Method, device and equipment for diagnosing fault of centrifugal compressor

Abstract

Embodiments of the present disclosure provide a centrifugal compressor fault diagnosis method, device and equipment. The method comprises: acquiring a current vibration signal of a centrifugal compressor to be diagnosed; extracting a current fault characteristic parameter of the current vibration signal based on a linear transformation time-frequency analysis method; acquiring a pre-stored fault characteristic sample library; and performing analogy analysis on the current fault characteristic parameter and the fault characteristic sample library to determine a fault type of the centrifugal compressor. In this way, the current fault characteristic parameter and the fault characteristic sample library can be automatically subjected to analogy analysis, so as to accurately and quickly locate the fault type of the centrifugal compressor. Since this fault analysis method does not require model training, the fault analysis result is not affected by the model precision, and the algorithm is simple to use, so the identification accuracy of the fault type of the centrifugal compressor can be ensured.

Description

Technical Field

[0001] This disclosure relates to the field of centrifugal compressors, and more particularly to the field of fault diagnosis technology. Background Technology

[0002] Centrifugal compressors are key equipment in many large-scale chemical production plants, and their normal operation is crucial for the safe and stable production of chemical enterprises. Fault diagnosis technology for centrifugal compressors mainly involves collecting signal data generated under different operating conditions and comparing and analyzing this data with models constructed using machine learning algorithms (such as convolutional neural networks) to determine the location of the anomaly and resolve the fault. However, this fault analysis method suffers from difficulties in accurately training models, resulting in long training cycles and unreliable accuracy. Furthermore, the highly specialized nature of model training makes it unsuitable for large-scale deployment. Summary of the Invention

[0003] This disclosure provides a method, apparatus, equipment, and storage medium for diagnosing faults in a centrifugal compressor.

[0004] According to a first aspect of this disclosure, a method for diagnosing faults in a centrifugal compressor is provided. The method includes:

[0005] Acquire the current vibration signal of the centrifugal compressor to be diagnosed;

[0006] The current fault characteristic parameters of the current vibration signal are extracted using a time-frequency analysis method based on linear transformation.

[0007] Obtain a pre-stored database of fault feature samples;

[0008] The current fault characteristic parameters are compared with the fault characteristic sample library to determine the fault type of the centrifugal compressor.

[0009] In addition to the aspects described above and any possible implementations, a further implementation is provided in which the fault feature sample library is constructed through the following steps:

[0010] Obtain vibration signals corresponding to each fault type of the faulty centrifugal compressor;

[0011] The time-frequency analysis method based on linear transformation is used to extract fault features from the vibration signals corresponding to each fault type, and to obtain the initial fault features corresponding to each fault type.

[0012] The initial fault features are clustered using the target particle swarm optimization algorithm and a preset clustering algorithm to obtain sample fault features corresponding to each fault type.

[0013] The fault feature sample library is constructed using the sample fault features corresponding to each fault type.

[0014] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the method further includes, prior to fault feature extraction:

[0015] The vibration signals corresponding to each fault type are processed using a multi-point averaging method and a smoothing algorithm.

[0016] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the time-frequency analysis method based on linear transformation extracts fault features from the vibration signals corresponding to each fault type to obtain initial fault features corresponding to each fault type, including:

[0017] The vibration signals corresponding to each fault type are converted into Morlet wavelets;

[0018] Select wavelet basis functions based on the Morlet wavelet;

[0019] The wavelet basis functions are subjected to frequency domain Fourier transform using a preset Fourier function;

[0020] The wavelet shape control parameters of the wavelet basis function are obtained using the Shannon entropy method.

[0021] The preset Fourier function and the vibration signals corresponding to each fault type are converted into Fourier transforms in the frequency domain through Fourier inverse transform to obtain the wavelet scale parameters of the wavelet basis function;

[0022] The wavelet shape control parameters and the wavelet scale parameters are determined as the initial fault features corresponding to each fault type.

[0023] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the step of using a target particle swarm optimization algorithm and a preset clustering algorithm to perform cluster analysis on the initial fault features to obtain sample fault features corresponding to each fault type includes:

[0024] Based on the initial fault characteristics corresponding to each fault type, the target particle swarm optimization algorithm and the preset clustering algorithm are used to continuously iterate and perform clustering analysis, thereby determining the sample fault characteristics corresponding to each fault type.

[0025] In addition to the aspects and any possible implementations described above, a further implementation is provided, which, based on the initial fault characteristics corresponding to each fault type, uses the target particle swarm optimization algorithm and the preset clustering algorithm to iteratively perform clustering analysis to determine the sample fault characteristics corresponding to each fault type, including:

[0026] Based on the initial fault characteristics corresponding to each fault type, during the continuous iteration of the target particle swarm algorithm, the particle fitness value obtained in each iteration is calculated by combining the particle fitness function composed of the preset clustering algorithm.

[0027] The particle fitness value obtained in each iteration is compared with the preset particle fitness value until the difference between the two is less than the preset threshold or the preset maximum number of iterations is reached. Then the iteration stops to determine the sample fault characteristics corresponding to each fault type.

[0028] In addition to the aspects and any possible implementations described above, a further implementation is provided, which constructs a nonlinearly changing particle swarm inertia weight function by utilizing a preset minimum inertia weight, a preset maximum inertia weight, the current iteration number, and the maximum iteration number.

[0029] The target particle swarm algorithm is constructed using the nonlinearly changing particle swarm inertia weight function.

[0030] According to a second aspect of this disclosure, a fault diagnosis device for a centrifugal compressor is provided. The device includes:

[0031] The first acquisition module is used to acquire the current vibration signal of the centrifugal compressor to be diagnosed;

[0032] The extraction module is used to extract the current fault characteristic parameters of the current vibration signal based on the time-frequency analysis method of linear transformation;

[0033] The second acquisition module is used to acquire a pre-stored fault feature sample library;

[0034] The analysis module is used to perform comparative analysis between the current fault characteristic parameters and the fault characteristic sample library to determine the fault type of the centrifugal compressor.

[0035] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0036] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the methods according to the first and / or second aspects of this disclosure.

[0037] In this disclosure, the current fault characteristic parameters of the current vibration signal are automatically extracted by using the time-frequency analysis method of linear transformation. Then, a pre-stored fault characteristic sample library is obtained. The current fault characteristic parameters and the fault characteristic sample library can be automatically compared and analyzed to accurately and quickly locate the fault type of the centrifugal compressor. Since this fault analysis method does not require model training, the fault analysis results are not affected by the accuracy of the model. Moreover, the algorithm is simple to use. Therefore, it can ensure the accuracy of fault type identification of centrifugal compressor. Since it does not require model training and does not have high professional requirements, it is suitable for large-scale use.

[0038] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0039] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0040] Figure 1 A flowchart of a centrifugal compressor fault diagnosis method according to an embodiment of the present disclosure is shown;

[0041] Figure 2 A flowchart of another centrifugal compressor fault diagnosis method according to an embodiment of the present disclosure is shown;

[0042] Figure 3 A block diagram of a centrifugal compressor fault diagnosis device according to an embodiment of the present disclosure is shown;

[0043] Figure 4 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0045] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0046] Figure 1 A flowchart of a centrifugal compressor fault diagnosis method 100 according to an embodiment of the present disclosure is shown. Method 100 may include:

[0047] Step 110: Obtain the current vibration signal of the centrifugal compressor to be diagnosed;

[0048] The current vibration signal is obtained by using a vibration velocity sensor and an eddy current displacement sensor to collect the vibration velocity signal and axial displacement signal, respectively. The current vibration signal is the vibration signal of the compressor in the horizontal and vertical directions.

[0049] Step 120: Extract the current fault characteristic parameters of the current vibration signal using a time-frequency analysis method based on linear transformation;

[0050] The current fault characteristic parameters include, but are not limited to, the wavelet shape control parameter β and the scale parameter a of the current vibration signal. β is used to control the wavelet shape (e.g., the wavelet can be controlled to be a sine or cosine waveform, a pulse waveform, etc.), and a is the scale parameter, which determines the scaling characteristics of the wavelet.

[0051] Step 130: Obtain the pre-stored fault feature sample library;

[0052] Step 140: Compare the current fault characteristic parameters with the fault characteristic sample library to determine the fault type of the centrifugal compressor. Fault types include, but are not limited to, rotor bending, rotor-to-blade rubbing, and blade failure.

[0053] By using time-frequency analysis with linear transformation to automatically extract the current fault characteristic parameters of the current vibration signal, and then obtaining a pre-stored fault characteristic sample library, the current fault characteristic parameters can be automatically compared and analyzed with the fault characteristic sample library to accurately and quickly locate the fault type of the centrifugal compressor. Since this fault analysis method does not require model training, the fault analysis results are not affected by the accuracy of the model, and the algorithm is simple to use. Therefore, it can ensure the accuracy of fault type identification of centrifugal compressors, and the lack of model training and low professional requirements make it suitable for large-scale deployment.

[0054] In some embodiments, the fault feature sample library is constructed through the following steps:

[0055] Obtain vibration signals corresponding to each fault type of the faulty centrifugal compressor;

[0056] The time-frequency analysis method based on linear transformation is used to extract fault features from the vibration signals corresponding to each fault type, and to obtain the initial fault features corresponding to each fault type.

[0057] The initial fault features are clustered using the target particle swarm optimization algorithm and a preset clustering algorithm to obtain sample fault features corresponding to each fault type.

[0058] The fault feature sample library is constructed using the sample fault features corresponding to each fault type.

[0059] By acquiring the vibration signals corresponding to each fault type of the faulty centrifugal compressor, fault features can be extracted from the vibration signals corresponding to each fault type based on the time-frequency analysis method of linear transformation, and the initial fault features corresponding to each fault type can be obtained. Then, by combining the target particle swarm algorithm with the preset clustering algorithm, cluster analysis can be performed on the initial fault features, thereby obtaining the sample fault features corresponding to each fault type. Then, by using the sample fault features corresponding to each fault type, a fault feature sample library that can identify multiple different fault types can be automatically constructed.

[0060] In some embodiments, the method further includes, prior to fault feature extraction:

[0061] The vibration signals corresponding to each fault type are processed using a multi-point averaging method and a smoothing algorithm.

[0062] Since the vibration signals corresponding to each fault type may have many spikes and unevenness, the multi-point averaging method and the five-point cubic smoothing method can be used for data cleaning to reduce the influence of spikes and high-frequency noise, greatly reduce the influence of interference signals, and improve the accuracy of fault signal resolution and the efficiency of fault feature parameter extraction.

[0063] In some embodiments, the time-frequency analysis method based on linear transformation extracts fault features from the vibration signals corresponding to each fault type to obtain initial fault features corresponding to each fault type, including:

[0064] The vibration signals corresponding to each fault type are converted into Morlet wavelets; Morlet wavelets are a type of wavelet function.

[0065] Select wavelet basis functions based on the Morlet wavelet;

[0066] The wavelet basis functions are subjected to frequency domain Fourier transform using a preset Fourier function;

[0067] The wavelet shape control parameters of the wavelet basis function are obtained using the Shannon entropy method.

[0068] The preset Fourier function and the vibration signals corresponding to each fault type are converted into Fourier transforms in the frequency domain through Fourier inverse transform to obtain the wavelet scale parameters of the wavelet basis function;

[0069] The wavelet shape control parameters and the wavelet scale parameters are determined as the initial fault features corresponding to each fault type.

[0070] Since Morlet wavelets can simulate signals of arbitrary waveforms after continuous transformations in the time and frequency domains, different types of signals will correspond to different Morlet wavelet feature parameters. Therefore, vibration signals corresponding to each fault type can be converted into Morlet wavelets. Since Morlet wavelets are complex expressions composed of real and imaginary parts, considering the requirement of tight support of wavelet basis functions, the real part of the Morlet wavelet can be used as the wavelet basis function. Then, using a preset Fourier function, the wavelet basis function is subjected to a frequency domain Fourier transform. Subsequently, the wavelet shape control parameters of the wavelet basis function are obtained using the Shannon entropy method. Finally, based on the convolution property, the Fourier transform in the time domain is converted into a Fourier transform in the frequency domain through an inverse Fourier transform. The scaling parameters in the frequency domain Fourier transform are obtained by correlating the center frequency with the frequency resolution. This allows for a comprehensive analysis of the vibration signal characteristics in both the time and frequency domains, avoiding the information loss problem caused by performing time or frequency domain analysis alone, and thus obtaining more realistic and effective initial fault characteristics corresponding to each fault type.

[0071] In some embodiments, the step of using a target particle swarm optimization algorithm and a preset clustering algorithm to perform cluster analysis on the initial fault features to obtain sample fault features corresponding to each fault type includes:

[0072] Based on the initial fault characteristics corresponding to each fault type, the target particle swarm optimization algorithm and the preset clustering algorithm are used to iteratively perform clustering analysis to determine the sample fault characteristics corresponding to each fault type. The preset clustering algorithm can be a fuzzy C-means clustering algorithm.

[0073] The target particle swarm algorithm is an improved version of the particle velocity inertia weighted exponent. Based on the initial fault characteristics corresponding to the fault type, the target particle swarm algorithm can quickly find the optimal cluster center of the preset clustering algorithm in the global scope, avoiding getting trapped in local optima and excessive iterative calculations during the fault classification process, thereby improving the computational efficiency and the reliability of the calculation results.

[0074] In addition, this embodiment enables the preprocessed compressor vibration signal to undergo Morlet wavelet transform, and then perform cluster analysis using the improved particle swarm optimization algorithm and fuzzy C-means clustering algorithm. This allows the Morlet wavelet waveform control parameter β and scale parameter a corresponding to each fault signal to be quickly and accurately classified into a specific fault type globally, thus forming a fault feature sample library with more comprehensive information and more accurate and reasonable classification.

[0075] In some embodiments, based on the initial fault characteristics corresponding to each fault type, the target particle swarm optimization algorithm and the preset clustering algorithm are used to iteratively perform clustering analysis to determine the sample fault characteristics corresponding to each fault type, including:

[0076] Based on the initial fault characteristics corresponding to each fault type, during the continuous iteration of the target particle swarm algorithm, the particle fitness value obtained in each iteration is calculated by combining the particle fitness function composed of the preset clustering algorithm.

[0077] The particle fitness value obtained in each iteration is compared with the preset particle fitness value until the difference between the two is less than the preset threshold or the preset maximum number of iterations is reached. Then the iteration stops to determine the sample fault characteristics corresponding to each fault type.

[0078] Based on the initial fault features corresponding to the fault types, the optimal cluster center for each fault type can be obtained by iteratively using the target particle swarm optimization algorithm at each iteration. Then, the particle fitness value at each iteration is calculated using a preset clustering algorithm. It is determined whether the difference between the particle fitness value and the preset particle fitness value is less than a preset threshold. If it is less, it means that the optimal cluster center for each fault type calculated in this iteration satisfies the global optimum and is already ideal. Alternatively, if the number of iterations reaches the preset maximum number of iterations, it means that it is impossible to continue to the next iteration. Therefore, the iteration can be stopped. Thus, the optimal cluster center corresponding to each fault type is automatically determined as the sample fault feature corresponding to each fault type through an appropriate number of iterations.

[0079] In some embodiments, a nonlinearly changing particle swarm inertia weight function is constructed using a preset minimum inertia weight, a preset maximum inertia weight, the current iteration number, and the maximum iteration number.

[0080] The target particle swarm algorithm is constructed using the nonlinearly changing particle swarm inertia weight function.

[0081] By utilizing preset minimum inertia weights, preset maximum inertia weights, the current iteration number, and the maximum iteration number, a nonlinearly changing particle swarm inertia weight function is constructed. This allows the constructed target particle swarm algorithm to use a larger step size to update the position of each particle (the cluster center corresponding to each fault type) in the early stage of the search, in order to achieve global search and reduce the number of iterations; while in the later stage of the search, a smaller update step size is used to achieve local search. Thus, when the target particle swarm algorithm is combined with the preset clustering algorithm, it can quickly find the optimal cluster center in the global scope.

[0082] The following will combine Figure 2 Further details of the technical solution disclosed herein:

[0083] like Figure 2 As shown, this disclosure provides a fault diagnosis method for centrifugal compressors based on Morlet wavelet continuous Fourier transform and an improved particle swarm optimization and fuzzy C-means clustering algorithm, including the following steps:

[0084] The vibration signal data of the centrifugal compressor collected by field sensors were cleaned using the multi-point averaging method and the five-point cubic smoothing method.

[0085] Fault characteristic parameters are extracted from the cleaned data using Morlet continuous wavelet Fourier transform.

[0086] An improved particle swarm optimization and fuzzy C-means clustering ensemble algorithm was used to perform cluster analysis on the extracted feature signal data, and a fault sample database was established based on the cluster analysis results.

[0087] Cluster analysis is performed on the test samples and the data in the fault sample database. By comparing the fault sample data with the data in the fault sample database, the corresponding fault type is determined.

[0088] The collected vibration signal data from the centrifugal compressor were processed using a multi-point averaging method (calculating the average value at each location) and a five-point cubic smoothing method. Specifically, the vibration signal data obtained at a certain sampling frequency were first averaged at multiple points to obtain a series of average value points: x1, x2, x3…xn, where n is the number of average value points. The obtained average value points were then subjected to five-point cubic smoothing, using a cubic polynomial Y = a0 + a1x + a2x2 + a3x3 for approximation. The coefficients a1, a2, and a3 in the cubic polynomial were primarily determined using the least squares principle.

[0089] Based on multi-point averaging and five-point cubic smoothing of the signal data, fault features are extracted using the Morlet continuous wavelet Fourier transform method, specifically including:

[0090] S1, considering the compact support requirement of wavelet basis functions, the real part of the Morlet wavelet (the Morlet wavelet function consists of a real part and an imaginary part, expressed as a complex number) is used as the wavelet basis function (it is also a wavelet, that is, a bunch of irregular vibration signals of various fault types are simulated as wavelet basis functions), and its form is as follows:

[0091]

[0092] S2, perform a Fourier transform on the above wavelet basis functions in the following form (Formula 2 is a Fourier function):

[0093]

[0094] The Morlet wavelet basis functions, after Fourier transform (transformed from the time domain to the frequency domain), take the following form:

[0095]

[0096] Substituting formula (1) into formula (2) yields formula (3). In formula (3), ω0 is the center frequency of the wavelet, which is a constant; β is the waveform control parameter, which controls the shape of the wavelet. The larger β is, the faster the waveform decays. When it approaches infinity, the wavelet shape is a pulse signal. The smaller β is, the slower the waveform decays. When it equals 0, the wavelet shape is a cosine signal with constant amplitude oscillation.

[0097] S3 selects the optimal β value by minimizing the Shannon entropy. The Shannon entropy is defined as follows:

[0098]

[0099] In the formula, f i (i = 1, 2, ..., n) is a probability sequence (determined by w in Formula 3; once w is determined, the corresponding β is determined), and satisfies ∑f i =1. The steps to determine the value of β are as follows:

[0100] (1) Select an appropriate step size s within the variation range of β [m,n] (an empirical threshold);

[0101] (2) β takes values ​​in the interval [m,n] with step size s, and performs wavelet transform on the vibration signal (here, assuming a β and a w, we get a Φ(ω));

[0102] (3) Standardize the wavelet coefficients according to the following formula (5);

[0103]

[0104] Among them, {c i}(i=1,2,…N) is a sequence of wavelet coefficients (determined by w in Formula 3); {d j} (j=1,2,…N) is the standardized wavelet coefficient sequence. Since the vibration signal simulated by Equation 3 may have deviations, it is necessary to use the wavelet coefficients obtained from Equation 5 to perform addition and subtraction operations on Equation 3 in order to correct Φ(ω).

[0105] (4) Calculate the entropy corresponding to each β value according to equation (4), and then find the minimum wavelet entropy. The corresponding β value is the optimal β value.

[0106] S4, based on the properties of convolution, transforms the convolution operation in the time domain into a multiplication operation in the frequency domain:

[0107] W x (a,b,Φ)=F -1 {X(ω)Φ a,b (ω)}, a>0 (6)

[0108] Where 'a' is the scaling parameter, which determines the wavelet's scaling and translation characteristics (stretching and translation may be required when moving from the time domain to the frequency domain). Once the value of β is determined, the wavelet's time-domain shape and frequency response are usually determined by the scaling parameter 'a'; 'b' is the positioning parameter, which determines the wavelet's time-shifting characteristics and is a constant; F -1 It is the inverse Fourier transform, X(ω) and Φ a,b (ω) represent the signal X(t) (the smoothed vibration signal obtained before Equation 1) and the wavelet function Φ, respectively. a,b Fourier transform of (t).

[0109] S5, the scale parameter a is solved by the correlation between the center frequency (w0) and the frequency resolution (a default value is given, which is a virtual parameter that characterizes the sensitivity between the converted time domain and frequency domain, approximately in the range of 0-1, the closer to 1 the better, therefore, it is generally taken as 0.5-1). The specific steps are as follows:

[0110] (1) Based on the characteristics of the analyzed signal, the value of w in Formula 3 is multiplied by a certain coefficient to determine f = w * k, and then the maximum and minimum analysis frequencies f in f are determined. max f min ;

[0111] (2) Place the frequency axis in the analysis range [f min ,f max The system is divided into n center frequencies using an equal logarithmic division, and each center frequency is denoted as ω. i(i = 1, 2, ..., n);

[0112] (3) The scale parameter a = f for each center frequency max / ω i ,i=(1,2,…n).

[0113] An improved particle swarm optimization and fuzzy C-means clustering algorithm was used to perform cluster analysis on the extracted feature signal data, and a standard sample database of centrifugal compressors was established based on the cluster analysis results, specifically including:

[0114] S1, based on the original particle swarm algorithm expression (7), the inertial weight index of the particle velocity taken from empirical values ​​is adjusted. Nonlinear adaptive modification was performed, and the modified The expression is shown in equation (8). In this case, the movement of the particle is the movement of the cluster center; in other words, the change in the particle's position vector in each iteration is the change in the position of the cluster center.

[0115]

[0116] Among them, v id k The x represents the d-dimensional element of the velocity vector of the i-th particle in the k-th iteration; id k pb represents the d-dimensional element of the position vector of the i-th particle in the k-th iteration; id k The optimal position pb of the i-th particle in the k-th iteration. k d-dimensional elements; gb id k The optimal position for the group (gb) k The d-dimensional elements (initially determined by a manually set optimal fitness, and then continuously updated based on the maximum fitness over k iterations); c1 and c2 are learning factors, which are positive constants and are mainly used for information sharing and interaction between particles; r1 and r2 are random numbers that follow a uniform distribution in the interval [0,1] and are used to maintain the diversity of the particle population.

[0117]

[0118] in, The inertial weighting exponent for particle velocity is mainly used to balance global and local search capabilities. The global optimality characteristic of the particle swarm optimization algorithm is mainly achieved through... It is worth highlighting. Commonly used. The empirical value is between [0.4, 0.9]. The maximum inertia weight (custom-defined, such as 1); The minimum inertia weight (customizable, such as 0 or 0.2); kmax is the maximum number of iterations, and k is the current number of iterations.

[0119] S2, using the improved particle swarm optimization algorithm and fuzzy C-means clustering algorithm for cluster analysis, specifically includes:

[0120] 1) Read in the dataset of waveform control parameters β and scale parameter a corresponding to the fault characteristic signals obtained by multi-point averaging, five-point cubic smoothing, and continuous Morlet wavelet transform. Set the initial number of clusters c and the number of particles M, and initialize the cluster centers p. i Set the maximum number of iterations t max Set learning factors c1 and c2 and two random numbers r1 and r2 to initialize the particle velocity information. and location information

[0121] 2) Calculate the objective function of the fuzzy C-means clustering algorithm according to the following formula (9), and then substitute it into the fitness function formula (10) to obtain the fitness value of the current generation of particles in the particle swarm.

[0122]

[0123] Among them, J m (U,P) is the objective function of the fuzzy C-means clustering algorithm. The dataset used in the fuzzy C-means clustering algorithm is: n is the total number of elements in X. The k-th element of X is an s-dimensional vector, x k =x k1 ,x k2 ,…x ks Each x k Each represents a set of values ​​for β and a. X contains c classes (the minimum number of classes that the clustering algorithm model can identify compared to the actual number of classes), and pi represents the center of the i-th class. ik Let U represent the membership degree of the k-th element of X to the i-th center. The membership matrix is ​​U = μ. ik ,1≤i≤c,1≤k≤n,μ ij ∈[0,1]. c cluster centers are P = p i m∈[1,∞) is the fuzzy weighting exponent; typically, m=2. d ik =||x i -p i || represents the cluster centers of sample point xi (a set of cluster centers of β and a randomly given at the k-th iteration of the particle swarm optimization algorithm) and the optimal cluster centers p of each category. i The Euclidean distance between them is mainly determined by the particle swarm optimization algorithm through the analysis of p. i The optimization minimizes the Euclidean distance, at which point the values ​​of β and a will be related by the membership degree μ. ikIt belongs to a certain type of fault.

[0124] The fitness function of a particle in particle swarm optimization (used to characterize the position x of the i-th particle). id k and velocity v id k The suitability is shown in equation (10):

[0125]

[0126] 3) Based on the velocity and position of the previous generation of particles, update and calculate the velocity and position of the next generation of particles, and set a threshold to allow them to vary within a specified maximum range. In this way, as the fault signal changes, the cluster center is in a state of continuous optimization and change, ultimately achieving an adaptive match between the fault signal and the fault type.

[0127] 4) Update the best position of the particle in the current particle swarm individual and the best position of the particle swarm according to equation (11) (where t and (t+1) are the subscripts of equation 11, t represents time, if time t is the kth iteration, then time t+1 is the k+1th iteration).

[0128]

[0129] 5) When the optimization result falls within the set minimum threshold range (fitness threshold) or the maximum number of iterations has been reached, the optimization process ends; otherwise, the above calculation steps will be followed to perform iterative iterations. Once the optimization process ends, the optimized cluster centers are the best cluster centers.

[0130] The test samples and the fault sample database are combined for cluster analysis. By comparing the fault samples with provided standard faults, the corresponding fault types are determined, specifically including:

[0131] By continuously optimizing the cluster centers of the fuzzy C-means clustering algorithm through the improved particle swarm optimization algorithm, the fault type corresponding to each fault signal is finally obtained, that is, the dataset of β and a data of the fault feature signals corresponding to the fault type is obtained.

[0132] By comparing the β and a data of the fault sample with the β and a datasets corresponding to different fault types, the fault type corresponding to the current fault sample can be obtained, thus enabling rapid and accurate identification of fault problems.

[0133] The technical solution disclosed herein will be further explained in detail below with specific examples:

[0134] This disclosure is based on vibration signal data of 10 common faults provided by a manufacturer of a large centrifugal pyrolysis gas compressor unit, as shown in Table 1.

[0135] Table 1 Common Fault Types of Centrifugal Pyrolysis Gas Compressor Units

[0136] serial number Fault type serial number Fault type 1 Rotor bending 6 Blade failure 2 Rotor rubbing 7 Oil film oscillation 3 Rotor imbalance 8 transverse cracks 4 Rotor misalignment 9 Loose support system 5 Axial movement 10 Rolling bearing failure

[0137] As shown in Table 1, there are approximately 2000 vibration signal datasets for each fault type. From these 2000 datasets for each fault type, 1600 datasets were randomly selected as training samples for the algorithm, and the remaining 400 datasets were used as test samples to evaluate the algorithm's performance. Data processing, modeling analysis, and the centrifugal compressor fault diagnosis process were implemented using the Matlab platform. Modeling analysis and comparisons were performed on three algorithms: the method proposed in this paper, the improved particle swarm optimization and fuzzy C-means clustering algorithm, and the particle swarm optimization and fuzzy C-means clustering algorithm.

[0138] After processing the compressor signal data using the three-point averaging method and the five-point cubic smoothing method, the vibration signal data is subjected to continuous Fourier transform using Morlet wavelets according to the above equations (1)-(3). The waveform parameter β and the scale parameter a are obtained through the minimum Shannon entropy and the correlation between the center frequency and the frequency resolution, respectively. Thus, the dataset of waveform parameter β and scale parameter a related to the vibration signal data is obtained. The initial number of clusters c = 50 and the number of particles M = 100 are set for this dataset, and the particle velocity inertia weight exponent is set to 50. Initialize cluster centers p i Set the maximum number of iterations t max =100, set learning factors c1=2 and c2=2.25 and two random numbers r1=0.12 and r2=0.68, and initialize the particle velocity information. and location information The objective function of the fuzzy C-means clustering algorithm and the particle fitness value of the particle swarm optimization algorithm are obtained by equations (9) and (10). A partial dataset of waveform parameter β and scale parameter a corresponding to 10 common faults is shown in Table 2 below. (Since each fault type corresponds to approximately 2000 vibration signal datasets, and each dataset is approximately 1024 bytes long, resulting in a large number of datasets for β and a, only a portion of the datasets are shown.) The diagnostic analysis results of the 10 common faults under the above three methods are shown in Table 3 below. As can be seen from Table 3, using the fault diagnosis method provided in this disclosure for fault diagnosis and identification, the fault diagnosis accuracy is controlled at approximately 0.98, and the diagnosis time is controlled between 20-25 seconds. Both the model accuracy and computational speed are superior to the other two methods.

[0139] Table 2. Partial dataset of β and a parameters corresponding to 10 common faults in centrifugal cracked gas compressor units.

[0140]

[0141]

[0142] Table 3. Diagnostic analysis results of 10 common fault types in centrifugal pyrolysis gas compressor units using three methods.

[0143]

[0144]

[0145] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.

[0146] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.

[0147] Figure 3 A block diagram of a centrifugal compressor fault diagnosis device 300 according to an embodiment of the present disclosure is shown. Figure 3 As shown, the device 300 includes:

[0148] The first acquisition module 310 is used to acquire the current vibration signal of the centrifugal compressor to be diagnosed;

[0149] Extraction module 320 is used to extract the current fault characteristic parameters of the current vibration signal based on time-frequency analysis of linear transformation;

[0150] The second acquisition module 330 is used to acquire a pre-stored fault feature sample library;

[0151] The analysis module 340 is used to perform comparative analysis between the current fault characteristic parameters and the fault characteristic sample library to determine the fault type of the centrifugal compressor.

[0152] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0153] According to embodiments of the present disclosure, the present disclosure also provides an electronic device and a non-transitory computer-readable storage medium storing computer instructions.

[0154] Figure 4A schematic block diagram of an electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0155] Device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0156] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0157] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of method 100 described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform method 100 by any other suitable means (e.g., by means of firmware).

[0158] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0159] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0160] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0161] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0162] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0163] Computing systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0164] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0165] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for diagnosing faults in a centrifugal compressor, characterized in that, include: Acquire the current vibration signal of the centrifugal compressor to be diagnosed; The current fault characteristic parameters of the current vibration signal are extracted using a time-frequency analysis method based on linear transformation. Obtain a pre-stored database of fault feature samples; The current fault characteristic parameters are compared with the fault characteristic sample library to determine the fault type of the centrifugal compressor. The fault feature sample library is constructed through the following steps: Obtain vibration signals corresponding to each fault type of the faulty centrifugal compressor; The time-frequency analysis method based on linear transformation is used to extract fault features from the vibration signals corresponding to each fault type, and to obtain the initial fault features corresponding to each fault type. The initial fault features are clustered using the target particle swarm optimization algorithm and a preset clustering algorithm to obtain sample fault features corresponding to each fault type. The fault feature sample library is constructed using the sample fault features corresponding to each fault type. The time-frequency analysis method based on linear transformation extracts fault features from the vibration signals corresponding to each fault type, obtaining initial fault features corresponding to each fault type, including: The vibration signals corresponding to each fault type are converted into Morlet wavelets; Select wavelet basis functions based on the Morlet wavelet; The wavelet basis functions are subjected to frequency domain Fourier transform using a preset Fourier function; The wavelet shape control parameters of the wavelet basis function are obtained using the Shannon entropy method. The preset Fourier function and the vibration signals corresponding to each fault type are converted into Fourier transforms in the frequency domain through Fourier inverse transform to obtain the wavelet scale parameters of the wavelet basis function; The wavelet shape control parameters and the wavelet scale parameters are determined as the initial fault features corresponding to each fault type.

2. The method according to claim 1, characterized in that, Before performing fault feature extraction, the method further includes: The vibration signals corresponding to each fault type are processed using a multi-point averaging method and a smoothing algorithm.

3. The method according to claim 1, characterized in that, The step of using the target particle swarm optimization algorithm and a preset clustering algorithm to perform cluster analysis on the initial fault features to obtain sample fault features corresponding to each fault type includes: Based on the initial fault characteristics corresponding to each fault type, the target particle swarm optimization algorithm and the preset clustering algorithm are used to continuously iterate and perform clustering analysis, thereby determining the sample fault characteristics corresponding to each fault type.

4. The method according to claim 3, characterized in that, Based on the initial fault characteristics corresponding to each fault type, the target particle swarm optimization algorithm and the preset clustering algorithm are used to iteratively perform clustering analysis to determine the sample fault characteristics corresponding to each fault type, including: Based on the initial fault characteristics corresponding to each fault type, during the continuous iteration of the target particle swarm algorithm, the particle fitness value obtained in each iteration is calculated by combining the particle fitness function composed of the preset clustering algorithm. The particle fitness value obtained in each iteration is compared with the preset particle fitness value until the difference between the two is less than the preset threshold or the preset maximum number of iterations is reached. Then the iteration stops to determine the sample fault characteristics corresponding to each fault type.

5. The method according to any one of claims 1 to 4, characterized in that, A nonlinearly changing particle swarm inertia weight function is constructed using preset minimum inertia weight, preset maximum inertia weight, current iteration number, and maximum iteration number. The target particle swarm algorithm is constructed using the nonlinearly changing particle swarm inertia weight function.

6. A centrifugal compressor fault diagnosis device, characterized in that, The apparatus is used to implement the method as described in any one of claims 1 to 5, the apparatus comprising: The first acquisition module is used to acquire the current vibration signal of the centrifugal compressor to be diagnosed; The extraction module is used to extract the current fault characteristic parameters of the current vibration signal based on the time-frequency analysis method of linear transformation; The second acquisition module is used to acquire a pre-stored fault feature sample library; The analysis module is used to perform comparative analysis between the current fault characteristic parameters and the fault characteristic sample library to determine the fault type of the centrifugal compressor.

7. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

8. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

Images