Rotor crack fault diagnosis and model construction methods and apparatus

Abstract

Disclosed are rotor crack fault diagnosis and model construction methods and apparatus. The rotor crack fault diagnosis method comprises: collecting a vibration signal to be diagnosed of a rotor of a current rotating device during an operation process; after singular value decomposition is performed on said vibration signal, performing singular spectrum analysis, and extracting to obtain a plurality of principal component singular values to be diagnosed; and inputting said plurality of principal component singular values into an nSVDD model for rotor crack fault diagnosis, and obtaining a rotor crack fault diagnosis result of the current rotating device; the nSVDD model for rotor crack fault diagnosis is pre-trained and generated on the basis of a vibration signal sample set of a rotor of a rotating device during a historical operation process. According to the present invention, rotor crack fault diagnosis based on singular values and nSVDD may be achieved, improving the accuracy of rotor crack fault diagnosis.

Description

Rotor crack fault diagnosis, model building method and device

[0001] This application claims priority to Chinese patent application No. 202310952521.6 filed on July 31, 2023, entitled “Rotor Crack Fault Diagnosis, Model Building Method and Device,” the entire contents of which are incorporated herein by reference. Technical Field

[0002] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a rotor crack fault diagnosis and model building method and device. Background Art

[0003] This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by virtue of its inclusion in this section.

[0004] Rotors are key components in rotating machinery systems, such as pumps. Rotor cracks are among the most common rotor failures, characterized by high insidiousness and severe hazard. Rotors are prone to cracking due to a combination of inherent material defects, harsh operating environments, random loads, and the prolonged high-speed operation of rotating machinery systems. The presence of cracks can seriously impact the safe operation of the rotor system. If rotor cracks are not detected promptly, their propagation can lead to shaft fracture, potentially resulting in catastrophic failures, resulting in economic losses to equipment and property, and posing a significant threat to human safety. Therefore, analyzing the rotor's dynamic characteristics and performing real-time online monitoring of vibration signals to implement online fault diagnosis for rotating machinery with cracked shafts is crucial for ensuring safe and reliable operation and preventing major accidents. Currently, rotor crack faults are typically diagnosed using spectrum analysis and vibration amplitude monitoring, which suffer from significant hysteresis and low automation.

[0005] Summary of the Invention

[0006] An embodiment of the present invention provides a rotor crack fault diagnosis method for implementing rotor crack fault diagnosis based on singular values ​​and nSVDD. The method includes:

[0007] Collect the vibration signals to be diagnosed from the rotor of the current rotating equipment during operation;

[0008] After the vibration signal to be diagnosed is subjected to singular value decomposition, a singular spectrum analysis is performed to extract multiple singular values ​​of the principal components to be diagnosed;

[0009] Multiple principal component singular values ​​to be diagnosed are input into a rotor crack fault diagnosis nSVDD model to obtain a rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on rotor vibration signal sample data during the historical operation of the rotating equipment.

[0010] An embodiment of the present invention provides a method for constructing a rotor crack fault diagnosis model to implement rotor crack fault diagnosis based on singular values ​​and nSVDD. The method includes:

[0011] Obtain historical vibration signal sample data of the rotor of the rotating equipment during operation;

[0012] After the historical vibration signal is subjected to singular value decomposition, a singular spectrum analysis is performed to extract multiple historical principal component singular values;

[0013] The obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results are used as training samples to input into the nSVDD learning model for model training to obtain the rotor crack fault diagnosis model.

[0014] An embodiment of the present invention further provides a rotor crack fault diagnosis device for implementing rotor crack fault diagnosis based on singular values ​​and nSVDD, the device comprising:

[0015] The acquisition unit is used to collect the vibration signal to be diagnosed of the rotor of the current rotating equipment during operation;

[0016] The first extraction unit is used to perform singular spectrum analysis on the vibration signal to be diagnosed after singular value decomposition, and extract multiple singular values ​​of the principal components to be diagnosed;

[0017] The diagnosis unit is used to input multiple principal component singular values ​​to be diagnosed into a rotor crack fault diagnosis nSVDD model to obtain a rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on vibration signal sample data of the rotor during the historical operation of the rotating equipment.

[0018] An embodiment of the present invention further provides a device for constructing a rotor crack fault diagnosis model, for implementing rotor crack fault diagnosis based on singular values ​​and nSVDD, the device comprising:

[0019] An acquisition unit, used to acquire historical vibration signal sample data of the rotor of the rotating equipment during operation;

[0020] The second extraction unit is used to perform singular spectrum analysis on the historical vibration signal after singular value decomposition, and extract multiple historical principal component singular values;

[0021] The training unit is used to input the obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

[0022] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned rotor crack fault diagnosis method when executing the computer program.

[0023] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the rotor crack fault diagnosis method described above is implemented.

[0024] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the rotor crack fault diagnosis method is implemented.

[0025] The rotor crack fault diagnosis solution provided by an embodiment of the present invention, when in operation: collects the vibration signal to be diagnosed of the rotor during the operation of the current rotating equipment; performs singular spectrum analysis on the vibration signal to be diagnosed after singular value decomposition, and extracts multiple principal component singular values ​​to be diagnosed; inputs the multiple principal component singular values ​​to be diagnosed into the rotor crack fault diagnosis nSVDD model to obtain the rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on the vibration signal sample set of the rotor during the historical operation of the rotating equipment.

[0026] Compared with the prior art, the rotor crack fault diagnosis solution provided by the embodiment of the present invention has the following beneficial technical effects:

[0027] First, the present invention uses the singular values ​​of the principal components as features for fault diagnosis, eliminating the need for signal restoration. Compared to conventional spectrum signal extraction, the singular value features show greater differences between rotor cracks and normal rotors, leading to more accurate fault classification. The resulting model is less prone to overfitting and exhibits superior generalization performance, resulting in highly accurate rotor crack fault diagnosis.

[0028] Secondly, the single-classification nSVDD algorithm used in the embodiment of the present invention has the advantages of excellent generalization performance in single-class fault identification, low computational resources, easy model iteration and update, and simple implementation steps. Its comprehensive performance in practical applications in the industrial field is better than that of deep learning algorithms.

[0029] In summary, the rotor crack fault diagnosis scheme provided by the embodiment of the present invention can effectively extract the prominent energy components in the rotor vibration signal, represent them in the form of singular values, and use the improved single-class fault diagnosis learner nSVDD for modeling and learning. The recognition rate of rotor crack faults is high, and rotor crack fault diagnosis based on singular values ​​and nSVDD can be realized, thereby improving the accuracy of rotor crack fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention. In the drawings:

[0031] FIG1 is a schematic flow chart of a rotor crack fault diagnosis method according to an embodiment of the present invention;

[0032] FIG2 is a schematic diagram showing the principle of rotor crack fault diagnosis according to an embodiment of the present invention;

[0033] FIG3 is a flow chart of a method for constructing a rotor crack fault diagnosis model according to an embodiment of the present invention;

[0034] FIG4 is a schematic structural diagram of a rotor crack fault diagnosis device according to an embodiment of the present invention;

[0035] FIG5 is a schematic structural diagram of a rotor crack fault diagnosis device according to another embodiment of the present invention;

[0036] FIG6 is a schematic structural diagram of a device for constructing a rotor crack fault diagnosis model according to an embodiment of the present invention;

[0037] FIG7 is a schematic structural diagram of a device for constructing a rotor crack fault diagnosis model according to another embodiment of the present invention;

[0038] FIG8 is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0040] A technical problem with existing rotor crack detection technology is that frequency-domain analysis of rotor vibration signals manually observes changes in the rotor's rated frequency harmonic components, assuming a rotor crack fault has occurred. In automated implementation, manual determination of the rotor crack fault monitoring frequency band is performed, and then a mechanism-based or data-driven monitoring model is established based on the features extracted from specific frequency components.

[0041] In response to the technical problems existing in the prior art, the inventors proposed a rotor crack fault diagnosis scheme based on singular values ​​and nSVDD. This scheme is a rotor crack fault diagnosis scheme based on singular value decomposition and single-class anomaly recognition. This scheme performs singular spectrum analysis on the vibration signal after singular value decomposition (SVD), extracts the principal component singular values ​​as the feature vectors for fault diagnosis and inputs them into the nSVDD model, establishes a rotor crack fault diagnosis model, and realizes automatic diagnosis (detection) of rotor crack faults.

[0042] FIG1 is a flow chart of a rotor crack fault diagnosis method based on singular values ​​and nSVDD according to an embodiment of the present invention. As shown in FIG1 , the method includes the following steps:

[0043] Step 101: collecting a vibration signal to be diagnosed of a rotor of a current rotating device during operation;

[0044] Step 102: performing singular spectrum analysis on the vibration signal to be diagnosed after singular value decomposition, and extracting a plurality of singular values ​​of the principal components to be diagnosed;

[0045] Step 103: Input multiple principal component singular values ​​to be diagnosed into a rotor crack fault diagnosis nSVDD model to obtain a rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on rotor vibration signal sample data during historical operation of the rotating equipment.

[0046] The rotor crack fault diagnosis method based on singular values ​​and nSVDD provided by an embodiment of the present invention, when working: collects the vibration signal to be diagnosed of the rotor during the operation of the current rotating equipment; performs singular spectrum analysis on the vibration signal to be diagnosed after singular value decomposition, and extracts multiple singular values ​​of the principal components to be diagnosed; inputs the multiple singular values ​​of the principal components to be diagnosed into the rotor crack fault diagnosis nSVDD model to obtain the rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on the vibration signal sample set of the rotor during the historical operation of the rotating equipment.

[0047] Compared with the prior art, the rotor crack fault diagnosis method based on singular values ​​and nSVDD provided by the embodiment of the present invention has the following beneficial technical effects:

[0048] First, the present invention uses the singular values ​​of the principal components as features for fault diagnosis, eliminating the need for signal restoration. Compared to conventional spectrum signal extraction, the singular value features show greater differences between rotor cracks and normal rotors, leading to more accurate fault classification. The resulting model is less prone to overfitting and exhibits superior generalization performance, resulting in highly accurate rotor crack fault diagnosis.

[0049] Secondly, the single-classification nSVDD algorithm used in the embodiment of the present invention has the advantages of excellent generalization performance in single-class fault identification, low computational resources, easy model iteration and update, and simple implementation steps. Its comprehensive performance in practical applications in the industrial field is better than that of deep learning algorithms.

[0050] In summary, the rotor crack fault diagnosis method based on singular values ​​and nSVDD provided by the embodiments of the present invention can effectively extract prominent energy components from rotor vibration signals, characterize them in the form of singular values, and use the improved single-class fault diagnosis learner nSVDD for modeling and learning. This method has a high recognition rate for rotor crack faults, can implement rotor crack fault diagnosis based on singular values ​​and nSVDD, and improves the accuracy of rotor crack fault diagnosis. The following is a detailed introduction to this rotor crack fault diagnosis method based on singular values ​​and nSVDD.

[0051] In order to solve the problem of difficulty in identifying rotor cracks, the present invention proposes a rotor crack fault diagnosis method based on singular value and nSVDD (Negative Support vector data description, support vector data description with negative samples) algorithm. The vibration signal is subjected to singular value decomposition and then singular spectrum analysis is performed. The singular values ​​of the principal components are extracted as the feature vectors for fault diagnosis and input into the nSVDD model to establish a rotor crack fault diagnosis model. The SVDD model is an efficient single-classification model suitable for specialized identification of single faults. nSVDD introduces negative samples on the basis of SVDD, which can more accurately divide the optimal classification surface. After verification experiments on rotor fault data sets, the results show that the classification accuracy of the present invention is much higher than that of the existing technology, and it can be applied to the diagnosis of rotor crack faults in various types of rotating mechanical equipment, and has a relatively broad application prospect. The following is a detailed description.

[0052] The present invention proposes a rotor crack fault diagnosis method based on singular values ​​and nSVDD algorithms. As shown in Figure 2, this method is based on an existing rotor crack label case data set and involves two main processes: singular value decomposition feature extraction and nSVDD modeling. The left half of Figure 2 is the modeling process, and the right half of Figure 2 is the process of using the established model to perform rotor crack fault diagnosis (also known as prediction). Similarly, the right half also involves singular value decomposition feature extraction and inputting the extracted data to be predicted into the established nSVDD for prediction. The specific implementation steps of the rotor crack fault diagnosis method based on singular values ​​and nSVDD algorithms involved in the embodiment of the present invention are as follows:

[0053] Step 1: Collect vibration signals from rotating equipment and perform preprocessing.

[0054] Step 2 - Reconstruct the vibration signal in the form of a Hankel matrix and perform singular value decomposition.

[0055] That is, in one embodiment, after the singular value decomposition of the vibration signal to be diagnosed currently, singular spectrum analysis is performed, and multiple principal component singular values to be diagnosed are extracted, which may include: reconstructing the vibration signal to be diagnosed in the form of a Hankel matrix, and performing singular spectrum analysis after singular value decomposition to extract multiple principal component singular values to be diagnosed.

[0056] Specifically, the main purpose of the embodiment of the present invention is to provide a rotor crack fault diagnosis method based on singular values and the nSVDD algorithm, aiming to identify possible crack faults in the rotor components during the operation of rotating equipment. The key point is to perform singular value decomposition with a Hankel matrix and extract singular values as features; applying nSVDD as a dedicated model for rotor crack identification. When using a Hankel matrix, singular value decomposition can decompose the original signal into a simple linear superposition of a series of component signals, and each component has the characteristic of zero phase shift, that is, if several components are deleted, the phase of each other separated component signal remains unchanged, and the main component of the rotor crack is more easily identified.

[0057] Further, for the vibration signal S = [s(1), s(2), …, s(N)] ∈ R N×1 , where N is the signal length, the construction form of the constructed Hankel matrix is:

[0058] where 1 < n < N, n is called the decomposition order, m = N - n + 1, H s ∈ R m×n ;

[0059] After performing singular value decomposition on it and rewriting the column vectors u i and ν i in the represented form, we have: H s = USV T = u1σ1ν1 T + u2σ2ν2 T + + … + u i σ i ν i T + … + + u r σ r ν r T

[0060] where, σ1, σ2, … σ r decrease from large to small and are called the singular values of the H s matrix, u i is the left singular vector, νi is a right singular vector.

[0061] Furthermore, experiments have shown that when n is the center of symmetry, the singular value decomposition feature extraction effect under the Hankel matrix is ​​optimal. That is, in one embodiment, n is the center of symmetry to achieve the best singular value decomposition feature extraction effect under the Hankel matrix.

[0062] When the signal length N is an even number, n = N / 2, m = N / 2 + 1 should be used to construct the Hankel matrix. When the signal length N is an odd number, n = (N + 1) / 2, m = (N + 1) / 2 should be used to construct the Hankel matrix. That is, in one embodiment, when the signal length N is an even number, n = N / 2, m = N / 2 + 1 can be used to construct the Hankel matrix; when the signal length N is an odd number, n = (N + 1) / 2, m = (N + 1) / 2 can be used to construct the Hankel matrix, so that the singular value decomposition feature extraction effect under the Hankel matrix is ​​the best.

[0063] Step 3-Select singular values ​​to form feature vectors based on a certain feature selection strategy.

[0064] In one embodiment, the rotor crack fault diagnosis method based on singular values ​​and nSVDD may further include: selecting singular values ​​from the extracted plurality of principal component singular values ​​to be diagnosed to form a feature vector;

[0065] Inputting multiple principal component singular values ​​to be diagnosed into a rotor crack fault diagnosis model to obtain a rotor crack fault diagnosis result of the current rotating equipment, including: inputting a eigenvector composed of selected singular values ​​into the rotor crack fault diagnosis model to obtain a rotor crack fault diagnosis result of the current rotating equipment.

[0066] Furthermore, the selection strategy for singular value features (preset feature selection strategy) is subject to the following principles:

[0067] 1. The singular value component value is positively correlated with the energy contained in the frequency component it represents. The higher the ranking, the larger the value of the singular value. First, obtain multiple singular values ​​of the principal components to be diagnosed. Among them, the singular value component value is positively correlated with the energy contained in the frequency component it represents. The multiple singular values ​​of the principal components to be diagnosed are arranged in descending order.

[0068] 2. If a DC component exists in the signal, a significant value (typically 3-5 orders of magnitude larger than the next largest singular value, based on the signal mean) will appear at the top of the singular value list. Therefore, this value must be discarded during analysis, and the analysis must begin again with the next largest value. If the signal has already been zero-meaned, this value need not be discarded. Specifically, multiple singular values ​​of the principal components to be diagnosed, arranged in descending order, are screened: If a DC component exists in the vibration signal to be diagnosed, the first singular value is discarded, resulting in multiple singular values ​​of the principal components to be diagnosed after a single screening process.

[0069] 3. The singular values ​​obtained by singular value decomposition of Hankel matrix are paired down (e.g. σ1≈σ2>σ3≈σ4>σ5≈σ6>σ7≈σ 8… ) feature, singular values ​​with consecutive numbers and similar sizes must be generated by the same frequency, and singular values ​​with discontinuous numbers must originate from different frequency sources. That is, a secondary screening process is performed on the multiple main component singular values ​​to be diagnosed after the primary screening: multiple main component singular values ​​to be diagnosed with pairs of descending singular values ​​are selected from the multiple main component singular values ​​to be diagnosed after the primary screening, to obtain multiple main component singular values ​​to be diagnosed after the secondary screening process; wherein, the pair-wise descending singular values ​​are: the arrangement sequence numbers of every two singular values ​​are consecutive and the difference between the two singular values ​​is less than a preset value, and the multiple pairs of singular values ​​are arranged in descending order.

[0070] 4. Based on experience, the top 10% (A) of singular value pairs already cover 90% (B) of the signal energy (for example, the sum of the energy of the top five singular values ​​(10% of the total) is greater than or equal to the sum of the energy of the remaining 90% of the singular value signals). The remainder is considered noise redundancy. Specifically, among the multiple singular values ​​of the principal component to be diagnosed after the secondary screening process, the singular values ​​of the principal component to be diagnosed that cover a predetermined proportion (90%) of the signal energy are found (these singular values ​​are the singular values ​​of the top 10% (A) pairs) to form the feature vector.

[0071] Therefore, if the number of fixed features is 5, the singular value feature selection strategy (the strategy for the fixed number of features) is as follows:

[0072] When the DC component is discarded (see step 2 above), the selected singular value feature group is [σ2,σ4,σ 6, σ 8, σ 10 ];

[0073] If there is no discard (see step 2 above), the selected singular value feature group is [σ1,σ3,σ 5, σ 7, σ9];

[0074] Furthermore, the method of selecting singular value feature entries (singular value feature selection strategy) can be diverse and is not limited to the fixed feature number strategy. On the premise of deleting the DC component as appropriate, you can also choose a fixed number ratio (a certain proportion of the total number of singular values), a fixed total value ratio (a certain proportion of the total singular value sum), and other methods (strategies).

[0075] As can be seen from the above, in one embodiment, selecting singular values ​​from the extracted plurality of singular values ​​of the principal components to be diagnosed to form a feature vector may include:

[0076] Obtaining multiple singular values ​​of the principal components to be diagnosed; wherein the value of the singular value component is positively correlated with the energy contained in the frequency component it represents, and the multiple singular values ​​of the principal components to be diagnosed are arranged in descending order of value;

[0077] Perform a screening process on the multiple singular values ​​of the principal components to be diagnosed that are arranged in descending order: if there is a DC component in the vibration signal to be diagnosed, discard the singular value ranked first, and obtain multiple singular values ​​of the principal components to be diagnosed after a screening process;

[0078] A secondary screening process is performed on the plurality of singular values ​​of the principal components to be diagnosed after the primary screening: a plurality of singular values ​​of the principal components to be diagnosed whose singular values ​​decrease in pairs are selected from the plurality of singular values ​​of the principal components to be diagnosed after the primary screening, to obtain a plurality of singular values ​​of the principal components to be diagnosed after the secondary screening process; wherein the singular values ​​that decrease in pairs are: the arrangement sequence numbers of every two singular values ​​are continuous and the difference between the two singular values ​​is less than a preset value, and the plurality of pairs of singular values ​​are arranged in descending order;

[0079] Among the plurality of singular values ​​of the principal components to be diagnosed after the secondary screening process, a plurality of singular values ​​of the principal components to be diagnosed that cover a preset proportion of signal energy are found to form a feature vector.

[0080] Step 4 - Normalize the feature vector by zscore and record the normalization parameters.

[0081] Furthermore, the zscore is normalized as follows:

[0082] in σ x std are the characteristic columns σ x The mean and standard deviation of .

[0083] As can be seen from the above, in one embodiment, the rotor crack fault diagnosis method based on singular values ​​and nSVDD may further include: performing zscore normalization on the eigenvector composed of the selected singular values ​​to obtain a normalization parameter so as to optimize the singular value decomposition feature extraction effect under the Hankel matrix;

[0084] Inputting the eigenvector composed of the selected singular values ​​into the rotor crack fault diagnosis model to obtain the rotor crack fault diagnosis result of the current rotating equipment, including: inputting the standardized parameters into the rotor crack fault diagnosis model to obtain the rotor crack fault diagnosis result of the current rotating equipment.

[0085] Step 5 - Filter the rotor crack label case data set, execute steps 2 to 4, input the obtained singular value feature matrix as a training sample into the nSVDD learner (model), perform single fault classification model training, and obtain the rotor crack fault diagnosis model.

[0086] Furthermore, the rotor crack label case dataset described in the embodiments of the present invention refers to manually labeled and verified process data from the development of rotor cracks in rotating equipment (a set of rotor vibration signal samples from the rotating equipment during its historical operation), containing two types of labels: faulty and normal. The faulty labels are the positive class (positive samples) and the normal labels are the negative class (negative samples). The embodiments of the present invention primarily learn from rotor crack case data, so the faulty data are the positive samples. Some normal crack data are added as auxiliary data, i.e., negative samples. This allows full utilization of the labels to determine the support vector, perform correction and optimization on the optimal classification surface, and improve the accuracy of crack fault identification.

[0087] SVDD is a single classification method proposed based on SVM. Its principle is:

[0088] Suppose there is a set of training data sets {x i The purpose of SVDD is to find the i The smallest hypersphere, the smallest hypersphere S has a radius R 2 and the center of the sphere a, continuing the idea of ​​SVM soft interval, introducing the relaxation factor ξ i and a penalty factor C1 to avoid rigid spheres paying a huge penalty for individual samples.

[0089] nSVDD introduces the negative term based on SVDD, fully utilizes the label to determine the support vector, performs correction optimization on the optimal classification surface, and further introduces the negative class relaxation factor ξ l and penalty factor C2, then the hypersphere should satisfy:

[0090] The constraints are:

[0091] In order to solve the above optimization problem, the Lagrange multiplier α is introduced i ,γ i ,α l ,γ l ,≥0, transform the formula into a quadratic optimization problem:

[0092] Define a new variable α i ’ = γ i α i , there is Σ i α′ i = 1, a = Σ i α′ i x i ;

[0093] For the distance d of any sample z from the center of the hypersphere k is:

[0094] The radius R of the hypersphere 2 can be obtained from any α i ’ < C support vectors, and is

[0095] When performing classification discrimination, when d k ≤ R 2 the sample point z is accepted, and when d k > R 2 the sample z is rejected:

[0096] Furthermore, in one embodiment, the nSVDD model in the present invention adopts an RBF (Radial basis function) kernel and is optimized by the SMO (Sequential minimal optimization) algorithm to obtain an optimal classification surface. In one embodiment, the optimization of the kernel parameter g and the penalty factors C1C2 adopts a particle swarm optimization algorithm or a genetic algorithm for tuning to cross-validate the effectiveness of the (nSVDD model).

[0097] Step 6 - Perform steps 2 and 3 on the real-time data of the rotating equipment (data to be diagnosed) to extract features, and standardize it with the parameters obtained in step 4 to obtain the data set to be diagnosed. The data set to be diagnosed is input into the nSVDD model to determine whether the single-frame signal belongs to the positive class. When the positive class judgment occurs five times in a row, it is considered that a rotor crack fault has occurred. The diagnostic conclusion is prompted by the auxiliary alarm system and the data of the rotor crack development process of this section is recorded to facilitate the accumulation of fault cases and iterative update of the model. That is, in one embodiment, the above-mentioned rotor crack fault diagnosis method based on singular values ​​and nSVDD can also include: when the rotor crack fault diagnosis model runs for a preset period of time and a new rotor crack fault diagnosis result type is detected, the new rotor crack fault diagnosis result and its corresponding vibration signal are added to the sample data to obtain updated sample data; the rotor crack fault diagnosis model is retrained using the updated sample data to obtain an updated rotor crack fault diagnosis model, thereby further improving the accuracy of nSVDD model prediction.

[0098] The mainstream conventional technical approach currently used for rotor crack detection involves frequency domain analysis of the rotor vibration signal. Manually observing changes in the frequency components of the rotor's rated frequency harmonics indicates a rotor crack fault. In automated implementation, manual determination of the rotor crack fault monitoring frequency band is performed, followed by the establishment of a mechanism-based monitoring model or a data-driven monitoring model based on the extracted features of the specific frequency components. Due to the complexities of industrial environments and the numerous nonlinearities in the signal, these methods rely heavily on signal analysts visually identifying specific fault frequency components in the spectrum. Manually determined fault monitoring frequency bands are not very reliable for fault detection, making automated frequency analysis difficult to implement, and manual visual inspection followed by alarm generation can lead to significant lags. Rotor cracks or rotor fractures develop rapidly, and any missed detection by manual inspection can cause irreversible damage to the equipment.

[0099] In the embodiment of the present invention, the singular values ​​of the principal components are extracted as features of the fault diagnosis method, and signal restoration is no longer performed. Compared with the existing common spectrum signal extraction, the singular value features are more different when the rotor is cracked and when the rotor is in a normal state, the fault classification is more accurate, the resulting model is less prone to overfitting, and has better generalization performance. The embodiment of the present invention does not perform complex processing such as component signal reconstruction after singular value decomposition, which is relatively simple to implement, and the resulting model has a higher diagnostic accuracy. The single-class nSVDD algorithm used has the advantages of excellent generalization performance, low computational resources, easy model iteration and update, and simple implementation steps in single-class fault identification. Its comprehensive performance in actual industrial applications is better than that of deep learning algorithms.

[0100] Therefore, compared with the prior art, the beneficial effects of the embodiments of the present invention are as follows: The present invention provides a rotor crack fault diagnosis method based on singular values and nSVDD, which can effectively extract the prominent energy components in the rotor vibration signal, characterize them in the form of singular values, and use the improved one-class fault diagnosis learner nSVDD for modeling and learning, with a high specialization recognition rate for rotor crack faults.

[0101] The embodiments of the present invention also provide a method for constructing a rotor crack fault diagnosis model, as described in the following embodiments. Since the principle of solving problems by this construction method is similar to that of the rotor crack fault diagnosis method, the implementation of this construction method can refer to the implementation of the rotor crack fault diagnosis method, and the repeated parts will not be elaborated.

[0102] FIG. 3 is a schematic flowchart of the method for constructing a rotor crack fault diagnosis model in the embodiments of the present invention. As shown in FIG. 3, this construction method includes:

[0103] Step 201: Obtain the historical vibration signal sample data of the rotor during the operation of the rotating equipment;

[0104] Step 202: Perform singular spectrum analysis after singular value decomposition of the historical vibration signal, and extract multiple historical principal component singular values;

[0105] Step 203: Input the obtained historical principal component singular values and their corresponding historical diagnosis results of rotor crack faults as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

[0106] In one embodiment, performing singular spectrum analysis after singular value decomposition of the historical vibration signal to extract multiple historical principal component singular values may include: reconstructing the historical vibration signal in the form of a Hankel matrix, and performing singular spectrum analysis after singular value decomposition to extract multiple historical principal component singular values.

[0107] In one embodiment, for the vibration signal S = [s(1), s(2),..., s(N)] ∈ R N×1 , where N is the signal length, the construction form of the constructed Hankel matrix is:

[0108] where 1 < n < N, n is called the decomposition order, m = N - n + 1, H s ∈ R m×n ;

[0109] After performing singular value decomposition on it and rewriting the column vectors u i and ν i in the represented form, we have:

[0110] H s=USV T =u1σ1ν1 T +u2σ2ν2 T ++…+u i σ i ν i T +…++u r σ r ν r T

[0111] Among them, σ1, σ2, … σ r Decreasing from large to small, called H s Singular values ​​of the matrix, u i is the left singular vector, ν i is a right singular vector.

[0112] In one embodiment, n is taken as the symmetric center so as to achieve the best singular value decomposition feature extraction effect under the Hankel matrix.

[0113] In one embodiment, when the signal length N is an even number, n=N / 2, m=N / 2+1 is taken to construct the Hankel matrix; when the signal length N is an odd number, n=(N+1) / 2, m=(N+1) / 2 is taken to construct the Hankel matrix.

[0114] In one embodiment, the rotor crack fault diagnosis method may further include: selecting singular values ​​from the extracted plurality of historical principal component singular values ​​to form a feature vector;

[0115] The obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results are input as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model, including: inputting the eigenvector composed of the selected singular values ​​and their corresponding historical rotor crack fault diagnosis results as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

[0116] In one embodiment, selecting singular values ​​from the extracted multiple historical principal component singular values ​​to form a feature vector may include:

[0117] Obtain multiple historical principal component singular values; wherein the value of the singular value component is positively correlated with the energy contained in the frequency component it represents, and the multiple historical principal component singular values ​​are arranged in descending order of value;

[0118] Perform a screening process on multiple historical principal component singular values ​​arranged in descending order: if there is a DC component in the historical vibration signal, discard the singular value ranked first, and obtain multiple historical principal component singular values ​​after a screening process;

[0119] Performing a secondary screening process on the multiple historical principal component singular values ​​after the primary screening: selecting multiple historical principal component singular values ​​whose singular values ​​decrease in pairs from the multiple historical principal component singular values ​​after the primary screening, to obtain multiple historical principal component singular values ​​after the secondary screening process; wherein the pairwise decreasing singular values ​​are: the arrangement sequence numbers of every two singular values ​​are continuous and the difference between the two singular values ​​is less than a preset value, and the multiple pairs of singular values ​​are arranged in descending order;

[0120] Among the multiple historical principal component singular values ​​after the secondary screening process, multiple historical principal component singular values ​​covering a preset proportion of signal energy are found to form a feature vector.

[0121] In one embodiment, the rotor crack fault diagnosis method may further include: performing zscore normalization on the eigenvector composed of the selected singular values ​​to obtain a normalized parameter;

[0122] The obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results are input as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model, including: inputting the standardized parameters and their corresponding historical rotor crack fault diagnosis results as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

[0123] In one embodiment, the nSVDD model uses an RBF kernel and is optimized using the SMO algorithm to obtain the optimal classification surface.

[0124] In one embodiment, the kernel parameters and penalty factors of the RBF kernel are optimized using a particle swarm optimization algorithm or a genetic algorithm.

[0125] The present invention also provides a rotor crack fault diagnosis device, as described in the following embodiments. Since the principles of the diagnostic device are similar to those of the rotor crack fault diagnosis method, the implementation of the diagnostic device can refer to the implementation of the rotor crack fault diagnosis method, and the repeated parts will not be repeated.

[0126] FIG4 is a schematic diagram of the structure of a rotor crack fault diagnosis device according to an embodiment of the present invention. As shown in FIG4 , the diagnosis device includes:

[0127] The acquisition unit 011 is used to acquire the vibration signal to be diagnosed of the rotor of the current rotating equipment during operation;

[0128] The first extraction unit 012 is used to perform singular spectrum analysis on the vibration signal to be diagnosed after singular value decomposition, and extract multiple singular values ​​of the principal components to be diagnosed;

[0129] A diagnostic unit 015 is configured to input the singular values of multiple main components to be diagnosed into a rotor crack fault diagnosis nSVDD model, so as to obtain the rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated according to the vibration signal sample data of the rotor during the historical operation of the rotating equipment.

[0130] In one embodiment, the extraction unit is specifically configured to: reconstruct the vibration signal to be diagnosed in the form of a Hankel matrix, and perform singular spectrum analysis after singular value decomposition to extract multiple singular values of the main components to be diagnosed.

[0131] In one embodiment, for the vibration signal S = [s(1), s(2), …, s(N)] ∈ R N×1 , where N is the signal length, the constructed form of the Hankel matrix is:

[0132] where 1 < n < N, n is called the decomposition order, m = N - n + 1, and H s ∈ R m×n ;

[0133] After performing singular value decomposition on it and rewriting the form represented by the column vectors u i and ν i , we have: H s = USV T = u1σ1ν1 T + u2σ2ν2 T + + … + u i σ i ν i T + … + + u r σ r ν r T

[0134] where σ1, σ2, … σ r decrease from large to small, which are called the singular values of the H s matrix, u i is the left singular vector, and ν i is the right singular vector.

[0135] In one embodiment, n takes the symmetric center to make the singular value decomposition feature extraction effect under the Hankel matrix the best.

[0136] In one embodiment, when the signal length N is even, take n = N / 2, m = N / 2 + 1 to construct the Hankel matrix; when the signal length N is odd, take n = (N + 1) / 2, m = (N + 1) / 2 to construct the Hankel matrix.

[0137] In one embodiment, as shown in FIG5 , the rotor crack fault diagnosis device may further include: a first singular value selection unit 013 for selecting singular values ​​from the extracted plurality of principal component singular values ​​to be diagnosed to form a feature vector;

[0138] The diagnosis unit is specifically used to: input the characteristic vector composed of the selected singular values ​​into the rotor crack fault diagnosis model to obtain the rotor crack fault diagnosis result of the current rotating equipment.

[0139] In one embodiment, the first singular value selection unit 013 is specifically configured to:

[0140] Obtaining multiple singular values ​​of the principal components to be diagnosed; wherein the value of the singular value component is positively correlated with the energy contained in the frequency component it represents, and the multiple singular values ​​of the principal components to be diagnosed are arranged in descending order of value;

[0141] Perform a screening process on the multiple singular values ​​of the principal components to be diagnosed that are arranged in descending order: if there is a DC component in the vibration signal to be diagnosed, discard the singular value ranked first, and obtain multiple singular values ​​of the principal components to be diagnosed after a screening process;

[0142] A secondary screening process is performed on the plurality of singular values ​​of the principal components to be diagnosed after the primary screening: a plurality of singular values ​​of the principal components to be diagnosed whose singular values ​​decrease in pairs are selected from the plurality of singular values ​​of the principal components to be diagnosed after the primary screening, to obtain a plurality of singular values ​​of the principal components to be diagnosed after the secondary screening process; wherein the singular values ​​that decrease in pairs are: the arrangement sequence numbers of every two singular values ​​are continuous and the difference between the two singular values ​​is less than a preset value, and the plurality of pairs of singular values ​​are arranged in descending order;

[0143] Among the plurality of singular values ​​of the principal components to be diagnosed after the secondary screening process, a plurality of singular values ​​of the principal components to be diagnosed that cover a preset proportion of signal energy are found to form a feature vector.

[0144] In one embodiment, as shown in FIG5 , the rotor crack fault diagnosis device may further include: a first normalization processing unit 014 for performing zscore (standard score) normalization on the eigenvector composed of the selected singular values ​​to obtain a normalized parameter;

[0145] The diagnosis unit is specifically used to: input the standardized parameters into a rotor crack fault diagnosis model to obtain a rotor crack fault diagnosis result of the current rotating equipment.

[0146] In one embodiment, the nSVDD model may use an RBF kernel and may be optimized using an SMO algorithm to obtain an optimal classification surface.

[0147] In one embodiment, the optimization of the kernel parameters and penalty factors of the RBF kernel may be performed using a particle swarm optimization algorithm or a genetic algorithm to cross-validate their effectiveness.

[0148] In one embodiment, the rotor crack fault diagnosis device based on singular values ​​and nSVDD may further include:

[0149] an adding unit, configured to add a new rotor crack fault diagnosis result and its corresponding vibration signal to the sample data when a new rotor crack fault diagnosis result type is detected during a preset period of operation of the rotor crack fault diagnosis model, thereby obtaining updated sample data;

[0150] An updating unit is used to retrain the rotor crack fault diagnosis model using updated sample data to obtain an updated rotor crack fault diagnosis model.

[0151] The present invention also provides a device for constructing a rotor crack fault diagnosis model, as described in the following embodiments. Because the principles underlying the device are similar to those of the rotor crack fault diagnosis method, the implementation of the device can be referenced to the implementation of the rotor crack fault diagnosis method, and any repetitions will not be repeated.

[0152] FIG6 is a schematic diagram of the structure of a device for constructing a rotor crack fault diagnosis model according to an embodiment of the present invention. As shown in FIG6 , the device includes:

[0153] An acquisition unit 021 is used to acquire historical vibration signal sample data of the rotor of the rotating equipment during operation;

[0154] The second extraction unit 022 is used to perform singular spectrum analysis on the historical vibration signal after singular value decomposition, and extract multiple historical principal component singular values;

[0155] The training unit 025 is used to input the obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

[0156] In one embodiment, the second extraction unit is specifically configured to reconstruct the historical vibration signal in the form of a Hankel matrix, and perform singular spectrum analysis after singular value decomposition to extract a plurality of historical principal component singular values.

[0157] In one embodiment, for a vibration signal S=[s(1),s(2),…,s(N)]∈R N×1 , N is the signal length, and the constructed Hankel matrix is ​​constructed as follows:

[0158] Where 1 < n < N, n is called the decomposition order, m = N - n + 1, H s ∈R m×n ;

[0159] After performing singular value decomposition on it and rewriting the column vectors u i and ν i in the form of representation, we have: H s = USV T = u1σ1ν1 T + u2σ2ν2 T +... + u i σ i ν i T +... + + u r σ r ν r T

[0160] Where, σ1, σ2,... σ r decrease from large to small, and are called the singular values of the H s matrix, u i is the left singular vector, and ν i is the right singular vector.

[0161] In one embodiment, n takes the symmetric center to make the singular value decomposition feature extraction effect of the Hankel matrix the best.

[0162] In one embodiment, when the signal length N is even, take n = N / 2, m = N / 2 + 1 to construct the Hankel matrix; when the signal length N is odd, take n = (N + 1) / 2, m = (N + 1) / 2 to construct the Hankel matrix.

[0163] In one embodiment, as shown in Figure

[0163] , the above rotor crack fault diagnosis model construction device may further include: a second singular value selection unit 023, configured to select singular values from the extracted multiple historical principal component singular values to form a feature vector;

[0164] The training unit is specifically configured to: use the feature vector composed of the selected singular values and its corresponding rotor crack fault historical diagnosis results as training samples to input into the nSVDD learning model for model training to obtain the rotor crack fault diagnosis model.

[0165] In one embodiment, the second singular value selection unit is specifically configured to:

[0166] Obtain multiple historical principal component singular values; where, the value of the singular value component is positively correlated with the energy contained in the frequency component it represents, and the multiple historical principal component singular values are arranged in descending order of numerical value;

[0167] Perform a screening process on multiple historical principal component singular values ​​arranged in descending order: if there is a DC component in the historical vibration signal, discard the singular value ranked first, and obtain multiple historical principal component singular values ​​after a screening process;

[0168] Performing a secondary screening process on the multiple historical principal component singular values ​​after the primary screening: selecting multiple historical principal component singular values ​​whose singular values ​​decrease in pairs from the multiple historical principal component singular values ​​after the primary screening, to obtain multiple historical principal component singular values ​​after the secondary screening process; wherein the pairwise decreasing singular values ​​are: the arrangement sequence numbers of every two singular values ​​are continuous and the difference between the two singular values ​​is less than a preset value, and the multiple pairs of singular values ​​are arranged in descending order;

[0169] Among the multiple historical principal component singular values ​​after the secondary screening process, multiple historical principal component singular values ​​covering a preset proportion of signal energy are found to form a feature vector.

[0170] In one embodiment, as shown in FIG7 , the apparatus for constructing the rotor crack fault diagnosis model may further include: a second normalization processing unit 024 for performing zscore normalization on the eigenvector composed of the selected singular values ​​to obtain a normalized parameter;

[0171] The training unit is specifically used to: input the standardized parameters and their corresponding rotor crack fault historical diagnosis results as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

[0172] In one embodiment, the nSVDD model uses an RBF kernel and is optimized using the SMO algorithm to obtain the optimal classification surface.

[0173] In one embodiment, the kernel parameters and penalty factors of the RBF kernel are optimized using a particle swarm optimization algorithm or a genetic algorithm.

[0174] Based on the aforementioned inventive concept, as shown in FIG8 , the present invention further proposes a computer device 500, comprising a memory 510, a processor 520, and a computer program 530 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 530, the aforementioned rotor crack fault diagnosis method and / or a method for constructing a rotor crack fault diagnosis model are implemented.

[0175] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned rotor crack fault diagnosis method and / or the method for constructing a rotor crack fault diagnosis model.

[0176] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the above-mentioned rotor crack fault diagnosis method and / or the method for constructing a rotor crack fault diagnosis model.

[0177] Compared with the prior art, the rotor crack fault diagnosis solution provided by the embodiment of the present invention has the following beneficial technical effects:

[0178] First, the present invention uses the singular values ​​of the principal components as features for fault diagnosis, eliminating the need for signal restoration. Compared to conventional spectrum signal extraction, the singular value features show greater differences between rotor cracks and normal rotors, leading to more accurate fault classification. The resulting model is less prone to overfitting and exhibits superior generalization performance.

[0179] Secondly, the single-classification nSVDD algorithm used in the embodiment of the present invention has the advantages of excellent generalization performance in single-class fault identification, low computational resources, easy model iteration and update, and simple implementation steps. Its comprehensive performance in practical applications in the industrial field is better than that of deep learning algorithms.

[0180] In summary, the rotor crack fault diagnosis scheme based on singular values ​​and nSVDD provided by the embodiment of the present invention can effectively extract the prominent energy components in the rotor vibration signal, characterize them in the form of singular values, and use the improved single-class fault diagnosis learner nSVDD for modeling and learning. The specialized recognition rate of rotor crack faults is high, and rotor crack fault diagnosis based on singular values ​​and nSVDD can be realized, thereby improving the accuracy of rotor crack fault diagnosis.

[0181] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0182] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0183] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0184] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0185] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A rotor crack fault diagnosis method, characterized in that: include: Collect the vibration signals to be diagnosed of the rotor of the current rotating equipment during operation; After the vibration signal to be diagnosed is subjected to singular value decomposition, a singular spectrum analysis is performed to extract a plurality of singular values ​​of the principal components to be diagnosed; A plurality of principal component singular values ​​to be diagnosed are input into a rotor crack fault diagnosis nSVDD model to obtain a rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on rotor vibration signal sample data of the rotating equipment during historical operation.

2. The method according to claim 1, characterized in that The current vibration signal to be diagnosed is subjected to singular value decomposition and then singular spectrum analysis is performed to extract multiple singular values ​​of the principal components to be diagnosed, including: reconstructing the vibration signal to be diagnosed in the form of a Hankel matrix, and performing singular spectrum analysis after singular value decomposition to extract multiple singular values ​​of the principal components to be diagnosed.

3. The method according to claim 2, characterized in that For the vibration signal S = [s(1), s(2), ..., s(N)] ∈ R N× 1 , N is the signal length, and the constructed Hankel matrix is ​​in the form of: where 1 < n < N, n is called the decomposition order, m = N - n + 1, H s ∈ R m×n ; After performing singular value decomposition, rewrite the column vector u i and ν i The forms of expression are: H s =USV T =u1σ1ν1 T +u2σ2ν2 T ++…+u i s i n i T +…++u r s r n r T Among them, σ1, σ2, …σ r Decreasing from large to small, called H s Singular values ​​of the matrix, u i is the left singular vector, ν i are right singular vectors.

4. The method according to claim 3, characterized in that n takes the symmetric center to achieve the best singular value decomposition feature extraction effect under the Hankel matrix.

5. The method according to claim 3, characterized in that When the signal length N is an even number, n=N / 2, m=N / 2+1 is taken to construct the Hankel matrix; when the signal length N is an odd number, n=(N+1) / 2, m=(N+1) / 2 is taken to construct the Hankel matrix.

6. The method according to claim 2, characterized in that Also includes: Selecting singular values ​​from the extracted plurality of singular values ​​of the principal components to be diagnosed to form a feature vector; Inputting multiple principal component singular values ​​to be diagnosed into a rotor crack fault diagnosis model to obtain a rotor crack fault diagnosis result of the current rotating equipment, including: inputting a eigenvector composed of selected singular values ​​into the rotor crack fault diagnosis model to obtain a rotor crack fault diagnosis result of the current rotating equipment.

7. The method according to claim 6, characterized in that Select singular values ​​from the extracted multiple principal component singular values ​​to be diagnosed to form a feature vector, including: Acquire multiple singular values ​​of the principal components to be diagnosed; wherein the value of the singular value component is positively correlated with the energy contained in the frequency component it represents, and the multiple singular values ​​of the principal components to be diagnosed are arranged in order from large to small; Perform a screening process on the plurality of singular values ​​of the principal components to be diagnosed arranged in descending order: if there is a DC component in the vibration signal to be diagnosed, discard the singular value ranked first, and obtain a plurality of singular values ​​of the principal components to be diagnosed after a screening process; A secondary screening process is performed on the plurality of singular values ​​of the principal components to be diagnosed after the primary screening: a plurality of singular values ​​of the principal components to be diagnosed whose singular values ​​are descending in pairs are selected from the plurality of singular values ​​of the principal components to be diagnosed after the primary screening, so as to obtain a plurality of singular values ​​of the principal components to be diagnosed after the secondary screening process; wherein the singular values ​​that are descending in pairs are: the arrangement sequence numbers of every two singular values ​​are continuous and the difference between the two singular values ​​is less than a preset value, and the plurality of pairs of singular values ​​are arranged in descending order; Among the plurality of singular values ​​of the principal components to be diagnosed after the secondary screening process, a plurality of singular values ​​of the principal components to be diagnosed that cover a preset proportion of signal energy are found to form a characteristic vector.

8. The method according to claim 6, characterized in that Also includes: The eigenvector composed of the selected singular values ​​is zscore normalized to obtain the normalization parameter; Inputting the eigenvector composed of the selected singular values ​​into the rotor crack fault diagnosis model to obtain the rotor crack fault diagnosis result of the current rotating equipment, including: inputting the standardized parameters into the rotor crack fault diagnosis model to obtain the rotor crack fault diagnosis result of the current rotating equipment.

9. The method according to claim 1, characterized in that The nSVDD model uses the RBF kernel and the SMO algorithm is used to optimize the optimal classification surface.

10. The method according to claim 9, characterized in that The optimization of the kernel parameters and penalty factors of the RBF kernel is adjusted by particle swarm optimization algorithm or genetic algorithm.

11. The method according to claim 1, characterized in that Also includes: When a new rotor crack fault diagnosis result type is detected during the operation of the rotor crack fault diagnosis model for a preset period of time, the new rotor crack fault diagnosis result and its corresponding vibration signal are added to the sample data to obtain updated sample data; The rotor crack fault diagnosis model is retrained using the updated sample data to obtain an updated rotor crack fault diagnosis model.

12. A method for constructing a rotor crack fault diagnosis model, characterized in that: include: Obtain historical vibration signal sample data of the rotor of the rotating equipment during operation; After the historical vibration signal is subjected to singular value decomposition, a singular spectrum analysis is performed to extract multiple historical principal component singular values; The obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results are used as training samples to input into the nSVDD learning model for model training to obtain the rotor crack fault diagnosis model.

13. The method according to claim 12, characterized in that The historical vibration signal is subjected to singular value decomposition and then singular spectrum analysis is performed to extract multiple historical principal component singular values, including: reconstructing the historical vibration signal in the form of a Hankel matrix, and performing singular spectrum analysis after singular value decomposition to extract multiple historical principal component singular values.

14. The method according to claim 13, characterized in that For the vibration signal S = [s(1), s(2), ..., s(N)] ∈ R N×1 , N is the signal length, and the constructed Hankel matrix is ​​in the form of: where 1 < n < N, n is called the decomposition order, m = N - n + 1, H s ∈R m×n ; After performing singular value decomposition, rewrite the column vector u i and ν i The forms of expression are: H s =USV T =u1σ1ν1 T +u2σ2ν2 T ++…+u i s i n i T +…++u r s r n r T Among them, σ1, σ2, …σ r Decreasing from large to small, called H s Singular values ​​of the matrix, u i is the left singular vector, ν i are right singular vectors.

15. The method according to claim 14, characterized in that n takes the symmetric center to achieve the best singular value decomposition feature extraction effect under the Hankel matrix.

16. The method according to claim 14, characterized in that When the signal length N is an even number, n=N / 2, m=N / 2+1 is taken to construct the Hankel matrix; when the signal length N is an odd number, n=(N+1) / 2, m=(N+1) / 2 is taken to construct the Hankel matrix.

17. The method according to claim 13, characterized in that Also includes: Select singular values ​​from the extracted multiple historical principal component singular values ​​to form a feature vector; The historical principal component singular values ​​and their corresponding rotor crack fault historical diagnosis results are used as training samples to input into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model, including: The characteristic vector composed of outliers and its corresponding historical diagnosis results of rotor crack fault are used as training samples to input into the nSVDD learning model for model training to obtain the rotor crack fault diagnosis model.

18. The method according to claim 17, characterized in that Select singular values ​​from the extracted multiple historical principal component singular values ​​to form a feature vector, including: Acquire multiple historical principal component singular values; wherein the value of the singular value component is positively correlated with the energy contained in the frequency component it represents, and the multiple historical principal component singular values ​​are arranged in descending order of value; Perform a screening process on multiple historical principal component singular values ​​arranged in descending order: if there is a DC component in the historical vibration signal, discard the singular value ranked first, and obtain multiple historical principal component singular values ​​after a screening process; A secondary screening process is performed on the multiple historical principal component singular values ​​after the primary screening: multiple historical principal component singular values ​​whose singular values ​​decrease in pairs are selected from the multiple historical principal component singular values ​​after the primary screening, so as to obtain multiple historical principal component singular values ​​after the secondary screening process; wherein the singular values ​​that decrease in pairs are: the arrangement sequence numbers of every two singular values ​​are continuous and the difference between the two singular values ​​is less than a preset value, and the multiple pairs of singular values ​​are arranged in descending order; Among the multiple historical principal component singular values ​​after the secondary screening process, multiple historical principal component singular values ​​covering a preset proportion of signal energy are found to form a feature vector.

19. The method according to claim 17, characterized in that Also includes: The eigenvector composed of the selected singular values ​​is zscore normalized to obtain the normalization parameter; The obtained historical principal component singular values ​​and their corresponding historical diagnosis results of rotor crack faults are input as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model, including: inputting the standardized parameters and their corresponding historical diagnosis results of rotor crack faults as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

20. The method of claim 12, wherein: The nSVDD model uses the RBF kernel and the SMO algorithm is used to optimize the optimal classification surface.

21. The method of claim 20, wherein: The optimization of the kernel parameters and penalty factors of the RBF kernel is adjusted by particle swarm optimization algorithm or genetic algorithm.

22. A rotor crack fault diagnosis device, characterized in that: include: A collection unit, used to collect vibration signals to be diagnosed from the rotor of the current rotating equipment during operation; A first extraction unit is used to perform singular spectrum analysis on the vibration signal to be diagnosed after singular value decomposition, and extract a plurality of singular values ​​of the principal components to be diagnosed; The diagnosis unit is used to input multiple principal component singular values ​​to be diagnosed into a rotor crack fault diagnosis nSVDD model to obtain a rotor crack fault diagnosis result of the current rotating equipment; the rotor crack fault diagnosis nSVDD model is pre-trained and generated based on vibration signal sample data of the rotor of the rotating equipment during historical operation.

23. A device for constructing a rotor crack fault diagnosis model, characterized in that: include: An acquisition unit, used to acquire historical vibration signal sample data of the rotor of the rotating equipment during operation; The second extraction unit is used to perform singular spectrum analysis on the historical vibration signal after singular value decomposition, and extract multiple historical principal component singular values; The training unit is used to input the obtained historical principal component singular values ​​and their corresponding historical rotor crack fault diagnosis results as training samples into the nSVDD learning model for model training to obtain a rotor crack fault diagnosis model.

24. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the rotor crack fault diagnosis method described in any one of claims 1 to 11, and / or the rotor crack fault diagnosis model construction method described in any one of claims 12-21.

25. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the rotor crack fault diagnosis method described in any one of claims 1 to 11, and / or the rotor crack fault diagnosis model construction method described in any one of claims 12-21.

26. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the rotor crack fault diagnosis method described in any one of claims 1 to 11, and / or the method for constructing a rotor crack fault diagnosis model described in any one of claims 12-21.

Images