Intelligent fault diagnosis method and system for rotating machinery based on feature selection
By combining deep neural networks and manual feature selection methods, the optimal feature subset is selected, which solves the problem of relying on expert experience and the lack of interpretability of deep learning in the fault diagnosis of rotating machinery, and realizes accurate diagnosis of rotating machinery faults.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2024-04-29
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies rely on expert experience in the diagnosis of rotating machinery faults, which is difficult to adapt to the needs of large-scale industries. Furthermore, the diagnostic results of deep learning methods are difficult to interpret, affecting the judgment of equipment users.
By combining deep features extracted from deep neural networks and handcrafted features, the optimal feature subset is selected through feature selection methods and input into the fault diagnosis classification model to provide accurate fault diagnosis results.
It enables accurate diagnosis of rotating machinery faults. The deep features have good generalization and spatial discriminability, while the handcrafted features provide interpretable information, thus improving the reliability and accuracy of the diagnosis.
Smart Images
Figure CN118312832B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method and system for intelligent fault diagnosis of rotating machinery based on feature selection. Background Technology
[0002] Rotating machinery, as a widely used mechanical component, is prevalent in robots, medical equipment, ships, automobiles, and other mechanical equipment. Its health status affects the overall reliability of the equipment. With the development of industrial big data, sufficient information data can be collected from monitoring sensors. With the rapid development of data-driven technology, how to use this data for effective and reliable diagnosis and reduce the probability of dangerous accidents is of great research significance.
[0003] After prolonged operation, rotating machinery experiences continuous wear and tear on components such as bearings and gears, leading to a decline in their lifespan and reliability. Ignoring the real-time operating status of the equipment can result in incalculable losses. In practice, bearings in mechanical equipment generally have the highest probability of failure. Therefore, reliable and accurate monitoring and timely maintenance of rotating machinery prone to failure are of great significance to human production and daily life.
[0004] Traditionally, monitoring of rotating machinery has relied heavily on expert systems, but these methods are overly dependent on expert experience and ill-suited to the demands of modern large-scale industrial fault diagnosis. While deep learning methods have achieved some success in rotating machinery fault diagnosis in recent years, common challenges remain that require further investigation. As a black-box model, deep learning methods often produce diagnostic results that are difficult to interpret. Furthermore, modern industrial equipment is becoming increasingly complex, sophisticated, and highly integrated; unreliable diagnoses can interfere with user judgment. Therefore, it is essential to explore safer, more reliable fault diagnosis methods that can provide accurate results. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent fault diagnosis of rotating machinery based on feature selection;
[0006] On the one hand, a feature-selection-based intelligent fault diagnosis method for rotating machinery is provided, including:
[0007] Obtain the operating data of the rotating machinery to be diagnosed;
[0008] Based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features.
[0009] The operating data of the rotating machinery to be diagnosed is input into the trained deep neural network to obtain several deep features;
[0010] All handcrafted features and all deep features are merged to obtain an initial feature set; the initial feature set is then optimized and filtered to select the optimal feature subset.
[0011] The process of selecting the optimal feature subset includes: an initial population, where each individual in the population represents a selected feature subset; if a set number of cycles is reached, offspring individuals are directly generated; if the set number of cycles is not reached, half of the elite individuals are selected from the entire population to generate offspring individuals, and individuals are selected for the next generation, and finally the selected population is output, which includes the optimal feature subset.
[0012] The optimal feature subset is input into the trained fault diagnosis classification model, and the fault diagnosis result is output.
[0013] On the other hand, a feature-selective intelligent fault diagnosis system for rotating machinery is provided, including:
[0014] The acquisition module is configured to acquire the operating data of the rotating machinery to be diagnosed.
[0015] The manual feature extraction module is configured to extract several manual features based on the operating data of the rotating machinery to be diagnosed. The manual features include: time-domain features, frequency-domain features, and time-frequency-domain features.
[0016] The deep feature extraction module is configured to input the operating data of the rotating machinery to be diagnosed into the trained deep neural network to obtain several deep features.
[0017] The feature selection module is configured to: merge all handcrafted features and all deep features to obtain an initial feature set; optimize and filter the initial feature set to select the optimal feature subset;
[0018] The process of selecting the optimal feature subset includes: an initial population, where each individual in the population represents a selected feature subset; if a set number of cycles is reached, offspring individuals are directly generated; if the set number of cycles is not reached, half of the elite individuals are selected from the entire population to generate offspring individuals, and individuals are selected for the next generation, and finally the selected population is output, which includes the optimal feature subset.
[0019] The classification module is configured to input the optimal feature subset into the trained fault diagnosis classification model and output the fault diagnosis result.
[0020] Furthermore, an electronic device is also provided, including:
[0021] Memory, used for non-transitory storage of computer-readable instructions; and
[0022] Processor, for executing the computer-readable instructions,
[0023] When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.
[0024] In another aspect, a storage medium is also provided for non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the instructions of the method described in the first aspect are executed.
[0025] In another aspect, a computer program product is also provided, including a computer program that, when run on one or more processors, is used to implement the method described in the first aspect above.
[0026] The above technical solution has the following advantages or beneficial effects:
[0027] Deep neural networks possess powerful feature learning capabilities, and the extracted deep features typically exhibit good generalization and spatial discriminability. Features constructed based on signal analysis methods have clear and explicit mathematical descriptions, providing more interpretable information for fault diagnosis results. This invention combines deep features and handcrafted features, and employs a feature selection method to select the most critical features, achieving accurate diagnosis of rotating machinery faults. Attached Figure Description
[0028] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0029] Figure 1 This is a flowchart of the feature selection method in Example 1;
[0030] Figure 2 This is an illustration of the individual vector representation method in Example 1;
[0031] Figure 3(a) shows the visualization results of the original features of Example 1 on the PT500 dataset;
[0032] Figure 3(b) shows the visualization results of the original features of Example 1 on a real dataset;
[0033] Figure 4(a) shows the visualization results of the artificial features in Example 1 on the PT500 dataset;
[0034] Figure 4(b) shows the visualization results of the artificial features in Example 1 on a real dataset;
[0035] Figure 5(a) shows the visualization results of the depth features in Example 1 on the PT500 dataset;
[0036] Figure 5(b) shows the visualization results of the deep features in Example 1 on a real dataset;
[0037] Figure 6(a) shows the visualization results of the fusion features in Example 1 on the PT500 dataset;
[0038] Figure 6(b) shows the visualization results of the fusion features in Example 1 on a real dataset;
[0039] Figure 7(a) shows the visualization results of the 16-dimensional feature subset after feature selection in Example 1 on the PT500 dataset;
[0040] Figure 7(b) shows the visualization results of the 8-dimensional feature subset after feature selection in the PT500 dataset in Example 1;
[0041] Figure 7(c) shows the visualization results of the 12-dimensional feature subset after feature selection in Example 1 on the PT500 dataset;
[0042] Figure 7(d) shows the visualization results of the 9-dimensional feature subset after feature selection in Example 1 on the PT500 dataset;
[0043] Figure 7(e) shows the visualization results of the 13-dimensional feature subset after feature selection in Example 1 on the PT500 dataset;
[0044] Figure 7(f) shows the visualization results of the 14-dimensional feature subset after feature selection in Example 1 on the PT500 dataset;
[0045] Figure 8(a) shows the visualization results of the 7-dimensional feature subset after feature selection in the real dataset in Example 1;
[0046] Figure 8(b) shows the visualization results of the 8-dimensional feature subset after feature selection in the real dataset in Example 1;
[0047] Figure 8(c) shows the visualization results of the 6-dimensional feature subset after feature selection in the real dataset of Example 1;
[0048] Figure 8(d) shows the visualization results of the 8-dimensional feature subset after feature selection in the real dataset of Example 1.
[0049] Figure 9 This is an example diagram of the rule model in Implementation Example 1. Detailed Implementation
[0050] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0051] Example 1
[0052] This embodiment provides a feature-selection-based intelligent fault diagnosis method for rotating machinery;
[0053] Intelligent fault diagnosis methods for rotating machinery based on feature selection include:
[0054] S101: Obtain the operating data of the rotating machinery to be diagnosed;
[0055] S102: Based on the operating data of the rotating machinery to be diagnosed, extract several manual features, including: time-domain features, frequency-domain features, and time-frequency-domain features;
[0056] S103: Input the operating data of the rotating machinery to be diagnosed into the trained deep neural network to obtain several deep features;
[0057] S104: Merge all handcrafted features and all deep features to obtain an initial feature set; optimize and filter the initial feature set to select the optimal feature subset;
[0058] The process of selecting the optimal feature subset includes: an initial population, where each individual in the population represents a selected feature subset; if a set number of cycles is reached, offspring individuals are directly generated; if the set number of cycles is not reached, half of the elite individuals are selected from the entire population to generate offspring individuals, and individuals are selected for the next generation, and finally the selected population is output, which includes the optimal feature subset.
[0059] S105: Input the optimal feature subset into the trained fault diagnosis classification model and output the fault diagnosis result.
[0060] Further, S101: acquiring the operating data of the rotating machinery to be diagnosed includes: collecting the operating data of the rotating machinery using sensors, wherein the operating data includes: current signal, voltage signal, power, sound signal, vibration signal, etc.
[0061] Further, in step S102: based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features;
[0062] The time-domain features specifically include: maximum value, minimum value, peak-to-peak value, average value, standard deviation, average rate of change, kurtosis, skewness, root mean square, waveform factor, peak factor, impulse factor, and signal margin.
[0063] Furthermore, the time-domain features specifically include:
[0064] TF1=max|x(n)|; TF2=min|x(n)|; TF3=TF1-TF2;
[0065]
[0066] TF7=kurtosis(x(n)); TF8=skewness(x(n));
[0067] TF 10 =TF9 / |TF4|;TF 11 =TF3 / TF9; TF 12 =TF3 / |TF4|;
[0068] Among them, TF1~TF 13 The values in the following order represent the maximum value, minimum value, peak-to-peak value, average value, standard deviation, average rate of change, kurtosis, skewness, root mean square, waveform factor, peak factor, impulse factor, and signal margin, respectively. x(n) represents the signal sample data, max and min represent the maximum and minimum values in the sample, K represents the sample sequence length, kurtosis represents the kurtosis, skewness represents the skewness, and Δt represents the sequence time.
[0069] Further, in step S102: based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features;
[0070] The frequency domain features include: performing a fast Fourier transform on the original vibration signal to convert the time domain signal into a frequency domain signal, and then extracting the maximum value, minimum value, peak-to-peak value, average value, standard deviation, and average rate of change from the frequency domain signal.
[0071] The frequency domain features include:
[0072] TF 14 =max|X(k)|;TF 15 =min|X(k)|;TF 16 =TF 14 -TF 15 ;
[0073]
[0074] Among them, TF 14 TF represents the maximum value. 15 TF represents the minimum value. 16 TF represents peak-to-peak value. 17 TF represents the average value. 18 TF represents the standard deviation. 19 Let X(k) represent the average rate of change, X(k) represent the signal sample data, max|X(k)| and min|X(k)| represent the maximum and minimum values in the sample, and K represent the sample sequence length. Δk represents the sequence time.
[0075] For example, the step of performing a fast Fourier transform on the original vibration signal specifically includes:
[0076]
[0077]
[0078] Further, in step S102: based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features;
[0079] Specifically, the time-frequency domain features include: using a three-layer wavelet packet transform to perform in-depth analysis of high-frequency and low-frequency signals, mining hidden features in the data, and extracting the energy spectrum of wavelet packet nodes as time-frequency domain features. The energy calculation formula for the nodes after wavelet packet decomposition is as follows:
[0080]
[0081] In the formula, M represents the length of the j-th frequency band. This represents the nth coefficient corresponding to node (i,k) after wavelet packet decomposition.
[0082] In the time-frequency domain, after the original vibration signal is decomposed into three layers of wavelet packets, the ratio of the energy of each wavelet packet in the third layer to the total energy is used as the time-frequency domain feature of the signal, and a total of 8-dimensional features are extracted.
[0083] TF 20 =E1 / E total ;TF 21 =E2 / E total ;TF 22 =E3 / E total ;TF 23 =E4 / E total ;
[0084] TF 24 =E5 / E total;TF 25 =E6 / E total ;TF 26 =E7 / E total ;TF 27 =E8 / E total .
[0085] Among them, TF 20 This represents the energy E1 of the first wavelet packet and the total energy E. total The ratio; TF 21 This represents the energy E1 of the first wavelet packet and the total energy E. total The ratio; TF 22 This represents the energy E1 of the second wavelet packet and the total energy E. total The ratio; TF 23 This represents the energy E1 of the third wavelet packet and the total energy E. total The ratio; TF 24 This represents the energy E1 of the fourth wavelet packet and the total energy E. total The ratio; TF 25 This represents the energy E1 of the fifth wavelet packet and the total energy E. total The ratio; TF 26 This represents the energy E1 of the sixth wavelet packet and the total energy E. total The ratio; TF 27 This represents the energy E1 of the seventh wavelet packet and the total energy E. total The ratio of .
[0086] Further, in step S103: the operating data of the rotating machinery to be diagnosed is input into the trained deep neural network to obtain several deep features; wherein, the trained deep neural network includes:
[0087] The first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the third pooling layer, the first fully connected layer, and the second fully connected layer are connected in sequence. The features output by the first fully connected layer are selected as the deep features.
[0088] For example, the WDCNN architecture is adopted, and the network structure is shown in Table 1. It consists of three convolutional layers and pooling layers, and finally outputs the result through a fully connected layer. In this invention, after the deep neural network is trained and converged, the features after the first fully connected layer are used as deep features.
[0089] Table 1. Network Structure of WDCNN
[0090]
[0091] Furthermore, S104: Optimize and filter the initial feature set to select the optimal feature subset, and use an improved manifold-assisted multimodal multi-objective algorithm to remove redundant features from the initial feature set.
[0092] Furthermore, such as Figure 1 As shown, a multimodal, multi-objective algorithm based on an improved manifold-assisted method is used to remove redundant features from the initial feature set. Specifically, this includes:
[0093] S104-1: Initial population P;
[0094] S104-2: Determine if the number of cycles has been reached. If yes, proceed to S104-3; otherwise, proceed to S104-4.
[0095] S104-3: Generate offspring, proceed to S104-5;
[0096] S104-4: Select half of the elite individuals from the entire population to generate offspring, which then enter S104-5;
[0097] S104-5: Select individuals for the next generation, proceed to S104-6;
[0098] S104-6: Determine if the final condition has been met. If yes, output the final population P. If no, return to S104-2.
[0099] Further, S104-1: Initial population P includes: each individual in the population represents a selected subset of features;
[0100] S104-11: Optimization function setting. In the feature selection problem, the optimization objective is as follows:
[0101] Minimize F(x)=(f1(x),f2(x)) (4)
[0102] Where f1(x) represents the classification error rate and f2(x) represents the number of features selected.
[0103] S104-12: Individual vector representation method setting: There are multiple individuals, each represented by a vector whose dimension is equal to the length of the constructed feature.
[0104] like Figure 2 As shown, a threshold variable is set to 0.5 when evaluating individuals. When the population searches in the discrete space, if the value of a certain dimension in the vector representing an individual exceeds the threshold, it indicates that the corresponding feature has been selected.
[0105] S104-13: Set the initial parameters of the optimization algorithm: population size N, maximum number of iterations G, sampling period T, number of clusters N C;
[0106] S104-14: Initialize the position and velocity of all particles, using the following initialization formula:
[0107] X = X min +(X max -X min )*rand (5)
[0108] Where X represents the current individual in the population, and rand represents a random number between [0,1].
[0109] Furthermore, S104-3: Generating offspring, specifically including:
[0110] S104-31: Based on the principal component analysis algorithm, the population is clustered into N. C kind;
[0111] S104-32: Based on the clustering results, construct a rule model for each cluster. The rule model is considered as a probability-fitted distribution of the population samples (blue dots) in the selected region, such as... Figure 9 As shown, the red line represents the rule model of the regional population sample.
[0112] S104-33: Obtain offspring by uniformly sampling from a regular model and adding noise to the uniform sampling results.
[0113] Table 2. Pseudocode for manifold-assisted strategies
[0114]
[0115] After the algorithm enters the manifold modeling stage, it performs local PCA clustering based on the population distribution, then establishes multiple rule models based on the divided clusters, and finally obtains offspring by uniformly sampling from the rule models and adding noise.
[0116] Furthermore, S104-4: Selecting half of the elite individuals from the entire population to generate offspring specifically includes: selecting elite individuals from the population:
[0117] S104-41: Classify the entire population based on non-dominant ordering;
[0118] S104-42: Calculate the special crowding distance, sort individuals according to the special crowding distance from largest to smallest, and select the individuals that make up half of the population size at the top of the sort as elite individuals;
[0119] Calculation of special congestion distance:
[0120]
[0121] in, This represents the specific crowding distance for individual i. This represents the crowding distance in the decision space of individual i. d represents the target spatial crowding distance of individual i. x,avg and d f,avg represents the average crowding distance in the target space and the decision space, respectively.
[0122] S104-43: Elite individuals produce offspring through differential evolution operators, the formula of which is shown below:
[0123]
[0124] Among them, v i It is a difference vector. For individuals randomly selected from the population, F is the proportion factor.
[0125] Table 3. Pseudocode of the pre-selection strategy
[0126]
[0127] Further, in step S105: the optimal feature subset is input into the trained fault diagnosis classification model, and the fault diagnosis result is output. The fault diagnosis classification model is implemented using a support vector machine. The trained fault diagnosis classification model is obtained by training the support vector machine based on the training set, and the training set consists of signal features of known fault diagnosis classification results.
[0128] To verify the practical performance of the proposed method in fault diagnosis, experiments were conducted using the PT500 fault diagnosis platform and a bearing fault diagnosis platform developed by a certain company. To provide intuitive feature quality, the experiments recorded the t-SNE visualization results of the original data, manual features, deep features, fused features, and each feature subset after filtering.
[0129] Figures 3(a) to 3(b) The t-SNE visualization results of the original data samples are shown. In both datasets, the features of faulty and normal samples are mixed together, making it impossible to clearly distinguish the classification boundaries between faulty and normal states. Feature extraction is needed to increase the distinguishability of each category.
[0130] Visualization results of artificially extracted features from the two datasets are as follows: Figures 4(a) to 4(b)As shown, compared to the original data features, manually extracted features significantly improve spatial separability, and each dimension of the feature has a clear and explicit mathematical description, exhibiting excellent interpretability. However, some aliasing exists in the extracted manually extracted features. Preliminary manually extracted features alone cannot directly obtain highly discriminative features for each category. In the PT500 dataset, there is some aliasing in the data features for normal state, fault 1, and fault 5. In the real dataset, there is slight aliasing in the data for normal state, fault 4, and fault 3.
[0131] Figures 5(a) to 5(b) The results of deep feature extraction using WDCNN are presented for various datasets. In the PT500 dataset, except for slight aliasing between normal samples and fault 1, the remaining samples exhibit good spatial distribution. In the real dataset, except for a small number of fault 4 samples whose extracted features are aliased into the normal state, the features in other states show good spatial separability. This indicates that deep neural networks can provide well-separable features for fault diagnosis.
[0132] Figures 6(a) to 6(b) The visualization results of the fused features are shown. It can be seen that after feature fusion, the fused features inherit the good separability of deep features and retain the interpretable information of artificial features. However, the initially fused features often have redundant information and further feature selection is needed to obtain a higher quality feature subset.
[0133] After feature selection, fused features can remove redundant feature information and improve the overall feature quality. Figures 7(a) to 7(f) Table 5 shows the feature visualization results and classification accuracy of the feature subsets obtained after feature selection on the PT500 dataset. Experimental results show that even after removing 72.9% of the features through feature selection, accurate diagnostic results can still be maintained, and each category of features has good discriminative power in the spatial domain. Furthermore, artificial features are also retained in the feature subsets after feature selection, indicating that deep features and artificial features can complement each other, providing highly separable and interpretable features for fault diagnosis.
[0134] Table 5 shows the feature subsets obtained from the PT500 dataset.
[0135]
[0136] Figures 8(a) to 8(d) Visualization results of feature subsets after feature selection on a real dataset.
[0137] Example 2
[0138] This embodiment provides a feature-selection-based intelligent fault diagnosis system for rotating machinery;
[0139] A feature-selection-based intelligent fault diagnosis system for rotating machinery includes:
[0140] The acquisition module is configured to acquire the operating data of the rotating machinery to be diagnosed.
[0141] The manual feature extraction module is configured to extract several manual features based on the operating data of the rotating machinery to be diagnosed. The manual features include: time-domain features, frequency-domain features, and time-frequency-domain features.
[0142] The deep feature extraction module is configured to input the operating data of the rotating machinery to be diagnosed into the trained deep neural network to obtain several deep features.
[0143] The feature selection module is configured to: merge all handcrafted features and all deep features to obtain an initial feature set; optimize and filter the initial feature set to select the optimal feature subset;
[0144] The process of selecting the optimal feature subset includes: an initial population, where each individual in the population represents a selected feature subset; if a set number of cycles is reached, offspring individuals are directly generated; if the set number of cycles is not reached, half of the elite individuals are selected from the entire population to generate offspring individuals, and individuals are selected for the next generation, and finally the selected population is output, which includes the optimal feature subset.
[0145] The classification module is configured to input the optimal feature subset into the trained fault diagnosis classification model and output the fault diagnosis result.
[0146] It should be noted that the acquisition module, manual feature extraction module, deep feature extraction module, feature selection module, and classification module described above correspond to steps S101 to S105 in Embodiment 1. The examples and application scenarios implemented by these modules and their corresponding steps are the same, but they are not limited to the content disclosed in Embodiment 1. It should also be noted that these modules, as part of the system, can be executed in a computer system, such as a set of computer-executable instructions.
[0147] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0148] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.
[0149] Example 3
[0150] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.
[0151] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.
[0152] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.
[0153] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.
[0154] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.
[0155] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.
[0156] Example 4
[0157] This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.
[0158] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A feature-selection-based intelligent fault diagnosis method for rotating machinery, characterized by: include: Obtain the operating data of the rotating machinery to be diagnosed; Based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features. The operating data of the rotating machinery to be diagnosed is input into the trained deep neural network to obtain several deep features; All handcrafted features and all deep features are merged to obtain an initial feature set; the initial feature set is then optimized and filtered to select the optimal feature subset. The process of selecting the optimal feature subset includes: an initial population, where each individual in the population represents a selected feature subset; if a set number of cycles is reached, offspring individuals are directly generated; if the set number of cycles is not reached, half of the elite individuals are selected from the entire population to generate offspring individuals, and individuals are selected for the next generation, and finally the selected population is output, which includes the optimal feature subset. Directly generating offspring includes: Based on the principal component analysis algorithm, the population is clustered into NC classes; Based on the clustering results, construct a probability fitting distribution for each class; Offspring are obtained by uniformly sampling from a probability-fitted distribution and adding noise; Selecting half of the elite individuals from the entire population to generate offspring specifically includes: selecting elite individuals from the population: The entire population is classified according to the non-dominated ordering; Calculate the special crowding distance, sort individuals according to the special crowding distance from largest to smallest, and select the individuals that make up half of the population size at the top of the sort as elite individuals; Calculation of special congestion distance: in, Represents an individual i Special crowded distance, Represents an individual i The decision-making space is crowded and the distance is short. Represents an individual i The target space is crowded. and These represent the average crowding distances in the target space and the decision space, respectively. Elite individuals produce offspring using a differential evolution operator, the formula of which is shown below: in, It is a difference vector. ~ Individuals randomly selected from the population. F It is a scaling factor; The optimal feature subset is input into the trained fault diagnosis classification model, and the fault diagnosis result is output.
2. The intelligent fault diagnosis method for rotating machinery based on feature selection as described in claim 1, characterized in that, Based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features. The time-domain features specifically include: maximum value, minimum value, peak-to-peak value, average value, standard deviation, average rate of change, kurtosis, skewness, root mean square, waveform factor, peak factor, impulse factor, and signal margin. The time-domain features specifically include: ； ； ； ； ； ； ； ； ； ； ； ； ； in, ~ The values, in order, represent the maximum value, minimum value, peak-to-peak value, average value, standard deviation, average rate of change, kurtosis, skewness, root mean square, waveform factor, peak factor, impulse factor, and signal margin. Represents signal sample data, and This indicates taking the maximum and minimum values from the sample, where K represents the sample sequence length. Indicates kurtosis. Indicates skewness, Represents the time sequence.
3. The intelligent fault diagnosis method for rotating machinery based on feature selection as described in claim 1, characterized in that, Based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features. The frequency domain features include: performing a fast Fourier transform on the original vibration signal to convert the time domain signal into a frequency domain signal, and then extracting the maximum value, minimum value, peak-to-peak value, average value, standard deviation, and average rate of change from the frequency domain signal. The frequency domain features include: ； ； ； ； ； ； in, Indicates the maximum value. This represents the minimum value. Indicates peak-to-peak value. This represents the average value. Indicates standard deviation, Indicates the average rate of change. Represents signal sample data, and This indicates taking the maximum and minimum values in the sample, where K represents the sample sequence length; Represents the time sequence.
4. The intelligent fault diagnosis method for rotating machinery based on feature selection as described in claim 1, characterized in that, Based on the operating data of the rotating machinery to be diagnosed, several manual features are extracted, including: time-domain features, frequency-domain features, and time-frequency-domain features. Specifically, the time-frequency domain features include: using a three-layer wavelet packet transform to perform in-depth analysis of high-frequency and low-frequency signals, mining hidden features in the data, and extracting the energy spectrum of wavelet packet nodes as time-frequency domain features. The energy calculation formula for the nodes after wavelet packet decomposition is as follows: In the formula, M Indicates the first j Each frequency band length, Represents the nodes after wavelet packet decomposition ( i , k The corresponding number n One coefficient; In the time-frequency domain, after the original vibration signal is decomposed into three layers of wavelet packets, the ratio of the energy of each wavelet packet in the third layer to the total energy is used as the time-frequency domain feature of the signal, and a total of 8-dimensional features are extracted. ； ； ； ； ； ； ； ； in, This represents the energy of the first wavelet packet. With total energy The ratio; This represents the energy of the first wavelet packet. With total energy The ratio; This represents the energy of the second wavelet packet. With total energy The ratio; This represents the energy of the third wavelet packet. With total energy The ratio; This represents the energy of the fourth wavelet packet. With total energy The ratio; This represents the energy of the fifth wavelet packet. With total energy The ratio; This represents the energy of the sixth wavelet packet. With total energy The ratio; This represents the energy of the seventh wavelet packet. With total energy The ratio of .
5. A feature-selective intelligent fault diagnosis system for rotating machinery, characterized in that, include: The acquisition module is configured to acquire the operating data of the rotating machinery to be diagnosed. The manual feature extraction module is configured to extract several manual features based on the operating data of the rotating machinery to be diagnosed. The manual features include: time-domain features, frequency-domain features, and time-frequency-domain features. The deep feature extraction module is configured to input the operating data of the rotating machinery to be diagnosed into the trained deep neural network to obtain several deep features. The feature selection module is configured to: merge all handcrafted features and all deep features to obtain an initial feature set; optimize and filter the initial feature set to select the optimal feature subset; The process of selecting the optimal feature subset includes: an initial population, where each individual in the population represents a selected feature subset; if a set number of cycles is reached, offspring individuals are directly generated; if the set number of cycles is not reached, half of the elite individuals are selected from the entire population to generate offspring individuals, and individuals are selected for the next generation, and finally the selected population is output, which includes the optimal feature subset. Directly generating offspring includes: Based on the principal component analysis algorithm, the population is clustered into NC classes; Based on the clustering results, construct a probability fitting distribution for each class; Offspring are obtained by uniformly sampling from a probability-fitted distribution and adding noise; Selecting half of the elite individuals from the entire population to generate offspring specifically includes: selecting elite individuals from the population: The entire population is classified according to the non-dominated ordering; Calculate the special crowding distance, sort individuals according to the special crowding distance from largest to smallest, and select the individuals that make up half of the population size at the top of the sort as elite individuals; Calculation of special congestion distance: in, Represents an individual i Special crowded distance, Represents an individual i The decision-making space is crowded and the distance is short. Represents an individual i The target space is crowded. and These represent the average crowding distances in the target space and the decision space, respectively. Elite individuals produce offspring using a differential evolution operator, the formula of which is shown below: in, It is a difference vector. ~ Individuals randomly selected from the population. F It is a scaling factor; The classification module is configured to input the optimal feature subset into the trained fault diagnosis classification model and output the fault diagnosis result.
6. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. as well as Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-4.
7. A storage medium characterized by being non-transitory. The system stores computer-readable instructions, wherein, when the non-transitory computer-readable instructions are executed by a computer, the instructions of the method according to any one of claims 1-4 are executed.
8. A computer program product, characterized in that, Includes a computer program, which, when run on one or more processors, is used to implement the method described in any one of claims 1-4.