset change rate threshold, the copper ion concentration correction mechanism is automatically activated; the salt spray corrosion characteristic parameters are obtained, and the corrected corrosion product copper ion concentration is generated based on the coupling relationship between humidity change and resistance change: ; The corrected copper ion concentration of the corrosion product Replace the original corrosion product copper ion concentration Conductivity correction amount Calculate; where, is the concentration correction coefficient in the salt spray corrosion characteristic parameters; is the attenuation ratio of the copper ion concentration of the corrosion product caused by the sudden change of humidity.

[0056] For example, a case study on the diagnosis of a sudden drop in humidity in a cargo ship's motor: when a cold front passes through the sea, within 10 minutes From 320MΩ to 650MΩ; real-time change rate of insulation resistance: .

[0057] Value ; Values are adjusted according to the hygroscopic properties of the material; in particular in marine propulsion electric machines, .

[0058] ; ; mA / Hz.

[0059] Although the amplitude of 3.7 kHz after compensation is only from 0.41 mA / Hz to 0.4086 mA / Hz, the probability of bearing damage is increased to 91.5% after the elimination of the fault characteristic frequency band pseudo-resonance peak, and only 65.3% without correction, which results in 3 weeks earlier warning of bearing failure and avoids shutdown accidents.

[0060] Real-time insulation resistance Measurement: When the equipment is running, a 1 Hz low-frequency alternating current detection signal with superimposed amplitude ≤10 V is added, and the amplitude A and phase of the leakage current are calculated: . .

[0061] The insulation resistance change rate is calculated using the first-order difference method: , where the time unit is seconds, and the trigger condition is: Corresponding to A decrease of >25% within 5 minutes.

[0062] Further, the method for compensating the real-time starting current spectrum to eliminate the pseudo-resonance peak and baseline offset distortion caused by salt spray and its corrosion product copper ions includes: obtaining the real-time starting current spectrum and the salt spray corrosion characteristic parameters, and according to the conductivity correction amount The real-time starting current spectrum and the salt spray corrosion characteristic parameters are compensated to obtain the compensated starting current spectrum : .

[0063] , where is a frequency variable, representing the frequency domain coordinate axis of the current spectrum analysis; is the chloride ion influence factor in the salt spray corrosion characteristic parameters; is the resonance bandwidth in the salt spray corrosion characteristic parameters; is the resonance center frequency in the salt spray corrosion characteristic parameters; α is the corrosion product copper ion conductivity correction coefficient in the salt spray corrosion characteristic parameters; is the initial insulation resistance in the salt spray corrosion characteristic parameters; is an amplitude-frequency response that decays exponentially with as the center. is the pseudo-resonance peak amplitude; is the insulation resistance change rate; is the baseline offset compensation term.

[0064] Exemplarily, the compensation parameter takes the chloride ion influence factor mA·m³ / Hz·mg; resonance bandwidth Hz, resonance center frequency ; copper ion conductance S·m, initial insulation resistance ; real-time insulation resistance .

[0065] At the frequency , the compensation calculation is: ; the pseudo-resonance peak with an amplitude of 0.815 mA / Hz is eliminated, and the baseline is compressed by 0.24%.

[0066] Further, the correction amount according to the conductivity The real-time starting current spectrum and the salt spray corrosion characteristic parameter are obtained by compensation processing to obtain the compensated starting current spectrum Further, the compensation formula is: and , the degradation compensation formula is: ;

[0067] The updated initial insulation resistance is calculated: ; wherein, is the calibration coefficient in the salt spray corrosion characteristic parameter; replace with for subsequent compensation calculation, and when is detected, the calibration is ended.

[0068] Exemplarily, when the humidity suddenly drops, the real-time insulation resistance is greater than the initial value , mA / Hz, take =0.9, and perform insulation calibration .

[0069] Wherein, the calibration coefficient is determined by historical data regression.

[0070] Further, the calculation of the compensated starting current spectrum needs to meet ; when is detected, it is considered that the device is in a non-salt spray state, and the original spectrum is directly used for diagnosis.

[0071] The compensation processing only deducts the spectrum component with the center, bandwidth of the fault peak, and other frequency band features containing the real fault peak are not modified.

[0072] Further, after the pre-processing of the first current fault data, a one-dimensional convolutional neural network is used to train a starting current fault diagnosis model capable of outputting the probability of association between the current frequency band and the fault type; after the pre-processing of the second current fault data, a two-dimensional convolutional neural network is used to train a steady-state current fault diagnosis model capable of outputting the corresponding association probability, including: the starting current spectrum is divided into segments according to different frequency bands of different fault types j; the amplitude in the frequency domain is directly normalized after the starting current spectrum is segmented, as one-dimensional input data; the frequency band feature map output by the one-dimensional convolutional neural network is globally averaged and pooled along the frequency domain dimension to generate a fault sensitive feature vector for each frequency band; and a starting association probability matrix is mapped through a fully connected layer.

[0073] The steady-state current spectrum is divided into segments according to different frequency bands of different fault types j; the steady-state current spectrum is converted into a time-frequency matrix through a short-time Fourier transform, and the time domain and the frequency domain are independently normalized as two-dimensional input data; the frequency band feature map output by the two-dimensional convolutional neural network is globally averaged and pooled along the time domain and the frequency domain dimensions to generate a fault sensitive feature vector for each frequency band; and a steady-state association probability matrix is mapped through a fully connected layer.

[0074] For example, the starting current model inputs 4 frequency bands, normalized amplitude values; a one-dimensional convolutional neural network structure is used, which includes 3 layers of convolution (kernel size = 5, step = 1), global average pooling, a fully connected layer containing two layers of 128 to 64 neurons, and a cross-entropy loss function; and a starting association probability matrix is output, which represents the association probability of 4 frequency bands and 5 types of faults.

[0075] The steady-state current model inputs 5 frequency bands, an STFT time-frequency matrix, with a 100ms window length and a 50% overlap; a two-dimensional convolutional neural network structure is used, which includes 4 layers of convolution (kernel size = 3x3), time-frequency domain global average pooling, a fully connected layer containing two layers of 256 to 128 neurons, and a focal loss function; and a steady-state association probability matrix is output, which represents the real-time association probability of 5 frequency bands and 5 types of faults.

[0076] Further, the dynamic allocation of FPGA computing resources includes: according to the fault type probability weight obtained in the starting stage​ , first according to the fault type probability weight Allocate FPGA computing resources: ; Fault type probability weight Compare to the preset first threshold and the second threshold , divide the FPGA computing core; then update the steady-state correlation probability matrix according to the real-time diagnosis results of steady-state current fault diagnosis , generate the frequency band priority weight : ; Band priority weight Compare to the preset third threshold and the fourth threshold , divide the FPGA computing core; when 5 consecutive cycles are detected When the change rate of the intra-band priority weight exceeds the preset priority threshold, FPGA resource reallocation is triggered.

[0077] For example, the first threshold of resource allocation is determined by 200 sets of marine motor fault injection simulation experiments. =0.4, second threshold =0.7, the third threshold and the fourth threshold , experimental data are shown in Table 4:

[0078]

[0079] Startup phase: Band 3 First compare with the first threshold 0.4, > 0.4, continue to compare with the second threshold 0.7, > 0.7 to allocate the first-level monitoring channel to use 4 cores for real-time monitoring; Operation phase: Band 2 ; Priority ; Assign the secondary monitoring channel to use 2-core monitoring.

[0080] When frequency band 5 has 5 consecutive cycles middle When the monitoring channel is reached, it will be immediately reallocated to the first-level monitoring channel.

[0081] The preset priority threshold The following experiment was used to determine the priority: 12 types of faults were injected into the ship motor fault simulation platform, and after counting the priority change rate data 500 times, the lower limit of the 95% confidence interval was taken.

[0082] Example 2: Based on the same inventive concept, Figure 2As shown, the embodiment also provides a big data-based electrical equipment fault analysis system, which comprises a fault diagnosis model module, a data acquisition module, a corrosion compensation module, and a phased diagnosis module, and the modules are sequentially and communicatively connected.

[0083] The fault diagnosis model module extracts, from a historical fault library, a starting current spectrum and corresponding fault types in a non-salt spray environment to form first current fault data, and a steady-state current spectrum and corresponding fault types to form second current fault data; after preprocessing the first current fault data, a one-dimensional convolutional neural network is used to train a starting current fault diagnosis model that outputs current frequency bands and fault type association probabilities; after preprocessing the second current fault data, a two-dimensional convolutional neural network is used to train a steady-state current fault diagnosis model that outputs corresponding association probabilities.

[0084] The corrosion compensation module obtains the equipment runtime and environmental parameters to obtain the corrosion product copper ion concentration in a salt spray environment, obtains the conductivity correction amount according to the corrosion product copper ion concentration, the environmental parameters, and the salt spray corrosion characteristic parameters, performs compensation processing on the real-time starting current spectrum, eliminates the pseudo-resonance peak and baseline offset distortion caused by the salt spray and its corrosion product copper ions, and obtains the compensated starting current spectrum.

[0085] The phased diagnosis module is used to perform full-band scanning to obtain a real-time starting current spectrum, input the compensated starting current spectrum into the starting current fault diagnosis model, identify the fault type probability weight corresponding to the fault frequency band, dynamically allocate FPGA computing resources according to the fault type probability weight, collect a real-time steady-state current spectrum, and input the steady-state current fault diagnosis model for fault detection.

[0086] Further, the corrosion compensation module further comprises an insulation resistance correction unit configured to, when and , the degradation compensation formula is: ; calculate the updated initial insulation resistance: ; wherein, is a calibration coefficient in the salt spray corrosion characteristic parameters; replace with for subsequent compensation calculation, and when is detected, the calibration is ended.

[0087] The humidity impact correction unit is configured to monitor the insulation resistance change rate in real time: When the insulation resistance change rate is monitored to exceed a preset change rate threshold, the copper ion concentration correction mechanism is automatically activated; the salt spray corrosion characteristic parameters are obtained, and the corrected corrosion product copper ion concentration is generated according to the coupling relationship between humidity change and resistance change: ; the corrected corrosion product copper ion concentration Substitute original corrosion product copper ion concentration Participate in conductivity correction amount Calculate; wherein, is the concentration correction coefficient in the salt spray corrosion characteristic parameter; is the decay ratio of the corrosion product copper ion concentration caused by the humidity mutation.

[0088] It should be noted that, as for the system in the above-mentioned embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0089] Finally, it should be noted that: although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing electrical equipment failure based on big data, characterized in that: The method includes: extracting a starting current spectrum and its corresponding fault type in a non-salt fog environment from a historical fault library to form first current fault data, and a steady-state current spectrum and its corresponding fault type to form second current fault data; after preprocessing the first current fault data, using a one-dimensional convolutional neural network to train and generate a starting current fault diagnosis model that outputs the probability of association between current frequency bands and fault types; after preprocessing the second current fault data, using a two-dimensional convolutional neural network to train and generate a steady-state current fault diagnosis model that outputs the corresponding association probability; obtaining equipment operating time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt fog environment; obtaining a conductivity correction value based on the copper ion concentration of the corrosion product, the environmental parameters, and salt fog corrosion characteristic parameters; compensating the real-time starting current spectrum to eliminate pseudo-resonance peaks and baseline offset distortion caused by salt fog and its corrosion product copper ions to obtain a compensated starting current spectrum; performing a full-band scan to obtain a real-time starting current spectrum, inputting the compensated starting current spectrum into the starting current fault diagnosis model, and identifying the fault type probability weight corresponding to the fault frequency band; dynamically allocating FPGA computing resources based on the fault type probability weight, collecting the real-time steady-state current spectrum, and inputting it into the steady-state current fault diagnosis model for fault detection.

2. The electrical equipment fault analysis method based on big data according to claim 1, characterized in that: The method of obtaining the equipment running time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment includes: obtaining the equipment running time and environmental parameters, including chloride ion concentration and ambient temperature ;Concentration of copper ions as corrosion products in salt spray environment for: .

3. The electrical equipment fault analysis method based on big data according to claim 2, characterized in that: The method for obtaining the conductivity correction value according to the copper ion concentration of the corrosion product, the environmental parameters and the conductivity parameters of the conductive ions includes: obtaining the conductivity parameters of the conductive ions, including the chloride ion conductivity coefficient and the conductivity coefficient of copper ions of corrosion products The conductivity correction amount for: .

4. The electrical equipment fault analysis method based on big data according to claim 3, characterized in that: The method of obtaining the equipment operation time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment further includes: real-time monitoring of the insulation resistance change rate: When the monitored insulation resistance change rate exceeds the preset change rate threshold, the copper ion concentration correction mechanism is automatically activated; the salt spray corrosion characteristic parameters are obtained, and the corrected corrosion product copper ion concentration is generated based on the coupling relationship between humidity change and resistance change: ; The corrected copper ion concentration of the corrosion product Replace the original corrosion product copper ion concentration Conductivity correction amount Calculate; where, is the concentration correction coefficient in the salt spray corrosion characteristic parameters; is the attenuation ratio of the copper ion concentration of the corrosion product caused by the sudden change of humidity.

5. The electrical equipment fault analysis method based on big data according to claim 4, characterized in that: The method for compensating the real-time starting current spectrum to eliminate the pseudo-resonance peak and baseline offset distortion caused by salt spray and its corrosion product copper ions to obtain the compensated starting current spectrum includes: obtaining the real-time starting current spectrum and salt spray corrosion characteristic parameters, and correcting the conductivity according to the conductivity correction value. The real-time starting current spectrum The compensation starting current spectrum is obtained by compensating the characteristic parameters of salt spray corrosion for: ;in, is the frequency variable, representing the frequency domain coordinate axis of current spectrum analysis; is the chloride ion influencing factor in the characteristic parameters of salt spray corrosion; is the resonance bandwidth in the characteristic parameters of salt spray corrosion; is the resonant center frequency in the salt spray corrosion characteristic parameters; α is the corrosion product copper ion conductivity correction coefficient in the salt spray corrosion characteristic parameters; is the initial insulation resistance in the salt spray corrosion characteristic parameters; For The amplitude-frequency response is exponentially decaying at the center; is the pseudo-resonance peak amplitude; is the insulation resistance change rate; is the baseline offset compensation term.

6. The electrical equipment fault analysis method based on big data according to claim 5, characterized in that: The conductivity correction amount The real-time starting current spectrum The compensation starting current spectrum is obtained by compensating the characteristic parameters of salt spray corrosion Also includes: and When , the degradation compensation formula is: ; Calculate the updated initial insulation resistance: ;in, is the calibration coefficient in the characteristic parameters of salt spray corrosion; Replace with Perform subsequent compensation calculations, when it is detected When the calibration is completed.

7. The electrical equipment fault analysis method based on big data according to claim 1, characterized in that: After preprocessing the first current fault data, a one-dimensional convolutional neural network is used to train and generate a starting current fault diagnosis model that can output the probability of association between current frequency bands and fault types; After preprocessing the second current fault data, a two-dimensional convolutional neural network is used to train and generate a steady-state current fault diagnosis model that can output the corresponding correlation probability. The starting current spectrum is divided into different frequency bands according to different fault types j. After the starting current spectrum is segmented, the frequency domain amplitude is directly normalized as one-dimensional input data; for the frequency band feature map output by the one-dimensional convolutional neural network, global average pooling is performed along the frequency domain dimension to generate each frequency band. Fault-sensitive feature vector of ; Steady-state current spectrum is divided into different frequency bands according to different fault types j. The steady-state current spectrum is converted into a time-frequency matrix through short-time Fourier transform, and the time domain and frequency domain are independently normalized as two-dimensional input data. The frequency band feature map output by the two-dimensional convolutional neural network is subjected to global average pooling along the time domain and frequency domain dimensions to generate each frequency band. Fault-sensitive feature vector of .

8. The electrical equipment fault analysis method based on big data according to claim 1, characterized in that: The dynamic allocation of FPGA computing resources in step S3 includes: , first according to the fault type probability weight Allocate FPGA computing resources: ; Fault type probability weight Compare to the preset first threshold and the second threshold , divide the FPGA computing core; then update the steady-state correlation probability matrix according to the real-time diagnosis results of steady-state current fault diagnosis , generate the frequency band priority weight : ; Band priority weight Compare to the preset third threshold and the fourth threshold , divide the FPGA computing core; when 5 consecutive cycles are detected When the change rate of the intra-band priority weight exceeds the preset priority threshold, FPGA resource reallocation is triggered.

9. An electrical equipment fault analysis system based on big data, characterized in that: The system includes: a fault diagnosis model module, a data acquisition module, a corrosion compensation module, and a staged diagnosis module, and each module is sequentially connected to each other; the fault diagnosis model module extracts the starting current spectrum and its corresponding fault type in a non-salt spray environment from the historical fault library to form the first current fault data, and the steady-state current spectrum and its corresponding fault type to form the second current fault data; after preprocessing the first current fault data, a one-dimensional convolutional neural network is used to train and generate a starting current fault diagnosis model that outputs the probability of association between the current frequency band and the fault type; after preprocessing the second current fault data, a two-dimensional convolutional neural network is used to train and generate a steady-state current fault diagnosis model that outputs the corresponding association probability; the corrosion compensation module obtains the equipment operation frequency spectrum and the corresponding fault type. The starting current spectrum is compensated based on the running time and environmental parameters to obtain the copper ion concentration of the corrosion product in the salt spray environment. The conductivity correction amount is obtained according to the copper ion concentration of the corrosion product, environmental parameters and salt spray corrosion characteristic parameters. The pseudo-resonance peak and baseline offset distortion caused by the salt spray and its corrosion product copper ions are eliminated to obtain the compensated starting current spectrum. A staged diagnosis module is used to perform full-band scanning to obtain the real-time starting current spectrum, input the compensated starting current spectrum into the starting current fault diagnosis model, and identify the fault type probability weight corresponding to the fault frequency band; FPGA computing resources are dynamically allocated according to the fault type probability weight, and the real-time steady-state current spectrum is collected and input into the steady-state current fault diagnosis model for fault detection.

10. The electrical equipment fault analysis system based on big data according to claim 9, characterized in that: The corrosion compensation module also includes an insulation resistance correction unit for and When , the degradation compensation formula is: ; Calculate the updated initial insulation resistance: ;in, is the calibration coefficient in the characteristic parameters of salt spray corrosion; Replace with Perform subsequent compensation calculations, when it is detected When the humidity shock correction unit is used to monitor the insulation resistance change rate in real time: When the monitored insulation resistance change rate exceeds the preset change rate threshold, the copper ion concentration correction mechanism is automatically activated; the salt spray corrosion characteristic parameters are obtained, and the corrected corrosion product copper ion concentration is generated based on the coupling relationship between humidity change and resistance change: ; The corrected copper ion concentration of the corrosion product Replace the original corrosion product copper ion concentration Conductivity correction amount Calculate; where, is the concentration correction coefficient in the salt spray corrosion characteristic parameters; is the attenuation ratio of the copper ion concentration of the corrosion product caused by the sudden change of humidity.

Images