Nuclear power electric rotating equipment bearing early fault online monitoring and intelligent diagnosis method

Abstract

The invention belongs to the field of nuclear power electric rotating equipment bearings, and relates to a nuclear power electric rotating equipment bearing early fault on-line monitoring and intelligent diagnosis method comprising the following steps: collecting acoustic emission signals generated during bearing operation; performing filtering, noise reduction and gain preprocessing on the acoustic emission signal; parameters of the acoustic emission signals are extracted according to waveform features, statistical features and energy features of the acoustic emission signals, and change features, distribution features and correlation features of the parameters of the acoustic emission signals under different working conditions and fault types are analyzed; performing time-frequency analysis on the acquired acoustic emission signals by adopting wavelet transform, optimizing the scale of the wavelet transform and the waveform parameters of the base wavelet by utilizing a genetic algorithm, and extracting the time-frequency characteristics of the acoustic emission signals; building a bearing fault diagnosis model by using a convolutional neural network and a support vector machine, and training and testing the bearing fault diagnosis model; and carrying out online monitoring on the nuclear power electric rotating equipment bearing by using the trained fault diagnosis model.

Description

Technical Field

[0001] The present invention belongs to the field of nuclear power electric rotating equipment bearings, and in particular relates to an online monitoring and intelligent diagnosis method for early faults of nuclear power electric rotating equipment bearings. Background Art

[0002] Electric rotating equipment is an important component of nuclear power plants, and bearings are key. They are subjected to harsh working environments such as high temperature, high pressure, high speed, and high radiation. Their operating status directly affects the safety and reliability of nuclear power plants. Therefore, it is a key technical issue for nuclear power plants to conduct online early fault monitoring and intelligent diagnosis of nuclear electric rotating equipment bearings, and to promptly detect and handle electric rotating equipment bearing failures.

[0003] Acoustic emission (AE) is a fault detection and diagnosis technology that uses elastic wave signals generated by an object when subjected to external forces. It is non-destructive, contactless, sensitive, and real-time, making it suitable for online fault monitoring and intelligent diagnosis of bearings in nuclear power electric rotating equipment. AE signals contain rich information about bearing faults in electric rotating equipment. By collecting, preprocessing, analyzing, and identifying AE signals, different types of bearing faults in nuclear power electric rotating equipment, such as pitting, wear, cracks, and spalling, can be effectively detected and identified.

[0004] Existing AE-based online monitoring and intelligent diagnosis methods for bearing faults in nuclear power electric rotating equipment mainly include the following: AE parameter-based methods, AE waveform-based methods, AE time-frequency-based methods, and AE artificial intelligence-based methods. These methods each have their own advantages and disadvantages, but they also have the following shortcomings: AE parameter-based methods can only extract the statistical features of AE signals, ignoring the waveform and time-frequency features of AE signals, resulting in poor recognition results for complex AE signals; AE waveform-based methods can only extract the time domain features of AE signals, ignoring the frequency domain and time-frequency features of AE signals, resulting in poor analysis results for non-stationary AE signals; AE time-frequency-based methods can only extract the frequency domain and time-frequency features of AE signals, ignoring the time domain and parameter features of AE signals, resulting in poor extraction results for multi-scale AE signals; AE artificial intelligence-based methods can only classify AE signals using artificial intelligence algorithms, ignoring the feature extraction and optimization of AE signals, resulting in poor learning results for high-dimensional AE signals. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for online monitoring and intelligent diagnosis of early faults in bearings of nuclear power electric rotating equipment. The method comprehensively utilizes the parameter characteristics, time-frequency characteristics, and artificial intelligence technology of acoustic emission signals to effectively detect and identify different types of faults in nuclear power electric rotating equipment bearings, thereby improving the safety and reliability of nuclear power plants.

[0006] The technical solution for achieving the purpose of the present invention is as follows:

[0007] A method for online monitoring and intelligent diagnosis of early-stage bearing failures of nuclear power electric rotating equipment, the method comprising:

[0008] Step 1: Acoustic emission signal collection: Use the acoustic emission sensor to collect the acoustic emission signal generated by the bearing during operation;

[0009] Step 2: Acoustic emission signal preprocessing: filtering, noise reduction, and gain preprocessing of the acoustic emission signal;

[0010] Step 3: Acoustic emission signal parameter analysis: According to the waveform characteristics, statistical characteristics, and energy characteristics of the acoustic emission signal, the parameters of the acoustic emission signal are extracted from the acoustic emission signal, and the change characteristics, distribution characteristics, and correlation characteristics of the parameters of the acoustic emission signal under different working conditions and fault types are analyzed;

[0011] Step 4: Acoustic emission signal waveform analysis: Use wavelet transform to perform time-frequency analysis on the collected acoustic emission signal, use genetic algorithm to optimize the scale of wavelet transform and base wavelet waveform parameters, and extract the time-frequency characteristics of the acoustic emission signal;

[0012] Step 5: Bearing fault diagnosis model construction: Use convolutional neural network and support vector machine to build a bearing fault diagnosis model, and train and test the bearing fault diagnosis model;

[0013] Step 6: Application of bearing fault diagnosis model: Use the trained fault diagnosis model to conduct online monitoring of the bearings of nuclear power electric rotating equipment, receive acoustic emission signals in real time, extract features, input the model, output the fault type and severity, and provide fault alarms and maintenance suggestions.

[0014] The acoustic emission sensor in step 1 has a broadband response characteristic of above 100 kHz, a sensitivity of above 10 dB, and a heat-resistant temperature of above 150° C.

[0015] The sampling frequency range of the acoustic emission sensor in step 1 is 2.5 MHz to 5 MHz, the sampling length range is 10 ms to 100 ms, the sampling mode is continuous sampling or triggered sampling, and the sampling accuracy is 12 bits to 16 bits.

[0016] In the step 2, a bandpass filter is used to filter the acoustic emission signal. The passband range of the filter is set to 100kHz-400kHz, the stopband suppression is above 40dB, the passband flatness is within ±1dB, the filter type is IIR type, and the filter order is 4-8.

[0017] In the step 2, wavelet threshold noise reduction is used to reduce the noise of the acoustic emission signal.

[0018] In the step 2, the adaptive gain is used to amplify the acoustic emission signal.

[0019] The parameters of the acoustic emission signal in step three include: ring count, root mean square, signal amplitude, and kurtosis coefficient of the acoustic emission signal.

[0020] The analysis of the parameters of the acoustic emission signal in step three includes: time series analysis, frequency domain analysis, statistical analysis, and correlation analysis.

[0021] In the step 4, wavelet transform is used to perform time-frequency analysis on the collected acoustic emission signal, and the wavelet transform coefficient is calculated according to the collected acoustic emission signal. The calculation formula of the wavelet transform coefficient is:

[0022]

[0023] in, Represents the continuous-time acoustic emission signal; Indicates acoustic emission signal Wavelet transform coefficients of Represents the scale parameter in wavelet transform, which controls the frequency and time resolution of the wavelet; Represents the time shift parameter in wavelet transform, which controls the translation of wavelet on the time axis; Indicates the mother wavelet function in wavelet transform, the superscript represents the complex conjugate, Indicates that the mother wavelet function is scaled and time-shifted to obtain the daughter wavelet function; Indicates continuous integration, which performs integration operations on the signal over the entire time axis.

[0024] The formula of the mother wavelet function in the wavelet transform in step 4 is:

[0025]

[0026] in, is the bandwidth parameter of the wavelet, is the center frequency of the wavelet, is the imaginary unit ( = -1).

[0027] The time-frequency characteristics of the acoustic emission signal in step 4 include: time-frequency graph, Hilbert spectrum, and discrete wavelet transform coefficients.

[0028] The beneficial technical effects of the present invention are:

[0029] 1. The online monitoring and intelligent diagnosis method for early faults of nuclear power electric rotating equipment bearings provided by the present invention can effectively detect and identify different types of faults of nuclear power electric rotating equipment bearings, such as pitting, wear, cracks, spalling, etc., thereby improving the safety and reliability of nuclear power plants.

[0030] 2. The online monitoring and intelligent diagnosis method for early-stage bearing failures of nuclear power electric rotating equipment provided by the present invention can achieve comprehensive analysis of acoustic emission signals. It comprehensively utilizes the parameter characteristics, time-frequency characteristics, and artificial intelligence technology of acoustic emission signals to extract effective information from acoustic emission signals, reflect the essential characteristics of bearing failures, and improve the accuracy and reliability of bearing fault diagnosis.

[0031] 3. The online monitoring and intelligent diagnosis method for early-stage bearing failures of nuclear power electric rotating equipment provided by the present invention can optimize the processing of acoustic emission signals. It uses wavelet transform and genetic algorithm to perform time-frequency analysis and optimization on acoustic emission signals, thereby improving the time-frequency resolution, robustness, and adaptability of acoustic emission signals, and improving the efficiency and sensitivity of bearing fault diagnosis.

[0032] 4. The online monitoring and intelligent diagnosis method for early-stage bearing failures of nuclear power electric rotating equipment provided by the present invention can realize intelligent recognition of acoustic emission signals, construct a bearing fault diagnosis model using convolutional neural networks and support vector machines, use known fault data for training and testing, optimize the model's structure and parameters, improve the model's recognition accuracy, and enhance the intelligence and automation of bearing fault diagnosis.

[0033] 5. The online monitoring and intelligent diagnosis method for early faults of nuclear power electric rotating equipment bearings provided by the present invention can realize online monitoring of acoustic emission signals, apply the trained fault diagnosis model to the online monitoring of nuclear power electric rotating equipment bearings, receive acoustic emission signals in real time, extract features, input the model, output the fault type and severity, give fault alarms and maintenance suggestions, and improve the real-time and practicality of bearing fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of the online monitoring and intelligent diagnosis method for early-stage bearing failures of nuclear power electric rotating equipment provided by the present invention. DETAILED DESCRIPTION

[0035] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0036] like Figure 1 As shown, the present invention provides a method for online monitoring and intelligent diagnosis of early-stage faults of bearings of nuclear power electric rotating equipment, which specifically includes the following steps:

[0037] Step 1: Acoustic Emission Signal Collection

[0038] Install acoustic emission sensors on the outside of the bearings of nuclear power electric rotating equipment to collect acoustic emission signals generated when the bearings are running. (In this step, the continuous-time acoustic emission signal is expressed as ), and transmit the signal to the data acquisition card and computer.

[0039] The selection of an acoustic emission sensor requires comprehensive consideration of technical indicators such as the frequency range of the bearing fault acoustic emission signal, the sensor's sensitivity, dynamic range, and temperature stability. Because the frequency of the acoustic emission signal generated by a bearing fault is relatively high, the selected sensor should have a wide-band response characteristic exceeding 100kHz and a high sensitivity exceeding 10dB to ensure that weak early bearing fault signals can be effectively captured. At the same time, considering that the amplitude of the acoustic emission signal will undergo significant changes during the development of a bearing fault, the sensor's dynamic range must exceed 60dB. Furthermore, the harsh, high-temperature operating conditions of nuclear power bearings place higher demands on the sensor's temperature stability, requiring the use of a high-temperature sensor capable of withstanding temperatures exceeding 150°C.

[0040] Piezoelectric acoustic emission sensors are well-suited to meeting these technical requirements. They utilize the piezoelectric effect of piezoelectric materials to convert acoustic emission signals into electrical output signals. They offer advantages such as a wide frequency response range (up to several MHz), high sensitivity (up to 80 dB), a large dynamic range (up to 80-100 dB), and excellent temperature stability (able to withstand temperatures exceeding hundreds of degrees Celsius). They are a commonly used sensor type for acoustic emission detection of bearing faults.

[0041] Furthermore, when used in the nuclear power sector, attention must be paid to the sensor's electromagnetic compatibility and installation coupling method. It is recommended to use a piezoelectric acoustic emission sensor with a stainless steel housing and shielded cable to improve its electromagnetic interference suppression capabilities. Magnetic or vacuum coupling should be used for sensor installation, and grease coupling should be avoided to prevent the coupling agent from fading at high temperatures, potentially affecting acoustic emission signal transmission.

[0042] In a specific embodiment, the acoustic emission sensor adopts an R15α resonant piezoelectric acoustic emission sensor, and its main technical parameters are as follows:

[0043] Peak resonance frequency: 150kHz

[0044] Effective frequency range: 50-400kHz

[0045] Sensitivity: 80dB (ref 1V / μbar)

[0046] Dynamic range: 87dB

[0047] Operating temperature: -65-175°C

[0048] Shell material: stainless steel

[0049] Weight: 34g

[0050] Size: Φ19mm×22mm

[0051] This type of sensor meets the technical requirements for acoustic emission detection of nuclear power bearings in terms of frequency response range, sensitivity, dynamic range, temperature stability, etc., and is compact, light, and easy to install. It has been successfully applied in the field of nuclear power equipment condition monitoring.

[0052] The installation of acoustic emission sensors is based on the structure, size, position and other factors of the bearing. The appropriate installation position, number and angle of the acoustic emission sensors are selected to ensure the effective collection of acoustic emission signals and avoid attenuation, interference and distortion of the acoustic emission signals. For example, a symmetrical four-point arrangement method is adopted. The symmetrical four-point arrangement method is an installation method that evenly distributes the acoustic emission sensors in the four directions of the bearing. It can effectively cover the entire surface of the bearing, improve the collection efficiency of the acoustic emission signals, reduce the propagation loss of the acoustic emission signals, and enhance the signal strength of the acoustic emission signals.

[0053] The sampling of acoustic emission signals needs to comprehensively consider factors such as the frequency characteristics, time characteristics, amplitude characteristics of the acoustic emission signals, and reasonably select parameters such as sampling frequency, sampling length, sampling method and sampling accuracy to ensure that complete and high-quality acoustic emission signal data is collected.

[0054] According to the Nyquist sampling theorem, the sampling frequency must be at least twice the maximum frequency of the acoustic emission signal to avoid information distortion caused by spectral aliasing. Considering that the frequency range of the acoustic emission signal from bearing failure is 100kHz-1MHz, with a maximum frequency of 500kHz, to further improve sampling quality, the present invention adopts a sampling frequency that is at least five times higher than the maximum frequency of the acoustic emission signal, selecting a sampling frequency range of 2.5MHz-5MHz. This higher sampling frequency not only meets the requirements of the Nyquist sampling theorem and achieves higher sampling accuracy, but also more completely records the original information of the acoustic emission signal, providing a high-quality data foundation for subsequent processing.

[0055] The selection of sampling length requires a balance between signal integrity and data volume. On the one hand, the sampling length should cover the duration of a single acoustic emission pulse event. The duration of a bearing fault acoustic emission signal is typically on the order of several milliseconds. On the other hand, excessively long sampling lengths consume significant storage space, increasing the data transmission and computational burden. Balancing these two factors, the present invention selects an acoustic emission signal sampling length range of 10ms-100ms, which both captures the complete acoustic emission pulse waveform and maintains a reasonable data volume.

[0056] Sampling methods can be selected as continuous or triggered. Continuous sampling continuously records acoustic emission signals. Its advantage is that it provides complete time series data, but its disadvantage is that it can produce relatively redundant data. Triggered sampling selectively records data based on preset thresholds or external trigger signals. Its advantage is that it reduces data volume and extends storage time, but its disadvantage is that valuable signal fragments may be missed. In practical applications, continuous or triggered sampling can be flexibly selected based on the operating conditions of the bearing equipment, data storage space, and other conditions. Given the serious consequences of nuclear power bearing failures, it is recommended to primarily use continuous sampling, supplemented by triggered sampling when necessary.

[0057] Sampling accuracy, also known as the number of sampling bits or resolution, reflects the fineness of digital sampling. Higher sampling accuracy increases the signal's dynamic range, improves amplitude resolution, and reduces quantization noise. However, increased accuracy also increases data length and processing time. To match the acoustic emission sensor's dynamic range of approximately 80dB and balance sampling accuracy and data length, the present invention adopts a more moderate sampling accuracy of 12-16 bits.

[0058] In one specific embodiment, the acoustic emission sensor sampling parameters are: a sampling frequency of 2.5MHz-5MHz, a sampling length of 10ms-100ms, primarily continuous sampling with triggered sampling when necessary, and a sampling accuracy of 12-bit to 16-bit. These parameter selections comprehensively consider factors such as the spectral characteristics, duration, sampling quality, and data volume of bearing fault acoustic emission signals. This effectively balances sampling integrity and data redundancy, yielding high-quality acoustic emission signal data that meets fault diagnosis requirements. The specific parameter values ​​require preliminary testing and theoretical analysis, with optimal selection within the parameter range required to ultimately determine an optimal sampling scheme that meets the application requirements of nuclear power bearing fault diagnosis.

[0059] Step 2: Acoustic emission signal preprocessing

[0060] Preprocess the acoustic emission signal, including filtering, noise reduction, gain, etc., to improve the signal-to-noise ratio and dynamic range of the signal.

[0061] The filtering of acoustic emission signals is to select appropriate filtering methods, filter types and filter parameters based on factors such as the frequency distribution, noise distribution and signal characteristics of the acoustic emission signals, so as to effectively remove irrelevant noise interference in the acoustic emission signals, retain useful signal components that can reflect the characteristics of bearing faults, and improve the signal-to-noise ratio and quality of the signals.

[0062] Preferably, a bandpass filter is used to preprocess the acoustic emission signal. A bandpass filter is a frequency-selective filter that allows signals within a specific frequency range to pass through while attenuating or eliminating frequency signals outside of that range. By properly setting the bandpass filter's passband range to cover the primary frequency band of bearing fault acoustic emission signals (typically 100kHz-400kHz), the low-frequency background noise and high-frequency broadband noise in the acoustic emission signal can be effectively filtered out, while retaining the characteristic frequency components near the center frequency and highlighting the effective information of the bearing fault signal.

[0063] To select a high-performance bandpass filter, it is important to focus on the following technical indicators: passband frequency range, stopband suppression, passband flatness, transition band width, filter type, filter order, amplitude-frequency and phase-frequency response characteristics, stability, and robustness. The passband frequency range determines the filter's frequency selectivity and must match the characteristic frequency band of the bearing fault signal. Stopband suppression reflects the filter's ability to suppress noise interference and is required to be above 40dB. Passband flatness affects the distortion of the filtered signal and is required to be within ±1dB. The transition band width affects the filter's selectivity and implementation difficulty and should be as narrow as possible. The filter type involves the algorithm structure and computational complexity, so an IIR type is recommended. The filter order needs to balance performance and complexity, generally ranging from 4 to 8 orders. The amplitude-frequency response requires flatness within the passband, high attenuation within the stopband, and the phase-frequency response requires as much linearity as possible. Stability and robustness determine whether the filter can operate reliably.

[0064] The specific design and implementation of the filter also requires comprehensive consideration of factors such as the actual needs of bearing fault diagnosis, computer resources, and signal processing platforms. A method of estimation followed by optimization is employed to obtain the optimal filter parameters. Initial values ​​for the filter parameters are first determined based on theoretical analysis and empirical estimates. An intelligent optimization algorithm is then used to optimize the parameter space, and the filter's actual performance is evaluated through simulations and experimental testing. Finally, necessary adjustments are made to the optimization results to obtain the optimal filter that meets the requirements of bearing fault diagnosis.

[0065] This paper recommends an optimal bandpass filter design using a sixth-order elliptical IIR filter with a passband range of 100kHz-400kHz, a maximum passband attenuation of 1dB, a minimum stopband attenuation of 60dB, and a transition band width of 20kHz. This design ensures high noise suppression and low signal distortion while also balancing the engineering complexity of the filter. It offers a good balance between filtering performance and computational efficiency, making it suitable for real-time processing of acoustic emission signals.

[0066] The denoising of acoustic emission signals is to select appropriate denoising methods, algorithms, parameters, etc. of the acoustic emission signals based on factors such as the noise type, noise level, and signal characteristics of the acoustic emission signals, so as to further eliminate random noise in the acoustic emission signals and improve the signal-to-noise ratio of the acoustic emission signals, such as using wavelet threshold denoising. Wavelet threshold denoising is a denoising method that uses wavelet transform to decompose the acoustic emission signals into wavelet coefficients of different scales, then performs threshold processing on the wavelet coefficients to remove the noise components in the wavelet coefficients, and then uses inverse wavelet transform to reconstruct the wavelet coefficients into the denoised acoustic emission signals. It can effectively retain the detailed features of the acoustic emission signals, eliminate the random noise of the acoustic emission signals, and improve the signal-to-noise ratio of the acoustic emission signals.

[0067] The gain of the acoustic emission signal is to select the appropriate gain method, gain coefficient, gain range, etc. of the acoustic emission signal based on factors such as the amplitude distribution, dynamic range, and signal characteristics of the acoustic emission signal to adjust the amplitude of the acoustic emission signal so that the dynamic range of the acoustic emission signal adapts to the input requirements of the data acquisition card and computer, such as using adaptive gain. Adaptive gain is a gain method that dynamically adjusts the gain coefficient of the acoustic emission signal according to the amplitude change of the acoustic emission signal. It can effectively change the amplitude of the acoustic emission signal within a certain range, avoid the acoustic emission signal being too large or too small, causing distortion or weakening of the acoustic emission signal, and improve the dynamic range of the acoustic emission signal.

[0068] Step 3: Acoustic emission signal parameter analysis

[0069] Perform parameter analysis on the acoustic emission signal, extract typical acoustic emission parameters such as ring count, root mean square, signal amplitude, and kurtosis coefficient of the acoustic emission signal, and analyze the response trend and sensitivity of each parameter to different working conditions and fault types.

[0070] The parameter analysis of acoustic emission signals is to select appropriate parameter extraction methods, parameter calculation formulas and parameter thresholds based on the waveform characteristics, statistical characteristics, energy characteristics and other factors of the acoustic emission signals, so as to extract parameters from the acoustic emission signals that can sensitively reflect the characteristics of bearing faults.

[0071] Typical characteristic parameters of acoustic emission signals include ring count, root mean square, signal amplitude, kurtosis coefficient, etc. These parameters describe the waveform characteristics, statistical characteristics and energy characteristics of acoustic emission signals from the perspectives of time domain, amplitude and energy distribution, such as the following formula:

[0072]

[0073]

[0074]

[0075]

[0076] in, The acoustic emission signal sampling points (the sampled discrete-time signal is expressed as ), Indicates the number of sampling points of the acoustic emission signal, represents the threshold of the acoustic emission signal, Represents an indicator function, which is 1 when the condition is met and 0 otherwise;

[0077] The parameter analysis of the acoustic emission signal also includes time series analysis, frequency domain analysis, statistical analysis, correlation analysis, etc. of the parameters of the acoustic emission signal, drawing the curve of the change of the parameters of the acoustic emission signal over time, power spectrum density diagram, histogram, box plot, scatter plot, etc., observing the change characteristics, distribution characteristics, correlation characteristics, etc. of the parameters of the acoustic emission signal under different working conditions and fault types, such as the ring count and signal amplitude will show a sudden change when a fault occurs, the root mean square and kurtosis coefficient will show an upward trend before the fault occurs, the ring count and signal amplitude will show an increase in high-frequency components when a fault occurs, the root mean square and kurtosis coefficient will show an increase in low-frequency components before the fault occurs, the ring count and signal amplitude will show a skewed distribution when a fault occurs, the root mean square and kurtosis coefficient will show a normal distribution before the fault occurs, the ring count and signal amplitude will show a high positive correlation when a fault occurs, and the root mean square and kurtosis coefficient will show a high negative correlation before the fault occurs.

[0078] Step 4: Acoustic emission signal waveform analysis

[0079] The acoustic emission signal is analyzed by using Morlet continuous wavelet transform. Perform time-frequency analysis, use genetic algorithms to optimize the scale of wavelet transform and base wavelet waveform parameters, improve the performance of time-frequency analysis, and extract the time-frequency characteristics of acoustic emission signals, such as time-frequency diagram, Hilbert spectrum, discrete wavelet transform coefficients, etc.

[0080] The wavelet transform of the acoustic emission signal is based on the non-stationary, nonlinear, multi-scale and other factors of the acoustic emission signal. It selects the appropriate wavelet transform method, wavelet transform type, wavelet transform parameters, etc. of the acoustic emission signal to convert the acoustic emission signal from the time domain to the time-frequency domain and reveal the time-frequency structure of the acoustic emission signal. For example, the following formula is used:

[0081]

[0082] in, Indicates acoustic emission signal The wavelet transform coefficients of represents the scale parameter, which controls the frequency and time resolution of the wavelet, represents the time shift parameter, which controls the translation of the wavelet on the time axis, t represents the time variable, represents the mother wavelet function, the superscript represents the complex conjugate, Indicates that the mother wavelet function is scaled and time-shifted to obtain the daughter wavelet function; Indicates continuous integration, which performs integration operation on the signal on the entire time axis; It's in time The scale is , translate to The wavelet function of .

[0083] Specifically, in the given wavelet transform formula, some parameters are known, and some parameters need to be selected and determined according to the characteristics of the acoustic emission signal and the analysis requirements, as follows:

[0084] 1. : represents the continuous-time acoustic emission signal, which is the input of the wavelet transform. This parameter is obtained by collecting the acoustic emission signal generated by the bearing during operation using an acoustic emission sensor installed on the outside of the bearing in step 1, "Acoustic Emission Signal Acquisition." Specifically, the acoustic emission sensor converts the elastic stress wave signal generated during bearing operation into an electrical signal, and transmits the electrical signal to a data acquisition card and computer, thereby obtaining a continuous-time acoustic emission signal. .

[0085] 2. : Indicates acoustic emission signal The wavelet transform coefficient is the output of the wavelet transform. This parameter is obtained by analyzing the collected acoustic emission signal in step 4 "wavelet time-frequency analysis of acoustic emission signal". The specific calculation formula is as above, by changing and The value of can be used to obtain the wavelet coefficients of the acoustic emission signal at different scales and time shift positions, reflecting the time-frequency distribution characteristics of the signal.

[0086] 3. : represents the scale parameter in wavelet transform, which is used to adjust the frequency characteristics of wavelet. The choice of scale parameter should be determined according to the frequency range of the acoustic emission signal and the purpose of analysis. Usually the following methods can be used to select the appropriate scale parameter range:

[0087] (1) According to the sampling frequency of the acoustic emission signal and acquisition frequency range , using the frequency-scale relationship of wavelet, estimate the corresponding scale parameter range :

[0088]

[0089] in, is the center frequency of the wavelet.

[0090] (2) Within the estimated scale parameter range, the scale parameter is discretized by logarithmic scaling or linear scaling to obtain a series of discrete scale values.

[0091] (3) For each discrete scale value, calculate the wavelet transform coefficient, observe the time-frequency distribution characteristics at different scales, and select the scale parameter that can better reflect the signal characteristics as the focus of analysis.

[0092] (4) Within the selected scale parameter range, the frequency resolution of the wavelet transform is controlled by setting different scale steps. The smaller the scale step, the higher the frequency resolution, but the greater the computational effort.

[0093] 4. : Represents the time shift parameter in wavelet transform, which is used to adjust the position of wavelet on the time axis. The choice of needs to be determined according to the duration and sampling frequency of the acoustic emission signal. Usually, a time shift step corresponding to the signal sampling point can be used, that is:

[0094]

[0095] in, is the sampling interval of the signal, is the number of sampling points of the signal. This ensures that the entire signal interval is fully covered on the time axis.

[0096] 5. : Represents the mother wavelet function used in wavelet transform, which is a key parameter. The selection of the mother wavelet function needs to be determined based on the time-frequency characteristics of the acoustic emission signal and the type of fault to be analyzed. In bearing fault diagnosis, since the acoustic emission signal has the characteristics of burstiness, attenuation, and modulation, the Morlet complex wavelet is usually selected as the mother wavelet function. The Morlet complex wavelet has a waveform similar to an impulse response in the time domain and has bandpass filtering characteristics in the frequency domain. It can well match and extract the fault pulse components in the acoustic emission signal. The expression of the Morlet complex wavelet is:

[0097]

[0098] in, is the bandwidth parameter of the wavelet, is the center frequency of the wavelet, is the imaginary unit ( = -1). By adjusting and , the time domain width and frequency domain position of the Morlet wavelet can be changed to adapt to acoustic emission signals in different frequency ranges.

[0099] After calculating the wavelet transform coefficients, time-frequency analysis is performed as follows:

[0100] 1. Conversion of wavelet transform coefficients to time-frequency analysis results:

[0101] Wavelet transform coefficients It contains the signal at different times and different scales To obtain the time-frequency analysis results, the following steps are required:

[0102] (1) Scale-frequency conversion:

[0103] First, you need to scale Convert to the corresponding frequency f. The conversion relationship is:

[0104]

[0105] in, is the center frequency of the wavelet, is the sampling frequency.

[0106] (2) Energy calculation:

[0107] Calculate the square of the modulus of the wavelet transform coefficients , which represents the signal at time and scale (or frequency f) at which the energy density is

[0108] (3) Time-frequency diagram construction:

[0109] Will By time and frequency By performing a two-dimensional arrangement, we can obtain the time-frequency distribution diagram of the signal, also known as the time-frequency spectrum.

[0110] 2. Correlation between wavelet transform coefficients and time-frequency analysis results:

[0111] (1) Time resolution:

[0112] Wavelet transform coefficients in It directly corresponds to the time axis and provides the temporal localization information of the signal.

[0113] (2) Frequency resolution:

[0114] Scale in coefficients and frequency Inversely proportional. Corresponding to high frequency, larger This multi-resolution analysis is an important feature of wavelet transform.

[0115] (3) Energy distribution:

[0116] It reflects the energy distribution of the signal in the time-frequency plane and directly constitutes the intensity value of the time-frequency spectrum.

[0117] (4) Instantaneous frequency:

[0118] Through analysis Follow The instantaneous frequency information of the signal can be obtained by analyzing the changes in

[0119] 3. Specific time-frequency analysis results:

[0120] Based on the wavelet transform coefficients, we can obtain the following specific time-frequency analysis results:

[0121] (1) Time-frequency spectrum:

[0122] This is the most intuitive time-frequency analysis result. The x-axis represents time, the y-axis represents frequency, and the depth of color represents the strength of energy.

[0123] (2) Edge spectrum:

[0124] - Time edge spectrum: fix the frequency and observe the change of energy over time.

[0125] - Frequency edge spectrum: fix the time and observe the change of energy with frequency.

[0126] (3) Instantaneous frequency:

[0127] By calculating the frequency corresponding to the maximum energy at each time point, the instantaneous frequency curve of the signal can be obtained.

[0128] (4) Time-frequency ridges:

[0129] Connecting the local maximum energy points on the time-frequency plane forms a time-frequency ridge line, which reflects the change of the main frequency components of the signal over time.

[0130] (5) Wavelet entropy:

[0131] Calculate the entropy value of the wavelet coefficient to reflect the complexity and uncertainty of the signal.

[0132] 4. Implementation process:

[0133] In practical applications, we usually use the following steps to obtain time-frequency analysis results:

[0134] (1) Calculate wavelet transform coefficients .

[0135] (2) Scale Convert to frequency .

[0136] (3) Calculation As energy density.

[0137] (4) Construct a time-frequency spectrum diagram, with the x-axis being , Axis , color indicates size.

[0138] (5) Extract other time-frequency features as needed, such as edge spectrum, instantaneous frequency, etc.

[0139] (6) Visualize the time-frequency analysis results to facilitate intuitive understanding and further analysis.

[0140] Through these steps, we can obtain rich time-frequency analysis results from wavelet transform coefficients, providing an important basis for subsequent feature extraction and fault diagnosis. These time-frequency analysis results can effectively reveal the time-varying and frequency characteristics of the signal, and are particularly suitable for analyzing the fault characteristics of non-stationary signals such as nuclear power bearings.

[0141] The wavelet transform optimization of acoustic emission signals is to select appropriate wavelet transform optimization methods, optimization algorithms, optimization parameters, etc. of acoustic emission signals based on the performance indicators, optimization objectives, optimization constraints and other factors of the wavelet transform of acoustic emission signals, so as to optimize the scale and base wavelet waveform parameters of the wavelet transform of acoustic emission signals and improve the resolution, robustness and adaptability of the wavelet transform of acoustic emission signals. For example, the following methods are used:

[0142] A genetic algorithm (GA) is used to optimize the wavelet transform of the acoustic emission signal. The scale and base wavelet waveform parameters of the wavelet transform of the acoustic emission signal are used as optimization variables. The energy concentration, spectral entropy, and mutual information of the wavelet transform of the acoustic emission signal are used as optimization targets. The bandwidth, frequency overlap, and frequency coverage of the wavelet transform of the acoustic emission signal are used as optimization constraints. Through the initialization, selection, crossover, mutation, and evaluation operations of the genetic algorithm, the optimal scale and base wavelet waveform parameters of the wavelet transform of the acoustic emission signal are solved. The genetic algorithm is an optimization algorithm that simulates the natural evolution process. It has the advantages of global search capability, parallel computing capability, and self-adaptation, and is suitable for optimizing the wavelet transform of acoustic emission signals.

[0143] The extraction of time-frequency features of acoustic emission signals is based on the results of wavelet transform of acoustic emission signals, the type of time-frequency features, the selection of time-frequency features and other factors. The appropriate extraction method of time-frequency features of acoustic emission signals, the calculation formula of time-frequency features, the threshold of time-frequency features, etc. are selected to extract time-frequency features that can reflect bearing faults from the wavelet transform of acoustic emission signals. For example, the following methods are used:

[0144] Use the wavelet transform coefficients of the acoustic emission signal to draw the time-frequency diagram of the acoustic emission signal, and observe the morphological characteristics of the time-frequency diagram of the acoustic emission signal under different working conditions and fault types, such as the color, brightness, distribution, shape, etc. of the time-frequency diagram. For example, when a fault occurs, the time-frequency diagram will show bright spots or stripes.

[0145] The Hilbert spectrum of the acoustic emission signal is calculated using the wavelet transform coefficients of the acoustic emission signal. The amplitude characteristics of the Hilbert spectrum of the acoustic emission signal under different working conditions and fault types are observed, such as the peak, valley, mean, and variance of the Hilbert spectrum. For example, the Hilbert spectrum will show sudden changes or fluctuations in amplitude when a fault occurs.

[0146] The discrete wavelet transform is performed using the wavelet transform coefficients of the acoustic emission signal to observe the energy characteristics of the discrete wavelet transform coefficients of the acoustic emission signal under different working conditions and fault types, such as the energy, energy ratio, energy entropy, etc. of the discrete wavelet transform coefficients. For example, the discrete wavelet transform coefficients will show energy concentration or dispersion when a fault occurs.

[0147] Step 5: Bearing fault diagnosis model construction

[0148] A bearing fault diagnosis model is constructed using convolutional neural networks (CNN) and support vector machines (SVM). The parameter characteristics and time-frequency characteristics of acoustic emission signals are used as input, and different types of faults are used as output. Known fault data is used for training and testing to optimize the model structure and parameters and improve the model's recognition accuracy.

[0149] The structure of the bearing fault diagnosis model is based on the dimension, type, complexity and other factors of the acoustic emission signal characteristics. The appropriate structure, number of layers, number of units, etc. of the bearing fault diagnosis model are selected to achieve effective extraction and learning of the characteristics of the acoustic emission signal. For example, the following structure is adopted:

[0150] The bearing fault diagnosis model consists of two components: a CNN-based module for extracting and classifying the time-frequency features of acoustic emission signals, and a SVM-based module for classifying the parameter features of acoustic emission signals. The CNN model takes the time-frequency image as input, extracts local time-frequency features through convolutional layers, and performs classification through fully connected layers. The SVM model takes multiple statistical parameter features as input and identifies the fault type through kernel function mapping and hyperplane classification.

[0151] During the model training phase, the CNN and SVM models were trained and their performance evaluated. Using methods such as cross-validation, the classification accuracy of the two models for different bearing fault types was evaluated, and the diagnostic confidence of the CNN and SVM models was calculated based on this accuracy.

[0152] In the model testing phase, the acoustic emission signal samples to be diagnosed are input into the trained CNN and SVM models respectively to obtain two diagnostic results, namely the probability or confidence that the sample belongs to each fault type.

[0153] To fuse the diagnostic results of the CNN and SVM, a weighted average method is used. Weight coefficients are calculated based on the classification accuracy of each fault type for each model. Models with higher accuracy receive higher weights, while models with lower accuracy receive lower weights. The outputs of the CNN and SVM are then weighted averaged to produce a fused fault probability vector, representing the combined probability of a sample belonging to each fault type.

[0154] Finally, the fused fault probability vector is used to diagnose the fault type and severity of the sample. The fault type corresponds to the category with the highest probability; the severity is determined by comparing the maximum fault probability value with a preset threshold, and is divided into three levels: mild, moderate, and severe.

[0155] While outputting the diagnostic results, it also provides corresponding maintenance decision suggestions to handle bearing faults in a timely and effective manner and ensure the safe operation of the equipment.

[0156] The diagnostic model fused with CNN and SVM can extract and utilize bearing fault feature information from both the time-frequency and statistical domains, leveraging the complementary advantages of the two approaches. While ensuring real-time diagnostic performance, it also improves diagnostic accuracy, stability, and applicability, providing a more reliable and comprehensive basis for intelligent diagnosis and predictive maintenance of bearing faults. Furthermore, the weighted average fusion strategy is simple and easy to implement, facilitating engineering implementation. The weights of diagnostic decisions are clearly defined and well interpretable, making them more conducive to gaining the understanding and trust of engineers and technicians in practical applications. The construction principles of this fused diagnostic model can also be extended to multi-source information fusion diagnosis in other fields, demonstrating its universal value.

[0157] The structure of the CNN-based time-frequency feature extraction and classification module for acoustic emission signals is as follows: the input layer receives the time-frequency map of the acoustic emission signal as input, the convolution layer performs a convolution operation on the time-frequency map to extract local features of the time-frequency map, the pooling layer performs a pooling operation on the output of the convolution layer to reduce the dimension of the features and increase the robustness of the features, the fully connected layer performs a nonlinear transformation on the output of the pooling layer to increase the expressive power of the model, and the output layer classifies the output of the fully connected layer and outputs the probability of different types of faults.

[0158] The structure of the parameter feature classification module of the acoustic emission signal based on SVM is as follows: the input layer receives the parameter features of the acoustic emission signal as input, the SVM layer classifies the parameter features, and outputs the probability of different types of faults.

[0159] The parameters of the bearing fault diagnosis model are selected according to the structure, performance, data and other factors of the bearing fault diagnosis model, such as the size, number, and step size of the convolution kernel, the size, type, and step size of the pooling kernel, the number of units and activation function of the fully connected layer, the kernel function and penalty parameter of the SVM, etc. For example, the following parameters are used:

[0160] The size of the convolution kernel is 3x3, the number is 32, and the stride is 1;

[0161] The pooling kernel size is 2x2, the type is maximum pooling, and the stride is 2;

[0162] The number of units in the fully connected layer is 128, and the activation function is ReLU;

[0163] The kernel function of SVM is radial basis function, and the penalty parameter is 1;

[0164] The training and testing of the bearing fault diagnosis model is based on the parameters, data, objectives and other factors of the bearing fault diagnosis model. The appropriate training and testing methods, optimization algorithms, loss functions, evaluation indicators, etc. of the bearing fault diagnosis model are selected to optimize the parameters of the bearing fault diagnosis model and improve the recognition accuracy of the bearing fault diagnosis model. For example, the following methods are used:

[0165] A dataset of acoustic emission signals from bearings in nuclear power electric rotating equipment is used for training and testing. This dataset contains time-frequency plots and parameter features of acoustic emission signals under different operating conditions and fault types, as well as corresponding fault labels such as normal, pitting, wear, cracks, and spalling.

[0166] Use the stochastic gradient descent (SGD) algorithm to optimize the bearing fault diagnosis model, setting appropriate learning rate, batch size, number of iterations, etc. to minimize the loss function of the bearing fault diagnosis model, such as the cross entropy loss function;

[0167] The bearing fault diagnosis model is evaluated using indicators such as accuracy, recall rate, and F1 value, and the consistency between the prediction results of the bearing fault diagnosis model and the actual results is compared, such as using the following formula:

[0168]

[0169]

[0170]

[0171] Among them, TP represents true positive examples, TN represents true negative examples, FP represents false positive examples, and FN represents false negative examples.

[0172] Step 6: Application of bearing fault diagnosis model

[0173] The trained fault diagnosis model is applied to the online monitoring of bearings of nuclear power electric rotating equipment. The acoustic emission signal is received in real time, features are extracted, input into the model, the fault type and severity are output, and fault alarms and maintenance suggestions are given.

[0174] The application process of the bearing fault diagnosis model requires system design based on factors such as the input, output, and function of the diagnosis model. It includes key links such as data acquisition, data preprocessing, feature extraction and selection, model reasoning, knowledge reasoning and decision fusion, decision visualization and human-computer interaction, online learning and incremental updates, system deployment and operation and maintenance management, so as to achieve online monitoring, intelligent diagnosis and predictive maintenance of bearing faults. Preferably, the application process of the present invention is as follows:

[0175] 1. Data collection: Use the acoustic emission sensor installed on the periphery of the bearing to collect the acoustic emission signal of the bearing in real time during operation. , and simultaneously record the working parameters of the bearing. The sensor uses a wide-band, high-sensitivity piezoelectric sensor with a sampling frequency of 2.5MHz-5MHz, a sampling length of 10ms-100ms, and continuous sampling.

[0176] 2. Data preprocessing: acoustic emission signal Adaptive noise reduction and feature enhancement are performed. The time-frequency features are extracted by combining wavelet packet decomposition and empirical mode decomposition to remove background noise; singular value decomposition and other methods are used to highlight the fault pulse features; the time-frequency feature matrix is ​​normalized to obtain a standardized time-frequency feature matrix. , which is convenient for subsequent processing.

[0177] 3. Feature extraction and selection: Extract multi-domain, multi-scale sensitive features from the acoustic emission signal, including time-domain statistical features, frequency-domain features, and time-frequency domain features. Using the maximum relevance minimum redundancy and recursive feature elimination algorithms, the feature subset most relevant to the bearing fault status is selected as input to the diagnostic model.

[0178] 4. Model Inference: The selected acoustic emission features are input into a pre-trained CNN-LSTM diagnostic model, and the fault type probability is calculated through forward propagation. The time-frequency image features are extracted using a CNN to extract local texture features, and then the LSTM is used to learn time series features and map them to fault categories. Statistical features are directly input into the SVM classifier for classification.

[0179] 5. Knowledge Reasoning and Decision Fusion: Deep learning diagnostic results are integrated with knowledge-based diagnostic results. The bearing fault diagnosis knowledge graph is used to perform semantic reasoning on the two diagnostic results to form a fault cause chain. DS evidence theory is used to integrate the two diagnostic results, update the fault cause confidence, and output the fault diagnosis conclusion.

[0180] 6. Decision Visualization and Human-Computer Interaction: Diagnostic results are visualized and interactive via a web platform. Bearing health status and fault diagnosis reports are presented in dashboards, fault trees, and other formats. Alerts are sent when warning thresholds are triggered, and repair recommendations are provided. Users can query historical reports, set diagnostic tasks, annotate sample data, and optimize diagnostic models.

[0181] 7. Online Learning and Incremental Updates: The diagnostic model can be learned and updated online based on newly acquired acoustic emission data. When new bearing acoustic emission data is received, incremental learning is automatically triggered to fine-tune and update model parameters using the new samples while retaining existing knowledge to adapt to dynamic changes in bearing condition.

[0182] 8. System Deployment and Operations Management: A microservices-based system architecture with front-end and back-end separation enables decoupling and elastic scalability of diagnostic algorithms, data services, and application services. Modules such as acoustic emission signal acquisition, edge computing, cloud storage, and container scheduling are integrated to support rapid deployment, dynamic scaling, and remote operations and maintenance of fault diagnosis services. Comprehensive operational mechanisms, including system monitoring, log analysis, and anomaly alerting, are established to ensure high availability and reliability of diagnostic services.

[0183] The above application process integrates advanced acoustic emission perception, deep learning, knowledge reasoning, cloud-edge collaboration and other technologies to build an online bearing fault diagnosis system that integrates data collection, intelligent diagnosis, knowledge accumulation, and decision optimization. It realizes a closed loop of the entire process from data to decision-making, which can significantly improve the early warning capability, classification diagnosis accuracy, knowledge accumulation capability and decision-making service quality of bearing fault diagnosis, and provides new technical means and application models for the health management of bearings throughout their life cycle. It is of great value to improving the reliability, safety and economy of rotating machinery equipment.

[0184] The present invention has been described in detail above with reference to the accompanying drawings and embodiments. However, the present invention is not limited to the above embodiments. Various modifications can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention. Any content not described in detail in the present invention may be adapted from existing technologies.

Claims

1. A method for online monitoring and intelligent diagnosis of early faults of bearings in nuclear power electric rotating equipment, It is characterized in that The method comprises: Step 1: Acoustic emission signal collection: Use the acoustic emission sensor to collect the acoustic emission signal generated during the operation of the bearing; Step 2: Acoustic emission signal preprocessing: filtering, noise reduction, and gain preprocessing of the acoustic emission signal; Step 3: AE signal parameter analysis: According to the waveform characteristics, statistical characteristics, and energy characteristics of the AE signal, extract the parameters of the AE signal from the AE signal, and analyze the change characteristics, distribution characteristics, and correlation characteristics of the parameters of the AE signal under different working conditions and fault types; Step 4: Acoustic emission signal waveform analysis: Use wavelet transform to perform time-frequency analysis on the collected acoustic emission signal, use genetic algorithm to optimize the scale of wavelet transform and base wavelet waveform parameters, and extract the time-frequency characteristics of the acoustic emission signal; Step 5: Bearing fault diagnosis model construction: Use convolutional neural network and support vector machine to build a bearing fault diagnosis model, and train and test the bearing fault diagnosis model; Step 6: Application of bearing fault diagnosis model: Use the trained fault diagnosis model to conduct online monitoring of nuclear power electric rotating equipment bearings, receive acoustic emission signals in real time, extract features, input the model, output the fault type and severity, and give fault alarms and maintenance suggestions.

2. According to claim 1, a method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment, It is characterized in that In the step 1, the acoustic emission sensor has a broadband response characteristic of more than 100 kHz, a sensitivity of more than 10 dB, and a heat-resistant temperature of more than 150° C.

3. According to claim 1, a method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment, It is characterized in that In the step 1, the sampling frequency range of the acoustic emission sensor is 2.5 MHz-5 MHz, the sampling length range is 10 ms-100 ms, the sampling mode is continuous sampling or triggered sampling, and the sampling accuracy is 12 bits-16 bits.

4. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that In the step 2, a bandpass filter is used to filter the acoustic emission signal, the passband range of the filter is set to 100kHz-400kHz, the stopband suppression is above 40dB, the passband flatness is within ±1dB, the filter type is IIR type, and the filter order is 4-8.

5. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that In the step 2, wavelet threshold noise reduction is used to reduce the noise of the acoustic emission signal.

6. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that In the step 2, an adaptive gain is used to gain the acoustic emission signal.

7. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that The parameters of the acoustic emission signal in step three include: ring count, root mean square, signal amplitude, and kurtosis coefficient of the acoustic emission signal.

8. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that The analysis of the parameters of the acoustic emission signal in step three includes: time series analysis, frequency domain analysis, statistical analysis, and correlation analysis.

9. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that In the step 4, wavelet transform is used to perform time-frequency analysis on the collected acoustic emission signal, and the wavelet transform coefficient is calculated according to the collected acoustic emission signal. The calculation formula of the wavelet transform coefficient is: ， in, Represents the acoustic emission signal of continuous time; Indicates acoustic emission signal The wavelet transform coefficients of Represents the scale parameter in wavelet transform, which controls the frequency and time resolution of the wavelet; It represents the time shift parameter in wavelet transform, which controls the translation of wavelet on the time axis; represents the mother wavelet function in wavelet transform, the superscript represents the complex conjugate, It means that the mother wavelet function is scaled and time-shifted to obtain the daughter wavelet function; Indicates continuous integration, which integrates the signal on the entire time axis.

10. A method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 9, It is characterized in that The formula of the mother wavelet function in the wavelet transform in step 4 is: ， in, is the bandwidth parameter of the wavelet, is the center frequency of the wavelet, Is an imaginary unit.

11. The method for online monitoring and intelligent diagnosis of early faults of bearings of nuclear power electric rotating equipment according to claim 1, It is characterized in that The time-frequency characteristics of the acoustic emission signal in step 4 include: time-frequency diagram, Hilbert spectrum, and discrete wavelet transform coefficients.

Images