A bearing fault diagnosis method based on an improved integrated stack denoising autoencoder

Abstract

The application discloses a bearing fault diagnosis method based on an improved integrated stack denoising autoencoder, and belongs to the technical field of rotating machinery fault diagnosis. The method comprises the following steps: collecting vibration signals of a bearing and extracting time domain data and frequency domain data; improving a stack denoising autoencoder to obtain a first feature learning model for extracting deep time sequence features of the time domain data and a second feature learning model for extracting deep frequency domain features of the frequency domain data; feeding the time domain data into the first feature learning model for training to learn deep time domain features; feeding the frequency domain data into the second feature learning model for training to learn deep frequency domain features; combining the learned deep time domain features and deep frequency domain features as a deep feature set and feeding the deep feature set into a preset model for training to establish a fault diagnosis model; and using the trained model to perform fault diagnosis on a bearing to be diagnosed to obtain a diagnosis result. The application can reduce network training time and improve fault diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of rotating machinery fault diagnosis technology, and in particular to a bearing fault diagnosis method based on an improved integrated stacked noise reduction self-encoder. Background Technology

[0002] Industry is a hallmark of a nation's development, determining the modernization of the national economy and serving as its leading force. Through continuous accumulation, modern industry has further advanced. With increasing product demand, rotating machinery has been widely used in industrial production. As key components of rotating machinery, the condition of rolling bearings, rotors, and other parts directly affects the operation of the entire machine. Ensuring the efficient and stable operation of equipment is the primary task of industrial production and a crucial factor in guaranteeing the economic benefits of enterprises.

[0003] Due to the harsh production environment, complex operating conditions, and heavy production tasks in industrial processes, the components of mechanical equipment inevitably deviate from their normal operating state or even malfunction after prolonged operation. This can range from minor issues like delays and reduced production efficiency to serious safety problems causing injuries or fatalities. Therefore, timely and accurate diagnosis of equipment malfunctions allows for prompt repair and replacement, ensuring the safe and stable operation of industrial processes.

[0004] Thanks to the development of intelligent sensor technology, data-driven methods have emerged and become an important technology for fault diagnosis. The massive amounts of data collected on-site can be used for equipment status analysis. Since vibration signals carry the status information of mechanical equipment components and are easy to collect, vibration signal-based equipment fault diagnosis methods have become the main means of equipment fault diagnosis. Traditional data-driven fault diagnosis methods include three steps: data acquisition, feature extraction, and fault classification. Researchers achieve mechanical equipment fault diagnosis based on vibration signals by feeding learned time-domain features, frequency-domain features, and time-frequency-domain features into a classifier. The quality of both data and features affects the performance of the diagnosis. Due to harsh working environments and complex transmission methods, the acquired vibration signals often contain background noise, which can mask useful information in the signal. Redundant feature information can also affect the diagnostic results. Therefore, how to effectively remove noise from vibration signals, extract useful features, and achieve timely and effective fault diagnosis of rotating machinery has gradually become a technical problem worthy of research in industry and academia.

[0005] Traditional manual feature extraction methods are labor-intensive, inefficient, and their shallow time-domain and frequency-domain features are easily affected by environmental interference. Therefore, a stable end-to-end diagnostic model that can directly connect raw monitoring data to the corresponding machine health status is needed. Deep learning methods, unlike traditional shallow techniques, simulate the information transmission pattern of neurons in the human brain. By deploying multi-layered, non-linear perception and mapping, they extract deep features hidden in the data, reducing reliance on signal processing and manual feature extraction, allowing end-to-end learning, achieving a breakthrough in the field of intelligent diagnosis and being applied to fault diagnosis. In recent years, deep learning algorithms such as Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), and Autoencoders (AE) have been widely used in mechanical equipment fault diagnosis. Among these deep learning methods, AE stands out due to its ease of implementation and automatic learning. As a variant of AE, the Stacked Denoising Autoencoder (SDAE) exhibits stronger robustness in noisy environments and is suitable for analyzing data containing environmental noise in harsh working environments. For example, Lü et al. used a noise-reducing autoencoder and Softmax to achieve fault diagnosis of rolling bearings; Hou et al. used particle swarm optimization to adjust the parameters of SDAE and adaptively determine the network structure to achieve fault diagnosis of rolling bearings; Shu et al. combined mouse swarm optimization and gray wolf optimization algorithms to select the optimal hyperparameters for bearing fault diagnosis using SDAE. However, in the above works, when SDAE adds mask noise to the input data to construct "damaged data", it randomly sets the input data to 0 at a certain proportion. This may not only ignore obvious noise, but may also destroy useful information.

[0006] Furthermore, most current work uses single deep learning models, which suffer from limitations such as limited model information and low generalization ability in fault diagnosis applications. How to rationally apply multiple deep learning models to bearing fault diagnosis is a worthy research question. Simultaneously, the high-dimensional deep features extracted by deep learning models are highly redundant or highly correlated, which can lead to problems such as the curse of dimensionality and affect diagnostic performance. Therefore, specific methods are needed to select representative low-dimensional features from high-dimensional features for fault diagnosis. Feature selection methods are mainly divided into filtering, wrapping, and embedding methods. Feature selection methods independent of deep network training, such as Principal Component Analysis (PCA), Fisher's Linear Discriminant Analysis (LDA), and Maximum Relevance Minimum Redundancy (mRMR), are widely used for dimensionality reduction of high-dimensional deep features. Although most current work can achieve dimensionality reduction of high-dimensional features to some extent, the steps independent of deep network features increase training complexity, thus increasing the diagnostic process time. There is a lack of feature dimensionality reduction training models that embed deep networks. Summary of the Invention

[0007] This invention provides a bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder, which solves the technical problems of existing bearing fault diagnosis technologies, such as the single deep learning model having limited information, low generalization ability, environmental noise in industrial acquisition signals, and redundancy in high-dimensional features.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] On one hand, the present invention provides a bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder, the bearing fault diagnosis method based on the improved integrated stacked noise-reducing autoencoder comprising:

[0010] The vibration signal of the bearing is collected, and the time domain data and frequency domain data of the vibration signal are extracted.

[0011] The stacked denoising autoencoder is improved for the time domain data and the frequency domain data respectively, resulting in a first feature learning model for extracting deep temporal features of the time domain data and a second feature learning model for extracting deep frequency domain features of the frequency domain data.

[0012] The time-domain data is fed into a first feature learning model for training to learn deep time-domain features; the frequency-domain data is fed into a second feature learning model for training to learn deep frequency-domain features.

[0013] The deep time-domain features learned by the first feature learning model and the deep frequency-domain features learned by the second feature learning model are combined into a deep feature set and fed into a preset model for training to establish a fault diagnosis model.

[0014] Using the trained first feature learning model, second feature learning model, and fault diagnosis model, the bearing to be diagnosed is fault diagnosed based on the vibration signal of the bearing to be diagnosed, and the diagnosis result is obtained.

[0015] Furthermore, the acquisition of the bearing vibration signal and the extraction of the time-domain and frequency-domain data of the vibration signal include:

[0016] Vibration signals of the bearing are collected, and the collected vibration signals are divided into several segments based on framing and windowing techniques as time-domain data. Then, the time-domain data is divided into training set and test set according to a preset ratio.

[0017] Perform a Fast Fourier Transform on the time-domain data in the training set to convert it into frequency-domain data.

[0018] Further, the time-domain data is fed into a first feature learning model for training to learn deep time-domain features, including:

[0019] Anomaly detection is performed on the time-domain data at a rate of 5%. The detected anomalies are then set to 0 as corrupted data and used for training the first feature learning model to learn deep time-domain features.

[0020] The frequency domain data is fed into a second feature learning model for training to learn deep frequency domain features, including:

[0021] The frequency domain data is randomly set to 0 at a rate of 5% to be corrupted data, which is then used to train the second feature learning model to learn deep frequency domain features.

[0022] Further, the improvement of the stacked denoising autoencoder for the time-domain data and the frequency-domain data respectively, to obtain a first feature learning model for extracting deep temporal features of the time-domain data, and a second feature learning model for extracting deep frequency-domain features of the frequency-domain data, includes:

[0023] A metric based on mutual information is designed to measure the relationship between features and embedded into the loss function of the deep network as a soft constraint for training the deep network to learn low-redundancy deep features. At the same time, a metric based on mutual information is designed to measure the relationship between features and class labels and embedded into the loss function of the deep network as a soft constraint for training the deep network to learn deep features that are highly correlated with the class, which is used for fault classification.

[0024] Furthermore, the loss function of the first feature learning model Represented as:

[0025]

[0026] Here, λ1 and λ2 are hyperparameters. By adjusting the size of the hyperparameters, the balance of each term in the loss function is maintained, thereby improving the performance of the model.

[0027]

[0028] Where N represents the number of sampling points, L MSE (X (i) Z (i) ) is the input time-domain data X (i) With the reconstructed signal Z (i) The mean square error between them.

[0029]

[0030]

[0031] Where N represents the number of sampling points, and |S| represents the total number of neurons in the l-th hidden layer, i.e., the total number of deep features; |S| represents the weight connecting the q-th neuron in the l-th hidden layer and the j-th neuron in the (l+1)-th hidden layer. l+1 This represents the total number of neurons in the (l+1)th hidden layer; I(Y (p) ,Y (q) ) represents the feature set Mutual information between different features, Y (p) and Y (q) They represent different types of features; I(Y) (q) ,T (i) T represents the mutual information between features and categories. (i) This indicates the target category to which the i-th sampling point belongs.

[0032] Furthermore, the loss function of the second feature learning model Represented as:

[0033]

[0034] Here, λ1 and λ2 are hyperparameters. By adjusting the size of the hyperparameters, the balance of each term in the loss function is maintained, thereby improving the performance of the model.

[0035]

[0036] Where N represents the number of sampling points, L MSE (F (i) Z (i) ) represents the input frequency domain data F (i) With the reconstructed signal Z (i) Mean square error between; F (i) This represents different input frequency domain data.

[0037] Furthermore, the preset model is an extreme learning machine model.

[0038] Furthermore, by utilizing the trained first feature learning model and second feature learning model, as well as the fault diagnosis model, based on the vibration signal of the bearing to be diagnosed, fault diagnosis is performed on the bearing to be diagnosed to obtain the diagnosis result, including:

[0039] The vibration signal of the bearing to be diagnosed is collected, and the collected vibration signal is divided into several segments based on framing and windowing techniques, which are used as the time domain data of the bearing to be diagnosed. The time domain data of the bearing to be diagnosed is then subjected to a fast Fourier transform to obtain the frequency domain data corresponding to the time domain data of the bearing to be diagnosed.

[0040] The time-domain data of the bearing to be diagnosed is input into the trained first feature learning model, and the trained first feature learning model is used to extract the deep time-frequency features corresponding to the time-domain data of the bearing to be diagnosed.

[0041] The frequency domain data of the bearing to be diagnosed is input into the trained second feature learning model, and the trained second feature learning model is used to extract the deep frequency domain features corresponding to the frequency domain data of the bearing to be diagnosed.

[0042] The extracted deep time-frequency features and deep frequency domain features are merged, and the merged features are input into the trained fault diagnosis model. The fault diagnosis results of the bearing to be diagnosed are obtained using the trained fault diagnosis model.

[0043] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0044] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above method.

[0045] The beneficial effects of the technical solution provided by this invention include at least the following:

[0046] The technical solution provided by this invention can effectively reduce the impact of noise on bearing fault diagnosis. It uses multi-domain information for training and embeds the feature redundancy index into the network training loss function, thus eliminating the feature selection step that is independent of network training, reducing network training time, and improving the accuracy of fault diagnosis. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of the noise reduction strategy provided in an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the execution flow of the bearing fault diagnosis method based on an improved integrated stack noise reduction autoencoder provided in an embodiment of the present invention;

[0050] Figure 3 This is a system block diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0052] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0053] Furthermore, in the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0054] Furthermore, in embodiments of the present invention, sometimes a subscript (such as W1) may be mistakenly written as a non-subscript form (such as W1). Without emphasizing the difference, the meaning they express is the same.

[0055] First Embodiment

[0056] This embodiment provides a bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder. Addressing the issue that randomly setting zeros and adding mask noise to the original stacked noise-reducing autoencoder may ignore obvious noise and destroy useful information, this method designs a novel approach for the stacked noise-reducing autoencoder based on a Lookout Detection (LOF) algorithm. Figure 1The targeted noise reduction strategy shown achieves targeted noise reduction in both the time and frequency domains, preserving more useful information to guide the model in selectively reducing noise from time and frequency domain data, thereby improving the model's noise resistance. Simultaneously, since the high-dimensional deep features extracted by the integrated deep learning model are highly redundant or highly correlated, this method designs a mutual information-based feature metric embedded in the deep network loss function as a soft constraint for training the deep network to learn low-redundancy deep features. Furthermore, to improve fault classification accuracy, a mutual information-based feature-to-class label metric is embedded in the deep network loss function as a soft constraint for training the deep network to learn deep features highly correlated with the class for fault classification. By embedding two mutual information-based metrics into the model training cost function as soft constraints to learn deep features with maximum correlation to the class target and minimum redundancy, the feature selection step, independent of deep model training, is eliminated. Finally, the learned multi-domain deep features are fed into an Extreme Learning Machine (ELM) for fault classification, achieving timely and accurate fault diagnosis and improving the accuracy of bearing fault diagnosis in noisy environments. The overall process of this method is as follows: Figure 2 As shown.

[0057] This method can be implemented by an electronic device, and specifically, it includes the following steps:

[0058] S1, Collect the vibration signal of the bearing, and extract the time domain data and frequency domain data of the vibration signal;

[0059] Specifically, in this embodiment, the implementation process of S1 is as follows:

[0060] S11, collect the vibration signal of the bearing, divide the collected vibration signal into several segments as the original time domain signal based on framing and windowing technology, and then divide the data into training set and test set according to a certain ratio;

[0061] S12 performs a Fast Fourier Transform on the time-domain data in the training set, converting it into frequency-domain data.

[0062] S2, for the time domain data and the frequency domain data respectively, the stacked noise reduction autoencoder is improved to obtain a first feature learning model for extracting deep temporal features of the time domain data and a second feature learning model for extracting deep frequency domain features of the frequency domain data.

[0063] S3, the time-domain data is fed into the first feature learning model for training to learn deep time-domain features; the frequency-domain data is fed into the second feature learning model for training to learn deep frequency-domain features.

[0064] Specifically, for the improved stacked denoising autoencoder trained on time-domain data, outlier detection is first performed on 5% of the time-domain data. These outliers are then set to 0 as "corrupted data" and used for training the improved stacked denoising autoencoder to learn deep time-domain features. For the improved stacked denoising autoencoder trained on frequency-domain data, 5% of the frequency-domain data is randomly set to 0 as "corrupted data" and used for training the improved stacked denoising autoencoder to learn deep frequency-domain features. The denoising strategy is as follows: Figure 1 As shown.

[0065] For the original time-domain data containing N sets of sampling points Calculate its Local Outlier Factor (LOF):

[0066]

[0067] Where, N k (x p ) represents point x p The number of points in the k-nearest neighbors of Ird k It is the locally reachable density, specifically calculated by the following formula:

[0068]

[0069] In the formula, reach_dist k It is the reachable distance, point x p and point x o The reachable distance between them can be calculated as follows:

[0070] reach_dist k (x p ,x o )=max(d(x p ,x o ),d k (x o (3)

[0071] Where d() represents the Euclidean distance between two points, and the local outlier factor is compared to see if it is greater than 1. If LOF k (x p If ) > 1, then the decision point x is determined. p If the point is an outlier, it is set to 0 and used as "damaged data" for training deep temporal features in the temporal stack denoising autoencoder.

[0072] At the same time, the original time domain data Perform a Fast Fourier Transform to convert the data to the frequency domain:

[0073]

[0074] in, N is the total number of sampling points.

[0075] Then, 5% of the data in the frequency domain is set to 0 as “damaged data” for training the frequency domain stacked denoising autoencoder to learn deep frequency domain features.

[0076] Furthermore, in addition to designing a targeted denoising strategy for the stacked denoising autoencoder, this embodiment also redesigns the loss function of the stacked denoising autoencoder based on mutual information. The redesigned loss function in this embodiment includes two metrics based on mutual information, as follows:

[0077] Redundancy metrics for feature sets:

[0078]

[0079] Where N represents the number of sampling points, and |S| represents the total number of neurons in the l-th hidden layer, i.e. the total number of deep features; |S| represents the weight connecting the q-th neuron in the l-th hidden layer and the j-th neuron in the (l+1)-th hidden layer. l+1 This represents the total number of neurons in the (l+1)th hidden layer; I(Y (p) ,Y (q) ) represents the feature set Mutual information between different features, Y (p) and Y (q) These represent different types of features, and the calculation formulas are as follows:

[0080]

[0081] Correlation index between feature set and category:

[0082]

[0083] Where I(Y) (q) ,T (i) T represents the mutual information between features and categories. (i) This indicates the target category to which the i-th sampling point belongs.

[0084]

[0085] For the input raw time-domain dataset After undergoing targeted noise reduction processing using specific noise reduction strategies, it becomes "damaged data". Mapped to hidden layer Y and reconstructed into output Z:

[0086]

[0087] Z = g θ′ (Y)=g(W′Y+b′) (10)

[0088] To make the reconstructed output signal as close as possible to the input signal, one term of the training loss function is the average error, expressed as follows:

[0089]

[0090] Where N represents the number of sample points, L MSE (X (i) Z (i) ) represents the input time-domain data X (i) With the reconstructed signal Z (i) The mean square error between them.

[0091] Based on the above introduction, the loss function of the improved temporal stacked denoising autoencoder is rewritten as follows:

[0092] J total =J+λ1J MIR -λ2J MIT (12)

[0093] For the input frequency domain dataset After undergoing targeted noise reduction processing using specific noise reduction strategies, it becomes "damaged data". Mapped to hidden layer Y and reconstructed into output Z:

[0094]

[0095] Z = g θ′ (Y)=g(W′Y+b′) (14)

[0096] To make the reconstructed output signal as close as possible to the input signal, one term of the training loss function is the average error, expressed as follows:

[0097]

[0098] Where N represents the number of sample points, L MSE (F (i) Z (i) ) represents the input frequency domain data F (i) With the reconstructed signal Z (i) Mean square error between

[0099] In summary, the loss function of the improved frequency-domain stacked denoising autoencoder is rewritten as follows:

[0100] J total =J td +λ1J MIR -λ2J MIT(16)

[0101] Here, λ1 and λ2 are hyperparameters. Adjusting the size of these hyperparameters maintains the balance of terms in the loss function, thereby improving model performance. After training, a deep temporal feature set is finally learned. and depth frequency domain feature set m T and m F These represent the number of types of deep time-domain features and the number of types of deep frequency-domain features, respectively.

[0102] S4, the deep time-domain features learned by the first feature learning model and the deep frequency-domain features learned by the second feature learning model are merged into a deep feature set and fed into the preset model for training to establish a fault diagnosis model.

[0103] In this embodiment, the model used is the Extreme Learning Machine algorithm, and the specific steps are as follows:

[0104] The deep temporal feature set and the deep frequency feature set are combined and used as input to the Extreme Learning Machine (ELM), which is then mapped to the hidden layer H of the ELM.

[0105]

[0106] Among them, U=[u1,u2,…,u L ] T ∈R L×n It is the weight matrix connecting the input layer and the hidden layer, b = [b1, b2, ... b2]. L ] T ∈R L This refers to the neuron bias in the hidden layer. The classifier's output T can be expressed as:

[0107] Hβ=T (18)

[0108] Where β is the weight matrix connecting the hidden layer and the output layer, and the training objective of the classifier is to minimize the objective function:

[0109]

[0110] Where ξ i It is the training error vector.

[0111] S5. Using the trained first feature learning model, second feature learning model, and fault diagnosis model, the bearing to be diagnosed is diagnosed based on the vibration signal of the bearing to be diagnosed, and the diagnosis result is obtained.

[0112] Specifically, the implementation process of S5 is as follows: after extracting time domain data and frequency domain data from the test set data, the time domain data and frequency domain data are respectively fed into the improved stacked noise reduction autoencoder for feature learning, and the learned deep feature test set is fed into the fault diagnosis model for classification to achieve fault diagnosis.

[0113] In summary, this embodiment addresses the limitation of using single-domain data in rotating machinery component fault diagnosis by proposing an improved integrated stacked denoising autoencoder bearing fault diagnosis method that integrates time-domain and frequency-domain data. Since noise in the signal affects fault diagnosis performance, an anomaly detection-based denoising strategy is designed for the encoding stage, guiding the integrated model to selectively denoise the time-domain and frequency-domain data, thus improving the model's noise resistance. Two mutual information-based metrics are embedded into the loss function as constraints for network training, extracting multi-domain deep features with minimal redundancy and maximum correlation to the target class, eliminating the feature selection step independent of network training and shortening the diagnosis time.

[0114] Second Embodiment

[0115] This embodiment provides an electronic device, such as... Figure 3 As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0116] Below, in conjunction with Figure 3 A detailed introduction to each component of this electronic device is provided below:

[0117] The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0118] In a specific implementation, as one example, the processor may include one or more CPUs, for example... Figure 3 CPU0 and CPU1 shown are, of course, merely illustrative examples.

[0119] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0120] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or may exist independently, and may be accessed through the interface circuit of the electronic device (…). Figure 3 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0121] The transceiver may include a receiver and a transmitter. Figure 3 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and is connected through the interface circuit of the electronic device (…). Figure 3 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0122] In addition, it should be noted that, Figure 3 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0123] Third Embodiment

[0124] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0125] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0126] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0128] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0129] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0130] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0131] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0132] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part 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 the present 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.

[0133] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A bearing fault diagnosis method based on an improved integrated stack noise-reducing self-encoder, characterized in that, The bearing fault diagnosis method based on the improved integrated stacked noise reduction autoencoder includes: The vibration signal of the bearing is collected, and the time domain data and frequency domain data of the vibration signal are extracted. The stacked denoising autoencoder is improved for the time domain data and the frequency domain data respectively, resulting in a first feature learning model for extracting deep temporal features of the time domain data and a second feature learning model for extracting deep frequency domain features of the frequency domain data. The time-domain data is fed into a first feature learning model for training to learn deep time-domain features; the frequency-domain data is fed into a second feature learning model for training to learn deep frequency-domain features. The deep time-domain features learned by the first feature learning model and the deep frequency-domain features learned by the second feature learning model are combined into a deep feature set and fed into a preset model for training to establish a fault diagnosis model. Using the trained first feature learning model, second feature learning model, and fault diagnosis model, the bearing to be diagnosed is fault diagnosed based on the vibration signal of the bearing to be diagnosed, and the diagnosis result is obtained. The improvement of the stacked denoising autoencoder for the time-domain data and the frequency-domain data respectively yields a first feature learning model for extracting deep temporal features from the time-domain data and a second feature learning model for extracting deep frequency-domain features from the frequency-domain data, including: A metric based on mutual information is designed to measure the relationship between features and embedded into the loss function of the deep network as a soft constraint for training the deep network to learn low-redundancy deep features. At the same time, a metric based on mutual information is designed to measure the relationship between features and class labels and embedded into the loss function of the deep network as a soft constraint for training the deep network to learn deep features that are highly correlated with the class for fault classification. The time-domain data is fed into a first feature learning model for training to learn deep time-domain features, including: Anomaly detection is performed on the time-domain data at a rate of 5%, and the detected anomalies are set to 0 as corrupted data, which is used to train the first feature learning model to learn deep time-domain features. The frequency domain data is fed into a second feature learning model for training to learn deep frequency domain features, including: The frequency domain data is randomly set to 0 at a rate of 5% to be considered corrupted data, which is then used to train the second feature learning model to learn deep frequency domain features.

2. The bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder as described in claim 1, characterized in that, The process of acquiring the vibration signal of the bearing and extracting the time-domain and frequency-domain data of the vibration signal includes: Vibration signals of the bearing are collected, and the collected vibration signals are divided into several segments based on framing and windowing techniques as time-domain data. Then, the time-domain data is divided into training set and test set according to a preset ratio. Perform a Fast Fourier Transform on the time-domain data in the training set to convert it into frequency-domain data.

3. The bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder as described in claim 1, characterized in that, The loss function of the first feature learning model Represented as: ； in, and These are hyperparameters. By adjusting the size of the hyperparameters, the balance of each term in the loss function is maintained, thereby improving the performance of the model. ； in, Indicates the number of sampling points It is time-domain data With reconstructed signal The mean square error between them; ； ； in, Indicates the first l The total number of neurons in the hidden layer, i.e., the total number of deep features; , Indicates the connection of the first l The first hidden layer q The first neuron and the second l +1 hidden layer j The weights of each neuron, Representing the l +1 Total number of neurons in the hidden layer; , Representation feature set Mutual information between different features and They represent different types of characteristics; This represents the mutual information between features and categories. Indicates the first i The target category to which each sampling point belongs.

4. The bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder as described in claim 3, characterized in that, The loss function of the second feature learning model Represented as: ； ； in, N Indicates the number of sample points; Representing frequency domain data With reconstructed signal The mean square error between them.

5. The bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder as described in claim 1, characterized in that, The preset model is an Extreme Learning Machine model.

6. The bearing fault diagnosis method based on an improved integrated stacked noise-reducing autoencoder as described in claim 1, characterized in that, The method utilizes the trained first feature learning model and second feature learning model, along with the fault diagnosis model, to perform fault diagnosis on the bearing under test based on its vibration signal, obtaining the diagnosis result, including: The vibration signal of the bearing to be diagnosed is collected. Based on framing and windowing techniques, the collected vibration signal is divided into several segments as the time domain data of the bearing to be diagnosed. The time domain data of the bearing to be diagnosed is then subjected to a fast Fourier transform to obtain the frequency domain data corresponding to the time domain data of the bearing to be diagnosed. The time-domain data of the bearing to be diagnosed is input into the trained first feature learning model, and the trained first feature learning model is used to extract the deep time-frequency features corresponding to the time-domain data of the bearing to be diagnosed. The frequency domain data of the bearing to be diagnosed is input into the trained second feature learning model, and the trained second feature learning model is used to extract the deep frequency domain features corresponding to the frequency domain data of the bearing to be diagnosed. The extracted deep time-frequency features and deep frequency domain features are merged, and the merged features are input into the trained fault diagnosis model. The fault diagnosis results of the bearing to be diagnosed are obtained using the trained fault diagnosis model.

Images