A high-temperature eddy current signal adaptive filtering and feature extraction method and system

Abstract

The high-temperature eddy current signal adaptive filtering and feature extraction method and system provided by the present disclosure adopts an improved adaptive noise complete ensemble empirical mode decomposition to adaptively decompose a noisy eddy current signal, and obtains a series of intrinsic mode functions; uses a spectral correlation density function in the cyclostationarity theory to perform cyclostationary coherence analysis on each IMF component, identifies fault-related and interference-related cyclic frequency components; realizes adaptive IMF selection and cyclic domain filtering based on the cyclic feature difference, reconstructs the signal after noise reduction and enhances the fault features. The technology can significantly improve the signal-to-noise ratio in a strong electromagnetic interference environment, effectively extract weak fault features submerged by noise, and has preliminary identification ability of interference sources.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal processing and fault diagnosis, and particularly relates to a high-temperature eddy current signal adaptive filtering and feature extraction method and system. BACKGROUND

[0002] In modern industrial production, high-temperature eddy current sensors are widely used in shaft vibration monitoring of key rotating equipment such as steam turbines, generators and compressors. However, these sensors face serious electromagnetic interference challenges in actual industrial field. Especially in the installation position close to large power equipment, the sensor and its signal transmission link are easily polluted by various complex electromagnetic noises, including power frequency and its high harmonic interference, high-frequency pulse cluster interference, wideband random noise, switching frequency modulation interference, etc. These interferences have the characteristics of time-varying and non-stationary, and often seriously overwhelm the characteristic signals related to early micro-faults of the equipment, such as the vibration characteristics generated by micro-crack propagation, initial imbalance, early rub-impact and bearing pitting. Traditional fixed parameter filtering methods, such as band-pass filter, band-stop filter, mean filter, etc., have poor effect in processing such complex interferences, which may cause excessive attenuation of useful signals or incomplete filtering of noises, seriously affecting the measurement accuracy and the reliability and sensitivity of subsequent fault diagnosis.

[0003] Although the existing adaptive filtering technology can process time-varying noise to some extent, it is still difficult to achieve ideal signal-noise separation effect in a strong electromagnetic interference environment due to the overlap of signal and noise in frequency domain and the complexity of noise. Therefore, there is an urgent need to develop a new technology that can effectively extract weak fault features in high-temperature eddy current signals in a strong electromagnetic interference environment. SUMMARY

[0004] The first aspect of the present disclosure provides a high-temperature eddy current signal adaptive filtering and feature extraction method, comprising the following steps:

[0005] S1: using an improved adaptive noise complete ensemble empirical mode decomposition ICEEMDAN Adaptive decomposition is performed on the noisy eddy current signal, and the amplitude of the auxiliary white noise is adaptively adjusted according to the signal-to-noise ratio estimation result of the input signal σ and the number of integrations N , to obtain intrinsic mode functions IMF 1 to IMF n ;

[0006] S2: calculating the spectral correlation density function IMF of each S(f,α) component, wherein f is the spectral frequency, α is the cyclic frequency, the expected fault feature cyclic frequency set is determined according to the device rotating speed and the fault type, and eachIMF Energy distribution characteristics at the fault cycle frequency and the interference cycle frequency;

[0007] S3: Fault correlation selection based on cyclic spectrum energy threshold criterion IMF Components, for mixing IMF The components are adaptively filtered in the cyclic frequency domain to retain fault-related cyclic frequency components and suppress interference-related cyclic frequency components.

[0008] S4: Filter and select the results IMF The components are superimposed and reconstructed to obtain the denoised eddy current signal, and the type of interference source is identified based on the cyclic frequency characteristics of the suppressed components.

[0009] In conjunction with the first aspect, S1 employs an improved adaptive noise-complete ensemble empirical mode decomposition. ICEEMDAN Adaptive decomposition of noisy eddy current signals includes:

[0010] When estimating the signal-to-noise ratio SNR < 0 dB At that time, number of integrations N= 500+100 × |SNR| Noise standard deviation σ= 0.2 × std (x) ;

[0011] When 0 dB ≤ SNR <10 dB hour, N= 300+50 ×( 10 -SNR) , σ= 0.1 × std(x) ;

[0012] when SNR ≥10 dB hour, N =200, σ=0.05× std ( x ),in std ( x ) represents the standard deviation of the input signal.

[0013] Combining the first aspect, the spectral correlation density function in S2 S ( f The calculation of α) employs an algorithm combining time smoothing and frequency smoothing, where the number of time smoothing windows... L satisfy L ≥4×( f h / f l ), f h and fl These are the highest and lowest characteristic frequencies of the signal, respectively, to ensure that the variance of the cyclic spectrum estimation is less than 0.1.

[0014] In conjunction with the first aspect, the set of expected fault characteristic cycle frequencies determined in S2 based on equipment speed and fault type includes:

[0015] For rotation frequency is f r Equipment, imbalance fault corresponding f r Misalignment fault correspondence f r and 2 f r Corresponding to bearing inner ring fault

[0016] Bearing outer ring fault corresponding

[0017] ,in N For the number of rolling elements, d The diameter of the rolling element, D The bearing pitch circle diameter, φ It represents the contact angle.

[0018] In conjunction with the first aspect, S3 addresses the hybridization... IMF Adaptive filtering of components in the cyclic frequency domain based on the cyclic spectral coherence function γ ( f Filtering criteria for α):

[0019] when γ ( f ,α)>γ th And α∈ Α fault When the frequency component is retained, when γ( f ,α)>γ th And α∈ Α noise Suppress this frequency component, where γ th The coherence threshold, Α fault For the set of fault cycle frequencies, Α noise This is a set of interference cycle frequencies.

[0020] In conjunction with the first aspect, the method further includes, after S3, when a certain cycle frequency α is continuously M Within a given analysis period, frequencies identified as interference and with a probability exceeding the threshold were considered interference frequencies. P th When, add it Α noise ,inM =5, P th =0.8.

[0021] In combination with the first aspect, S4 specifically includes:

[0022] When detecting 50 Hz or 60 Hz and integer multiple harmonics thereof, it is determined as power frequency interference;

[0023] When detecting carrier frequencies and sidebands thereof in the range of 1 kHz kHz to 10 kHz, it is determined as frequency converter switching interference;

[0024] When detecting wideband random features, it is determined as arc or discharge interference.

[0025] In combination with the second aspect of the present disclosure, a high-temperature eddy current signal adaptive filtering and feature extraction system is provided, comprising:

[0026] A signal acquisition module is configured to acquire a high-temperature eddy current sensor output signal;

[0027] A preprocessing module is configured to perform signal conditioning and signal-to-noise ratio estimation;

[0028] ICEEMDAN A decomposition module is configured to perform adaptive signal decomposition;

[0029] A cyclic spectrum analysis module is configured to IMF cyclic feature recognition of components;

[0030] An adaptive filtering module is configured to perform signal reconstruction based on cyclic features;

[0031] A feature extraction module is configured to perform fault feature enhancement and interference source identification.

[0032] In combination with the third aspect of the present disclosure, an electronic device is provided, comprising:

[0033] one or more processors;

[0034] a storage unit configured to store one or more programs, which when executed by the one or more processors, enable the one or more processors to implement the high-temperature eddy current signal adaptive filtering and feature extraction method.

[0035] In combination with the fourth aspect of the present disclosure, a computer-readable storage medium is provided, which stores a computer program, which when executed by a processor, enables the high-temperature eddy current signal adaptive filtering and feature extraction method.

[0036] Beneficial Effects: This disclosure provides an adaptive filtering and feature extraction method and system for high-temperature eddy current signals. It employs an improved adaptive noise-complete integrated empirical mode decomposition (ICEEMDAN) to adaptively decompose noisy eddy current signals, obtaining a series of intrinsic mode functions (IMFs). Cyclic spectral coherence analysis is performed on each IMF component using the spectral correlation density function from cyclostationarity theory to identify fault-related and interference-related cyclic frequency components. Adaptive IMF selection and cyclic domain filtering are implemented based on cyclic feature differences, reconstructing the denoised signal and enhancing fault characteristics. This technology can significantly improve the signal-to-noise ratio under strong electromagnetic interference environments, effectively extract weak fault features submerged by noise, and possess preliminary interference source identification capabilities. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating an adaptive filtering and feature extraction method for high-temperature eddy current signals according to an embodiment of this disclosure.

[0038] Figure 2 This is a schematic diagram of the structure of a high-temperature eddy current signal adaptive filtering and feature extraction system according to an embodiment of the present disclosure;

[0039] Figure 3 An electronic device according to an embodiment of this disclosure. Detailed Implementation

[0040] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those disclosed herein.

[0041] The specific implementation of this invention adopts a system architecture that combines hardware and software.

[0042] The hardware system mainly includes core components such as high-temperature eddy current sensors, signal conditioning circuits, high-speed data acquisition cards, and industrial computers. These hardware devices work together to ensure the stability and reliability of the entire signal chain from acquisition to processing.

[0043] The software algorithm part has implemented an improved version. ICEEMDAN The core functional modules, such as decomposition, cyclic spectrum coherence analysis, and adaptive filtering reconstruction, achieve intelligent processing of eddy current signals under strong electromagnetic interference environments through a carefully designed algorithm process.

[0044] The system architecture adopts a layered design concept, with clear functions and standardized interfaces at each layer, from the underlying hardware drivers to the upper-level user interface. The data flow starts from the sensor, goes through signal conditioning, digital acquisition, preprocessing, and core algorithm processing, and finally outputs the noise-reduced signal and fault diagnosis results.

[0045] like Figure 1 The diagram shown is a flowchart illustrating an adaptive filtering and feature extraction method for high-temperature eddy current signals according to an embodiment of this disclosure, including:

[0046] S1: Employing an improved adaptive noise-complete ensemble empirical mode decomposition. ICEEMDAN Adaptive decomposition is performed on the noisy eddy current signal, and the amplitude of the auxiliary white noise is adaptively adjusted based on the signal-to-noise ratio estimation result of the input signal. σ and number of integrations N Obtain the intrinsic mode functions IMF 1 to IMF n ;

[0047] The improved adaptive noise complete integration empirical mode decomposition ICEEMDAN The adaptive decomposition in the model includes:

[0048] When estimating the signal-to-noise ratio SNR < 0 dB At that time, number of integrations N= 500+100 × |SNR| Noise standard deviation σ= 0.2 × std (x) ;

[0049] When 0 dB ≤ SNR <10 dB hour, N= 300+50 ×( 10 -SNR) , σ= 0.1 × std(x) ;

[0050] when SNR ≥10 dB hour, N =200, σ=0.05× std ( x ),in std ( x ) represents the standard deviation of the input signal.

[0051] Specifically, improved ICEEMDAN The algorithm adaptively adjusts the decomposition parameters based on the characteristics of the input signal. First, based on the signal-to-noise ratio estimation obtained in the preprocessing stage, the algorithm intelligently determines the standard deviation of the auxiliary noise. σ and number of integrations N When the signal-to-noise ratio is low, the system automatically increases the number of integration iterations to improve the stability and reliability of the decomposition; when the signal complexity is high, the noise amplitude is appropriately increased to effectively avoid mode aliasing. This adaptive strategy significantly improves the performance of traditional... ICEEMDANDecomposition performance of the algorithm in strong interference environment.

[0052] The specific implementation process of the algorithm includes generating N different white noise sequences, calculating the corresponding modal components for each noise implementation, and then obtaining the final eigenmode function through ensemble averaging. The calculation formula of the first modal is

[0053]

[0054] , wherein M 1 represents the first local mean function. Through the average of N different noise implementation results, a stable and reliable

[0055] is obtained.

[0056] Then, the first residual

[0057]

[0058] is calculated, and the above process is repeated until the residual becomes a monotonic function.

[0059] In order to further improve the decomposition effect, the key parameters of the algorithm are deeply optimized. The adaptive adjustment of noise amplitude follows a piecewise function strategy: when ICEEMDAN <0 SNR ≤ 1, set dB

[0060]

[0061]

[0062] When in the range of 0 ≤ SNR <10 dB , set

[0063]

[0064]

[0065] When SNR ≥ 10 dB , set

[0066]

[0067]

[0068] The stopping criterion adopts an improved standard deviation criterion, which stops the screening process when the standard deviation of the screening results of two consecutive times is less than a preset threshold of 0.2, which not only ensures the integrity of the decomposition, but also effectively avoids the phenomenon of over-decomposition.​

[0069] S2: calculating the spectral correlation density function of each IMF component S(f,α) , f where α is the spectral frequency, and the expected fault feature cyclic frequency set is determined according to the device rotating speed and the fault type, and the energy distribution characteristics at the fault cyclic frequency and the interference cyclic frequency are analyzed. IMF

[0070] The calculation of the spectral correlation density function S ( f , α) adopts an algorithm combining time smoothing and frequency smoothing, wherein the number of time smoothing windows L satisfies L ≥ 4 × ( f h / f l ), f h and f l are respectively the highest and lowest feature frequencies of the signal, so as to ensure that the cyclic spectrum estimation variance is less than 0.1.

[0071] The expected fault feature cyclic frequency set is determined according to the device rotating speed and the fault type, which includes:

[0072] For a device with a rotating frequency of f r , the unbalance fault corresponds to f r , the misalignment fault corresponds to f r and 2 f r , the bearing inner ring fault corresponds to the bearing inner ring fault

[0073] , the bearing outer ring fault corresponds to

[0074] , wherein N is the number of rolling bodies, d is the rolling body diameter, D is the bearing pitch diameter, φ is the contact angle.

[0075] Specifically, the cyclic spectrum coherence analysis utilizes the difference between the mechanical fault signal and the electromagnetic interference signal in the cyclic stationary characteristics for intelligent identification. For each IMF component, the system calculates its spectral correlation density function

[0076] ​

[0077] where X T f denotes the Fourier transform of the signal within a time window T α is the cyclic frequency parameter,

[0078]

[0079] denotes the correlation of the signal at frequency f with cyclic frequency a, divided by T for normalization, E（） denotes the expected value. When a = 0, SCD degenerates to the conventional power spectral density (PSD), which represents the total energy distribution of the signal; when a ≠ 0, SCD reflects the spectral characteristics of the periodic modulation components in the signal; SCD can be regarded as the "cyclic power spectrum" of the signal, which can highlight the signal components with periodic modulation characteristics.

[0080] In practical calculation, an efficient algorithm combining time smoothing and frequency smoothing is adopted, which divides the signal into L overlapping time windows, calculates the short-time Fourier transform for each time window, then calculates the Fourier transform of the cyclic correlation function, and finally performs statistical averaging on the results of multiple time windows.

[0081] Cyclic frequency feature recognition is the key link for the success of the algorithm. The system establishes a complete fault feature cyclic frequency database according to the operating parameters of the equipment. For common faults of rotating machinery, unbalance fault corresponds to the rotational frequency f r , misalignment fault corresponds to f r and 2 f r , mechanical looseness corresponds to f r , 2 f r , 3 f r and other multiple harmonic components, oil film whirl corresponds to the specific frequency range of 0.4 f r to 0.48 f r . The calculation of cyclic frequency of bearing fault is more complex, bearing inner ring fault corresponds to

[0082] , bearing outer ring fault corresponds to

[0083] , and rolling element fault corresponds to

[0084]

[0085] ​​Cage failure response

[0086]

[0087] ,in N For the number of rolling elements, d The diameter of the rolling element, D The bearing pitch circle diameter, φ It represents the contact angle.

[0088] The cyclic frequency characteristics of electromagnetic interference also exhibit obvious regularity; power frequency interference shows a 50° cycle frequency. Hz Or 60 Hz And its integer multiples of harmonics, the frequency converter interference manifests as the carrier frequency being added or subtracted. k In a frequency-doubling sideband structure, switching power supply interference manifests as the switching frequency and its harmonic components. Cyclic spectrum coherence function.

[0089]

[0090] The calculation results vary between 0 and 1, with values ​​closer to 1 indicating frequency. f The presence of strong α-cyclic frequency components provides a reliable basis for subsequent signal-noise separation.

[0091] Circular spectrum coherence function threshold γ th The determination is based on the Neyman-Pearson criterion, by setting the false alarm probability. P fa =0.01, calculated as follows

[0092]

[0093] ,in L The number of time smoothing windows.

[0094] For degrees of freedom L The chi-square distribution has a 0.01 quantile. This threshold setting method ensures the reliability of fault feature identification in noisy environments.

[0095] S3: Fault correlation selection based on cyclic spectrum energy threshold criterion IMF Components, for mixing IMF The components are adaptively filtered in the cyclic frequency domain to retain fault-related cyclic frequency components and suppress interference-related cyclic frequency components.

[0096] The pair of mixtures IMF Adaptive filtering of components in the cyclic frequency domain based on the cyclic spectral coherence function γ ( f Filtering criteria for α):

[0097] whenγ ( f ,α)>γ th And α∈ Α fault When the frequency component is retained, when γ( f ,α)>γ th And α∈ Α noise Suppress this frequency component, where γ th The coherence threshold, Α fault For the set of fault cycle frequencies, Α noise This is a set of interference cycle frequencies.

[0098] The method further includes, after S3, when a certain cycle frequency α is continuously... M Within a given analysis period, frequencies identified as interference and with a probability exceeding the threshold were considered interference frequencies. P th When, add it Α noise ,in M =5, P th =0.8.

[0099] Specifically, adaptive filtering and signal reconstruction achieve accurate signal and noise separation based on cyclic spectrum analysis results. The system first separates all signals based on the cyclic spectrum analysis results. IMF The components are intelligently classified into three main categories: fault-dominant, interference-dominant, and hybrid. Fault-dominant... IMF The characteristic is that γ²( at the fault characteristic cycle frequency) f ,α) is greater than the threshold γ th At the interference cycle frequency, γ²( f ,α) is less than the threshold; interference-dominant type IMF They exhibit the opposite characteristics; mixed type IMF It exhibits a strong response at both fault and disturbance cycle frequencies. Classification threshold γ th An adaptive adjustment strategy is adopted, and the calculation formula is as follows:

[0100]

[0101] This ensures that the best classification results can be obtained under different signal-to-noise ratio conditions.

[0102] For mixed types IMF The components are then subjected to refined filtering in the cyclic frequency domain. Cyclic frequency domain filter. H ( f The design of α) adopts a piecewise function form, when α Belongs to the set of fault cycle frequencies andγ²(f,α) When α is greater than the threshold, the filter output is 1, completely preserving that frequency component; when α belongs to the noise cycle frequency set and γ²(f,α) When the frequency exceeds the threshold, the filter output is 0, completely suppressing that frequency component; for the transition region, a smooth transition weighting function is used.

[0103]

[0104] This avoids abrupt changes during the filtering process. After filtering... IMF The components are then transformed back to the time domain via the inverse Fourier transform, i.e. IMF'(t) = IFFT(FFT(IMF(t)) × H(f,α)) .

[0105] The signal reconstruction process employs a weighted superposition strategy, combining the classified and filtered signals... IMF According to the portion

[0106] ,

[0107] Reconstruct the weights. w i The determination is based on each IMF The energy percentage at the fault cycle frequency is calculated using the following formula:

[0108]

[0109] ,in E i,fault and E i,noise They represent the first i indivual IMF Energy at the fault and noise cycle frequencies, ε To prevent division by zero by small constants, this weighting strategy ensures that fault-related components dominate the reconstructed signal while minimizing the influence of interfering components.

[0110] S4: Filter and select the results. IMF The components are superimposed and reconstructed to obtain the denoised eddy current signal, and the type of interference source is identified based on the cyclic frequency characteristics of the suppressed components.

[0111] The filtered and sorted IMF The components are superimposed and reconstructed to obtain the denoised eddy current signal, and the interference source type is identified based on the cyclic frequency characteristics of the suppressed components, including:

[0112] When 50 is detected Hz Or 60 Hz When the harmonics are integer multiples of their harmonics, they are determined to be power frequency interference;

[0113] When the number is detected kHzThe carrier frequency and its sideband within the range of 1 kHz to 20 kHz kHz is determined as the interference of frequency converter switch;

[0114] The arc or discharge interference is determined when the broadband random feature is detected.

[0115] Specifically, the interference source identification is realized by analyzing the features of the suppressed IMF component and the cyclic frequency component, and intelligently identifying the main electromagnetic interference source type.

[0116] Firstly, the multi-dimensional feature vector is extracted from the suppressed IMF component, including the frequency domain features such as the main frequency component, the spectral shape parameter, the bandwidth feature, etc., the cyclic domain features such as the main cyclic frequency, the cyclic spectrum peak value, the cyclic frequency interval, etc., and the time domain features such as the amplitude distribution statistics, the pulse characteristic index, the intermittence feature parameter, etc. These feature parameters constitute the basic data set for the identification of the interference source.

[0117] Based on years of engineering practice experience and theoretical analysis, the system establishes a complete interference source classification rule library. The identification rule of power frequency interference is that when the main cyclic frequency belongs to the power frequency harmonic sequence of 50 Hz , 100 Hz , 150 Hz , etc., or the 60 Hz , 120 Hz , 180 Hz , etc. 60 Hz system harmonic sequence, it is determined as power frequency and harmonic interference. The identification of frequency converter interference is more complex, which needs to meet the existence of carrier frequency kHz kHz within the range of 1 f c kHz, and the existence of sideband frequency structure in the form of f c ± k × f r , where k is a positive integer, f r is the frequency conversion. The feature of switch power supply interference is that the main frequency is concentrated in the range of 20 kHz to 100 kHz kHz, the spectrum presents obvious peak feature, and is accompanied by high harmonic components.

[0118] The arc or discharge interference has unique broadband random characteristics, the frequency spectrum presents a broadband distribution, the time domain signal has obvious pulse characteristics, and the amplitude distribution presents a heavy tail characteristic. The combination of these characteristics can effectively identify the interference source. The system also has learning ability, can continuously improve and update the identification rule library according to the actual interference situation on site, and improve the accuracy and adaptability of identification. The interference source identification result not only provides guidance for signal processing, but also provides important technical support for on-site electromagnetic compatibility improvement and interference source treatment.

[0119] As shown in Figure 2 Fig. 1 is a structural schematic diagram of a high-temperature eddy current signal adaptive filtering and feature extraction system according to an embodiment of the present disclosure, which comprises:

[0120] The signal acquisition module 210 is configured to acquire a high-temperature eddy current sensor output signal.

[0121] The preprocessing module 220 is configured to perform signal conditioning and signal-to-noise ratio estimation.

[0122] ICEEMDAN The decomposition module 230 is configured to perform adaptive signal decomposition.

[0123] The cyclic spectrum analysis module 240 is configured to IMF cyclic feature recognition of components;

[0124] The adaptive filtering module 250 is configured to perform signal reconstruction based on the cyclic features.

[0125] The feature extraction module 260 is configured to perform fault feature enhancement and interference source identification.

[0126] Specifically, the signal acquisition module 210 is responsible for directly connecting the high-temperature eddy current sensor to acquire the original analog electrical signal reflecting the vibration state of the device. The core of this module is to complete the acquisition of the signal with high fidelity. Usually, it includes a high-precision analog-to-digital converter (ADC) and necessary hardware interfaces to ensure that the original signal is not distorted during digitization and provides a reliable data source for all subsequent advanced processing.

[0127] The preprocessing module 220 receives the original digital signal from the acquisition module. It includes removing DC bias, eliminating trend items, and performing necessary amplification or scaling to standardize the signal. More importantly, this module needs to estimate the signal-to-noise ratio (SNR) of the input signal in real time. This estimate is the core basis for the adaptive parameter adjustment of the subsequent ICEEMDAN decomposition module and directly affects the effectiveness of the entire noise reduction process.

[0128] The ICEEMDAN decomposition module 230 receives the preprocessed signal and dynamically adjusts the number of integrations and noise amplitude according to the signal-to-noise ratio information provided by the preprocessing module, and performs improved adaptive noise complete ensemble empirical mode decomposition. The function of this module is to adaptively decompose the complex and non-stationary original signal into a series of intrinsic mode functions (IMF) with different physical meanings from high frequency to low frequency, so as to preliminarily separate different components in the signal.

[0129] The cyclic spectrum analysis module 240 performs deep feature mining on the series of IMF components after decomposition. It calculates the spectral correlation density function of each IMF component using the cyclostationary theory, analyzes the energy distribution and coherence at specific fault cycle frequencies (such as rotational frequency, bearing fault frequency) and known interference cycle frequencies (such as power frequency and its harmonics). The output of this module is essentially a "fault-related" or "interference-related" label for each IMF component, providing direct evidence for the next filtering decision.

[0130] The adaptive filtering module 250 is the decision and reconstruction center based on the results of the cyclic spectrum analysis. According to the feature labels provided by the cyclic spectrum analysis, this module performs intelligent filtering strategy: retains and strengthens the IMF components judged as fault dominant, directly eliminates the IMF components judged as interference dominant, and performs fine filtering in the cycle frequency domain for mixed IMF components containing both fault and interference components. Finally, it superimposes and reconstructs all the processed effective IMF components to output the final denoised high-quality eddy current signal.

[0131] The feature extraction module 260 is the final link of the system value realization. On the one hand, it further enhances the fault features of the denoised reconstructed signal (such as envelope demodulation, order analysis) to highlight the early weak fault features for easy diagnosis and identification. On the other hand, it intelligently identifies the type of interference source (for example, whether it is power frequency interference, frequency converter interference or arc discharge interference) based on the cycle frequency characteristics of the interference components identified and suppressed in the entire processing process, provides the user with fault diagnosis conclusions and interference source analysis report, and forms a closed loop.

[0132] The electronic device 300 can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The electronic device 300 can include but is not limited to a processor 301 and a memory 302. Those skilled in the art can understand that the processor 301 can be a central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or the like. Figure 3 The electronic device 300 is only an example and does not constitute a limitation on the electronic device 300, and can include more or fewer components than illustrated, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, and the like.

[0133] The processor 301 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0134] The memory 302 can be an internal storage unit of the electronic device 300, for example, a hard disk or a memory of the electronic device 300. The memory 302 can also be an external storage device of the electronic device 300, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 300. Further, the memory 302 can also include both the internal storage unit and the external storage device of the electronic device 300. The memory 302 is used to store the computer program 303 and other programs and data required by the electronic device. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0135] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than limit them; although the present disclosure has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent ones; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure.

Claims

1. A method for adaptive filtering and feature extraction of high-temperature eddy current signals, characterized in that, Includes the following steps: S1: Employing an improved adaptive noise-complete ensemble empirical mode decomposition. ICEEMDAN Adaptive decomposition is performed on the noisy eddy current signal, and the amplitude of the auxiliary white noise is adaptively adjusted based on the signal-to-noise ratio estimation result of the input signal. σ and number of integrations N Obtain the intrinsic mode functions IMF 1 to IMF n ; S2: For each IMF Component calculation of spectral correlation density function S(f,α) ,in f For spectral frequency, α As the cycle frequency, a set of expected fault characteristic cycle frequencies is determined based on equipment speed and fault type, and the various cycles are analyzed. IMF Energy distribution characteristics at the fault cycle frequency and the interference cycle frequency; S3: Fault correlation selection based on cyclic spectrum energy threshold criterion IMF Components, for mixing IMF The components undergo adaptive filtering in the cyclic frequency domain, retaining fault-related cyclic frequency components and suppressing interference-related cyclic frequency components. The adaptive filtering criterion for the hybrid IMF components in the cyclic frequency domain, based on the cyclic spectral coherence function γ(f,α), is as follows: when γ ( f ,α)>γ th And α∈ A fault When the frequency component is retained, when γ( f ,α)>γ th And α∈ A noise Suppress this frequency component, where γ th The coherence threshold, A fault For the set of fault cycle frequencies, A noise For the set of interfering cycle frequencies; S4: Filter and select the results IMF The components are superimposed and reconstructed to obtain the denoised eddy current signal. The interference source type is then identified based on the cyclic frequency characteristics of the suppressed components, specifically including: When 50 is detected Hz Or 60 Hz When the harmonics are integer multiples of their harmonics, they are considered power frequency interference. When 1 is detected kHz Up to 20 kHz The carrier frequency fc within the specified range, and the presence of a sideband frequency structure in the form of fc±k×fr, are identified as inverter switching interference, where k is a positive integer and fr is the switching frequency. When broadband random features are detected, they are identified as arc or discharge interference.

2. The method according to claim 1, characterized in that, S1 employs an improved adaptive noise complete ensemble empirical mode decomposition. ICEEMDAN Adaptive decomposition of noisy eddy current signals includes: When estimating the signal-to-noise ratio SNR<0dB At that time, number of integrations N= 500+100 ×|SNR| Noise standard deviation σ= 0.2 ×std(x) ; When 0 dB ≤ SNR <10 dB hour, N= 300+50 ×( 10 -SNR) , σ= 0.1 ×std(x) ; when SNR ≥10 dB hour, N =200, σ=0.05× std ( x ),in std ( x ) represents the standard deviation of the input signal.

3. The method according to claim 1, characterized in that, Spectral correlation density function in S2 S ( f The calculation of α) employs an algorithm combining time smoothing and frequency smoothing, where the number of time smoothing windows... L satisfy L ≥4×( f h / f l ), f h and f l These are the highest and lowest characteristic frequencies of the signal, respectively, to ensure that the variance of the cyclic spectrum estimation is less than 0.

1.

4. The method according to claim 1, characterized in that, S2 determines the set of expected fault characteristic cycle frequencies based on equipment speed and fault type, including: For rotation frequency is f r Equipment, imbalance fault corresponding f r Misalignment fault correspondence f r and 2 f r Corresponding to bearing inner ring fault Bearing outer ring fault corresponding ,in N For the number of rolling elements, d The diameter of the rolling element, D The bearing pitch circle diameter, φ It represents the contact angle.

5. The method according to claim 1, characterized in that, The method further includes, after S3, when a certain cycle frequency α is continuously... M Within a given analysis period, frequencies identified as interference and with a probability exceeding the threshold were considered interference frequencies. P th When, add it A noise ,in M =5, P th =0.

8.

6. A high-temperature eddy current signal adaptive filtering and feature extraction system, used to implement the method described in any one of claims 1-5, characterized in that, include: The signal acquisition module is used to acquire the output signal of the high-temperature eddy current sensor; The preprocessing module is used for signal conditioning and signal-to-noise ratio estimation; ICEEMDAN Decomposition module for adaptive signal decomposition; Cyclic spectrum analysis module, used for IMF Cyclic feature identification of components; An adaptive filtering module is used for signal reconstruction based on cyclic features; The feature extraction module is used for fault feature enhancement and interference source identification.

7. An electronic device, characterized in that, include: One or more processors; A storage unit is used to store one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the adaptive filtering and feature extraction method for high-temperature eddy current signals as described in claim 1.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can implement the adaptive filtering and feature extraction method for high-temperature eddy current signals according to claim 1.

Images