High-speed motor bearing rotor system fault detection method and system

Abstract

The present application provides a high-speed motor bearing rotor system fault detection method and system, the method comprising: installing vibration displacement sensors with mutually perpendicular directions on multiple sections, and extracting instantaneous characteristics of non-stationary vibration signals of multi-section rotors; constructing complex signals; converting the complex signals into forward signals and backward signals with physical meanings through Hilbert transform; using MNCMD algorithm to realize signal decomposition, and obtaining multiple groups of multi-channel signals with the same frequency, each group of multi-channel signals containing multiple harmonic components; using multiple groups of multi-channel signals to extract track features; outputting fault types based on multi-channel neural networks to solve the technical problem that when working in harsh environments such as multi-field coupling, heavy loads and even speed changes, the vibration signals exhibit non-stationary characteristics, which increases the difficulty of fault detection and diagnosis.

Description

Technical Field

[0001] The present application relates to the technical field of fault detection, and in particular to a method and system for detecting faults in a high-speed motor bearing rotor system. Background Art

[0002] High-speed motors are increasingly used in aerospace, weapons, deep-sea exploration and other fields due to their advantages such as high speed, high affinity and low weight. The bearing rotor system of a high-speed motor is a core component inside the motor. Its main function is to support the rotor so that it can rotate smoothly and efficiently when the motor is running. This system has a direct impact on the performance and life of the entire motor.

[0003] As the core component of high-speed motors, the rotor bearing system is of vital importance for its safety and reliability. Once a rotor bearing system fails, it will not only cause the motor performance to deteriorate, but may also cause serious safety accidents. Therefore, timely and effective detection and diagnosis of the status of the rotor bearing system is the key to ensuring the stable operation of high-speed motors.

[0004] However, the rotor bearing system faces some challenges, such as bearing component damage, rotor imbalance, rotor misalignment, etc., which may cause abnormal system vibration, overheating and even shutdown. Therefore, we must deeply analyze the fault signals of the high-speed motor rotor bearing system. Since the rotor bearing system often works in harsh environments such as multi-field coupling, heavy load and even variable speed, the vibration signal shows non-stationary characteristics, which increases the difficulty of fault detection and diagnosis. Summary of the invention

[0005] The present application provides a high-speed motor bearing rotor system fault detection method and system to solve the technical problem that when working in harsh environments such as multi-field coupling, heavy load and even speed change, the vibration signal exhibits non-stationary characteristics, which increases the difficulty of fault detection and diagnosis.

[0006] The present application provides a high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN, comprising:

[0007] Install vibration displacement sensors in mutually perpendicular directions on multiple sections, and extract the instantaneous characteristics of the non-stationary vibration signals of the multi-section rotor;

[0008] construct complex signals;

[0009] The complex signal is converted into forward signal and backward signal with physical meaning through Hilbert transform;

[0010] The MNCMD algorithm is used to decompose the signal and obtain multiple groups of multi-channel signals with the same frequency. Each group of multi-channel signals contains multiple harmonic components.

[0011] Extracting the characteristics of the track using multiple sets of multi-channel signals;

[0012] Output fault type based on multi-channel neural network.

[0013] Optionally, the step of constructing a complex signal includes:

[0014] According to two or more vibration displacement sensors orthogonally installed on each bearing, the vibration signals in mutually perpendicular directions are combined into complex signals, and the complex signal is:

[0015] g i (t) = x i (t)+jy i (t)

[0016] Among them, g i (t) represents a complex signal, x i (t) and y i (t) represents the signal obtained by two vibration displacement sensors installed on a bearing, j is an imaginary unit, t represents time, i=1, 2...Q, indicating that Q complex signals are constructed.

[0017] Optionally, the complex signal is converted into forward and backward signals having physical meaning by Hilbert transform, including:

[0018] Decompose the extracted complex signal and design the following operation to convert the complex signal g i (t) is divided into forward signal and reverse signal

[0019]

[0020] Among them, g i (t) represents a complex signal, Indicates a positive signal, represents the reverse signal, j is an imaginary unit, H(*) refers to the Hilbert transform, t represents time, f and b represent forward and reverse respectively, i=1, 2...Q, indicating that Q complex signals are constructed;

[0021] Get the spectrum corresponding to the forward signal and the backward signal:

[0022]

[0023] Where ω represents frequency, G(ω) represents the full spectrum of g(t), and G f (ω) represents the forward spectrum, G b (ω) represents the reverse spectrum, f and b represent the forward and reverse directions, respectively, and sgn(*) represents the sign function;

[0024] Extract the real part of the forward and reverse signals:

[0025] g + (t) = R(F -1 (G f (ω)

[0026] g - (t) = R(F -1 (G b (ω)

[0027] Among them, g + (t) represents the total positive signal, g - (t) represents the total reverse signal, t represents time, f and b represent forward and reverse directions respectively, ω represents frequency, and F -1 (G f (ω)) represents G f The inverse Fourier transform of (ω), F -1 (G b (ω)) represents G b The inverse Fourier transform of (ω), R(*) refers to the process of extracting the real part information.

[0028] Optionally, in the step of using the MNCMD algorithm to decompose the signal and obtain multiple groups of multi-channel signals having the same frequency, each group of multi-channel signals containing multiple harmonic components:

[0029] The signals in each group of channels are decomposed into M groups of signals:

[0030] A reconstructed complex signal is obtained based on the decomposed signal, and the reconstructed complex signal is:

[0031]

[0032] Among them, gi(t) is the signal of the i-th channel, The modal component of the mth forward signal, is the modal component of the mth reverse signal, f and b represent the forward and reverse directions respectively, t represents the time, and M represents the number of components in the channel.

[0033] Optionally, the step of extracting the features of the track using multiple groups of multi-channel signals specifically includes the following steps:

[0034] Transform the decomposed forward and backward components into complex exponential form:

[0035]

[0036] in, and Represents the instantaneous amplitude of the decomposed forward signal component and the instantaneous amplitude of the backward signal component; and represents the phase of the forward signal component and the phase of the backward signal component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, j is an imaginary unit, t represents time, f and b represent forward and reverse directions respectively;

[0037] Add the forward signal to the backward signal to get the instantaneous harmonic trajectory IOF:

[0038]

[0039] in, represents the sum of the forward and backward components,

[0040] represents the complex exponential form of the forward component, represents the complex exponential form of the backward component, and represents the instantaneous amplitude of the decomposed forward signal component and the instantaneous amplitude of the backward signal component, and represents the phase of the forward signal component and the phase of the backward signal component, i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, j is an imaginary unit, t represents time, f and b represent forward and reverse directions respectively;

[0041] Define an index SDI that includes the shape and direction of the frequency components:

[0042]

[0043] Among them, gim(t) represents the complex form of the current component, and It represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from this component. represents the real part, Indicates taking the imaginary part, H(*) indicates Hilbert transform, i indicates the number of channels of the harmonic component, m indicates the mth component decomposed in the channel, j is the imaginary unit, and t indicates time;

[0044] Get the major axis L of the elliptical harmonic trajectory im :

[0045]

[0046] Among them, L im represents the major axis of the orbital ellipse, and represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from the component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, and t represents time;

[0047] The angle between the horizontal probe and the elliptical main axis in the direction of the main axis of the elliptical harmonic trajectory is the average value θ of the initial phase of the forward signal and the initial phase of the reverse signal. im :

[0048]

[0049] Among them, α im and β im They represent the initial phase of the forward signal and the initial phase of the reverse signal respectively, i represents the channel number of the harmonic component, and m represents the mth component decomposed in the channel.

[0050] Optionally, the phase of the forward signal component and the phase of the backward signal component are:

[0051]

[0052] in, and represents the phase of the forward signal component and the phase of the backward signal component, ω im represents the filtering frequency, α im and β im is the initial phase of the forward signal and the initial phase of the reverse signal, i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, t represents time, f and b represent forward and reverse respectively.

[0053] Optionally, in the step of adding the forward signal to the backward signal to obtain an instantaneous harmonic trajectory IOF:

[0054] when When , the elliptical trajectory precesses positively;

[0055] when When , the elliptical trajectory moves in the opposite direction;

[0056] in, and It represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from the component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, and t represents time.

[0057] Optionally, in the step of defining an index SDI comprising the shape and direction of the frequency components:

[0058] When SDI∈[-1,1], the sign of SDI reflects the direction. When SDI>0, the frequency component is moving in the forward direction; when SDI<0, the frequency component is moving in the reverse direction. When |SDI|=0, the frequency component is moving in a straight line. When |SDI|=1, the frequency component is moving in a circular motion.

[0059] Optionally, the step of outputting the fault type based on the multi-channel neural network specifically includes the following steps:

[0060] Frequency division input fault classification characteristics;

[0061] Perform convolution pooling on the input features of each frequency separately to obtain the feature vector of each frequency band;

[0062] The features in each frequency are fused through the feature fusion layer;

[0063] The fault classification result is obtained through the softmax layer.

[0064] Correspondingly, the present application also provides a high-speed motor bearing rotor system fault detection system based on MNCSCMD-CNN, which includes a memory and a processor, the memory is used to store executable program code; the processor is connected to the memory, and runs a computer program corresponding to the executable program code by reading the executable program code to execute any step in the aforementioned high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN.

[0065] The present application provides a high-speed motor bearing rotor system fault detection method and system based on MNCSCMD-CNN. According to the rotation characteristics of the bearing rotor system, the MNCSCMD method first performs Hilbert transform on the complex signal, divides the signal into forward components and backward components with actual physical meaning, and then extracts the real part of the analytical signal to construct the objective function. After that, the alternating multiplier algorithm is used to iterate the parameters, and the center frequency of the multivariate signal is weighted to make multiple groups of IOFs with different frequencies. This method takes into account the correlation between multivariate signals, has a good effect on the decomposition of multi-channel non-stationary signals, and has modal neatness characteristics.

[0066] Based on the feature extraction and construction of POTFR, the key features that can represent the bearing-rotor system are extracted from IOFs. SDI can represent the shape and direction of orbital motion. mk represents the length of the major axis of the track, θ mkIt reflects the angle between the track and the main shaft. The shape of the track can reveal whether the rotor has runout or imbalance problems, while the direction helps determine the stability and possible abnormalities of the rotor movement. The measurement of the major axis length can reflect the size of the track of the bearing during movement, which is related to the stability of the bearing operation and possible wear conditions. The angle between the track and the main shaft can reveal the degree of alignment between the main shaft and the bearing, and improper alignment is a common cause of early failures. By extracting features of different frequencies of different bearing segments and constructing complete bearing-rotor system features, we can have a more comprehensive understanding of the operation of the entire system and achieve high-speed motor fault prediction.

[0067] The FAML-CNN-FC network can fuse the different features of fault signals at multiple frequencies, fully utilize the coupling relationship between them and perform fault classification. Using different convolution kernels for each frequency band allows the network to better specialize and learn the features of the frequency band, which is more effective than using a single convolution kernel to process the features of all frequencies; then, the features are fused through a dual feature fusion layer and a fully connected layer. This design allows the model to learn complex feature expressions from local to global, which helps capture subtle differences in the data and improves classification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0069] Figure 1 It is a flow chart of a high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by the present application;

[0070] Figure 2 It is a flow chart of step S300 in the high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by the present application;

[0071] Figure 3 It is a flow chart of step S500 in the high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by the present application;

[0072] Figure 4 It is an implementation diagram of a high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by the present application;

[0073] Figure 5It is a schematic diagram of a high-speed motor bearing in a high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided in this application;

[0074] Figure 6 It is a flow chart of the MNCSCMD algorithm in the high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by the present application;

[0075] Figure 7 It is a schematic diagram of signal decomposition in the high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by this application;

[0076] Figure 8 It is a signal synthesis diagram on a complex plane in the high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided by the present application;

[0077] Fig. 9 This is a schematic diagram of the network architecture of the high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN provided in this application. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application. In addition, it should be understood that the specific implementation methods described herein are only used to illustrate and explain the present application and are not used to limit the present application. In the present application, unless otherwise stated, the directional words used, such as "up", "down", "left", and "right", generally refer to the up, down, left, and right of the device in actual use or working state, specifically the drawing direction in the accompanying drawings.

[0079] The present application provides a high-speed motor bearing rotor system fault detection method and system based on MNCSCMD-CNN, which are described in detail below. It should be noted that the description order of the following embodiments is not intended to limit the preferred order of the embodiments of the present application. In the following embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0080] See also Figure 1The vibration response of the high-speed motor rotor bearing system often shows a strong time-varying frequency component. The analysis based on a single vibration signal is not enough to accurately extract the characteristics of the system motion, which will further affect the accuracy of the fault detection results. In order to realize the fault detection of the motor bearing rotor system, it is necessary to design a high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN. This method requires effective feature extraction of the separated harmonic components, and the extracted features are input into the fault detection network to obtain the fault category.

[0081] See also Figure 1-Figure 9 The present application provides a high-speed motor bearing rotor system fault detection method based on MNCSCMD-CNN, which specifically includes the following steps:

[0082] S100, installing vibration displacement sensors in mutually perpendicular directions on multiple sections, and extracting instantaneous characteristics of non-stationary vibration signals of the multi-section rotor;

[0083] In order to more comprehensively understand the vibration state of the complex rotor bearing system, vibration displacement sensors are installed in multiple sections in perpendicular directions to measure the vibration of the rotor shaft. There is a strong coupling relationship between the vibration signals of multiple bearing sections. If the signals are directly input into the fault classification network, it may cause interference between the signals and generate feature confusion. To solve this problem, we decided to decompose the multivariate signal and separate the harmonic components with different frequencies.

[0084] In order to detect the vibration state of multiple bearing sections, vibration displacement sensors in mutually perpendicular directions are installed on multiple sections to measure the vibration of the rotor shaft. The spectrum obtained separately from each sensor is called a half-spectrum. In the half-spectrum analysis, only a single-direction signal is used at a time, so the forward and reverse precession information of the rotor cannot be obtained. Therefore, the present invention utilizes the relative phase correlation of the vibration signal in two orthogonal directions, and can extract more features from the vibration signal.

[0085] S200, constructing a complex signal;

[0086] See also Figure 1-Figure 5 Step S200 specifically includes the following steps:

[0087] According to the two or more vibration displacement sensors orthogonally installed on each bearing, the vibration signals in mutually perpendicular directions are combined into a complex signal, which is:

[0088] g i (t) = x i (t)+jy i (t)

[0089] Among them, g i(t) represents the constructed i-th complex signal, x i (t) and y i (t) represents the signal obtained by two vibration displacement sensors installed on a bearing, j is an imaginary unit, t represents time, i=1, 2…Q, indicating that Q complex signals are constructed.

[0090] S300, converting the complex signal into a forward signal and a backward signal having physical meaning through Hilbert transform;

[0091] See also Figure 6 The Multivariate Nonlinear Complex Signals Chirp Mode Decomposition (MNCSCMD) algorithm performs multi-channel complex signal decomposition in two steps. The algorithm first converts the complex signal into forward and backward signals with physical meaning through Hilbert transform; then the multivariate nonlinear chirp mode decomposition (MNCMD) algorithm is used to analyze the multi-channel decomposition of the processed signal. The algorithm process is as follows:

[0092] Among them, the objective function is represents the minimum mode bandwidth, Q is the number of complex signals, and M groups of harmonic components are decomposed, u" i,m (t) and v" i,m (t) are two demodulated signals respectively; λ is the Lagrange multiplier; the signal decomposed by k+1 iterations And the signal decomposed by k iterations The squared norm of the difference divided by The square norm of this relative change, together with the set threshold 6, controls the cycle of the algorithm. The algorithm finally obtains M groups of harmonic components with different center frequencies.

[0093] Step S300 specifically includes the following steps:

[0094] S310, decompose the extracted complex signal, and design the following operation to convert the complex signal g i (t) is divided into forward signal and reverse signal

[0095]

[0096] in, Indicates a positive signal. represents the reverse signal, j is an imaginary unit, H(*) refers to the Hilbert transform, t represents time, f and b represent forward and reverse respectively, i=1, 2...Q, indicating that Q complex signals are constructed;

[0097] The extracted complex signal is decomposed. Considering that the signal spectrum of the complex signal is different in the forward and reverse directions and has obvious physical meaning, it can reflect the direction and angle of rotation of the object. In order to avoid mutual interference, we designed the following operation to divide the signal g(t) into the forward signal g f (t) and the reverse signal g b (t).

[0098] S320, obtaining the frequency spectrum corresponding to the forward signal and the backward signal:

[0099]

[0100] Among them, Gf(ω) represents the spectrum of the forward signal, G b (ω) represents the spectrum of the backward signal, ω represents the frequency, G(ω) represents the full spectrum of g(t), G f (ω) represents the forward spectrum, G b (ω) represents the reverse spectrum, f and b represent the forward and reverse directions respectively, and sgn(*) is the sign function;

[0101] S330, extracting the real part information of the forward signal and the reverse signal:

[0102] g + (i) = R(F -1 (G f (ω)

[0103] g - (i) = R(F -1 (G b (ω)

[0104] Among them, R(*) refers to the extraction of real part information processing, g + (t) represents the total positive signal, g - (t) represents the total reverse signal, t represents time, f and b represent forward and reverse directions respectively, ω represents frequency, and F -1 (G f (ω)) represents G f The inverse Fourier transform of (ω), F -1 (G b (ω)) represents G b (ω) is the inverse Fourier transform of

[0105] gf(t) is zero in the negative frequency region, while g bThe conjugate complex number of (t) is also zero in the negative frequency region. Both have only positive frequency parts, so g f (t) and g b (t) are all analytical signals. Since the real part of the analytical signal contains all the information, the real part information is extracted.

[0106] S400, using the MNCMD algorithm to decompose the signal, obtaining multiple groups of multi-channel signals with the same frequency, each group of multi-channel signals containing multiple harmonic components;

[0107] For the multi-channel analytical signal obtained in step S300, the MNCMD algorithm is used to decompose the signal, and multiple groups of multi-channel signals with the same frequency are obtained, and each group of signals has modal uniformity. The core idea is to define a multivariable nonlinear chirp mode in the variational optimization framework and use the ADMM (alternating direction multiplier method) algorithm to optimize the objective function to achieve signal decomposition. Figure 7 This is a schematic diagram of the breakdown of one of the channels.

[0108] Step S400 specifically includes the following steps:

[0109] S410, each group of channels is decomposed into multiple groups of signals:

[0110] S420, obtaining a reconstructed complex signal based on the decomposed signal, the reconstructed complex signal is:

[0111]

[0112] Among them, gi(t) is the signal of the i-th channel, The modal component of the mth forward signal, is the modal component of the mth reverse signal, f and b represent the forward and reverse directions respectively, t represents the time, and M represents the number of components in the channel.

[0113] According to the rotation characteristics of the bearing rotor system, the MNCSCMD method first performs Hilbert transform on the complex signal, divides the signal into forward and backward components with actual physical significance, and then extracts the real part of the analytical signal to construct the objective function. After that, the alternating multiplier algorithm is used to iterate the parameters and weight the center frequency of the multivariate signal to obtain multiple groups of IOFs with different frequencies. This method takes into account the correlation between multivariate signals, has a good effect on the decomposition of multi-channel non-stationary signals, and has modal neatness characteristics.

[0114] S500, extracting track features using multiple groups of multi-channel signals;

[0115] This application proposes a harmonic orbit instantaneous feature extraction method based on purified orbit time-frequency representation (POTFR). The method obtains a vector composite image of the instantaneous orbit by constructing forward and backward components, and extracts the motion characteristics of the orbit from it.

[0116] Step S500 specifically includes the following steps:

[0117] S510, after the signal is processed by the multivariate signal decomposition algorithm, M groups of harmonic components can be obtained, each group contains Q harmonic components from different channels, and the instantaneous orbit characteristics are constructed using these M groups of components; the decomposed forward component and backward component are transformed into a complex exponential form:

[0118]

[0119] in, and represents the instantaneous amplitude of the decomposed forward signal component and the instantaneous amplitude of the backward signal component, represents the complex exponential form of the forward component, represents the complex exponential form of the backward component, and represents the phase of the forward signal component and the phase of the backward signal component, i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, j is an imaginary unit, t represents time, f and b represent forward and reverse directions respectively;

[0120] The phase of the forward signal component and the phase of the backward signal component are:

[0121]

[0122] in, and represents the phase of the forward signal component and the phase of the backward signal component, ω im represents the filtering frequency, α im and β im is the initial phase of the forward signal and the initial phase of the reverse signal, i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, t represents time, f and b represent forward and reverse respectively.

[0123] S520, adding the forward signal and the backward signal, the instantaneous harmonic trajectory IOF can be obtained:

[0124]

[0125] in, represents the sum of the forward and backward components,

[0126] represents the complex exponential form of the forward component, represents the complex exponential form of the backward component, and represents the instantaneous amplitude of the decomposed forward signal component and the instantaneous amplitude of the backward signal component, and represents the phase of the forward signal component and the phase of the backward signal component; i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, j is an imaginary unit, t represents time, f and b represent forward and reverse directions respectively;

[0127] like Figure 8 As shown, the elliptical orbit can be expressed as the sum of two circular orbits: one is the trajectory of the vector in the direction of rotation (forward), and the other is the trajectory of the vector in the opposite direction of rotation (reverse). The two vectors rotate at the same frequency, and the instantaneous position of the rotor on its filtered orbit can be expressed as the sum of the instantaneous position vectors on the forward and reverse orbits.

[0128] when When , the elliptical trajectory precesses positively;

[0129] when When , the elliptical trajectory moves in the opposite direction;

[0130] in, and It represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from the component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, and t represents time.

[0131] S530, define an index SDI including the shape and direction of the frequency component:

[0132]

[0133] Among them, g im (t) represents the plural form of the current component, and It represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from this component. represents the real part, Indicates taking the imaginary part, H(*) indicates Hilbert transform, i indicates the number of channels of the harmonic component, m indicates the mth component decomposed in the channel, j is the imaginary unit, and t indicates time;

[0134] It is easy to know that SDI∈[-1,1], the sign of SDI reflects the direction, when SDI>0, the frequency component is moving in the positive direction; when SDI<0, the frequency component is moving in the negative direction. The value determines the shape of the movement, when |SDI|=0, it means that the frequency component is moving in a straight line, |SDI|=1, it means that the component on the frequency is moving in a perfect circle, and the closer the value is to 1, the closer the movement is to a perfect circle.

[0135] S540, obtaining the major axis L of the elliptical harmonic trajectory im :

[0136]

[0137] Among them, L im represents the major axis of the orbital ellipse, and represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from the component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, and t represents time;

[0138] S550, obtain the main axis direction of the elliptical harmonic trajectory. The angle between the horizontal probe and the main axis of the ellipse is the average value θ of the initial phase of the forward and reverse signals. im :

[0139]

[0140] Among them, α im and β im They represent the initial phase of the forward signal and the initial phase of the reverse signal respectively, i represents the channel number of the harmonic component, and m represents the mth component decomposed in the channel.

[0141] The multi-channel signal decomposed by MNCSCMD is extracted by POTFR, and the characteristic components obtained are SDI, L im and θ im , these characteristic components include the rotation shape, direction and angle between the main axis and the orbit of each frequency in the bearing rotor motion system.

[0142] Based on the feature extraction and construction of POTFR, the key features that can represent the bearing-rotor system are extracted from IOFs. SDI can represent the shape and direction of orbital motion. im represents the length of the major axis of the track, θ imIt reflects the angle between the track and the main shaft. The shape of the track can reveal whether the rotor has runout or imbalance problems, while the direction helps determine the stability and possible abnormalities of the rotor movement. The measurement of the major axis length can reflect the size of the track of the bearing during movement, which is related to the stability of the bearing operation and possible wear conditions. The angle between the track and the main shaft can reveal the degree of alignment between the main shaft and the bearing, and improper alignment is a common cause of early failures. By extracting features of different frequencies of different bearing segments and constructing complete bearing-rotor system features, we can have a more comprehensive understanding of the operation of the entire system and achieve high-speed motor fault prediction.

[0143] S600, output fault type based on multi-channel neural network;

[0144] FAML-CNN-FC (Frequency Aware Multi Layer Convolutional NeuralNetwork Fusion Classifier) ​​is designed to perform fault classification of high-speed motor bearing rotor system.

[0145] The network structure diagram of fault classification is as follows Fig. 9 The network has two characteristics. First, in order to extract the correlation between the same-frequency signals of each bearing segment, the features are input into the network by frequency division. Second, in order to comprehensively consider the signals of different frequencies of each bearing segment, a dual feature fusion layer is designed to perform feature fusion on the signals after convolution pooling.

[0146] The network mainly includes convolutional pooling layer, feature fusion layer, fully connected layer and softmax layer. Figure 7 , the left frame is the separate processing of the divided frequency signal, and the right frame is the processing of all signals. First, the fault classification features are divided into different frequencies, and the input features of each frequency are convoluted and pooled separately to obtain the feature vector of each frequency band; then, the features in each frequency are fused through the feature fusion layer to provide more comprehensive information support for the next softmax layer; finally, the fault classification result is obtained through the softmax layer.

[0147] In the dual feature fusion layer, the features within each frequency band are first processed, and the features of m channels are fused; then the features of all frequency bands are further fused. In the fusion of features within each frequency band, we first define a trainable weight variable for all features. During the training process, these weights are automatically updated and weighted for each feature vector. The weighted feature vectors are added together to form a fused feature vector. After the initial fusion is completed, the fused vectors generated by each frequency band are fused again. This design can make full use of the information of each frequency band and different feature channels to improve the accuracy and comprehensiveness of feature expression.

[0148] The FAML-CNN-FC network can fuse the different features of fault signals at multiple frequencies, make full use of the coupling relationship between them and perform fault classification. First, all the features are divided into different frequencies and input into the convolutional neural network, and the features of each frequency are separately convolved and pooled. Using different convolution kernels for each frequency band can make the network better specialize in learning the features of the frequency band, which is more effective than using a single convolution kernel to process the features of all frequencies. After that, the features are fused through the dual feature fusion layer and the fully connected layer. This design allows the model to learn complex feature expressions from local to global, which helps to capture the subtle differences in the data and improve the accuracy of classification.

[0149] The above is a detailed introduction to the high-speed motor bearing rotor system fault detection method and system based on MNCSCMD-CNN provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A high-speed motor bearing rotor system fault detection method, characterized in that: include: Install vibration displacement sensors in mutually perpendicular directions on multiple sections, and extract the instantaneous characteristics of the non-stationary vibration signals of the multi-section rotor; construct complex signals; The complex signal is converted into forward signal and backward signal with physical meaning through Hilbert transform; The MNCMD algorithm is used to decompose the signal and obtain multiple groups of multi-channel signals with the same frequency. Each group of multi-channel signals contains multiple harmonic components. Extracting the characteristics of the track using multiple sets of multi-channel signals; Based on the multi-channel neural network output fault type, Among them, the complex signal is converted into forward signal and backward signal with physical meaning through Hilbert transform, including: Decompose the extracted complex signal and design the following operation to convert the complex signal g i (t) is divided into forward signal and reverse signal in, Indicates a positive signal, represents the reverse signal, j is an imaginary unit, H(*) refers to the Hilbert transform, t represents time, f and b represent forward and reverse respectively, i=1,2......Q, indicating that Q complex signals are constructed; Get the spectrum corresponding to the forward signal and the backward signal: Where ω represents frequency, G(ω) represents the full spectrum of g(t), and G f (ω) represents the forward spectrum, G b (ω) represents the reverse spectrum, f and b represent the forward and reverse directions, respectively, and sgn(*) represents the sign function; Extract the real part of the forward and reverse signals: g + (t)=R(F -1 (G f (ω)) g - (t)=R(F -1 (G b (ω)) Among them, g + (t) represents the total forward signal, g-(t) represents the total reverse signal, t represents time, f and b represent forward and reverse respectively, ω represents frequency, and F -1 (G f (ω)) represents G f The inverse Fourier transform of (ω), F -1 (G b (ω)) represents G b (ω) is the inverse Fourier transform, R(*) refers to the extraction of real part information processing, Among them, the MNCMD algorithm is used to realize signal decomposition, and multiple groups of multi-channel signals with the same frequency are obtained. Each group of multi-channel signals contains multiple harmonic components, including: The signals in each group of channels are decomposed into M groups of signals: A reconstructed complex signal is obtained based on the decomposed signal, and the reconstructed complex signal is: Among them, g i (t) is the signal of the ith channel, The modal component of the mth forward signal, is the modal component of the mth reverse signal, f and b represent the forward and reverse directions respectively, t represents the time, and M represents the number of components in the channel. Among them, multiple groups of multi-channel signals are used to extract the characteristics of the track, including: Transform the decomposed forward and backward components into complex exponential form: in, and Represents the instantaneous amplitude of the decomposed forward signal component and the instantaneous amplitude of the backward signal component; and represents the phase of the forward signal component and the phase of the backward signal component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, j is an imaginary unit, t represents time, f and b represent forward and reverse directions respectively; Add the forward signal to the backward signal to get the instantaneous harmonic trajectory IOF: in, represents the sum of the forward and backward components, and represents the instantaneous amplitude of the decomposed forward signal component and the instantaneous amplitude of the backward signal component, represents the complex exponential form of the forward component, represents the complex exponential form of the backward component, and represents the phase of the forward signal component and the phase of the backward signal component, i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, j is an imaginary unit, t represents time, f and b represent forward and reverse directions respectively; Define an index SDI that includes the shape and direction of the frequency components: Among them, g im (t) represents the plural form of the current component, and It represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from this component. represents the real part, Indicates taking the imaginary part, H(*) indicates Hilbert transform, i indicates the number of channels of the harmonic component, m indicates the mth component decomposed in the channel, j is the imaginary unit, and t indicates time; Get the major axis L of the elliptical harmonic trajectory im : Among them, L im represents the major axis of the orbital ellipse, and represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from the component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, and t represents time; The angle between the horizontal probe and the elliptical main axis in the direction of the main axis of the elliptical harmonic trajectory is the average value θ of the initial phase of the forward signal and the initial phase of the reverse signal. im : Among them, α im and β im They represent the initial phase of the forward signal and the initial phase of the reverse signal respectively, i represents the channel number of the harmonic component, and m represents the mth component decomposed in the channel.

2. The high-speed motor bearing rotor system fault detection method according to claim 1, characterized in that: Constructing a complex signal includes: According to two or more vibration displacement sensors orthogonally installed on each bearing, the vibration signals in mutually perpendicular directions are combined into complex signals, and the complex signal is: g i (t)=x i (t)+you i (t) Among them, g i (t) represents the constructed i-th complex signal, x i (t) and y i (t) represents the signal obtained by two vibration displacement sensors installed on a bearing, j is an imaginary unit, t represents time, i=1,2...Q, indicating that Q complex signals are constructed.

3. The high-speed motor bearing rotor system fault detection method according to claim 1, characterized in that: The phase of the forward signal component and the phase of the backward signal component are: in, and represents the phase of the forward signal component and the phase of the backward signal component, ω im represents the filtering frequency, α im and β im is the initial phase of the forward signal and the initial phase of the reverse signal, i represents the channel number of the harmonic component, m represents the mth component decomposed in the channel, t represents time, f and b represent forward and reverse respectively.

4. The high-speed motor bearing rotor system fault detection method according to claim 1, characterized in that: In the step of adding the forward signal to the backward signal to obtain the instantaneous harmonic trajectory IOF: when When , the elliptical trajectory precesses positively; when When , the elliptical trajectory moves in the opposite direction; in, and It represents the instantaneous amplitude of the forward signal component and the instantaneous amplitude of the backward signal component decomposed from the component, i represents the channel number of the harmonic component, m represents the mth component decomposed from the channel, and t represents time.

5. The high-speed motor bearing rotor system fault detection method according to claim 1, characterized in that: In the step of defining an index SDI comprising the shape and direction of the frequency components: When SDI∈[-1,1], the sign of SDI reflects the direction. When SDI>0, the frequency component is moving in the forward direction; when SDI<0, the frequency component is moving in the reverse direction. When |SDI|=0, the frequency component is moving in a straight line. When |SDI|=1, the frequency component is moving in a circle.

6. The high-speed motor bearing rotor system fault detection method according to claim 1, characterized in that: Output fault types based on multi-channel neural network, including: Frequency division input fault classification characteristics; Perform convolution pooling on the input features of each frequency separately to obtain the feature vector of each frequency band; The features in each frequency are fused through the feature fusion layer; The fault classification result is obtained through the softmax layer.

7. High-speed motor bearing rotor system fault detection system, characterized in that: include: A memory for storing executable program codes; and a processor connected to the memory, which runs a computer program corresponding to the executable program code by reading the executable program code, so as to execute the high-speed motor bearing rotor system fault detection method according to any one of claims 1-6.

Images