Method for diagnosing a fault of a rotating machine and device therefor

Abstract

The application provides a fault diagnosis method and device of a rotating machinery equipment, and relates to the field of fault diagnosis. Each original vibration time domain signal of the rotating machinery equipment collected by a plurality of sensors is decomposed to obtain a plurality of candidate intrinsic mode function (IMF) components obtained after decomposition of each original vibration time domain signal. For any original vibration time domain signal, a plurality of target IMF components are determined from the plurality of candidate IMF components corresponding to the original vibration time domain signal, and a reconstructed vibration time domain signal of the original vibration time domain signal is obtained through signal reconstruction based on the target IMF components. A symmetric polar coordinate (SDP) image of the scatter point fusion of all original vibration time domain signals and the respective reconstructed vibration time domain signals corresponding to the original vibration time domain signals is obtained based on the SDP method. The SDP image is input into a target convolutional neural network to obtain a fault classification output by the target convolutional neural network. The application selects reasonable target IMF components for signal reconstruction, avoids the participation of false components in signal reconstruction, eliminates background noise, and fully extracts equipment state features.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis, and in particular to a fault diagnosis method and apparatus for rotating machinery. Background Technology

[0002] Rotating machinery, as a crucial component, is widely used in modern industrial equipment. Its harsh working environment often leads to malfunctions. In actual production, a large number of equipment failures are related to faults in rotating machinery such as bearings, gears, and rotors, significantly reducing the stability and safety of equipment operation. Furthermore, due to the complex noise environment, the impact components characterizing fault information are easily drowned out, affecting the accuracy of fault diagnosis. Therefore, achieving accurate and rapid fault diagnosis of rotating machinery in complex environments is of paramount importance.

[0003] In related technologies, many fault diagnosis techniques for rotating machinery often use information from a single sensor as the data source, ignoring the benefits of multi-sensor information in ensuring data complementarity in the fault identification process. Furthermore, many signal feature extraction methods cannot separate and capture fault features under strong noise and multi-source interference in the field. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art.

[0005] Therefore, one objective of this application is to propose a fault diagnosis method for rotating machinery. This method involves acquiring raw vibration time-domain signals of the rotating machinery from multiple sensors, and decomposing each raw vibration time-domain signal to obtain multiple candidate intrinsic mode functions (IMF) components. For any given raw vibration time-domain signal, multiple target IMF components are determined from the multiple candidate IMF components corresponding to that raw vibration time-domain signal for signal reconstruction. Signal reconstruction is then performed based on the target IMF components to obtain the reconstructed vibration time-domain signal corresponding to the raw vibration time-domain signal. Using the Symmetric Dot Pattern (SDP) method, all raw vibration time-domain signals and their corresponding reconstructed vibration time-domain signals are transformed into scatter points in polar coordinates to obtain a scatter-point fused SDP image. The SDP image is then input into a trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network.

[0006] The second objective of this application is to provide a fault diagnosis device for rotating machinery.

[0007] The third objective of this application is to propose an electronic device.

[0008] The fourth objective of this application is to provide a non-transitory computer-readable storage medium.

[0009] The fifth objective of this application is to provide a computer program product.

[0010] To achieve the above objectives, the first aspect of this application proposes a fault diagnosis method for rotating machinery, comprising: acquiring original vibration time-domain signals of the rotating machinery collected by multiple sensors, and decomposing each original vibration time-domain signal to obtain multiple candidate intrinsic mode factor (IMF) components obtained after decomposition of each original vibration time-domain signal; for any original vibration time-domain signal, determining multiple target IMF components for signal reconstruction from the multiple candidate IMF components corresponding to the original vibration time-domain signal, and reconstructing the signal based on the target IMF components to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal; based on the symmetric polar coordinate SDP method, converting all original vibration time-domain signals and their respective corresponding reconstructed vibration time-domain signals into scatter points in polar coordinates to obtain a scatter point fused SDP image; inputting the SDP image into a trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network.

[0011] According to one embodiment of this application, for any original vibration time-domain signal, determining multiple target IMF components for reconstructing the signal from multiple candidate IMF components corresponding to the original vibration time-domain signal includes: for any original vibration time-domain signal, obtaining the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal and the correlation coefficient between each candidate IMF component corresponding to the original vibration time-domain signal and the original vibration time-domain signal; filtering all candidate IMF components corresponding to the original vibration time-domain signal according to the kurtosis index and the correlation coefficient to obtain the filtered target IMF components.

[0012] According to one embodiment of this application, all candidate IMF components corresponding to the original vibration time-domain signal are screened based on kurtosis index and correlation coefficient to obtain the screened target IMF component, including: obtaining a preset kurtosis threshold and correlation coefficient threshold; for any original vibration time-domain signal, the candidate IMF component that simultaneously satisfies the kurtosis index being less than the kurtosis threshold and the correlation coefficient being greater than the correlation coefficient threshold is taken as the target IMF component corresponding to the original vibration time-domain signal.

[0013] According to one embodiment of this application, a training method for a target convolutional neural network includes: acquiring multiple sampled SDP images based on sampled original vibration time-domain signals of a rotating mechanical device collected by multiple sampling sensors; and iteratively training an initial convolutional neural network based on the sampled SDP images and preset fault labels of the rotating mechanical device corresponding to the sampled SDP images to obtain a target convolutional neural network generated after training.

[0014] According to one embodiment of this application, multiple sampled SDP images are obtained based on the sampled original vibration time-domain signals of a rotating mechanical device collected by multiple sampling sensors. The process includes: acquiring the sampled original vibration time-domain signals of the rotating mechanical device collected by multiple sampling sensors; resampling the sampled original vibration time-domain signals at different time lengths with a set window size to obtain a resampled signal; for any resampled signal, determining multiple target resampled IMF components from multiple candidate resampled IMF components corresponding to the resampled signal for signal reconstruction, and reconstructing the signal based on the target resampled IMF components to obtain a resampled reconstructed signal corresponding to the resampled signal; grouping the resampled signal and the resampled reconstructed signal accordingly, and based on the SDP method, converting the resampled signal in each group and its corresponding resampled reconstructed signal into scatter points in polar coordinates to obtain a sampled SDP image of scatter point fusion corresponding to each group.

[0015] To achieve the above objectives, a second aspect of this application proposes a fault diagnosis device for rotating machinery, comprising: an acquisition module for acquiring original vibration time-domain signals of the rotating machinery collected by multiple sensors, and decomposing each original vibration time-domain signal to obtain multiple candidate intrinsic mode factor (IMF) components obtained after decomposition of each original vibration time-domain signal; a reconstruction module for determining multiple target IMF components for reconstructing the signal from the multiple candidate IMF components corresponding to any original vibration time-domain signal, and reconstructing the signal based on the target IMF components to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal; a transformation module for converting all original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals into scatter points in polar coordinates based on the symmetric polar coordinate SDP method to obtain a scatter point fused SDP image; and a classification module for inputting the SDP image into a trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network.

[0016] According to one embodiment of this application, the reconstruction module is further configured to: for any original vibration time-domain signal, obtain the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal, and the correlation coefficient between each candidate IMF component corresponding to the original vibration time-domain signal and the original vibration time-domain signal; and filter all candidate IMF components corresponding to the original vibration time-domain signal according to the kurtosis index and the correlation coefficient to obtain the filtered target IMF component.

[0017] According to one embodiment of this application, the reconstruction module is further configured to: obtain a preset kurtosis threshold and a correlation coefficient threshold; for any original vibration time-domain signal, for all candidate IMF components, the candidate IMF component that simultaneously satisfies the kurtosis index being less than the kurtosis threshold and the correlation coefficient being greater than the correlation coefficient threshold is used as the target IMF component corresponding to the original vibration time-domain signal.

[0018] According to one embodiment of this application, the device further includes a model training module, which is used to acquire multiple sampled SDP images based on the sampled original vibration time-domain signals of the sampled rotating machinery collected by multiple sampling sensors, and to iteratively train an initial convolutional neural network based on the sampled SDP images and the preset fault labels of the rotating machinery corresponding to the sampled SDP images, so as to obtain the target convolutional neural network generated after training.

[0019] According to one embodiment of this application, the model training module is further configured to: acquire the original vibration time-domain signal of the sampled rotating machinery collected by multiple sampling sensors; resample the original vibration time-domain signal at different time lengths with a set window size to obtain the resampled signal; for any resampled signal, determine multiple target resampled IMF components for reconstructing the signal from multiple candidate resampled IMF components corresponding to the resampled signal, and reconstruct the signal based on the target resampled IMF components to obtain the resampled reconstructed signal corresponding to the resampled signal; group the resampled signal and the resampled reconstructed signal accordingly, and based on the SDP method, convert the resampled signal in each group and its corresponding resampled reconstructed signal into scatter points in polar coordinates to obtain the sampled SDP image of scatter point fusion corresponding to each group.

[0020] To achieve the above objectives, a third aspect of this application provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to implement the fault diagnosis method for rotating machinery as described in the first aspect of this application.

[0021] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the fault diagnosis method for rotating machinery as described in the first aspect of this application.

[0022] To achieve the above objectives, a fifth aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the fault diagnosis method for rotating machinery as described in the first aspect of this application.

[0023] This application achieves at least the following beneficial effects: This application selects reasonable target IMF components for signal reconstruction, avoids pseudo-components from participating in signal reconstruction, eliminates background noise, fully extracts equipment status features, and finally obtains a single SDP image that simultaneously encompasses the original vibration information and feature extraction information of multiple sensors, thus containing rich equipment operating status features, highlighting the subtle differences between different faults. The method has low complexity but strong effectiveness and robustness, and has good application prospects. Attached Figure Description

[0024] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0025] Figure 1 This is a schematic diagram illustrating an exemplary implementation of a fault diagnosis method for rotating machinery, as shown in one embodiment of this application.

[0026] Figure 2 This is a schematic diagram illustrating the principle of an SDP method according to one embodiment of this application.

[0027] Figure 3 This is a schematic diagram of an SDP image formed by scatter fusion, as shown in one embodiment of this application.

[0028] Figure 4 This is a schematic diagram illustrating resampling with different time lengths in one embodiment of this application.

[0029] Figure 5 This is a schematic diagram illustrating an exemplary implementation of a fault diagnosis method for rotating machinery, as shown in one embodiment of this application.

[0030] Figure 6 This is a general concept diagram illustrating a fault diagnosis method for rotating machinery equipment according to one embodiment of this application.

[0031] Figure 7 This is a schematic diagram of a fault diagnosis device for rotating machinery, as shown in one embodiment of this application.

[0032] Figure 8 This is a schematic diagram of an electronic device according to one embodiment of this application. Detailed Implementation

[0033] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0034] Figure 1 This is a schematic diagram of an exemplary embodiment of a fault diagnosis method for rotating machinery equipment shown in this application, as follows: Figure 1 As shown, the fault diagnosis method for this rotating machinery includes the following steps:

[0035] S101: Acquire the original vibration time-domain signals of rotating machinery collected by multiple sensors, and decompose each original vibration time-domain signal to obtain multiple candidate intrinsic mode (IMF) components after the decomposition of each original vibration time-domain signal.

[0036] The original vibration time-domain signal of rotating machinery is acquired using multiple sensors.

[0037] The Complementary Ensemble Empirical Mode Decomposition (CEEMD) method is used to decompose each original vibration time-domain signal to obtain multiple candidate intrinsic mode functions (IMF) components after the decomposition of each original vibration time-domain signal.

[0038] S102, for any original vibration time-domain signal, determine multiple target IMF components for reconstructing the signal from multiple candidate IMF components corresponding to the original vibration time-domain signal, and reconstruct the signal based on the target IMF components to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal.

[0039] Traditional CEEMD decomposition algorithms, after decomposing the original vibration time-domain signal, do not allow all candidate IMF components to characterize the vibration features of the original signal. Some candidate IMF components contain "pseudo-components" with low correlation to the original vibration time-domain signal, affecting diagnostic accuracy. To address this, this application identifies multiple target IMF components from all candidate IMF components corresponding to each original vibration time-domain signal for reconstructing the original signal. Signal reconstruction is then performed to obtain the reconstructed vibration time-domain signal for each original signal. This avoids pseudo-components from participating in signal reconstruction, eliminates background noise, and fully extracts equipment state features. While completely preserving the original vibration time-domain signal, it significantly reduces the interference of noise in the signal on fault analysis.

[0040] It is easy to understand that each original vibration time-domain signal corresponds to a set of target IMF components, that is, each original vibration time-domain signal corresponds to a reconstructed vibration time-domain signal.

[0041] S103, based on the symmetric polar coordinate SDP method, transforms all original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals into scattered points in polar coordinates to obtain a scattered point fused SDP image.

[0042] Based on the Symmetrized Dot Pattern (SDP) method, all original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals are transformed into scatter points in polar coordinates. Then, the scatter plots of these polar coordinate scatter points are fused to generate a single SDP image containing multi-source information, highlighting the differences between different faults.

[0043] Figure 2 This is a schematic diagram illustrating the principle of an SDP method shown in this application, such as... Figure 2 As shown, in the SDP method, the formulas corresponding to the parameters of each scatter point are as follows:

[0044]

[0045]

[0046]

[0047] In the above formula: (i) is the polar coordinate radius of the scatter point; θ(i) and These represent the angles by which the scattered points rotate about the mirror symmetry plane in the counterclockwise and clockwise directions, respectively; the maximum and minimum values ​​of the waveform amplitude of the original signal correspond to x and x, respectively. max and x min ; rθ is the rotation angle of the specified mirror symmetry plane (θ = 360 / n, m = 1, 2, ..., n, n is the number of mirror symmetry planes); ζ is the gain coefficient (ζ < ).

[0048] Figure 3 This is a schematic diagram of an SDP image formed by scatter fusion as shown in this application, such as... Figure 3 As shown, this application uses the SDP imaging method to effectively fuse the original vibration time-domain signals and reconstructed vibration time-domain signals from multiple sensors simultaneously, achieving multi-source information complementarity. This allows a single SDP image to encompass both the original vibration information and feature extraction information from multiple sensors, thereby containing rich equipment operating status characteristics, highlighting subtle differences between different faults, and improving the distinguishability of fault features.

[0049] S104: Input the SDP image into the trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network.

[0050] The SDP image is input into a trained target convolutional neural network (CNN), and the SDP image features are extracted layer by layer by the target convolutional neural network to obtain the fault classification of rotating machinery output by the target convolutional neural network.

[0051] This application proposes a fault diagnosis method for rotating machinery. The method involves acquiring raw vibration time-domain signals of the rotating machinery from multiple sensors, and decomposing each raw vibration time-domain signal to obtain multiple candidate intrinsic mode factor (IMF) components. For any given raw vibration time-domain signal, multiple target IMF components are determined from the multiple candidate IMF components for signal reconstruction. Signal reconstruction is then performed based on the target IMF components to obtain the reconstructed vibration time-domain signal corresponding to the raw vibration time-domain signal. Using a symmetric polar coordinate (SDP) method, all raw vibration time-domain signals and their corresponding reconstructed vibration time-domain signals are converted into scatter points in polar coordinates to obtain a scatter-point fused SDP image. The SDP image is then input into a trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network.

[0052] This application selects appropriate target IMF components for signal reconstruction, avoiding the participation of spurious components, eliminating background noise, and fully extracting equipment status features. The resulting single SDP image simultaneously encompasses the raw vibration information and feature extraction information from multiple sensors, thus containing rich equipment operating status features and highlighting subtle differences between different faults. The method has low complexity but strong effectiveness and robustness, showing promising application prospects. Furthermore, this application utilizes the powerful feature dimensionality reduction and extraction learning of the target convolutional neural network to achieve rapid and automated recognition, avoiding subjective interference from manual recognition and truly achieving intelligent processing. This effectively diagnoses on-site equipment faults, truly improving efficiency and reducing costs.

[0053] Furthermore, in this application, the target convolutional neural network needs to be pre-trained. The training method for the target convolutional neural network includes the following steps:

[0054] Multiple sampled SDP images are obtained from the original vibration time-domain signals of the rotating machinery collected by multiple sampling sensors. Similar to the above process, specifically, the original vibration time-domain signals of the rotating machinery collected by multiple sampling sensors are acquired, and the original vibration time-domain signals are resampled at different time lengths with a set window size to obtain resampled signals. For any resampled signal, multiple target resampled IMF components are determined from multiple candidate resampled IMF components corresponding to the resampled signal for signal reconstruction. Signal reconstruction is performed based on the target resampled IMF components to obtain the resampled reconstructed signal corresponding to the resampled signal. The resampled signals and resampled reconstructed signals are grouped accordingly, and based on the SDP method, the resampled signals and their corresponding resampled reconstructed signals within each group are converted into scatter points in polar coordinates to obtain a sampled SDP image of scatter point fusion for each group. Figure 4 This application illustrates a schematic diagram of resampling with different time lengths, as shown below. Figure 4 As shown, a fixed window size of 1024 is set, and the original vibration time-domain signal is resampled with window movement steps of 256, 512, and 768. The data after multi-scale resampling are combined to obtain a multi-scale resampled signal with a single learning sample length of 1024.

[0055] After acquiring a sampled SDP image, the initial convolutional neural network is iteratively trained based on the sampled SDP image and the corresponding preset rotating machinery equipment fault labels to obtain the target convolutional neural network generated after training.

[0056] The convolutional neural network training method disclosed in this application ensures that the subsequent data structure is the same by setting a fixed window, and achieves multi-scale overlapping data resampling by setting different window movement steps, multi-scale cropping of the original data and expansion of the dataset, thereby avoiding the uniformity of data patterns and effectively suppressing the risk of overfitting in CNN network learning.

[0057] Optionally, nonlinear data dimensionality reduction methods (t-distributed stochastic neighbor embedding, T-SNE) clustering methods and confusion matrices can be used as evaluations of the classification performance of the target convolutional neural network model.

[0058] Figure 5 This is a schematic diagram of an exemplary embodiment of a fault diagnosis method for rotating machinery equipment shown in this application, as follows: Figure 5 As shown, the fault diagnosis method for this rotating machinery includes the following steps:

[0059] S501 acquires the original vibration time-domain signals of rotating machinery collected by multiple sensors, and decomposes each original vibration time-domain signal to obtain multiple candidate intrinsic mode (IMF) components after the decomposition of each original vibration time-domain signal.

[0060] The original vibration time-domain signals of rotating machinery are collected by multiple sensors, and each original vibration time-domain signal is decomposed using the CEEMD method to obtain multiple candidate IMF components after the decomposition of each original vibration time-domain signal.

[0061] S502, for any original vibration time-domain signal, obtain the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal, and the correlation coefficient between each candidate IMF component corresponding to the original vibration time-domain signal and the original vibration time-domain signal.

[0062] Traditional CEEMD decomposition algorithms, after decomposing the original vibration time-domain signal, do not allow all candidate IMF components to characterize the vibration features of the original signal. Some candidate IMF components contain "pseudo-components" with low correlation to the original vibration time-domain signal, affecting diagnostic accuracy. To address this, this application introduces kurtosis criteria and correlation coefficients to determine multiple target IMF components from all candidate IMF components corresponding to each original vibration time-domain signal for reconstructing the original signal. This signal reconstruction obtains the reconstructed vibration time-domain signal for each original signal, avoiding pseudo-components from participating in the reconstruction, eliminating background noise, and fully extracting equipment state characteristics. While completely preserving the original vibration time-domain signal, this significantly reduces the interference of noise in the signal on fault analysis.

[0063] Specifically, for any original vibration time-domain signal, the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal is obtained. The formula for calculating the kurtosis index is as follows:

[0064]

[0065] In the above formula, K represents the kurtosis index of a candidate IMF component, E(x-μ). 4 Let σ represent the fourth-order mathematical expectation of a candidate IMF component, let μ represent the standard deviation of a candidate IMF component, let x represent the mean of a candidate IMF component, and let x represent the signal of a candidate IMF component.

[0066] Furthermore, the correlation coefficient method is used to calculate the correlation coefficient between each candidate IMF component corresponding to the original vibration time-domain signal and the original vibration time-domain signal.

[0067] S503, based on the kurtosis index and correlation coefficient, all candidate IMF components corresponding to the original vibration time-domain signal are screened to obtain the screened target IMF component.

[0068] Obtain the preset kurtosis threshold and correlation coefficient threshold. For any original vibration time-domain signal, select the candidate IMF component that simultaneously satisfies the kurtosis index being less than the kurtosis threshold and the correlation coefficient being greater than the correlation coefficient threshold as the target IMF component corresponding to the original vibration time-domain signal.

[0069] S504, based on the target IMF component, the signal is reconstructed to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal.

[0070] For each original vibration time-domain signal, the corresponding target IMF component is reconstructed to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal.

[0071] It is easy to understand that each original vibration time-domain signal corresponds to a set of target IMF components, that is, each original vibration time-domain signal corresponds to a reconstructed vibration time-domain signal.

[0072] S505, based on the symmetric polar coordinate SDP method, transforms all original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals into scattered points in polar coordinates to obtain a fused SDP image.

[0073] S506, Input the SDP image into the trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network.

[0074] For details on the specific implementation of steps S505 to S506, please refer to the relevant parts of the above embodiments, which will not be repeated here.

[0075] This application's embodiments select reasonable target IMF components for signal reconstruction, avoid pseudo-components from participating in signal reconstruction, eliminate background noise, and fully extract equipment status features. The final single SDP image simultaneously encompasses the original vibration information and feature extraction information from multiple sensors, thus containing rich equipment operating status features and highlighting subtle differences between different faults. The method has low complexity but strong effectiveness and robustness, and has good application prospects.

[0076] Figure 6 This is a general conceptual diagram of a fault diagnosis method for rotating machinery equipment shown in this application, as follows: Figure 6 As shown, a fixed window size of 1024 is set, and the original vibration time-domain signal is resampled with window movement steps of 256, 512, and 768. The data after multi-scale resampling are combined to obtain a multi-scale resampled signal with a single learning sample length of 1024. Multiple target resampled IMF components for signal reconstruction are determined from multiple candidate resampled IMF components corresponding to this resampled signal. Signal reconstruction is performed based on the target resampled IMF components to obtain the resampled reconstructed signal corresponding to the resampled signal. The resampled signal and the resampled reconstructed signal are grouped accordingly, and based on the SDP method, the resampled signal in each group and its corresponding resampled reconstructed signal are converted into scatter points in polar coordinates to obtain the scatter point fusion sampling SDP image corresponding to each group. After obtaining the sampling SDP image, the initial convolutional neural network is iteratively trained based on the sampling SDP image and the preset rotating machinery equipment fault labels corresponding to the sampling SDP image to obtain the target convolutional neural network generated after training.

[0077] In the actual use of rotating machinery, raw vibration time-domain signals are collected from multiple sensors. The CEEMD method is used to decompose each raw vibration time-domain signal, obtaining multiple candidate IMF components. For any given raw vibration time-domain signal, the kurtosis index of each candidate IMF component and its correlation coefficient with the original vibration time-domain signal are obtained. Preset kurtosis and correlation coefficient thresholds are acquired. For any given raw vibration time-domain signal, the candidate IMF component that simultaneously satisfies both a kurtosis index less than the kurtosis threshold and a correlation coefficient greater than the correlation coefficient threshold is selected as the target IMF component. For each raw vibration time-domain signal, its corresponding target IMF component is reconstructed to obtain the reconstructed vibration time-domain signal. Based on the symmetric polar coordinate SDP method, all raw vibration time-domain signals and their corresponding reconstructed vibration time-domain signals are converted into scatter points in polar coordinates to obtain a scatter-point fused SDP image. The SDP image is input into the trained target convolutional neural network to obtain the fault classification of rotating machinery output by the target convolutional neural network.

[0078] Figure 7 This is a schematic diagram of a fault diagnosis device for rotating machinery shown in this application, such as... Figure 7 As shown, the fault diagnosis device 700 for rotating machinery includes an acquisition module 701, a reconstruction module 702, a conversion module 703, and a classification module 704, wherein:

[0079] The acquisition module 701 is used to acquire the original vibration time-domain signals of rotating machinery collected by multiple sensors, and to decompose each original vibration time-domain signal to obtain multiple candidate intrinsic mode (IMF) components after the decomposition of each original vibration time-domain signal.

[0080] The reconstruction module 702 is used to determine multiple target IMF components for reconstructing the signal from multiple candidate IMF components corresponding to any original vibration time-domain signal, and to reconstruct the signal based on the target IMF components to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal.

[0081] The conversion module 703 is used to convert all the original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals into scattered points in polar coordinates based on the symmetric polar coordinate SDP method, so as to obtain the scattered point fused SDP image.

[0082] The classification module 704 is used to input the SDP image into the trained target convolutional neural network to obtain the fault classification of the rotating machinery equipment output by the target convolutional neural network.

[0083] In this device, reasonable target IMF components are selected for signal reconstruction to avoid pseudo-components from participating in signal reconstruction, eliminate background noise, and fully extract equipment status features. The final single SDP image simultaneously encompasses the original vibration information and feature extraction information from multiple sensors, thus containing rich equipment operating status features and highlighting the subtle differences between different faults. The method has low complexity but strong effectiveness and robustness, and has good application prospects.

[0084] According to one embodiment of this application, the reconstruction module 702 is further configured to: for any original vibration time-domain signal, obtain the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal, and the correlation coefficient between each candidate IMF component corresponding to the original vibration time-domain signal and the original vibration time-domain signal; and filter all candidate IMF components corresponding to the original vibration time-domain signal according to the kurtosis index and the correlation coefficient to obtain the filtered target IMF component.

[0085] According to one embodiment of this application, the reconstruction module 702 is further configured to: obtain a preset kurtosis threshold and a correlation coefficient threshold; for any original vibration time-domain signal, for all candidate IMF components, the candidate IMF component that simultaneously satisfies the kurtosis index being less than the kurtosis threshold and the correlation coefficient being greater than the correlation coefficient threshold is used as the target IMF component corresponding to the original vibration time-domain signal.

[0086] According to one embodiment of this application, the fault diagnosis device 700 for rotating machinery further includes a model training module 705, which is used to acquire multiple sampled SDP images based on the sampled original vibration time-domain signals of the rotating machinery collected by multiple sampling sensors, and to iteratively train an initial convolutional neural network based on the sampled SDP images and the preset rotating machinery fault labels corresponding to the sampled SDP images, so as to obtain the target convolutional neural network generated after training.

[0087] According to one embodiment of this application, the model training module 705 is further configured to: acquire the original vibration time-domain signal of the sampled rotating machinery collected by multiple sampling sensors; resample the original vibration time-domain signal at different time lengths with a set window size to obtain the resampled signal; for any resampled signal, determine multiple target resampled IMF components for reconstructing the signal from multiple candidate resampled IMF components corresponding to the resampled signal, and reconstruct the signal based on the target resampled IMF components to obtain the resampled reconstructed signal corresponding to the resampled signal; group the resampled signal and the resampled reconstructed signal accordingly, and convert the resampled signal in each group and its corresponding resampled reconstructed signal into scatter points in polar coordinates based on the SDP method to obtain the sampled SDP image of scatter point fusion corresponding to each group.

[0088] To implement the above embodiments, this application also proposes an electronic device 800, such as... Figure 8 As shown, the electronic device 800 includes a processor 801 and a memory 802 communicatively connected to the processor. The memory 802 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 801 to implement the fault diagnosis method for rotating machinery as shown in the above embodiment.

[0089] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to implement the fault diagnosis method for rotating machinery as shown in the above embodiments.

[0090] To implement the above embodiments, this application also proposes a computer program product, including a computer program that, when executed by a processor, implements the fault diagnosis method for rotating machinery as shown in the above embodiments.

[0091] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0093] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0094] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A fault diagnosis method for rotating machinery, characterized in that, include: The original vibration time-domain signals of rotating machinery collected by multiple sensors are acquired, and each of the original vibration time-domain signals is decomposed to obtain multiple candidate intrinsic mode (IMF) components after the decomposition of each original vibration time-domain signal. For any of the original vibration time-domain signals, obtain the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal, and the correlation coefficient between each candidate IMF component corresponding to the original vibration time-domain signal and the original vibration time-domain signal; Based on the kurtosis index and the correlation coefficient, all the candidate IMF components corresponding to the original vibration time-domain signal are filtered to obtain the filtered target IMF components. Based on the target IMF component, the signal is reconstructed to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal. Based on the symmetric polar coordinate SDP method, all the original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals are transformed into scattered points in polar coordinates to obtain the SDP image fused from the scattered points. The SDP image is input into a trained target convolutional neural network to obtain the fault classification of the rotating machinery output by the target convolutional neural network. The training method for the target convolutional neural network includes: Acquire the original vibration time-domain signals of the rotating mechanical equipment collected by multiple sampling sensors; The original vibration time-domain signal is resampled at different time lengths with a set window size to obtain the resampled signal. For any of the resampled signals, multiple target resampled IMF components for reconstructing the signal are determined from multiple candidate resampled IMF components corresponding to the resampled signal, and the signal is reconstructed based on the target resampled IMF components to obtain the resampled reconstructed signal corresponding to the resampled signal. The resampled signal and the resampled reconstructed signal are grouped accordingly, and based on the SDP method, the resampled signal and its corresponding resampled reconstructed signal in each group are converted into scatter points in polar coordinates to obtain a sampled SDP image of scatter point fusion for each group. Based on the sampled SDP image and the preset rotating machinery equipment fault labels corresponding to the sampled SDP image, the initial convolutional neural network is iteratively trained to obtain the target convolutional neural network generated after training.

2. The method according to claim 1, characterized in that, The step of filtering all candidate IMF components corresponding to the original vibration time-domain signal according to the kurtosis index and the correlation coefficient to obtain the filtered target IMF components includes: Obtain the preset kurtosis threshold and correlation coefficient threshold; For any candidate IMF component corresponding to any original vibration time-domain signal, the candidate IMF component that simultaneously satisfies the kurtosis index being less than the kurtosis threshold and the correlation coefficient being greater than the correlation coefficient threshold is taken as the target IMF component corresponding to the original vibration time-domain signal.

3. A fault diagnosis device for rotating machinery, characterized in that, include: The acquisition module is used to acquire the original vibration time-domain signals of rotating machinery collected by multiple sensors, and to decompose each of the original vibration time-domain signals to obtain multiple candidate intrinsic mode (IMF) components after the decomposition of each original vibration time-domain signal. The reconstruction module is configured to, for any given original vibration time-domain signal, obtain the kurtosis index of each candidate IMF component corresponding to the original vibration time-domain signal, and the correlation coefficient between each candidate IMF component and the original vibration time-domain signal; filter all candidate IMF components corresponding to the original vibration time-domain signal according to the kurtosis index and the correlation coefficient to obtain the filtered target IMF component, and reconstruct the signal based on the target IMF component to obtain the reconstructed vibration time-domain signal corresponding to the original vibration time-domain signal; The conversion module is used to convert all the original vibration time-domain signals and their corresponding reconstructed vibration time-domain signals into scatter points in polar coordinates based on the symmetric polar coordinate SDP method, so as to obtain the SDP image fused from the scatter points. The classification module is used to input the SDP image into a trained target convolutional neural network to obtain the fault classification of the rotating machinery equipment output by the target convolutional neural network. The model training module is used to acquire the original vibration time-domain signals of the rotating machinery collected by multiple sampling sensors; resample the original vibration time-domain signals at different time lengths with a set window size to obtain the resampled signals; for any resampled signal, determine multiple target resampled IMF components for signal reconstruction from multiple candidate resampled IMF components corresponding to the resampled signal, and reconstruct the signal based on the target resampled IMF components to obtain the resampled reconstructed signal corresponding to the resampled signal; group the resampled signals and the resampled reconstructed signals accordingly, and convert the resampled signals and their corresponding resampled reconstructed signals in each group into scatter points in polar coordinates based on the SDP method to obtain the scatter point fusion sampling SDP image corresponding to each group; and iteratively train the initial convolutional neural network based on the sampling SDP image and the preset rotating machinery fault labels corresponding to the sampling SDP image to obtain the target convolutional neural network generated after training.

4. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of claim 1 or 2.

5. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to claim 1 or 2.

6. A computer program product comprising a computer program that, when executed by a processor, implements the steps according to claim 1 or 2.

Images