Fault diagnosis method for rotating equipment based on multivariate parameter estimation and image recognition

Abstract

The present invention proposes a method for fault diagnosis of rotating equipment based on multivariate parameter estimation and image recognition, comprising the following steps: S1, using a filtering method to implement signal noise reduction processing on the original vibration acceleration signal, and integrating the acceleration signal in the frequency domain to obtain a velocity signal; S2, determining the fault frequency band, extracting the signal in the fault frequency band and performing envelope spectrum transformation to achieve feature extraction; S3, calculating the feature weight value corresponding to each feature according to the equipment failure probability and the characteristic frequency; S4, calculating the estimated value and residual value of each characteristic parameter in the verification data through a multi-parameter state estimation technology; S5, establishing a two-dimensional grid map based on frequency and residual or a three-dimensional grid map based on time, frequency and residual, identifying key change information in the image, and realizing fault diagnosis according to key features. The fault diagnosis accuracy is greatly improved, and the diagnosis efficiency is also improved.

Description

Technical Field

[0001] The invention relates to the technical field of fault diagnosis, in particular to a rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition. Background Art

[0002] There are a large number of rotating equipment in various industrial sites. The rotating equipment has a complex structure and is prone to failure due to the environment and noise. Any failure of any type of rotating equipment may cause unplanned system downtime, and the staff will need to spend a lot of time to locate the fault, repair, order replacement parts, etc., resulting in huge economic losses. Therefore, under strong noise and multiple interference conditions, it is of great significance to timely detect equipment operation abnormalities and accurately diagnose the type of rotating equipment failure for the status monitoring of rotating machinery systems.

[0003] In recent years, with the deepening of machine learning research, data models have gradually been applied to rotating equipment fault diagnosis. The common method is to extract features from the collected vibration signals, label different fault type data, and then classify the data through various classifier methods, such as support vector machines, nearest neighbors, neural networks, decision trees, etc. There is also a multivariate state estimation algorithm, which compares the current operating data with the generated historical operating state, calculates and compares the similarity between multi-state variables, and thus realizes fault diagnosis. For the extraction method of feature parameters, there are time domain feature parameters, frequency domain feature parameters, and signal processing methods such as envelope spectrum analysis, wavelet analysis and empirical mode decomposition are applied to process the vibration signal to obtain each component signal, and then the time domain and frequency domain feature parameters of different component signals are extracted. If the equipment working condition is unstable, there is a deviation between the abnormal characteristic frequency of the equipment and the theoretical characteristic frequency, and the method of directly extracting features is easy to ignore some key frequencies. In addition, for early fault signals, the noise signal in the signal cannot be effectively eliminated, resulting in the inability to extract the key fault signal in the equipment, resulting in low fault identification accuracy.

[0004] The "A Fault Diagnosis Method and Fault Diagnosis System" disclosed in the Chinese patent literature has a publication number of CN108803573A. The embodiment of the present application discloses a fault diagnosis method and a fault diagnosis system, so that the fault diagnosis personnel can quickly obtain fault data, thereby realizing timely diagnosis of the vibration fault of the rotating equipment and improving the accuracy of the vibration fault diagnosis of the rotating equipment. The method of the embodiment of the present application is applied to the fault diagnosis system of the rotating equipment. The fault diagnosis system includes at least one rotating equipment vibration diagnosis system TDM and at least one client DC. At least one TDM and at least one DC have mutual communication functions. The fault data of the rotating equipment is stored in at least one TDM as a target block. The specific implementation method is as follows: when the target rotating equipment has a vibration fault, the fault diagnosis system obtains the fault data of the target rotating equipment; the fault diagnosis system sends the fault data of the target rotating equipment to at least one DC; the fault diagnosis system receives the diagnosis result fed back by at least one DC, and the diagnosis result is obtained by analyzing the fault data. However, the invention does not solve the specific problems encountered in the process of rotating equipment fault diagnosis and data processing, and does not propose corresponding algorithms for the corresponding problems. Summary of the invention

[0005] In order to solve the problems of difficulty in extracting characteristic parameters, consistent characteristic weights, and low fault diagnosis accuracy in the existing rotating equipment fault diagnosis methods, the present invention proposes a rotating equipment fault diagnosis method combining multivariate parameter estimation and image recognition. Signal noise removal and feature extraction are achieved through filtering, fault frequency band selection, and envelope spectrum analysis methods. The model is established with the spectrum of the full frequency band of the signal as the characteristic parameter, and all the features of the original data are retained; a characteristic parameter weight calculation model is established according to the equipment type and failure mode, and different weight values ​​are assigned to the characteristic parameters. The estimated value and residual calculation of the sample to be observed are realized through the multi-parameter state estimation method, and the estimation accuracy of each parameter is improved. Finally, a three-dimensional grid diagram of time, frequency and residual or a two-dimensional grid diagram of frequency and residual is established to identify key change information in the image and realize fault identification. The fault diagnosis accuracy is greatly improved, and the diagnosis efficiency is also improved.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: a rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition, comprising the following steps:

[0007] S1, using a filtering method to implement signal noise reduction processing on the original vibration acceleration signal, and performing frequency domain integration on the acceleration signal to obtain a velocity signal;

[0008] S2, determine the fault frequency band, extract the signal within the fault frequency band and perform envelope spectrum transformation to achieve feature extraction;

[0009] S3, calculating the feature weight value corresponding to each feature according to the equipment failure probability and feature frequency;

[0010] S4, calculating the estimated value and residual value of each characteristic parameter in the verification data by multi-parameter state estimation technology;

[0011] S5, establishing a two-dimensional grid map based on frequency and residual or a three-dimensional grid map based on time, frequency and residual, identifying key change information in the image, and realizing fault diagnosis according to key features.

[0012] In the present invention, feature extraction is performed by combining Wiener filtering, spectral kurtosis and envelope spectrum analysis, and weights are established according to the type of rotating equipment, the operating condition of the rotating equipment and the failure mode. Signal envelope spectrum data is used to establish a multi-parameter state estimation method model, and finally a parameter state estimation method residual model is established using a three-dimensional grid diagram. Fault diagnosis is performed by identifying key change information of the image through an adaptive threshold. By identifying the key change information in the image, fault identification is achieved. In view of the problems in the prior art that some key features are easily missed and the influence of feature parameter weights is not considered, a method is proposed that considers feature parameter weights, can adaptively extract fault features in feature signals, and can calculate the weights of each feature parameter under the model for different fault models. Fault mode identification is achieved through multi-parameter state estimation technology. By identifying key image information, the accuracy of fault diagnosis is improved, and the diagnostic efficiency is also improved.

[0013] Preferably, the step S1 comprises the following steps:

[0014] S11, vibration acceleration signal s(t) can be expressed as:

[0015] s(t)=x(t)+d(t)

[0016] Among them, x(t) represents the equipment vibration signal, and d(t) represents the noise signal;

[0017] Perform discrete S transform on s(t):

[0018]

[0019] Where ω(t) represents the window function;

[0020] S12, after defining the domain radius and window function, adjust the resolution of the S transform and derive the digital filter H(z):

[0021]

[0022] Among them, P xy is the cross power spectral density of x(t) and s(t), P yis the power spectral density of s(t), H(z) is the digital filter transfer function in the z domain, and γ is the domain radius;

[0023] S13, calculate the optimal filter according to the minimum mean square error criterion, and perform discrete S inverse transform on the signal, and the vibration signal x(t) can be identified as:

[0024]

[0025] S14, the vibration signal x(t) is integrated in the frequency domain to obtain the velocity signal v(t). The numerical calculation formula of the first integration is:

[0026]

[0027]

[0028] f l , f r They represent the lower cutoff frequency and the upper cutoff frequency respectively; X(k) represents the Fourier transform of x(t), Δf represents the frequency resolution, k is the data point corresponding to the vibration signal x(t), H(k) is the frequency coefficient, N is the number of data of x(t), and γ is the domain radius.

[0029] In the present invention, the signal noise reduction is performed by using the Wiener filtering method, and then the acceleration signal after noise reduction is integrated to obtain the velocity signal, thereby realizing signal noise removal.

[0030] Preferably, step S2 comprises the following steps:

[0031] S21, set the decomposition layer number k from 2 to n, construct multiple bandpass and bandstop filters respectively, filter the signal, and obtain 2 k component signals, calculate the kurtosis value of each component signal, assuming that the length of the vibration signal is N, the component signal is represented by d, and the formula for the kurtosis value is:

[0032]

[0033] S22, determining the optimal decomposition layer number and the optimal filter coefficient of the signal according to the maximum value of the kurtosis value, performing convolution calculation on the signal v(t) and the filter coefficient to obtain a fault characteristic signal;

[0034] S23, performing envelope spectrum transformation on the fault characteristic signal to obtain frequency domain information of the fault characteristic signal.

[0035] In the present invention, a spectral kurtosis method is used to select the fault frequency band, and then the signal feature extraction is realized through envelope spectrum transformation.

[0036] Preferably, step S3 comprises the following steps:

[0037] S31, calculating the fault characteristic frequency according to the equipment rotation speed and equipment type;

[0038] S32, count the probability of equipment failure p1 and the probability of abnormality of the characteristic frequency of equipment failure p2, and calculate the weight value k of each frequency in the frequency domain information of the fault characteristic signal according to the formula f ;

[0039]

[0040] Where, f is the frequency; m is the number of equipment failure types;

[0041] K f Perform normalization to obtain the final weight value of each frequency

[0042]

[0043] In the present invention, the equipment failure probability and characteristic frequency can be quickly calculated through some common parameters or state quantities of the rotating equipment.

[0044] Preferably, step S4 comprises the following steps:

[0045] S41, taking the frequency domain information of the characteristic signal as the characteristic parameter, establishing a sample parameter matrix D, in which each row represents the parameter value in different samples, and each column represents the frequency domain information of a set of vibration data. Assuming that the number of characteristic parameters is n and the number of samples is m, the sample matrix D i It can be expressed as:

[0046]

[0047] S42, calculate the estimated value of the test set data under the sample parameter matrix, assuming that the test set data is x obs , estimated value x est Calculated by the formula:

[0048] x est =D·W

[0049] Where W represents the weight value vector;

[0050] The weight value vector W can be expressed as:

[0051]

[0052] in, Refers to Euclidean distance, Pearson correlation coefficient, Manhattan distance,

[0053] After the weight value vector is determined, the estimated value x of the test set dataest It can be expressed as:

[0054]

[0055] S43, calculate the estimated value x est With the test set data x obs The residual res is:

[0056] res=x obs -x est .

[0057] In the present invention, residual values ​​and estimated values ​​are used to lay the foundation for establishing a grid diagram.

[0058] Preferably, step S5 comprises the following steps:

[0059] S51, establishing a three-dimensional grid diagram of the residual values, frequencies and times of all test set data, where the horizontal axis corresponds to time and the vertical axis corresponds to the frequency value, and the corresponding sizes of the frequency residual values ​​are distinguished by color, or establishing a grid diagram of the frequencies and residual values ​​of all test set data, where the horizontal axis corresponds to the frequency value;

[0060] S52, performing grayscale processing on the established three-dimensional grid image to obtain a grayscale image;

[0061] S53, establishing an image threshold, and then performing filtering processing on the image;

[0062] S54, determining the frequency value corresponding to the high brightness area in the image, and performing fault diagnosis according to the characteristic frequency value.

[0063] In the present invention, the fault can be accurately diagnosed by means of the grid map and the image processing of the grid map.

[0064] The beneficial effects of the present invention are as follows: the present invention proposes a method for fault diagnosis of rotating equipment based on multivariate parameter estimation and image recognition, realizes signal noise removal and signal processing through filtering method, fault characteristic frequency band acquisition and envelope spectrum analysis method, establishes a model with the spectrum of the full frequency band of the signal as the feature, and retains all the features of the original data; establishes a characteristic parameter weight calculation model according to the equipment type and failure mode, assigns different weight values ​​to the characteristic parameters, realizes the estimated value and residual calculation of the sample to be observed through the multi-parameter state estimation method, and improves the estimation accuracy of each parameter. Finally, a three-dimensional grid diagram of time, frequency and residual or a two-dimensional grid diagram of frequency and residual is established, and fault identification is realized by identifying the key change information in the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a flow chart of this embodiment;

[0066] Figure 2 Schematic diagram of estimated values ​​and residual values ​​of the 700th set of data in this embodiment;

[0067] Figure 3 is a three-dimensional grid diagram corresponding to the time, frequency and frequency amplitude residual value of this embodiment;

[0068] Figure 4 is a three-dimensional grid diagram corresponding to time, frequency and frequency amplitude of this embodiment;

[0069] Figure 5 is a result diagram of the three-dimensional grid diagram corresponding to the time, frequency and frequency amplitude residual value of this embodiment after filtering;

[0070] Figure 6 It is the result of filtering the three-dimensional grid graph corresponding to time, frequency and frequency amplitude. DETAILED DESCRIPTION

[0071] Example:

[0072] This embodiment proposes a rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition. Figure 1 , a rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition, in this embodiment, taking a water pump as an example, the acceleration data of the pump drive end is collected at a rotation speed of 1485r / min, the sampling frequency is 20kHz, the single data collection time is 1s, and a total of 1330 groups of vibration data are selected. It includes the following steps:

[0073] Step S1, first use the filtering method to perform signal noise reduction on the original vibration acceleration signal, and then perform frequency domain integration on the acceleration signal to obtain the velocity signal; the Wiener filtering method is one of the signal noise reduction methods, and other methods are also applicable. In this step, there are mainly step S11, first the acceleration signal s(t) contains two parts: the equipment vibration signal x(t) and the noise signal d(t), namely:

[0074] s(t)=x(t)+d(t),

[0075] The acceleration signal s(t) is then processed by discrete S transform:

[0076]

[0077] In the above formula, ω(t) refers to the window function, and the Gaussian window function is selected here.

[0078] In step S12, after defining the domain radius and the window function, the resolution of the S transform is adjusted and the digital filter H(z) is derived:

[0079]

[0080] In the above formula, P xy Refers to the cross power spectral density of x(t) and s(t), P y refers to the power spectral density of s(t), H(z) refers to the digital filter transfer function in the z-domain, and γ refers to the domain radius.

[0081] Then, step S13 is performed to calculate the optimal filter according to the minimum mean square error criterion, and then the signal is processed by discrete S inverse transform. At this time, the vibration signal x(t) is expressed as:

[0082]

[0083] Step S14 mainly uses frequency domain integration to obtain the velocity signal. Specifically, the vibration signal x(t) is processed by frequency domain integration to obtain the velocity signal v(t). For details, refer to the following formula:

[0084]

[0085]

[0086] f l , refers to the lower cutoff frequency, f r refers to the upper cutoff frequency, X(k) refers to the Fourier transform of x(t), Δf refers to the frequency resolution, k refers to the data point corresponding to the vibration signal x(t), H(k) refers to the frequency coefficient, N refers to the number of data points of x(t), and γ specifically refers to the domain radius. Here, f l Select 0, f r Select 10k and Δf as 1.

[0087] Step S2, determine the fault frequency band, extract the signal in the fault frequency band and perform envelope spectrum transformation to achieve feature extraction; the spectral kurtosis method is a method for selecting the fault frequency band, and other fault frequency band methods are also applicable. In this step, there are mainly step S21, first, it is necessary to set the decomposition layer number k from 2 to n, build several bandpass and bandstop filters, and then filter the signal. After the processing is completed, 2 k component signals, and the kurtosis value k of each component signal is obtained by referring to the following formula:

[0088]

[0089] In the above formula, N refers to the length of the vibration signal, and d refers to the component signal.

[0090] Then, the calculation of step S22 is performed. First, the optimal number of decomposition layers and the optimal filter of the signal are determined, both of which are determined based on the maximum value of the kurtosis value. Then, the signal v(t) and the filter coefficient are convolved and processed. Finally, the fault characteristic signal can be obtained. The optimal number of decomposition layers is 3, and the frequency range corresponding to the optimal filter coefficient is 2500~3750Hz.

[0091] Step S23, transforming the envelope spectrum of the fault characteristic signal obtained in the previous step, and finally obtaining the frequency domain information of the fault characteristic signal.

[0092] Step S3, calculate the characteristic weight value corresponding to each characteristic according to the equipment failure probability and characteristic frequency; in this step, it mainly includes step S31, which mainly calculates the frequency of the fault characteristic based on the rotation speed of the rotating equipment and the type of the rotating equipment. According to the calculation, the equipment rotation frequency value is 24.75Hz, the corresponding double frequency is 45.5Hz, the characteristic frequency of the inner ring of the bearing is 122.19Hz, the characteristic frequency of the outer ring of the bearing is 75.8Hz, the characteristic frequency of the rolling element of the bearing is 99.8Hz, and the characteristic frequency of the bearing retainer is 9.45Hz.

[0093] In the subsequent step S32, the failure probability p1 of the rotating equipment and the probability p2 of abnormality of the characteristic frequency of the rotating equipment failure are firstly counted to obtain the weight value k of each frequency in the frequency domain information of the fault characteristic signal. f :

[0094]

[0095] In the above formula, f refers to the frequency; m refers to the number of equipment failure types;

[0096] Then the weight value k is normalized f , find the final weight value of each frequency

[0097]

[0098] Step S4, calculate the estimated value and residual value of each characteristic parameter in the verification data by multi-parameter state estimation technology; specifically, this step mainly includes step S41, first create a sample parameter matrix D, the characteristic parameter is the frequency domain information of the characteristic signal, select 500 groups of vibration data under normal operation of the equipment to establish the sample parameter matrix D, each row of the matrix represents the parameter value in different samples, each column of the matrix represents the frequency domain information of a group of vibration data, select the frequency within 0 to 200 Hz in the envelope spectrum as the characteristic parameter, the frequency resolution is 1 Hz, then the number of characteristic parameters is 200, the number of samples is 500, and the sample matrix D i Please refer to the following formula for details:

[0099]

[0100] In the above formula, n refers to the number of feature parameters, and m refers to the number of samples.

[0101] Step S42, this step is mainly to find the estimated value of the test set data under the sample parameter matrix, which is found by the following formula:

[0102] x est =D·W

[0103] In the above formula, W specifically refers to the weight value vector, x obs refers to the test set data, x est Refers to the estimated value of the test set data;

[0104] Since the weight value vector is specifically:

[0105]

[0106] In the above formula, Refers to Euclidean distance, Pearson correlation coefficient, and Manhattan distance;

[0107] Substitute the above formula into x est =D·W, we can get:

[0108]

[0109] Finally, step S43 is performed to obtain the residual res by using the estimated value and the test set data:

[0110] res=x obs -x est

[0111] For details, please refer to Figure 2 are the estimated and residual values ​​for the 700th set of data.

[0112] S5, establish a two-dimensional grid map based on frequency and residual or a three-dimensional grid map based on time, frequency and residual, identify key change information in the image, identify key change information in the image through adaptive thresholds, and implement fault diagnosis based on key features. In this step, there are mainly steps S51: create a three-dimensional grid map containing all test set data with the horizontal axis being time, the vertical axis being frequency value, and the frequency residual value being distinguished by color, refer to Figure 3 and Figure 4 , Figure 3 is the three-dimensional grid diagram corresponding to the time, frequency and frequency amplitude residual values, Figure 4 It is a three-dimensional grid diagram corresponding to time, frequency and frequency amplitude. The time covers the data of the entire cycle of the equipment from normal to failure, and a two-dimensional grid diagram can also be created.

[0113] In step S52, the three-dimensional grid image is converted into a grayscale image after grayscale processing.

[0114] Step S53, according to the results of the above steps, use the Gaussian mean of the neighborhood to customize the image threshold, and after completion, filter the image. For details, refer to Figure 5 and Figure 6 .

[0115] In the final step S54, the frequency value corresponding to the high brightness area in the image is judged, and the fault diagnosis is based on the characteristic frequency value. The frequency of the highlighted area is 25Hz, which corresponds to the rotation frequency. That is, the amplitude of the rotation frequency changes significantly, and it can be inferred that the fault is related to the rotation frequency change. Figure 5 and Figure 6 contrast, Figure 6 There is no obvious distinction between the vibration data of the equipment under normal operation and the vibration data under fault. Figure 5 ) can effectively identify the time when equipment is abnormal and the frequency information of changes when equipment is abnormal, which is helpful for fault diagnosis.

[0116] The specific process of the present invention is mainly as follows: combining Wiener filtering, spectral kurtosis and envelope spectrum analysis to extract features, establishing weights according to the type of rotating equipment, the operating condition of the rotating equipment and the failure mode, using signal envelope spectrum data to establish a multi-parameter state estimation method model, and finally using a three-dimensional grid diagram to establish a parameter state estimation method residual model, and using adaptive thresholds to identify key image change information for fault diagnosis. By identifying key change information in the image, fault identification is achieved. In view of the problems in the prior art that some key features are easily missed and the influence of feature parameter weights is not considered, a method that considers feature parameter weights, can adaptively extract fault features in feature signals, and can calculate the weights of each feature parameter under the model for different fault models, and realizes fault mode recognition through multi-parameter state estimation technology. By identifying key image information, the accuracy of fault diagnosis is improved, and the diagnostic efficiency is also improved.

[0117] In the present invention, the signal noise reduction is performed by using the Wiener filtering method, and then the acceleration signal after noise reduction is integrated to obtain the velocity signal, thereby realizing signal noise removal.

[0118] In the present invention, a spectral kurtosis method is used to select the fault frequency band, and then the signal feature extraction is realized through envelope spectrum transformation.

[0119] In the present invention, the equipment failure probability and characteristic frequency can be quickly calculated through some common parameters or state quantities of the rotating equipment.

[0120] In the present invention, residual values ​​and estimated values ​​are used to lay the foundation for establishing a grid diagram.

[0121] In the present invention, the fault can be accurately diagnosed by means of the grid map and the image processing of the grid map.

[0122] The above embodiments are further elaborations and illustrations of the present invention for ease of understanding, and are not limitations of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition, characterized in that: The following steps are involved: S1, using a filtering method to implement signal noise reduction processing on the original vibration acceleration signal, and integrating the acceleration signal in the frequency domain to obtain a velocity signal; S2, determine the fault frequency band, extract the signal in the fault frequency band and perform envelope spectrum transformation to achieve feature extraction; use the spectral kurtosis method to select the fault frequency band; S3, calculating the feature weight value corresponding to each feature according to the equipment failure probability and feature frequency; S4, calculating the estimated value and residual value of each characteristic parameter in the verification data by multi-parameter state estimation technology; The estimated value of the test set data under the sample parameter matrix is ​​the product of the sample parameter matrix and the weight value vector, and the residual value is the difference between the test set data and the estimated value of the test set data; each row of the matrix is ​​the parameter value in different samples, and each column of the matrix is ​​the frequency domain information of a group of vibration data; S5, establishing a two-dimensional grid map based on frequency and residual or a three-dimensional grid map based on time, frequency and residual, identifying key change information in the image, and realizing fault diagnosis according to key features.

2. The rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition according to claim 1 is characterized in that: The step S1 comprises the following steps: S11, vibration acceleration signal s(t) can be expressed as: s(t)=x(t)+d(t) Among them, x(t) represents the equipment vibration signal, and d(t) represents the noise signal; Perform discrete S transform on s(t): Where ω(t) represents the window function; S12, after defining the domain radius and window function, adjust the resolution of the S transform and derive the digital filter H(z): Among them, P xy is the cross power spectral density of x(t) and s(t), P y is the power spectral density of s(t), H(z) is the digital filter transfer function in the z-domain, and γ is the domain radius. S13, calculate the optimal filter according to the minimum mean square error criterion, and perform discrete S inverse transform on the signal, and the vibration signal x(t) can be identified as: S14, the vibration signal x(t) is integrated in the frequency domain to obtain the velocity signal v(t). The numerical calculation formula of the first integration is: f l , f r They represent the lower cutoff frequency and the upper cutoff frequency respectively; X(k) represents the Fourier transform of x(t), Δf represents the frequency resolution, k is the data point corresponding to the vibration signal x(t), H(k) is the frequency coefficient, N is the number of data of x(t), and γ is the domain radius.

3. The rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition according to claim 1 is characterized in that: The step S2 comprises the following steps: S21, set the decomposition layer number k from 2 to n, construct multiple bandpass and bandstop filters respectively, filter the signal, and obtain 2 k component signals, calculate the kurtosis value of each component signal, assuming that the length of the vibration signal is N, the component signal is represented by d, and the formula for the kurtosis value is: S22, determining the optimal decomposition layer number and the optimal filter coefficient of the signal according to the maximum value of the kurtosis value, performing convolution calculation on the signal v(t) and the filter coefficient to obtain a fault characteristic signal; S23, performing envelope spectrum transformation on the fault characteristic signal to obtain frequency domain information of the fault characteristic signal.

4. The rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition according to claim 3 is characterized in that: The step S3 comprises the following steps: S31, calculating the fault characteristic frequency according to the equipment rotation speed and equipment type; S32, count the probability of equipment failure p1 and the probability of abnormality of the characteristic frequency of equipment failure p2, and calculate the weight value k of each frequency in the frequency domain information of the fault characteristic signal according to the formula f : Where, f is the frequency; m is the number of equipment failure types; K f Perform normalization to obtain the final weight value of each frequency 5. The rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition according to claim 1 is characterized in that: The step S4 comprises the following steps: S41, taking the frequency domain information of the characteristic signal as the characteristic parameter, establishing a sample parameter matrix D, in which each row represents the parameter value in different samples, and each column represents the frequency domain information of a set of vibration data. Assuming that the number of characteristic parameters is n and the number of samples is m, the sample matrix D i It can be expressed as: S42, calculate the estimated value of the test set data under the sample parameter matrix, assuming that the test set data is x obs , estimated value x est Calculated by the formula: x est =D·W Where W represents the weight value vector; The weight value vector W can be expressed as: in, Represents Euclidean distance, Pearson correlation coefficient, Manhattan distance, After the weight value vector is determined, the estimated value x of the test set data est It can be expressed as: S43, calculate the estimated value x est With the test set data x obs The residual res is: res=x obs -x est 。 6. The rotating equipment fault diagnosis method based on multivariate parameter estimation and image recognition according to claim 1 is characterized in that: The step S5 comprises the following steps: S51, establishing a three-dimensional grid diagram of the residual values, frequencies and times of all test set data, where the horizontal axis corresponds to time and the vertical axis corresponds to the frequency value, and the corresponding sizes of the frequency residual values ​​are distinguished by color, or establishing a grid diagram of the frequencies and residual values ​​of all test set data, where the horizontal axis corresponds to the frequency value; S52, performing grayscale processing on the established three-dimensional grid image to obtain a grayscale image; S53, establishing an image threshold, and then performing filtering processing on the image; S54, determining the frequency value corresponding to the high brightness area in the image, and performing fault diagnosis according to the characteristic frequency value.

Images