Rolling bearing fault diagnosis method and system

Abstract

The present invention discloses a rolling bearing fault diagnosis method and system, relating to the field of rolling bearing fault diagnosis. The method comprises obtaining a current one-dimensional time series signal of the rolling bearing; determining a corresponding frequency domain signal and envelope spectrum signal based on the current one-dimensional time series signal; and determining a diagnosis result using a rolling bearing fault diagnosis model based on an improved DenseNet and deep feature fusion based on the current one-dimensional time series signal, the corresponding frequency domain signal, and the envelope spectrum signal. The present invention can improve the accuracy of rolling bearing fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of rolling bearing fault diagnosis, and in particular to a rolling bearing fault diagnosis method and system. Background Art

[0002] Rolling bearings are prone to failures such as fracture, wear, and fatigue spalling during operation, which can lead to property losses and even casualties. Therefore, timely diagnosis of rolling bearing failures is of great significance to safe production.

[0003] Currently, most fault diagnosis methods use single-domain signal information for deep learning, such as time, frequency, or time-frequency domains. However, due to interference from factors such as noise, different fault types have varying sensitivities to signals in different domains, which can result in poor robustness of the diagnostic model. Therefore, research on fault diagnosis methods that utilize input from multiple feature domains is of great significance.

[0004] With the advancement of computer technology, networks such as convolutional neural networks, deep belief networks, and long short-term memory networks have rapidly emerged. Convolutional neural networks have been widely used due to their powerful learning capabilities. However, as the network becomes deeper, it can lead to network degradation. Therefore, many researchers have experimented with adding skip connections between convolutional layers. In particular, DenseNet utilizes dense connections to maximize information between layers within the network, alleviating learning issues while also enabling feature reuse. While DenseNet offers significant advantages, it also has significant drawbacks. First, each layer of the network derives feature maps from previous layers by combining them, failing to consider the interdependencies between different channels. Second, rolling bearings are susceptible to noise during operation, and using deeper networks can easily introduce noise, which can affect feature extraction. Third, DenseNet's complex structure with numerous layers results in slow computational speed, making it difficult to quickly identify faults and provide timely feedback, leading to machine downtime and property damage. Fourth, the network consists of dense blocks alternately connected with transition layers. The dense blocks are only responsible for feature extraction, and the size of the feature maps between layers remains unchanged. The feature maps need to be reduced in dimension in the transition layers, resulting in redundancy in most of the calculations of the dense blocks.

[0005] Therefore, based on the above problems, there is an urgent need for a new rolling bearing fault diagnosis method or system to improve the accuracy of rolling bearing fault diagnosis. Summary of the Invention

[0006] The purpose of the present invention is to provide a rolling bearing fault diagnosis method and system, which can improve the accuracy of rolling bearing fault diagnosis.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A rolling bearing fault diagnosis method, comprising:

[0009] Obtain the current one-dimensional time series signal of the rolling bearing;

[0010] Determine the corresponding frequency domain signal and envelope spectrum signal respectively according to the current one-dimensional time series signal;

[0011] Based on the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal, a rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion is used to determine the diagnosis result; the rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion includes: a convolution layer, an improved DenseNet, a deep feature fusion unit, a tiling layer, a fully connected layer and a softmax classification layer; the diagnosis result includes: normal state, inner race fault, outer race fault or rolling element fault;

[0012] The process of determining the diagnosis result of the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion is as follows:

[0013] The current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal are respectively extracted through the convolution layer to obtain the corresponding shallow features;

[0014] The shallow features are extracted through the improved DenseNet to obtain the corresponding deep features; the improved DenseNet includes: an improved dense block and a transition layer; the improved dense block introduces a pooling layer and a multi-scale convolution kernel into the dense block; a squeeze excitation module is added after each convolution layer in the improved dense block;

[0015] The deep features are weightedly fused through the attention mechanism in the deep feature fusion unit to obtain global features;

[0016] The global features are sequentially passed through the flattening layer, the fully connected layer and the softmax classification layer to output the results and obtain the diagnosis results.

[0017] Optionally, the determining the corresponding frequency domain signal and envelope spectrum signal respectively according to the current one-dimensional time series signal specifically includes:

[0018] performing standardization processing on the current one-dimensional time series signal;

[0019] Perform Fourier transform on the current one-dimensional time series signal after normalization to determine the corresponding frequency domain signal;

[0020] Perform envelope spectrum analysis on the current one-dimensional time series signal after normalization to determine the corresponding envelope spectrum signal.

[0021] Optionally, the method of determining the diagnosis result by adopting a rolling bearing fault diagnosis model based on an improved DenseNet and deep feature fusion according to the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal may further include:

[0022] Obtain one-dimensional time series data of different state types of rolling bearings;

[0023] Determine the frequency domain signal and envelope spectrum signal corresponding to the one-dimensional time series data of different state types respectively;

[0024] Determine a data set based on one-dimensional time series data of different state types and corresponding frequency domain signals and envelope spectrum signals;

[0025] Divide the dataset into training, validation, and test sets;

[0026] Use the training set to train a rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion;

[0027] The validation set and test set were used to verify and test the trained rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion respectively.

[0028] Optionally, the loss function of the rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion is a cross entropy loss function.

[0029] Optionally, the cross entropy loss function is:

[0030]

[0031] Among them, H(p,q) is the cross entropy loss function, p(x j ) is the true label value of the j-th class input data, q(x j ) is the predicted probability value of the j-th category input data, and N is the number of categories.

[0032] Optionally, the step of performing weighted fusion of deep features through an attention mechanism in a deep feature fusion unit to obtain global features specifically includes:

[0033] Calculate the weight of each deep feature through multi-level feature dimensionality reduction;

[0034] Determine the importance weight of each deep feature;

[0035] The global features are determined based on the weights of the layer features and the importance weights of the deep features.

[0036] A rolling bearing fault diagnosis system, comprising:

[0037] A one-dimensional time series signal acquisition module is used to obtain the current one-dimensional time series signal of the rolling bearing;

[0038] A transform domain signal determination module, configured to determine the corresponding frequency domain signal and envelope spectrum signal according to the current one-dimensional time series signal;

[0039] A diagnosis result determination module is configured to determine a diagnosis result based on the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal using a rolling bearing fault diagnosis model based on an improved DenseNet and deep feature fusion; the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion includes: a convolutional layer, an improved DenseNet, a deep feature fusion unit, a tiling layer, a fully connected layer, and a softmax classification layer; the diagnosis result includes: normal state, inner race fault, outer race fault, or rolling element fault;

[0040] The process of determining the diagnosis result of the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion is as follows:

[0041] The current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal are respectively extracted through the convolution layer to obtain the corresponding shallow features;

[0042] The shallow features are extracted through the improved DenseNet to obtain the corresponding deep features; the improved DenseNet includes: an improved dense block and a transition layer; the improved dense block introduces a pooling layer and a multi-scale convolution kernel into the dense block; a squeeze excitation module is added after each convolution layer in the improved dense block;

[0043] The deep features are weightedly fused through the attention mechanism in the deep feature fusion unit to obtain global features;

[0044] The global features are sequentially passed through the flattening layer, the fully connected layer and the softmax classification layer to output the results and obtain the diagnosis results.

[0045] Optionally, the transform domain signal determination module specifically includes:

[0046] A standardization processing unit, configured to perform standardization processing on the current one-dimensional time series signal;

[0047] A Fourier transform unit is used to perform Fourier transform on the current one-dimensional time series signal after normalization to determine the corresponding frequency domain signal;

[0048] The envelope spectrum analysis unit is used to perform envelope spectrum analysis on the current one-dimensional time series signal after standardization to determine the corresponding envelope spectrum signal.

[0049] A rolling bearing fault diagnosis system comprises: at least one processor, at least one memory and computer program instructions stored in the memory, and the method described is implemented when the computer program instructions are executed by the processor.

[0050] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0051] The present invention provides a rolling bearing fault diagnosis method and system, which utilizes multi-domain signals to perform rolling bearing fault diagnosis. Compared with the deep learning algorithm of single-dimensional data, it overcomes the problem of single information of the original rolling bearing signal. The multi-domain signal can simultaneously extract valuable information from multiple transform domains, minimize information loss, provide more important feature information for fault diagnosis, and enable the model to maintain good accuracy under noisy and variable working conditions. An improved DenseNet is used to extract deep features, the pooling layer is integrated with the dense block, and multi-scale convolution kernels and squeeze excitation module (SEM) are introduced. This not only promotes feature reuse and reduces the dimension of the feature map, but also automatically calculates the importance weight of each feature channel, enhances the useful features of the current task and suppresses useless features, improves the accuracy of data classification, has good generalization ability, accelerates network convergence, improves the model's fitting ability and variable working condition diagnosis performance; the deep features of multiple signals are weighted and fused through the attention mechanism method, thereby ensuring the accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A schematic flow chart of a rolling bearing fault diagnosis method provided by the present invention;

[0054] Figure 2 A schematic diagram of the process of determining a rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion provided by the present invention;

[0055] Figure 3 Schematic diagram of the rolling bearing data center dataset test bench at Case Western Reserve University in the United States;

[0056] Figure 4 This is a schematic diagram of the rolling bearing fault diagnosis model structure based on improved DenseNet and deep feature fusion adopted by the present invention;

[0057] Figure 5 Schematic diagram of the improved DenseNet structure in the method of the present invention;

[0058] Figure 6 This is a SEM diagram of the squeeze excitation module for improving DenseNet in the method of the present invention;

[0059] Figure 7 Schematic diagram of the attention mechanism feature fusion method in the method of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] The purpose of the present invention is to provide a rolling bearing fault diagnosis method and system, which can improve the accuracy of rolling bearing fault diagnosis.

[0062] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] like Figure 1 and Figure 2 As shown, the present invention provides a rolling bearing fault diagnosis method, comprising:

[0064] S101, obtaining the current one-dimensional time series signal of the rolling bearing;

[0065] S102 , determining corresponding frequency domain signals and envelope spectrum signals respectively according to the current one-dimensional time series signal; that is, obtaining a plurality of transform domain signals.

[0066] S102 specifically includes:

[0067] The current one-dimensional time series signal is normalized to convert all data points into values between 0 and 1. The normalization formula is as follows:

[0068]

[0069] Among them, x i is the value of a point in the one-dimensional time series data, x min with x max are the minimum and maximum values in one-dimensional time series data;

[0070] After that, the current one-dimensional time series signal after normalization is expanded to multiple feature domains, such as Figure 4 As shown, the original one-dimensional time series data is subjected to Fourier transform and envelope spectrum analysis to obtain frequency signal and envelope spectrum signal, and the two signals together with the time domain signal constitute the input of the model.

[0071] Perform Fourier transform on the current one-dimensional time series signal after normalization to determine the corresponding frequency domain signal;

[0072] Perform envelope spectrum analysis on the current one-dimensional time series signal after normalization to determine the corresponding envelope spectrum signal.

[0073] S103, determining a diagnosis result using a rolling bearing fault diagnosis model based on an improved DenseNet and deep feature fusion based on the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal; the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion includes: a convolutional layer, an improved DenseNet, a deep feature fusion unit, a tiling layer, a fully connected layer, and a softmax classification layer; the diagnosis result includes: normal state, inner race fault, outer race fault, or rolling element fault;

[0074] The process of determining the diagnosis result of the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion is as follows:

[0075] The current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal are respectively extracted through the convolution layer to obtain the corresponding shallow features;

[0076] The shallow features are extracted through the improved DenseNet to obtain the corresponding deep features; the improved DenseNet includes: an improved dense block and a transition layer; the improved dense block introduces a pooling layer and a multi-scale convolution kernel into the dense block; a squeeze excitation module is added after each convolution layer in the improved dense block;

[0077] The deep features are weightedly fused through the attention mechanism in the deep feature fusion unit to obtain global features;

[0078] The specific process of obtaining global features is:

[0079] Calculate the weight of each deep feature through multi-level feature dimensionality reduction;

[0080] Determine the importance weight of each deep feature;

[0081] The global features are determined based on the weights of the layer features and the importance weights of the deep features.

[0082] The global features are sequentially passed through the flattening layer, the fully connected layer and the softmax classification layer to output the results and obtain the diagnosis results.

[0083] Before S103, it also included:

[0084] Obtain one-dimensional time series data of different state types of rolling bearings; Figure 3 The data shown above comes from the Rolling Bearing Data Center at Case Western Reserve University in the United States. The data uses drive-end bearing status data collected at a frequency of 12kHz. The data includes ten bearing status types: normal, outer race fault, inner race fault, and rolling element fault. The damage diameters for outer race, inner race, and rolling element faults are 7, 14, and 21 mils, respectively. 100,000 data points are collected for each bearing status, for a total of 1,000,000 data points. This dataset only contains one bearing status type at a time, and there is no concurrent data.

[0085] Determine the frequency domain signal and envelope spectrum signal corresponding to the one-dimensional time series data of different state types respectively;

[0086] 500 samples are randomly extracted from each bearing state of the original signal, with 1024 points in each sample. The one-dimensional original signal is then preprocessed to obtain the frequency domain signal and envelope spectrum signal. Due to the symmetry of the sampling frequency, the sample data points of the frequency domain signal and envelope spectrum signal are only half the sample length of the original signal, and the number of sample data points is 512. A total of 5000 samples are extracted from the three signals, for a total of 15,000 samples. The sample composition information of the time domain signal is shown in Table 1.

[0087] Table 1 Sample composition information table of timing signal

[0088]

[0089]

[0090] Determine a data set based on one-dimensional time series data of different state types and corresponding frequency domain signals and envelope spectrum signals;

[0091] The dataset is divided into training set, validation set and test set; the training set, validation set and test set account for 60%, 20% and 20% respectively.

[0092] Use the training set to train a rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion;

[0093] The training process is:

[0094] First, the convolutional layer uses 32 large convolution kernels of size 1×64, with strides of 16, 8, and 8 for the three signals, respectively, to extract shallow signal features. The improved DenseNet extracts deeper signal features, primarily consisting of four convolutional layers and six pooling layers. The first three convolutional layers use 1×3 and 1×5 convolution kernels with a stride of 2 and 16 kernels each. The last convolutional layer uses a 1×3 convolution kernel with a stride of 2 and 32 kernels. The network introduces pooling layers and multi-scale convolution kernels within dense blocks, and adds SEM after each convolution layer to adjust channel weights and enhance feature representation.

[0095] Secondly, the attention mechanism is used in the deep feature fusion unit to perform weighted fusion of the deep features of the three signals;

[0096] Then, the fused feature map is flattened into a one-dimensional vector, input into the softmax output layer through the fully connected layer, and the predicted probability value of the data is obtained;

[0097] Finally, the error between the true value and the predicted value is calculated by the cross entropy loss function, and the gradient descent algorithm is used to perform back propagation calculation on the error, update the model parameters, and repeat the above steps. When the number of training iterations reaches the maximum number, the model training ends and the trained model parameters are saved. Figure 4 The specific process is as follows:

[0098] The convolution layer uses 32 large convolution kernels of size 1×64, and the step sizes of the three signals are 16, 8, and 8 respectively, to extract the shallow features of the signal. The expression of the convolution process is as follows:

[0099]

[0100] in, is the output of the convolutional layer, is the input of the convolutional layer, is the convolutional layer weight, is the convolution layer bias, * is the convolution operation, f is the ReLU nonlinear activation function, and the expression is Rulu(x)=max(0,x).

[0101] Improvements to DenseNet mainly include improvements to dense blocks and SEM, which are used to extract deep features of signals.

[0102] The improved dense block is improved from the DenseNet network. The DenseNet network is mainly composed of dense blocks and transition layers. The dense block is used to extract features and fuse multi-layer features. The pooling layer reduces the dimension of the feature map. However, the network has many shortcomings. Based on this, the present invention introduces the pooling layer into the dense block and proposes an improved dense block structure, such as Figure 5 As shown in the figure, it mainly includes 4 convolutional layers and 6 pooling layers. The first three convolutional layers use 1×3 and 1×5 convolution kernels, with a stride of 2 and 16 convolution kernels respectively. The last convolutional layer uses a 1×3 convolution kernel with a stride of 2 and 32 convolution kernels. This structure inherits the advantages of the DenseNet network while reducing the complexity of the model, improving the training speed, and enhancing the network's noise resistance.

[0103] SEM is connected to the improved dense block convolution layer to adjust the channel weights and enhance the interdependence between different channels, such as Figure 6 As shown, it is helpful to improve the diagnostic capability under variable working conditions.

[0104] The deep feature fusion unit is to fuse the deep features extracted by the improved DenseNet. The main processes of the feature weighted fusion algorithm using the attention mechanism include Excitation, Proportion and Fusion. Figure 7 As shown, the specific process is as follows:

[0105] Excitation: Calculate the feature weight u through multi-level feature dimensionality reduction i

[0106] u i =σ(w j h i +b j );

[0107] Among them, h i is the depth feature corresponding to the time domain, frequency domain and envelope spectrum signal, i is the number of features, w j and b j are weights and biases, σ is the sigmoid nonlinear activation function

[0108] Proportion: Calculate the importance weight a of each feature in the feature map i .

[0109]

[0110] Where n is the number of features in each feature map.

[0111] Fusion: Perform weighted fusion operations on deep features to obtain global features f.

[0112]

[0113] The calculation formula of the fully connected layer is as follows:

[0114]

[0115] in, is the value of the hth neuron in the l+1th layer, is the value of the hth neuron in the lth layer, and bl are the weights and biases of the fully connected layer, and f is the ReLU nonlinear activation function.

[0116] The softmax layer is mainly used to calculate the predicted probability value of the bearing status. The softmax expression is as follows:

[0117]

[0118] Among them, P(z m ) is the predicted probability value of the mth class, z m is the value of the mth output neuron of the fully connected layer, and N is the number of categories.

[0119] The expression of the cross entropy loss function is as follows:

[0120]

[0121] Among them, p(x j ) is the true label value of the jth class, q(x j ) is the predicted probability value of the jth class, and N is the number of classifications.

[0122] Use the validation set to verify the trained network model, output the network model parameters, and use the test set to test the recognition accuracy of the model.

[0123] In order to test the accuracy of the model, it is compared with DenseNet, ResNet and multi-scale DenseNet. The network accuracy is shown in Table 2. It can be seen from the table that the proposed method has the highest accuracy, reaching 99.772%.

[0124] Table 2 Recognition accuracy of different methods / %

[0125]

[0126]

[0127] The method of the present invention can effectively fuse multiple transform domain information, increasing the diversity and integrity of information.

[0128] Moreover, the improved Densenet not only simplifies the model and reduces training time, but also reduces noise interference and improves the accuracy of rolling bearing fault diagnosis. While realizing multi-domain information fusion, it also realizes the fusion of multi-layer features, accelerates network convergence, and improves the model's fitting ability and variable working condition diagnosis performance.

[0129] Corresponding to the above method, the present invention further provides a rolling bearing fault diagnosis system, comprising:

[0130] A one-dimensional time series signal acquisition module is used to obtain the current one-dimensional time series signal of the rolling bearing;

[0131] A transform domain signal determination module, configured to determine the corresponding frequency domain signal and envelope spectrum signal according to the current one-dimensional time series signal;

[0132] A diagnosis result determination module is configured to determine a diagnosis result based on the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal using a rolling bearing fault diagnosis model based on an improved DenseNet and deep feature fusion; the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion includes: a convolutional layer, an improved DenseNet, a deep feature fusion unit, a tiling layer, a fully connected layer, and a softmax classification layer; the diagnosis result includes: normal state, inner race fault, outer race fault, or rolling element fault;

[0133] The process of determining the diagnosis result of the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion is as follows:

[0134] The current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal are respectively extracted through the convolution layer to obtain the corresponding shallow features;

[0135] The shallow features are extracted through the improved DenseNet to obtain the corresponding deep features; the improved DenseNet includes: an improved dense block and a transition layer; the improved dense block introduces a pooling layer and a multi-scale convolution kernel into the dense block; a squeeze excitation module is added after each convolution layer in the improved dense block;

[0136] The deep features are weightedly fused through the attention mechanism in the deep feature fusion unit to obtain global features;

[0137] The global features are sequentially passed through the flattening layer, the fully connected layer and the softmax classification layer to output the results and obtain the diagnosis results.

[0138] The transform domain signal determination module specifically includes:

[0139] A standardization processing unit, configured to perform standardization processing on the current one-dimensional time series signal;

[0140] A Fourier transform unit is used to perform Fourier transform on the current one-dimensional time series signal after normalization to determine the corresponding frequency domain signal;

[0141] The envelope spectrum analysis unit is used to perform envelope spectrum analysis on the current one-dimensional time series signal after standardization to determine the corresponding envelope spectrum signal.

[0142] In order to execute the method corresponding to the above-mentioned embodiment 1 to achieve the corresponding functions and technical effects, the present invention also provides a rolling bearing fault diagnosis system, including: at least one processor, at least one memory and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, the described method is implemented.

[0143] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0144] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A rolling bearing fault diagnosis method, characterized in that: include: Obtain the current one-dimensional time series signal of the rolling bearing; Determine the corresponding frequency domain signal and envelope spectrum signal respectively according to the current one-dimensional time series signal; According to the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal, the rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion is used to determine the diagnosis result; The rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion includes: a convolution layer, an improved DenseNet, a deep feature fusion unit, a tiling layer, a fully connected layer, and a softmax classification layer; the diagnosis results include: normal state, inner race fault, outer race fault, or rolling element fault; The process of determining the diagnosis result of the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion is as follows: The current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal are respectively extracted through the convolution layer to obtain the corresponding shallow features; The shallow features are extracted through the improved DenseNet to obtain the corresponding deep features; the improved DenseNet includes: an improved dense block and a transition layer; the improved dense block introduces a pooling layer and a multi-scale convolution kernel into the dense block; a squeeze excitation module is added after each convolution layer in the improved dense block; The deep features are weightedly fused through the attention mechanism in the deep feature fusion unit to obtain global features; The global features are sequentially passed through the flattening layer, the fully connected layer and the softmax classification layer to output the results and obtain the diagnosis results.

2. A rolling bearing fault diagnosis method according to claim 1, characterized in that: The determining of the corresponding frequency domain signal and envelope spectrum signal according to the current one-dimensional time series signal specifically includes: performing standardization processing on the current one-dimensional time series signal; Perform Fourier transform on the current one-dimensional time series signal after normalization to determine the corresponding frequency domain signal; Perform envelope spectrum analysis on the current one-dimensional time series signal after normalization to determine the corresponding envelope spectrum signal.

3. A rolling bearing fault diagnosis method according to claim 1, characterized in that: The method of determining the diagnosis result by adopting the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion according to the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal also includes: Obtain one-dimensional time series data of different state types of rolling bearings; Determine the frequency domain signal and envelope spectrum signal corresponding to the one-dimensional time series data of different state types respectively; Determine a data set based on one-dimensional time series data of different state types and corresponding frequency domain signals and envelope spectrum signals; Divide the dataset into training, validation, and test sets; Use the training set to train a rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion; The validation set and test set were used to verify and test the trained rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion respectively.

4. A rolling bearing fault diagnosis method according to claim 3, characterized in that: The loss function of the rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion is a cross entropy loss function.

5. A rolling bearing fault diagnosis method according to claim 4, characterized in that: The cross entropy loss function is: Among them, H(p,q) is the cross entropy loss function, p(x j ) is the true label value of the j-th class input data, q(x j ) is the predicted probability value of the j-th category input data, and N is the number of categories.

6. A rolling bearing fault diagnosis method according to claim 1, characterized in that: The deep features are weightedly fused using the attention mechanism in the deep feature fusion unit to obtain global features, specifically including: Calculate the weight of each deep feature through multi-level feature dimensionality reduction; Determine the importance weight of each deep feature; The global features are determined based on the weights of the layer features and the importance weights of the deep features.

7. A rolling bearing fault diagnosis system, characterized in that: include: A one-dimensional time series signal acquisition module is used to obtain the current one-dimensional time series signal of the rolling bearing; A transform domain signal determination module, configured to determine the corresponding frequency domain signal and envelope spectrum signal according to the current one-dimensional time series signal; The diagnosis result determination module is used to determine the diagnosis result based on the current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal, using the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion; The rolling bearing fault diagnosis model based on improved DenseNet and deep feature fusion includes: a convolution layer, an improved DenseNet, a deep feature fusion unit, a tiling layer, a fully connected layer, and a softmax classification layer; the diagnosis results include: normal state, inner race fault, outer race fault, or rolling element fault; The process of determining the diagnosis result of the rolling bearing fault diagnosis model based on the improved DenseNet and deep feature fusion is as follows: The current one-dimensional time series signal and the corresponding frequency domain signal and envelope spectrum signal are respectively extracted through the convolution layer to obtain the corresponding shallow features; The shallow features are extracted through the improved DenseNet to obtain the corresponding deep features; the improved DenseNet includes: an improved dense block and a transition layer; the improved dense block introduces a pooling layer and a multi-scale convolution kernel into the dense block; a squeeze excitation module is added after each convolution layer in the improved dense block; The deep features are weightedly fused through the attention mechanism in the deep feature fusion unit to obtain global features; The global features are sequentially passed through the flattening layer, the fully connected layer and the softmax classification layer to output the results and obtain the diagnosis results.

8. A rolling bearing fault diagnosis system according to claim 7, characterized in that: The transform domain signal determination module specifically includes: A standardization processing unit, configured to perform standardization processing on the current one-dimensional time series signal; A Fourier transform unit is used to perform Fourier transform on the current one-dimensional time series signal after normalization to determine the corresponding frequency domain signal; The envelope spectrum analysis unit is used to perform envelope spectrum analysis on the current one-dimensional time series signal after standardization to determine the corresponding envelope spectrum signal.

9. A rolling bearing fault diagnosis system, characterized in that: include: At least one processor, at least one memory, and computer program instructions stored in the memory, which implement the method according to any one of claims 1 to 6 when the computer program instructions are executed by the processor.

Images