A nuclear fuzzy C-means clustering-based self-learning method for fault diagnosis of oil delivery pumps

Abstract

The present application relates to a kind of oil pump fault diagnosis technical field, more particularly to a kind of oil pump fault diagnosis self-learning method based on kernel fuzzy C means clustering, comprising: the normalized six-dimensional feature vector corresponding to each historical vibration data is obtained, and according to the vibration data of the running process of the oil pump to be diagnosed, the normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed is obtained;According to the normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed, the similarity of the vibration data and each cluster center is calculated, whether the vibration data of the oil pump to be diagnosed corresponds known fault is judged, and the judgment result is obtained;If yes, determine the vibration data of the oil pump to be diagnosed corresponding fault, if no, record as a new fault mode data and store. Can be in time to the oil pump is diagnosed for fault, and on the basis of fault diagnosis, make model have self-learning ability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil pump fault diagnosis, and particularly relates to a self-learning method for oil pump fault diagnosis based on kernel fuzzy C-means clustering. BACKGROUND

[0002] Oil and gas pipelines are important infrastructure, which are related to energy security and social stability. Oil pumps are key equipment of oil and gas pipelines. With the increase of service life, the probability of oil pump failure increases. If hidden dangers cannot be found and eliminated in time, it will have a serious impact on oil production. SUMMARY

[0003] The present application relates to the technical field of oil pump fault diagnosis, and particularly relates to a self-learning method for oil pump fault diagnosis based on kernel fuzzy C-means clustering.

[0004] The technical scheme of the self-learning method for oil pump fault diagnosis based on kernel fuzzy C-means clustering is as follows:

[0005] Obtain the normalized six-dimensional feature vector corresponding to each historical vibration data in the historical vibration data set of the oil pump, and use all the normalized six-dimensional feature vectors to construct a training set and a test set;

[0006] Based on the training set and the test set, a trained kernel fuzzy C-means clustering model is obtained;

[0007] Collect and obtain the normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed according to the vibration data of the running process of the oil pump to be diagnosed;

[0008] According to the normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed, the similarity of the vibration data to each cluster center is calculated, and it is judged whether the vibration data of the oil pump to be diagnosed corresponds to a known fault, and a judgment result is obtained;

[0009] When the judgment result is yes, the vibration data of the oil pump to be diagnosed is divided into the cluster center with the highest similarity, and the fault label of the cluster center is combined to determine the fault corresponding to the vibration data of the oil pump to be diagnosed. When the judgment result is no, the number of unknown faults is recorded, and when the number reaches a set threshold, it is marked as a new fault mode. The vibration data under the fault mode is made as a sample and added to the training set, and the kernel fuzzy C-means clustering model is retrained.

[0010] The technical scheme of the self-learning system for oil pump fault diagnosis based on kernel fuzzy C-means clustering is as follows:

[0011] The method comprises an acquisition construction module, a training module, an acquisition acquisition module, a first judgment module and a determination module.

[0012] The acquisition construction module is configured to acquire a normalized six-dimensional feature vector corresponding to each historical vibration data in a historical vibration data set of the oil pump, and construct a training set and a test set by using all the normalized six-dimensional feature vectors.

[0013] The training module is configured to obtain a trained kernel fuzzy C-means clustering model based on the training set and the test set.

[0014] The acquisition acquisition module is configured to acquire vibration data of a to-be-diagnosed oil pump during operation and obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil pump.

[0015] The first judgment module is configured to calculate the similarity between the vibration data of the to-be-diagnosed oil pump and each cluster center according to the normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil pump, and determine whether the vibration data of the to-be-diagnosed oil pump corresponds to a known fault to obtain a judgment result.

[0016] The determination module is configured to, when the judgment result is yes, divide the vibration data of the to-be-diagnosed oil pump to the cluster center with the highest similarity to the vibration data, and determine the fault corresponding to the vibration data of the to-be-diagnosed oil pump in combination with the fault label of the cluster center, and when the judgment result is no, record the number of times of occurrence of the unknown fault, and when the number of times of occurrence reaches a set threshold, mark the fault as a new fault mode, make the vibration data under the fault mode as a sample, and add the sample to the training set to retrain the kernel fuzzy C-means clustering model.

[0017] The storage medium of the application stores instructions, and when a computer reads the instructions, the computer executes the fault diagnosis self-learning method of the oil pump based on the kernel fuzzy C-means clustering.

[0018] The electronic device of the application comprises a processor and the storage medium described above, and the processor executes the instructions in the storage medium.

[0019] The technical scheme of the application has the following advantages:

[0020] The oil pump can be diagnosed in time, and the model has self-learning ability based on fault diagnosis, which can improve and expand the types of identifiable faults of the model, make more faults identifiable and prewarnable, improve the safe use of the oil pump, and better improve efficiency and save cost. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of a self-learning method for oil pump fault diagnosis based on kernel fuzzy C-means clustering according to an embodiment of the present application is shown in FIG. 1.

[0022] Figure 2 A waveform diagram of vibration data is shown in FIG. 2.

[0023] Figure 3 An IMF component obtained after ensemble empirical mode decomposition is shown in FIG. 3.

[0024] Figure 4 A schematic diagram of experimental results for identifying new fault patterns is shown in FIG. 4.

[0025] Figure 5 Model self-learning capability experimental results are shown in FIG. 5.

[0026] Figure 6 A structural schematic diagram of a self-learning system for oil pump fault diagnosis based on kernel fuzzy C-means clustering according to an embodiment of the present application is shown in FIG. 6. DETAILED DESCRIPTION

[0027] As shown in FIG. 1, a self-learning method for oil pump fault diagnosis based on kernel fuzzy C-means clustering according to an embodiment of the present application includes the following steps: Figure 1

[0028] S1, obtaining a normalized six-dimensional feature vector corresponding to each piece of historical vibration data in a historical vibration data set of an oil pump, and using all normalized six-dimensional feature vectors to construct a training set and a test set;

[0029] S2, obtaining a trained kernel fuzzy C-means clustering model based on the training set and the test set;

[0030] S3, collecting vibration data of an oil pump to be diagnosed during operation and obtaining a normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed;

[0031] S4, calculating a similarity between the vibration data of the oil pump to be diagnosed and each cluster center according to the normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed, and determining whether the vibration data of the oil pump to be diagnosed corresponds to a known fault to obtain a determination result;

[0032] S5, when the determination result is yes, the vibration data of the oil pump to be diagnosed is divided into a cluster center with the highest similarity to the vibration data, and a fault corresponding to the vibration data of the oil pump to be diagnosed is determined in combination with a fault label of the cluster center, when the determination result is no, a number of occurrences of the unknown fault is recorded, when the number of occurrences reaches a set threshold, the fault is marked as a new fault pattern, vibration data under the fault pattern is made as a sample, and the training set is added to retrain the kernel fuzzy C-means clustering model. ​

[0033] The application can diagnose faults of the oil delivery pump in time, and has self-learning ability on the basis of fault diagnosis, can improve, expand the type of recognizable faults of the model, can make more faults recognizable and prewarning, improve the safe use ability of the oil delivery pump, and better improve efficiency and save cost.

[0034] Optionally, in the technical solution above, in S1, the process of obtaining the normalized six-dimensional feature vector corresponding to each piece of historical vibration data is:

[0035] S10, performing ensemble empirical mode decomposition on each piece of historical vibration data in the historical vibration data set to obtain an IMF component of each piece of historical vibration data;

[0036] S11, selecting the first six components in the IMF component of each piece of historical vibration data, respectively performing singular value decomposition to obtain a six-dimensional feature vector corresponding to each piece of historical vibration data;

[0037] S12, performing normalization processing on each six-dimensional feature vector to obtain a normalized six-dimensional feature vector corresponding to each piece of historical vibration data.

[0038] Optionally, in the technical solution above, in S2, the trained kernel fuzzy C-means clustering model is obtained based on the training set and the test set, and includes:

[0039] S20, constructing a kernel fuzzy C-means clustering model, and initializing the cluster center and the membership matrix in the kernel fuzzy C-means clustering model;

[0040] S21, iteratively calculating the cluster center and updating the membership matrix;

[0041] S22, judging whether the change precision of the membership matrix meets the requirement or not, and terminating the training when the requirement is met, otherwise, continuing the iterative training, repeatedly executing S21 until the change precision of the membership matrix meets the requirement;

[0042] S23, performing fuzzy processing on the membership matrix meeting the change precision requirement, so that the membership of each sample in the training set to the latest obtained cluster center is 1 at most and 0 at least, and obtaining the trained kernel fuzzy C-means clustering model, the trained kernel fuzzy C-means clustering model including the membership matrix meeting the change precision requirement and the latest obtained cluster center.

[0043] Optionally, in the technical solution above, in S3, the normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil delivery pump is obtained according to the vibration data of the running process of the to-be-diagnosed oil delivery pump, and includes:

[0044] S30, performing ensemble empirical mode decomposition on the vibration data of the to-be-diagnosed fuel pump to obtain an IMF component of the vibration data of the to-be-diagnosed fuel pump;

[0045] S31, selecting the first six components in the IMF components of the vibration data of the to-be-diagnosed fuel pump, respectively performing singular value decomposition to obtain a six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump;

[0046] S32, performing normalization processing on the six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump to obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump.

[0047] The following describes a fuel pump fault diagnosis self-learning method based on kernel fuzzy C-means clustering of the application through a complete embodiment, including S100-S114, in particular:

[0048] S100, obtaining a historical vibration data set of a fuel pump;

[0049] S101, performing ensemble empirical mode decomposition on each piece of historical vibration data in the historical vibration data set to obtain an IMF component of each piece of historical vibration data, wherein the vibration data is as shown in Figure 2 .

[0050] The principle of the ensemble empirical mode decomposition is as follows:

[0051] On the basis of the original vibration signal, one or more groups of white noise signals are added, and the mixed complex signal is decomposed into a limited number of intrinsic mode functions (IMF). Each IMF component contains local characteristic information of different time scales of the original signal, representing the internal vibration mode of the data. At the same time, due to the addition of white noise signals, the end effect and mode aliasing phenomenon in the decomposition process are suppressed. Due to the introduction of white noise disturbance and ensemble averaging, the scale mixing problem is avoided, so that each component of the final decomposition maintains physical uniqueness and retains information with actual physical significance of the signal itself, including S1011-S1014:

[0052] S1011, adding a white noise sequence to the original signal:

[0053] x i (t)=x(t)+n i (t)

[0054] In the formula, x i (t) is a new signal obtained by adding the i-th white noise to the original signal data x(t), n i (t) is white noise.

[0055] S1012, for the signal after adding white noise, do empirical mode decomposition. First find all local maximum and minimum, then use cubic spline interpolation method to form the upper envelope line u1(t) and lower envelope line u2(t), then the local mean envelope line m1(t) can be expressed as:

[0056]

[0057] Add white noise after the signal x i (t) minus the local mean envelope line m1(t), can get the first vector h1:

[0058] h1(t)=x i (t)-m1(t)

[0059] If h1(t) does not meet the IMF condition, it is regarded as new x i (t), repeat the above calculation steps, after several times of repetition, get the required IMF component.

[0060] S1013, repeat the above two steps operation, namely repeat S1011 and S1012, each time to add the same amplitude of new white noise sequence to get different IMFs, as shown in Figure 3 .

[0061] S1014, the IMFs obtained by each decomposition are set to average, so that the added white noise cancels each other out, and the final decomposition IMF is obtained. In order to select the first 6 IMFs with higher correlation with the original vibration signal to represent the original vibration signal.

[0062] S102, in the IMF component of each historical vibration data, select the first 6 components, respectively, and perform singular value decomposition to obtain the six-dimensional feature vector corresponding to each historical vibration data.

[0063] Among them, the principle of singular value decomposition is: singular value decomposition can be used for matrix decomposition of any matrix. A m*n matrix can be decomposed into:

[0064] A=U∑V T

[0065] Wherein, A is an m*n matrix, U is an m*n orthogonal matrix, is an n*n diagonal matrix, V is an n*n orthogonal matrix, is also called singular value matrix, which is 0 except the diagonal line is singular value, for singular value, it is arranged from large to small in singular value matrix, and the singular value decreases fast, in most cases, the sum of the top 10% or even 1% singular values accounts for more than 99% of the total singular value, therefore, the corresponding left and right singular vectors in the matrix U, V using the largest k singular values can approximately describe the original matrix A. Due to the above properties, the singular value decomposition can be used to reduce the dimension of the matrix composed of each IMF obtained by ensemble empirical mode decomposition, and the calculated singular value is used as a feature to represent the original vibration signal. Specifically, S1021-S1023:

[0066] S1021, the transpose matrix A of the matrix A composed of each IMF T and A matrix multiplication, get square matrix A T A, and the following feature decomposition is carried out:

[0067] (A T A)v i =λ i v i

[0068] Thus A T A n eigenvectors v i , A T All eigenvectors of A T A form an n*n matrix V.

[0069] S1022, the matrix A composed of each IMF and its transpose matrix A T Matrix multiplication, get square matrix AA T , and the following feature decomposition is carried out:

[0070] (AA i )u i =λ i u T

[0071] Thus AA i m eigenvectors u T , all eigenvectors of AA T Form an m*m matrix U.

[0072] S1023, since the matrix is 0 except that the diagonal line is singular value, therefore, each singular value sigma needs to be calculated, through A=U∑V i , AV=U∑ can be obtained, therefore Av i =σ i, σ i = Av i / u i , thus the singular values can be obtained. Since the matrix A is composed of 6 IMF, 6 singular values can be obtained, which form a six-dimensional vector, which is used as the feature vector of the original vibration signal. Through ensemble empirical mode decomposition and singular value decomposition, the feature of the original vibration signal is extracted, and finally a six-dimensional feature vector is obtained.

[0073] S103, normalize each six-dimensional feature vector to obtain a normalized six-dimensional feature vector corresponding to each historical vibration data.

[0074] S104, using all normalized six-dimensional feature vectors, construct a training set and a test set.

[0075] S105, construct a kernel fuzzy C-means clustering model, and initialize the cluster center and membership matrix in the kernel fuzzy C-means clustering model.

[0076] The construction of the kernel fuzzy C-means clustering model mainly includes two aspects: on the one hand, a suitable kernel function needs to be selected, and on the other hand, the kernel function needs to be combined with the fuzzy C-means clustering model to construct a kernel fuzzy C-means clustering algorithm model for identifying and diagnosing known and unknown faults of the oil pump. The process of constructing the kernel fuzzy C-means clustering model is S1051-S1052:

[0077] S1051, select a kernel function:

[0078] The kernel function can convert the vector in the original space from a low-dimensional space to a high-dimensional space, so that the problem of non-linear classification in a low-dimensional space can be linearly classified in a high-dimensional space. Based on the above principle, the kernel function can improve the classification accuracy of the classification algorithm on complex data sets such as vibration data sets. The kernel function mainly includes linear kernel function, polynomial kernel function, Gaussian kernel function, etc. In specific problems, the kernel function is usually selected according to the dimension of the input data. Since the input of the algorithm model is the six-dimensional feature after feature extraction, the dimension of the feature vector is relatively small. Therefore, the Gaussian kernel function is selected, and the Gaussian kernel function is defined as follows:

[0079]

[0080] where x' is the center of the kernel function, and ||x-x'| 2 is the Euclidean distance between vector x and vector x', which monotonically decreases with the distance between the two vectors.

[0081] S1052, construct a kernel fuzzy C-means clustering algorithm model:

[0082] The kernel function is combined with the fuzzy C-means clustering model, the kernel function is added when calculating the distance between the sample and the cluster, the objective function of the fuzzy C-means clustering is optimized, and the final kernel fuzzy C-means clustering algorithm model objective function is:

[0083]

[0084] Wherein, o i is the cluster center, c is the cluster number, n is the sample number, μ ik is the membership degree of the kth sample relative to the ith class, and m is the weight factor.

[0085] S106, iteratively calculate the cluster center and update the membership matrix.

[0086] S107, judge whether the change precision of the membership matrix meets the requirements or not, terminate the training if the requirements are met, otherwise continue the iterative training, and repeat S21 until the change precision of the membership matrix meets the requirements.

[0087] S108, fuzzy processing is performed on the membership matrix meeting the change precision requirements, so that the membership degree of each sample in the training set to the latest obtained cluster center is maximized to 1, and the rest is 0, and a trained kernel fuzzy C-means clustering model is obtained, the trained kernel fuzzy C-means clustering model includes the membership matrix meeting the change precision requirements and the latest obtained cluster center.

[0088] S109, obtaining the vibration data of the running process of the oil pump to be diagnosed.

[0089] S110, performing ensemble empirical mode decomposition on the vibration data of the oil pump to be diagnosed, and obtaining the IMF component of the vibration data of the oil pump to be diagnosed.

[0090] S111, selecting the first 6 components in the IMF component of the vibration data of the oil pump to be diagnosed, respectively performing singular value decomposition, and obtaining a six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed.

[0091] S112, normalizing the six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed, and obtaining a normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed.

[0092] S113, calculating the similarity between the vibration data of the oil pump to be diagnosed and each cluster center according to the normalized six-dimensional feature vector corresponding to the vibration data of the oil pump to be diagnosed, judging whether the vibration data of the oil pump to be diagnosed corresponds to a known fault or not, and obtaining a judgment result.

[0093] When identifying and diagnosing the vibration data collected during the operation of the oil pump, a suitable similarity function needs to be constructed to calculate the similarity between the sample xnew The similarity of each cluster center o i outputted by training is used to distinguish known faults from unknown faults. The similarity function is constructed as follows:

[0094]

[0095] where d(x new , o i ) is the Euclidean distance of the new sample x new and the cluster center o i in the and space, d avg is the average Euclidean distance of all samples in the ith class to the cluster center o i .

[0096]

[0097] where |C i | is the number of samples classified in the ith class.

[0098] S114, when the judgment result is yes, the vibration data of the oil pump to be diagnosed is classified into the cluster center with the highest similarity to the vibration data, and the fault label of the cluster center is combined to determine the corresponding fault of the vibration data of the oil pump to be diagnosed, when the judgment result is no, the number of occurrences of the unknown fault is recorded, when the number of occurrences reaches a set threshold, it is marked as a new fault mode, the vibration data under the fault mode is made as a sample, and is added to the training set, and the kernel fuzzy C-means clustering model is retrained.

[0099] For verification of the model self-learning ability, the oil pump vibration data set is used for testing, the data set mainly includes four types of normal, inner ring fault, outer ring fault and rolling body fault, and the known fault diagnosis ability, unknown fault recognition and discovery ability and model self-learning ability are verified.

[0100] 1) Known fault diagnosis ability verification:

[0101] The kernel fuzzy C-means clustering algorithm model is trained by using the historical vibration data of the oil pump. As can be seen from Table 1, using the algorithm model, 99% classification accuracy can be achieved with relatively fewer iterations, which proves the effectiveness of the algorithm model for known fault diagnosis. The diagnosis ability for known faults is shown in Table 1.

[0102] Table 1:

[0103]

[0104] 2) Unknown fault recognition and discovery ability and self-learning ability verification:

[0105] The oil delivery pump normal, outer ring fault, rolling body fault three kinds of fault data are taken as training data, input into the algorithm model for classification, and the inner ring fault data is taken as test data, the similarity of the inner ring fault test data and the three kinds of historical known faults is calculated, and the similarity calculation result of a certain vibration data sample is as follows:

[0106] [2.52×10 -18 ,2.88×10 -8 ,2.86×10 -9 ]

[0107] The similarity threshold is set to 0.5, and the calculation result shows that the similarity of the test sample and the three kinds of historical known faults is less than the threshold 0.5, so the test sample is output as other unknown faults except the three kinds of historical known faults, which is consistent with the actual situation. The experimental result of the new fault recognition of the algorithm model is shown in Figure 4 , wherein "+" represents test data, and Figure 4 It can be seen that the test data is far away from the other three kinds of historical known faults, so the algorithm model can distinguish the known faults and unknown faults, and realize the discovery of unknown faults.

[0108] At the same time, since the inner ring fault data is identified as an unknown fault mode by the algorithm model, and the fault mode is detected by the model for many times, reaches the set threshold constant 10, therefore, the model identifies it as a new fault mode, and adds the fault mode data detected for many times to the training data set, combines the fault mode detection result of artificial field, re-trains the algorithm model, realizes the expansion of the diagnosable fault types of the model. After adding the new fault data, the re-training result of the algorithm model is shown in Figure 5 , the historical known faults that can be identified by the algorithm model are changed from three to four, which proves the self-learning ability of the algorithm model.

[0109] The technical effects of the present application are as follows:

[0110] 1) The method of the present application has self-learning ability, and can expand the diagnosable fault types of the fault diagnosis model semi-automatically.

[0111] 2) The set empirical mode decomposition and singular value decomposition are used for feature extraction of the vibration data, which reduces the dimension of the vibration data, and further reduces the complexity of the algorithm operation and the time cost of model training.

[0112] 3) On the basis of a large number of experiments, it is found that the fuzzy C-means clustering algorithm combines K-means hard clustering algorithm and fuzzy mathematical theory, and can obtain more flexible and accurate clustering results than hard clustering, but it performs poorly on the vibration data set, and the diagnosis accuracy is not high, therefore, the kernel function is added in the fuzzy C-means clustering algorithm, the vibration data features are transformed to high-dimensional space through the kernel function, and then the fuzzy C-means clustering method is used for clustering analysis, so that the final classification accuracy of known faults reaches 99%, and a model capable of discovering new fault modes is obtained.

[0113] 4) The application designs a fault diagnosis self-learning method based on kernel fuzzy C-means clustering, which is applied to the fault diagnosis of the oil pump. In view of the deficiency that the traditional fault diagnosis method can only diagnose the known faults in the training set, the self-learning mode is designed, unknown faults can be discovered, and the fault diagnosis model is semi-automatically expanded to diagnose fault types. The set empirical mode decomposition and singular value decomposition are used for feature extraction of the vibration data, so that the dimension of the vibration data is reduced, and the time cost of model training is reduced. The fuzzy C-means clustering algorithm fused with the kernel function is used, and the six-dimensional feature vector after feature extraction is used as the input of the model. The known fault diagnosis ability, unknown fault recognition and discovery ability and model self-learning ability are verified respectively, the effectiveness of the algorithm model is proved, and the finally obtained model is applied to the fault diagnosis of the oil pump.

[0114] In the above embodiments, although the steps are numbered S1, S2, etc., it is only a specific embodiment given by the present application, and those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation, which is within the protection scope of the present application. It can be understood that in some embodiments, some or all of the above embodiments can be included.

[0115] As shown in Figure 6 The oil pump fault diagnosis self-learning system 200 based on the kernel fuzzy C-means clustering of the embodiment of the application includes an acquisition and construction module 210, a training module 220, an acquisition module 230, a first judgment module 240 and a determination module 250.

[0116] The acquisition and construction module 210 is used for acquiring a normalized six-dimensional feature vector corresponding to each historical vibration data in the historical vibration data set of the oil pump, and constructing a training set and a test set by using all the normalized six-dimensional feature vectors.

[0117] The training module 220 is used for obtaining a trained kernel fuzzy C-means clustering model based on the training set and the test set.

[0118] The acquisition module 230 is configured to acquire vibration data of the to-be-diagnosed fuel pump and obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump;

[0119] The first determining module 240 is configured to calculate the similarity between the vibration data of the to-be-diagnosed fuel pump and each cluster center according to the normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump, and determine whether the vibration data of the to-be-diagnosed fuel pump corresponds to a known fault, to obtain a determination result;

[0120] The determining module 250 is configured to, when the determination result is yes, divide the vibration data of the to-be-diagnosed fuel pump to the cluster center with the highest similarity to the vibration data, and determine the fault corresponding to the vibration data of the to-be-diagnosed fuel pump in combination with the fault label of the cluster center, and when the determination result is no, record the number of occurrences of the unknown fault, and when the number of occurrences reaches a set threshold, mark the fault as a new fault mode, make the vibration data under the fault mode as a sample, and add the sample to a training set to retrain the kernel fuzzy C-means clustering model.

[0121] Optionally, in the technical solution above, the obtaining and constructing module 210 is specifically configured to:

[0122] perform ensemble empirical mode decomposition on each piece of historical vibration data in the historical vibration data set to obtain an IMF component of each piece of historical vibration data;

[0123] select the first six components in the IMF component of each piece of historical vibration data, and perform singular value decomposition on each of the first six components to obtain a six-dimensional feature vector corresponding to each piece of historical vibration data;

[0124] perform normalization processing on each six-dimensional feature vector to obtain a normalized six-dimensional feature vector corresponding to each piece of historical vibration data.

[0125] Optionally, in the technical solution above, the training module 220 includes an obtaining module, an iterative updating module, a second determining module, and a fuzzification processing module;

[0126] The obtaining module is configured to construct a kernel fuzzy C-means clustering model, and initialize the cluster centers and a membership matrix in the kernel fuzzy C-means clustering model;

[0127] The iterative updating module is configured to iteratively calculate the cluster centers and update the membership matrix;

[0128] The second determining module is configured to determine whether the change precision of the membership matrix meets the requirement, and terminate the training when the change precision meets the requirement, or continue the iterative training, repeatedly call the iterative updating module, and terminate the training when the change precision of the membership matrix meets the requirement.

[0129] The fuzzification processing module is configured to perform fuzzification processing on the membership matrix meeting the change precision requirement, so that the membership of each sample in the training set to the latest obtained clustering center is 1 and the membership to the rest is 0, and a trained kernel fuzzy C-means clustering model is obtained, wherein the trained kernel fuzzy C-means clustering model comprises the membership matrix meeting the change precision requirement and the latest obtained clustering center.

[0130] Optionally, in the above technical solution, the acquisition module 230 is specifically configured to:

[0131] The vibration data of the to-be-diagnosed fuel pump is subjected to ensemble empirical mode decomposition to obtain IMF components of the vibration data of the to-be-diagnosed fuel pump;

[0132] In the IMF components of the vibration data of the to-be-diagnosed fuel pump, the first six components are selected, and singular value decomposition is performed on each of the first six components to obtain a six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump;

[0133] The six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump is subjected to normalization processing to obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump.

[0134] The above steps about the implementation of the respective functions of the parameters and the respective unit modules in the self-learning system 200 for fuel pump fault diagnosis based on kernel fuzzy C-means clustering can refer to the parameters and steps in the above embodiments of the self-learning method for fuel pump fault diagnosis, and will not be described here.

[0135] The storage medium of the embodiment of the present application has instructions stored therein, and when a computer reads the instructions, the computer executes the self-learning method for fuel pump fault diagnosis according to any one of the above embodiments.

[0136] The electronic device of the embodiment of the present application comprises a processor and the above storage medium, and the processor executes the instructions in the storage medium, wherein the electronic device can be a computer, a mobile phone or the like.

[0137] Those skilled in the art know that the present application can be implemented as a system, a method or a computer program product.

[0138] Therefore, the present disclosure can be embodied in the form of a complete hardware, a complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, which is generally referred to as "circuitry", "module" or "system" herein. In addition, in some embodiments, the present application can also be embodied in the form of a computer program product in one or more computer readable media, which contains computer readable program codes.

[0139] Any combination of one or more computer readable medium can be employed. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium can be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

[0140] Although the embodiments of the present application have been shown and described above, it should be understood by those skilled in the art that the above embodiments are exemplary, and should not be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements, and variations to the above embodiments within the scope of the present application.

Claims

1. A nuclear fuzzy C-means clustering-based fault diagnosis self-learning method for a fuel delivery pump, characterized in that, The method comprises the following steps: obtaining a normalized six-dimensional feature vector corresponding to each piece of historical vibration data in a historical vibration data set of an oil pump, constructing a training set and a test set by using all the normalized six-dimensional feature vectors; the process of obtaining the normalized six-dimensional feature vector corresponding to each piece of historical vibration data is as follows: performing ensemble empirical mode decomposition on each piece of historical vibration data in the historical vibration data set to obtain an IMF component of each piece of historical vibration data; selecting the first six components in the IMF component of each piece of historical vibration data, respectively performing singular value decomposition to obtain a six-dimensional feature vector corresponding to each piece of historical vibration data; performing normalization processing on each six-dimensional feature vector to obtain a normalized six-dimensional feature vector corresponding to each piece of historical vibration data; obtaining a trained kernel fuzzy C-means clustering model based on the training set and the test set; collecting vibration data of a to-be-diagnosed oil pump during operation and obtaining a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil pump; calculating the similarity between the vibration data of the to-be-diagnosed oil pump and each cluster center according to the normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil pump, judging whether the vibration data of the to-be-diagnosed oil pump corresponds to a known fault, and obtaining a judgment result; when the judgment result is yes, the vibration data of the to-be-diagnosed oil pump is divided into the cluster center with the highest similarity to the vibration data, and the fault label of the cluster center is combined to determine the fault corresponding to the vibration data of the to-be-diagnosed oil pump; when the judgment result is no, the number of occurrences of an unknown fault is recorded, and when the number of occurrences reaches a set threshold, the fault mode is marked as a new fault mode, the vibration data under the fault mode is made as a sample, and the training set is added to retrain the kernel fuzzy C-means clustering model.

2. The method according to claim 1, wherein, The trained kernel fuzzy C-means clustering model based on the training set and the test set comprises the following steps: S20, constructing a kernel fuzzy C-means clustering model, and initializing the cluster centers and the membership matrix in the kernel fuzzy C-means clustering model; S21, iteratively calculating the cluster centers and updating the membership matrix; S22, judging whether the change precision of the membership matrix meets the requirements, and terminating the training when the change precision meets the requirements, otherwise, continuing the iterative training, repeatedly executing S21, and until the change precision of the membership matrix meets the requirements; S23, fuzzifying the membership matrix meeting the change precision meeting the requirements, so that the membership of each sample in the training set to the latest obtained cluster center is maximum and is 1, and the rest is 0, to obtain the trained kernel fuzzy C-means clustering model, and the trained kernel fuzzy C-means clustering model comprises the membership matrix meeting the change precision meeting the requirements and the latest obtained cluster centers.

3. The method according to claim 1, wherein, The normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil pump is obtained according to the vibration data of the to-be-diagnosed oil pump during operation, and the method comprises the following steps: performing ensemble empirical mode decomposition on the vibration data of the to-be-diagnosed oil pump to obtain an IMF component of the vibration data of the to-be-diagnosed oil pump; The first six components in the IMF component of the vibration data of the to-be-diagnosed fuel pump are selected, and singular value decomposition is performed on each of the selected components to obtain a six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump; The six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump is normalized to obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump.

4. A nuclear fuzzy C-means clustering-based fault diagnosis self-learning system for a fuel delivery pump, characterized by, The method comprises an acquisition and construction module, a training module, a collection and acquisition module, a first judgment module, and a determination module; The acquisition and construction module is configured to: acquire a normalized six-dimensional feature vector corresponding to each piece of historical vibration data in a historical vibration data set of the fuel pump, and use all the normalized six-dimensional feature vectors to construct a training set and a test set; The acquisition and construction module is specifically configured to: Perform ensemble empirical mode decomposition on each piece of historical vibration data in the historical vibration data set to acquire an IMF component of each piece of historical vibration data; The first six components in the IMF component of each piece of historical vibration data are selected, and singular value decomposition is performed on each of the selected components to obtain a six-dimensional feature vector corresponding to each piece of historical vibration data; Each six-dimensional feature vector is normalized to obtain a normalized six-dimensional feature vector corresponding to each piece of historical vibration data; The training module is configured to: based on the training set and the test set, obtain a trained kernel fuzzy C-means clustering model; The collection and acquisition module is configured to: collect vibration data of a running process of a to-be-diagnosed fuel pump to obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump; The first judgment module is configured to: based on the normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed fuel pump, calculate a similarity between the vibration data and each cluster center, and determine whether the vibration data of the to-be-diagnosed fuel pump corresponds to a known fault to obtain a judgment result; The determination module is configured to: when the judgment result is yes, divide the vibration data of the to-be-diagnosed fuel pump to a cluster center with the highest similarity to the vibration data, and determine a fault corresponding to the vibration data of the to-be-diagnosed fuel pump in combination with a fault label of the cluster center; when the judgment result is no, record a number of times of occurrence of an unknown fault, and when the number of times of occurrence reaches a set threshold, mark the unknown fault as a new fault mode, make vibration data in the fault mode as a sample, and add the sample to the training set to retrain the kernel fuzzy C-means clustering model.

5. The oil pump fault diagnosis self-learning system based on kernel fuzzy C-means clustering according to claim 4, characterized in that, The training module comprises a construction module, an iterative updating module, a second judgment module, and a fuzzification processing module; The construction module is configured to: construct a kernel fuzzy C-means clustering model, and initialize cluster centers and a membership matrix in the kernel fuzzy C-means clustering model; The iterative updating module is configured to: iteratively calculate the cluster centers, and update the membership matrix; The second judgment module is configured to: determine whether a change precision of the membership matrix meets a requirement, and terminate training when the change precision of the membership matrix meets the requirement, or continue iterative training when the change precision of the membership matrix does not meet the requirement, repeatedly call the iterative updating module, and stop until the change precision of the membership matrix meets the requirement. The fuzzification processing module is configured to perform fuzzification processing on the membership matrix meeting the change precision requirement, so that the membership of each sample in the training set to the latest obtained clustering center is 1 at most and 0 at least, and obtain a trained kernel fuzzy C-means clustering model, the trained kernel fuzzy C-means clustering model comprising the membership matrix meeting the change precision requirement and the latest obtained clustering center.

6. The oil pump fault diagnosis self-learning system based on kernel fuzzy C-means clustering according to claim 4, characterized in that, The collection module is specifically configured to: perform ensemble empirical mode decomposition on the vibration data of the to-be-diagnosed oil delivery pump to obtain an IMF component of the vibration data of the to-be-diagnosed oil delivery pump; select the first six components from the IMF component of the vibration data of the to-be-diagnosed oil delivery pump, and perform singular value decomposition on each of the first six components to obtain a six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil delivery pump; perform normalization processing on the six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil delivery pump to obtain a normalized six-dimensional feature vector corresponding to the vibration data of the to-be-diagnosed oil delivery pump.

7. A storage medium, characterized by The storage medium has instructions stored therein, and when a computer reads the instructions, the computer is caused to perform the method of claim 1 to 3.

8. An electronic device, comprising: The storage medium has instructions stored therein, and when a computer reads the instructions, the computer is caused to perform the method of claim 1 to 3.

Images