A method and system for rotating machinery fault analysis based on adaptive wavelet decomposition

Abstract

The application discloses a kind of based on adaptive wavelet decomposition's rotating machinery fault analysis method and system, wherein, method includes the vibration signal of the rotating machinery to be analyzed is collected and data preprocessing is carried out, obtains data sample;Depth feature extraction processing is carried out to the data sample by feature extraction module, and depth feature is obtained;The depth feature is input to adaptive wavelet decomposition model, and target decomposition feature is output and obtained;The target decomposition feature is subjected to fault type discrimination processing, and obtains fault analysis result.The adaptive wavelet decomposition model is carried out in the embodiment of the application wavelet filter and the automatic learning of wavelet decomposition route, and then separates fault component and interference component in rotating machinery fault analysis task, effectively improves the anti-noise performance of method, and can be widely applied in intelligent monitoring technical field.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to a method and system for analyzing rotating machinery faults based on adaptive wavelet decomposition. Background Technology

[0002] Rotating machinery has a complex structure, and troubleshooting malfunctions is often difficult, causing significant adverse impacts on production and daily life. Automated and efficient diagnostics for rotating machinery can effectively assist workers in designing maintenance and improvement plans, which is crucial for the reliability and safety of modern industrial systems. While noise-resistant rotating machinery fault diagnosis methods typically combine wavelet transform with deep learning, these methods often employ fixed wavelet bases and wavelet decomposition paths, making them ill-suited to the complex frequency distributions of fault modes and noise. This results in issues with fault diagnosis accuracy and noise robustness.

[0003] In summary, the technical problems existing in the relevant technologies urgently need to be solved. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a rotating machinery fault analysis method and system based on adaptive wavelet decomposition, to improve the accuracy and robustness of fault analysis.

[0005] On the one hand, the present invention provides a method for fault analysis of rotating machinery based on adaptive wavelet decomposition, the method comprising:

[0006] Vibration signals of the rotating machinery to be analyzed are collected and preprocessed to obtain data samples;

[0007] The data samples are processed by a feature extraction module to extract deep features, thereby obtaining deep features.

[0008] The deep features are input into an adaptive wavelet decomposition model, and the target decomposition features are output. The adaptive wavelet decomposition model includes a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module. The high-low frequency adaptive decomposition module includes a data-driven neural network.

[0009] The target decomposition features are subjected to fault type discrimination processing to obtain fault analysis results.

[0010] Optionally, the step of acquiring the vibration signal of the rotating machinery to be analyzed and performing data preprocessing to obtain data samples includes:

[0011] Sensors are deployed on the rotating machinery to be analyzed, and vibration signals of the rotating machinery to be analyzed are collected through the sensors.

[0012] The vibration signal is subjected to sliding window segmentation, and the segmented data are combined to obtain a data matrix;

[0013] The data matrix is ​​then subjected to standard normalization to obtain data samples.

[0014] Optionally, the step of inputting the depth features into an adaptive wavelet decomposition model and outputting the target decomposition features includes:

[0015] The deep features are decomposed by the high- and low-frequency adaptive decomposition module to obtain a set of frequency band components, which includes low-frequency components and high-frequency components.

[0016] The high-low frequency band adaptive selection module selects one component from the frequency band component set as a depth feature and returns to the step of decomposing the depth feature by the high-low frequency adaptive decomposition module until the number of layers of the high-low frequency adaptive decomposition module reaches a preset threshold.

[0017] The exit layer adaptive selection module selects a selection layer from the high and low frequency band adaptive selection modules as the exit layer, and outputs the target decomposition features.

[0018] Optionally, the step of decomposing the deep features through the high- and low-frequency adaptive decomposition module to obtain a set of frequency band components includes:

[0019] The depth features are arranged in chronological order to obtain a feature sequence, which includes odd sequences and even sequences.

[0020] The feature sequence is input into a data-driven neural network, which includes an updater and a predictor.

[0021] The odd sequence is input into the updater, and the output of the updater is added to the even sequence to obtain the low-frequency component;

[0022] The low-frequency component is input into the predictor, and the odd sequence is subtracted from the output of the predictor to obtain the high-frequency component.

[0023] Optionally, the high-low frequency band adaptive selection module includes a first encoder, and the step of selecting a component from the set of frequency band components as a deep feature through the high-low frequency band adaptive selection module includes:

[0024] The first encoder performs semantic feature extraction processing on the frequency band component set to obtain low-frequency band features and high-frequency band features.

[0025] The first encoder is updated based on the low-frequency band characteristics and the high-frequency band characteristics;

[0026] The updated first encoder selects one component from the set of frequency band components as a depth feature.

[0027] Optionally, the exit layer adaptive selection module includes a second encoder, wherein the exit layer adaptive selection module selects a selection layer from the high and low frequency band adaptive selection modules as the exit layer, and outputs the target decomposition features, including:

[0028] Deep feature calculation processing is performed on each decomposition layer in the high- and low-frequency adaptive decomposition module to obtain the decomposition layer feature set;

[0029] Each feature in the decomposition layer feature set is subjected to average pooling to obtain an average pooled feature set.

[0030] The second encoder is updated based on the average pooling feature set;

[0031] The updated second encoder selects a selection layer from the high and low frequency band adaptive selection module as the exit layer.

[0032] The target decomposition features are obtained through the output of the exit layer.

[0033] Optionally, the step of performing fault type discrimination processing on the target decomposition features to obtain fault analysis results includes:

[0034] Obtain the unselected frequency band features of all preceding layers of the exited layer;

[0035] Pooling is performed on the unselected frequency band features and the target decomposition features respectively, and the pooled features are then concatenated to obtain classification features;

[0036] The classification features are input into a linear classifier to determine the fault category, and the fault analysis results are obtained.

[0037] On the other hand, embodiments of the present invention also provide a rotating machinery fault analysis system based on adaptive wavelet decomposition, comprising:

[0038] The first module is used to collect vibration signals of the rotating machinery to be analyzed and perform data preprocessing to obtain data samples;

[0039] The second module is used to perform deep feature extraction processing on the data sample through the feature extraction module to obtain deep features;

[0040] The third module is used to input the deep features into the adaptive wavelet decomposition model and output the target decomposition features. The adaptive wavelet decomposition model includes a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module. The high-low frequency adaptive decomposition module includes a data-driven neural network.

[0041] The fourth module is used to perform fault type discrimination processing on the target decomposition features to obtain fault analysis results.

[0042] On the other hand, embodiments of the present invention also disclose an electronic device, including a processor and a memory;

[0043] The memory is used to store programs;

[0044] The processor executes the program to implement the method described above.

[0045] On the other hand, embodiments of the present invention also disclose a computer-readable storage medium storing a program that is executed by a processor to implement the methods described above.

[0046] On the other hand, embodiments of the present invention also disclose a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0047] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects: In the embodiments of the present invention, the adaptive wavelet decomposition model performs wavelet decomposition processing on the vibration signal of rotating machinery in a data-driven manner, which can avoid the use of fixed wavelet basis and wavelet decomposition route, and can better adapt to data characteristics, thereby improving the accuracy and robustness of fault analysis. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, without creative effort,

[0049] Other figures can be obtained from these figures.

[0050] Figure 1 This is a flowchart of a rotating machinery fault analysis method based on adaptive wavelet decomposition provided in an embodiment of this application;

[0051] Figure 2 This is a schematic diagram of the structure of an adaptive wavelet decomposition model provided in an embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the structure of a rotating machinery fault analysis system based on adaptive wavelet decomposition provided in an embodiment of this application;

[0053] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] First, let's analyze some of the terms used in this application:

[0056] Rotating machinery diagnostics: Rotating machinery refers to mechanical equipment that performs specific functions through the rotational motion of its internal components. Rotating machinery fault diagnosis refers to the task of detecting, isolating, and identifying faults in rotating machinery by collecting continuous state information of its operation (such as vibration signals). This task plays a crucial role in the reliability and safety of modern industrial systems.

[0057] Neural networks: A neural network is an algorithmic mathematical model that mimics the behavioral characteristics of animal neural networks to perform distributed parallel information processing. This type of network relies on the complexity of the system and adjusts the relationships between a large number of interconnected nodes to achieve the purpose of processing information.

[0058] Wavelet transform: Wavelet transform is a new transform analysis method that inherits and develops the idea of ​​localization of short-time Fourier transform, while overcoming the shortcomings such as the window size not changing with frequency. It can provide a "time-frequency" window that changes with frequency, making it an ideal tool for time-frequency analysis and processing of signals.

[0059] In related technologies, noise-resistant rotating machinery fault diagnosis methods usually combine wavelet transform with deep learning. However, these methods use fixed wavelet bases and wavelet decomposition paths, requiring the pre-selection of specific wavelet bases for time-frequency decomposition. The amplitude-frequency response of these wavelet bases is relatively fixed and cannot effectively adapt to the data characteristics of vibration signals. In addition, in actual fault diagnosis tasks, the frequency distribution of fault components is usually quite complex, and wavelet decomposition along a fixed decomposition path makes it difficult to effectively separate fault components from interference components.

[0060] In view of this, this application provides a rotating machinery fault analysis method based on adaptive wavelet decomposition. The analysis method in this application can be applied to a terminal, a server, or software running on either a terminal or a server. The terminal can be a tablet computer, a laptop computer, a desktop computer, etc., but is not limited to these. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0061] Reference Figure 1 This invention provides a method for analyzing rotating machinery faults based on adaptive wavelet decomposition, comprising:

[0062] S101. Collect the vibration signal of the rotating machinery to be analyzed and perform data preprocessing to obtain data samples;

[0063] S102. The data sample is subjected to deep feature extraction processing by the feature extraction module to obtain deep features;

[0064] S103. Input the deep features into the adaptive wavelet decomposition model and output the target decomposition features. The adaptive wavelet decomposition model includes a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module. The high-low frequency adaptive decomposition module includes a data-driven neural network.

[0065] S104. Perform fault type discrimination processing on the target decomposition features to obtain fault analysis results.

[0066] In this embodiment of the invention, vibration signals of the rotating machinery to be analyzed are acquired, and the acquired vibration signals are preprocessed to obtain data samples. Then, depth features of the data samples are extracted using a feature extraction module. This embodiment of the invention inputs the data samples into the feature extraction module for feature extraction processing. This feature extraction module includes multiple convolutional modules, each of which consists of a convolutional layer, a batch normalization layer, and a parameterized linear rectified activation layer. Furthermore, each convolutional layer uses a large-size convolutional kernel to better capture temporal correlations. Then, this embodiment of the invention inputs the depth features into an adaptive wavelet decomposition model, outputting the target decomposed features. This adaptive wavelet decomposition model can adaptively realize the wavelet decomposition of the vibration signal and its depth features in a data-driven manner through a neural network, thereby effectively separating fault components from interference components. (Refer to...) Figure 2The adaptive wavelet decomposition model includes a high- and low-frequency adaptive decomposition module, a high- and low-frequency band adaptive selection module, and an exit layer adaptive selection module. This embodiment of the invention extracts depth features from vibration signal data samples and performs adaptive wavelet decomposition on these features to separate fault components from noise components. Finally, the target decomposition features are processed for fault type discrimination to obtain fault analysis results. Compared with fault diagnosis methods in related technologies, this embodiment of the invention demonstrates significantly better fault diagnosis performance under six different signal-to-noise ratio conditions on three experimental datasets, with an average classification accuracy improvement of approximately 6.7%. Furthermore, this embodiment of the invention has a small network parameter count of approximately 61k, which improves the efficiency of rotating machinery fault analysis.

[0067] As a further optional implementation, in step S101 above, the step of acquiring the vibration signal of the rotating machinery to be analyzed and performing data preprocessing to obtain data samples includes:

[0068] Sensors are deployed on the rotating machinery to be analyzed, and vibration signals of the rotating machinery to be analyzed are collected through the sensors.

[0069] The vibration signal is subjected to sliding window segmentation, and the segmented data are combined to obtain a data matrix;

[0070] The data matrix is ​​then subjected to standard normalization to obtain data samples.

[0071] In this embodiment of the invention, a sensor, such as an accelerometer, is deployed on the rotating machinery to be analyzed. The accelerometer is placed at one or more key locations on the rotating machinery, and a sampling rate is set to record the vibration signal of the rotating machinery. Then, a sliding window of size L is used to segment the vibration signal recorded by the sensor in a non-overlapping manner, resulting in multiple samples of fixed length. L can be flexibly adjusted according to the sampling rate and the fault type. The multiple fixed-length samples are then combined to obtain a data matrix X, and the data matrix is ​​subjected to standard normalization to obtain data samples. The standard normalization formula is shown below:

[0072]

[0073] Among them, X norm This represents a data sample, where min and max are the minimum and maximum values, respectively.

[0074] By performing standard normalization on multiple fixed-length samples after segmentation, these data samples are then used as input to an adaptive wavelet decomposition neural network.

[0075] As a further optional implementation, in step S103 above, the step of inputting the depth features into an adaptive wavelet decomposition model and outputting the target decomposition features includes:

[0076] The deep features are decomposed by the high- and low-frequency adaptive decomposition module to obtain a set of frequency band components, which includes low-frequency components and high-frequency components.

[0077] The high-low frequency band adaptive selection module selects one component from the frequency band component set as a depth feature and returns to the step of decomposing the depth feature by the high-low frequency adaptive decomposition module until the number of layers of the high-low frequency adaptive decomposition module reaches a preset threshold.

[0078] The exit layer adaptive selection module selects a selection layer from the high and low frequency band adaptive selection modules as the exit layer, and outputs the target decomposition features.

[0079] In this embodiment of the invention, reference is made to Figure 2The adaptive wavelet decomposition model includes a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module. The high-low frequency adaptive decomposition module decomposes the depth features to obtain a set of frequency band components, which includes low-frequency and high-frequency components. This module, based on the improved wavelet decomposition, adds a data-driven learnable neural network to the deep features, obtaining low-frequency and high-frequency components. Then, the high-low frequency band adaptive selection module selects either a low-frequency or high-frequency component from the set as the depth feature. The purpose of this module is to select a suitable frequency band from the low-frequency and high-frequency components for further time-frequency decomposition. Both low-frequency and high-frequency bands may be selected for the next time-frequency decomposition layer. The process then returns to the step of decomposing the depth features using the high-low frequency adaptive decomposition module, until the number of layers in the high-low frequency adaptive decomposition module reaches a preset threshold, which can be set according to actual conditions. The high- and low-frequency adaptive decomposition module contains multiple decomposition layers, and the high- and low-frequency band adaptive selection module contains multiple selection layers, with the number of decomposition layers and selection layers being the same. The wavelet decomposition process consists of multiple cascaded high- and low-frequency adaptive decomposition steps. As the number of high- and low-frequency adaptive decomposition layers increases, the spectral decomposition becomes more refined. To end the entire wavelet decomposition process with an appropriate time-frequency decomposition granularity, an exit layer needs to be adaptively selected, and the decomposition features corresponding to that layer are used as the final classification features. In this embodiment of the invention, the exit layer adaptive selection module selects a selection layer from the high- and low-frequency band adaptive selection module as the exit layer, outputting the target decomposition features. These target decomposition features are the features after separating the fault components from the noise components, and are used for subsequent fault type discrimination.

[0080] As a further optional implementation, the step of decomposing the deep features through the high- and low-frequency adaptive decomposition module to obtain a set of frequency band components includes:

[0081] The depth features are arranged in chronological order to obtain a feature sequence, which includes odd sequences and even sequences.

[0082] The feature sequence is input into a data-driven neural network, which includes an updater and a predictor.

[0083] The odd sequence is input into the updater, and the output of the updater is added to the even sequence to obtain the low-frequency component;

[0084] The low-frequency component is input into the predictor, and the odd sequence is subtracted from the output of the predictor to obtain the high-frequency component.

[0085] In this embodiment of the invention, the high- and low-frequency adaptive decomposition module decomposes the depth features. First, the depth feature sequence output by the convolution module is divided into odd sequences s according to the time sequence. o And even sequence s e :

[0086] s e [n] = s[2n], s o [n] = s[2n+1]

[0087] Where n represents the timestamp number.

[0088] The feature sequence is then input into a data-driven neural network, which includes an updater and a predictor constructed from the neural network. The neural network can be a multilayer perceptron, convolutional network, recurrent neural network, or attention mechanism, etc. By using the odd sequence s... o The input is given to the updater U, and the output of the updater is compared with the even sequence s. e By performing addition, the low-frequency component c is obtained.

[0089] c[n] = s e [n]+U(s o [n])

[0090] The updater U is constructed using a neural network. In one feasible implementation, the updater can be constructed using a sub-network consisting of three convolutional layers, each consisting of a 1×3 convolutional kernel, a batch normalization layer, and a linear activation layer. During training, the sub-network used to construct the updater U is optimized along with the entire fault diagnosis network, thereby enabling automatic updating of the updater U in a data-driven manner.

[0091] The low-frequency component is input into the predictor, and the odd sequence is subtracted from the predictor's output to obtain the high-frequency component. The low-frequency component c is input into the predictor P, and the odd sequence s is... o Subtracting the predictor's output from the high-frequency component d yields the high-frequency component d:

[0092] d[n]=s o [n]-P(c[n]),

[0093] The predictor P is also constructed using a neural network. Similar to the updater U, predictor P can be built using a sub-network consisting of three convolutional layers. Each convolutional layer comprises multiple 1×3 convolutional kernels, batch normalization layers, and linear activation layers. During training, the sub-network used to construct predictor P is optimized along with the entire fault diagnosis network, thereby enabling automatic updates of P in a data-driven manner. The network structure of predictor P is typically consistent with that of updater U.

[0094] As a further optional implementation, the high-low frequency band adaptive selection module includes a first encoder, and the step of selecting a component as a deep feature from the frequency band component set through the high-low frequency band adaptive selection module includes:

[0095] The first encoder performs semantic feature extraction processing on the frequency band component set to obtain low-frequency band features and high-frequency band features.

[0096] The first encoder is updated based on the low-frequency band characteristics and the high-frequency band characteristics;

[0097] The updated first encoder selects one component from the set of frequency band components as a depth feature.

[0098] In this embodiment of the invention, the purpose of the high-low frequency band adaptive selection module is to select suitable frequency bands from the low-frequency component c and high-frequency component d of the frequency band component set for further time-frequency decomposition. First, semantic features are further extracted from c and d using the first encoder E to obtain low-frequency band features and high-frequency band features, which are then used for frequency band selection. Since the frequency band selection process is not differentiable, it is modeled as a sampling process during training, and the entire sampling process is optimized by updating the first encoder E.

[0099]

[0100] G i =-log(-log(U) i ));U i ~U(0,1)

[0101] Where, x i Let c represent the low-frequency band feature and d represent the high-frequency band feature, τ be a temperature parameter that gradually decays during training, U be a uniformly distributed value, and a be a value. i G represents the sampling distribution to be optimized for frequency band selection. i Let a represent the Gumbel noise vector following a Gumbel(0,1) distribution. Then, during the test, a... i Hardened into a single heat distribution for frequency band selection:

[0102] s i =argmax(a i ).

[0103] At this point, the updated first encoder selects a component from the set of frequency band components as a depth feature. Both the low-frequency band c and the high-frequency band d may be selected for time-frequency decomposition of the next decomposition layer.

[0104] As a further optional implementation, the exit layer adaptive selection module includes a second encoder. The step of selecting a selection layer from the high- and low-frequency band adaptive selection modules as the exit layer through the exit layer adaptive selection module, and outputting the target decomposition features, includes:

[0105] Deep feature calculation processing is performed on each decomposition layer in the high- and low-frequency adaptive decomposition module to obtain the decomposition layer feature set;

[0106] Each feature in the decomposition layer feature set is subjected to average pooling to obtain an average pooled feature set.

[0107] The second encoder is updated based on the average pooling feature set;

[0108] The updated second encoder selects a selection layer from the high and low frequency band adaptive selection module as the exit layer.

[0109] The target decomposition features are obtained through the output of the exit layer.

[0110] In this embodiment of the invention, deep feature calculation processing is performed on each decomposition layer in the high-low frequency adaptive decomposition module to obtain a decomposition layer feature set. The decomposition layer feature set includes multiple decomposition layer features, wherein the decomposition layer feature F of each high-low frequency adaptive decomposition layer is... j Unselected frequency band features for all preceding high and low frequency adaptive decomposition layers Frequency band characteristics of the current high and low frequency adaptive decomposition layer Sum:

[0111]

[0112] Then, average pooling is performed on each decomposition layer feature in the decomposition layer feature set to obtain an average pooled feature set. The second encoder is then updated based on the average pooled feature set. The sampling process is modeled as follows:

[0113]

[0114] G j =-log(-log(U)j ));U j ~U(0,1),

[0115] Where avg represents the average pooling operation, E is the second encoder set during the exit layer selection process (not shared with the first encoder parameters during the frequency band selection process), and b j G represents the sampling distribution to be optimized for exiting the layer selection. j Let b represent the Gumbel noise vector following a Gumbel(0,1) distribution. During the test, b... j Hardening is performed as a single heat distribution for exit layer selection:

[0116] l j =argmax(b j ).

[0117] At this point, the updated second encoder selects a selection layer from the high and low frequency band adaptive selection module as the exit layer. That is, a certain layer can be selected as the exit layer from all high and low frequency adaptive selection layers, and the target decomposition features are obtained through the output of the exit layer.

[0118] As a further optional implementation, the step of performing fault type discrimination processing on the target decomposition features to obtain fault analysis results includes:

[0119] Obtain the unselected frequency band features of all preceding layers of the exited layer;

[0120] Pooling is performed on the unselected frequency band features and the target decomposition features respectively, and the pooled features are then concatenated to obtain classification features;

[0121] The classification features are input into a linear classifier to determine the fault category, and the fault analysis results are obtained.

[0122] In this embodiment of the invention, the unselected frequency band features of all preceding layers of the exit layer are obtained. For example, in the above processing steps, the fifth layer is selected as the exit layer from all high-low frequency adaptive selection layers, and the preceding layers are the first to fourth layers in the high-low frequency adaptive selection module. Pooling is performed on the unselected frequency band features of all preceding layers and the target decomposition features output by the exit layer, and the pooled features are concatenated to obtain classification features. These classification features are used to determine the fault type of the rotating machinery. Finally, a linear classifier is constructed, and the classification features are input into the linear classifier to determine the fault type, thus obtaining the fault analysis result.

[0123] Reference Figure 3 This invention also provides a rotating machinery fault analysis system based on adaptive wavelet decomposition, the system comprising:

[0124] The first module 301 is used to collect the vibration signals of the rotating machinery to be analyzed and perform data preprocessing to obtain data samples;

[0125] The second module 302 is used to perform deep feature extraction processing on the data sample through the feature extraction module to obtain deep features;

[0126] The third module 303 is used to input the depth features into the adaptive wavelet decomposition model and output the target decomposition features. The adaptive wavelet decomposition model includes a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module. The high-low frequency adaptive decomposition module includes a data-driven neural network.

[0127] The fourth module 304 is used to perform fault type discrimination processing on the target decomposition features to obtain fault analysis results.

[0128] It is understood that the content of the above embodiments of the rotating machinery fault analysis method based on adaptive wavelet decomposition is applicable to the embodiments of the rotating machinery fault analysis system based on adaptive wavelet decomposition. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0129] Reference Figure 4 This invention also provides an electronic device, including a processor 401 and a memory 402; the memory is used to store a program; the processor executes the program to implement the method described above.

[0130] and Figure 1 Corresponding to the method described above, embodiments of the present invention also provide a computer-readable storage medium storing a program that is executed by a processor to implement the method described above.

[0131] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0132] In summary, the embodiments of the present invention have the following advantages: The embodiments of the present invention use neural networks to implement the wavelet decomposition process in a data-driven manner, avoiding the use of fixed wavelet bases and wavelet decomposition routes, and can better adapt to data characteristics. Furthermore, the embodiments of the present invention can adaptively separate fault components and interference components during the wavelet decomposition process, improving the method's noise resistance and demonstrating robustness under different signal-to-noise ratio conditions.

[0133] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0134] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0135] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0136] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0137] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0138] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0139] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. 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.

[0140] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0141] The above is a detailed description of the preferred embodiments of the present invention, but the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for analyzing the fault of a rotating machine based on adaptive wavelet decomposition, characterized in that, The method comprises: Collecting vibration signals of a rotating machine to be analyzed and performing data preprocessing to obtain data samples; Performing deep feature extraction processing on the data samples through a feature extraction module to obtain deep features; Inputting the deep features into an adaptive wavelet decomposition model to output target decomposition features, the adaptive wavelet decomposition model comprising a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module, the high-low frequency adaptive decomposition module comprising a data-driven neural network; Performing fault type discrimination processing on the target decomposition features to obtain a fault analysis result; The step of inputting the deep features into the adaptive wavelet decomposition model to output target decomposition features comprises: Performing decomposition processing on the deep features through the high-low frequency adaptive decomposition module to obtain a frequency band component set, the frequency band component set comprising low frequency components and high frequency components; Selecting a component from the frequency band component set as a deep feature through the high-low frequency band adaptive selection module, and returning to the step of performing decomposition processing on the deep features through the high-low frequency adaptive decomposition module until the number of layers of the high-low frequency adaptive decomposition module reaches a preset threshold; Selecting an exit layer from the high-low frequency band adaptive selection module through the exit layer adaptive selection module to output target decomposition features; The step of performing decomposition processing on the deep features through the high-low frequency adaptive decomposition module to obtain a frequency band component set comprises: Arranging the deep features in a time sequence to obtain a feature sequence, the feature sequence comprising odd sequences and even sequences; Inputting the feature sequence into a data-driven neural network, the data-driven neural network comprising an updater and a predictor; Inputting the odd sequences into the updater, and adding the output of the updater to the even sequences to obtain low frequency components; Inputting the low frequency components into the predictor, and subtracting the output of the predictor from the odd sequences to obtain high frequency components; The step of performing fault type discrimination processing on the target decomposition features to obtain a fault analysis result comprises: Obtaining unselected band features of all previous layers of the exit layer; Performing pooling processing on the unselected band features and the target decomposition features respectively, and connecting the pooled features to obtain classification features; Inputting the classification features into a linear classifier to perform fault category discrimination and obtain a fault analysis result.

2. The method of claim 1, wherein, The step of collecting vibration signals of a rotating machine to be analyzed and performing data preprocessing to obtain data samples comprises: Deploying sensors on the rotating machine to be analyzed, and collecting vibration signals of the rotating machine to be analyzed through the sensors; Performing sliding window segmentation processing on the vibration signals, and combining the segmented data to obtain a data matrix; Performing standard normalization processing on the data matrix to obtain data samples.

3. The method of claim 1, wherein, The high-low frequency band adaptive selection module comprises a first encoder, and the high-low frequency band adaptive selection module selects one component from the set of frequency band components as a deep feature, comprising: The first encoder performs semantic feature extraction processing on the set of frequency band components to obtain low frequency band features and high frequency band features; The first encoder is updated according to the low frequency band features and the high frequency band features; The first encoder selects one component from the set of frequency band components as a deep feature after being updated.

4. The method of claim 1, wherein, The exit layer adaptive selection module comprises a second encoder, and the exit layer adaptive selection module selects one selection layer from the high-low frequency band adaptive selection module as an exit layer to output a target decomposition feature, comprising: The deep feature calculation processing is performed on each decomposition layer in the high-low frequency adaptive decomposition module to obtain a set of decomposition layer features; The average pooling processing is performed on each decomposition layer feature in the set of decomposition layer features to obtain an average pooling feature set; The second encoder is updated according to the average pooling feature set; The second encoder selects one selection layer from the high-low frequency band adaptive selection module as an exit layer after being updated; The target decomposition feature is output by the exit layer.

5. A rotating machinery fault analysis system based on adaptive wavelet decomposition, characterized by, The system comprises: A first module is configured to collect vibration signals of a rotating machine to be analyzed and perform data preprocessing to obtain data samples; A second module is configured to perform deep feature extraction processing on the data samples by a feature extraction module to obtain deep features; A third module is configured to input the deep features into an adaptive wavelet decomposition model to output target decomposition features, wherein the adaptive wavelet decomposition model comprises a high-low frequency adaptive decomposition module, a high-low frequency band adaptive selection module, and an exit layer adaptive selection module, and the high-low frequency adaptive decomposition module comprises a data-driven neural network; A fourth module is configured to perform fault type discrimination processing on the target decomposition features to obtain a fault analysis result; The deep features are input into the adaptive wavelet decomposition model to output target decomposition features, comprising: The deep features are decomposed by the high-low frequency adaptive decomposition module to obtain a set of frequency band components, wherein the set of frequency band components comprises low frequency components and high frequency components; One component is selected from the set of frequency band components as a deep feature by the high-low frequency band adaptive selection module, and the step of decomposing the deep features by the high-low frequency adaptive decomposition module is returned until the number of layers of the high-low frequency adaptive decomposition module reaches a preset threshold; One selection layer is selected from the high-low frequency band adaptive selection module as an exit layer by the exit layer adaptive selection module to output target decomposition features; The deep features are decomposed by the high-low frequency adaptive decomposition module to obtain a set of frequency band components, comprising: The deep features are arranged in time sequence to obtain a feature sequence, wherein the feature sequence comprises an odd sequence and an even sequence; The characteristic sequence is input into a data-driven neural network, which includes an updater and a predictor; The odd sequence is input into the updater, and the output of the updater is added to the even sequence to obtain a low-frequency component; The low-frequency component is input into the predictor, and the odd sequence is subtracted from the output of the predictor to obtain a high-frequency component; The fault type discrimination processing on the target decomposition feature obtains a fault analysis result, including: Obtaining unselected band features of all previous layers of the exit layer; Respectively performing pooling processing on the unselected band features and the target decomposition feature, and connecting the pooled features to obtain classification features; The classification features are input into a linear classifier for fault category discrimination to obtain a fault analysis result.

6. An electronic device, comprising: The electronic device includes a memory and a processor; The memory is configured to store a program; The processor executes the program to implement the method in any one of claims 1 to 4.

7. A computer readable storage medium, the storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the method in any one of claims 1 to 4.

Images