A time-varying rotating speed bearing fault diagnosis method and device

Abstract

The application belongs to the technical field of variable speed bearing fault, and particularly relates to a time-varying rotating speed bearing fault diagnosis method and device. The application obtains a fault type result by inputting bearing operation data in a prediction time period into a time-varying rotating speed bearing fault diagnosis model; wherein, the operation data comprises vibration data and rotating speed pulse data; the time-varying rotating speed bearing fault diagnosis model comprises two channels and a classifier; the first channel is used for obtaining time fault information of the operation data, the second channel is used for obtaining space fault information of the operation data, and the classifier is used for combining the outputs of the first channel and the second channel to complete fault diagnosis. The time-varying rotating speed bearing fault diagnosis method of the application comprehensively considers the time information and space information of the input signal, enhances the mutual dependence between time sequence data, can obtain the space and time fault information of the input signal at the same time, and improves the accuracy of fault diagnosis of the time-varying rotating speed bearing.

Description

Technical Field

[0001] This invention belongs to the field of variable speed bearing fault technology, specifically relating to a method and device for diagnosing time-varying speed bearing faults. Background Technology

[0002] Rolling bearings play a crucial role in various industrial machinery. Due to the complexity of their operating conditions, they are also frequently prone to failure. Therefore, timely fault diagnosis is extremely important for the normal operation of mechanical equipment. The operating environment of rolling bearings is often complex and constantly changing. These operating conditions lead to changes in the vibration characteristics of rolling bearings, thus reducing the effectiveness of diagnostics in practical applications. Notably, speed variations are quite common. Considering torque pulsation, vibration loads, and speed fluctuations, rolling bearings often operate at speeds that vary over time in practical applications. Therefore, faults under variable speed conditions are difficult to detect and can cause unpredictable losses.

[0003] For fault diagnosis of time-varying speed bearings, the variation in speed causes changes in the vibration signal spectrum, increasing the difficulty of feature extraction and selection. At different speeds, fault features may exhibit different frequencies and amplitudes; therefore, traditional frequency domain analysis methods are used to diagnose time-varying speed bearings.

[0004] A Chinese invention patent application with publication number CN114112398A discloses a method for diagnosing rolling bearing faults under variable speed conditions. This method utilizes a deep learning-based convolutional neural network to diagnose rolling bearing faults under varying speeds by employing horizontal and vertical vibration signals from the rolling bearing at different speeds. The convolutional neural network extracts fault features through alternating convolutional and pooling layers. However, this method fails to consider external interference that may occur during actual operation, such as noise from surrounding equipment and the environment, and electromagnetic interference. These external interferences can obscure or mask the original signal, leading to signal distortion and affecting the accuracy and reliability of signal processing. Furthermore, this method only considers the spatial fault features of the input signal, which may fail to accurately identify fault feature frequencies, resulting in the inability to accurately extract fault features from time-varying speed bearings. This increases the difficulty of fault diagnosis and leads to inaccurate diagnostic results. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for diagnosing time-varying speed bearing faults, in order to solve the problem that the existing bearing fault diagnosis methods are inaccurate in diagnosing time-varying speed bearings.

[0006] To address the aforementioned technical problems, this invention provides a time-varying speed bearing fault diagnosis method. The method includes: acquiring bearing operating data for a predicted time period, the operating data including vibration data and speed pulse data; and inputting the acquired operating data into a trained time-varying speed bearing fault diagnosis model to obtain fault type results. The time-varying speed bearing fault diagnosis model includes two channels and a classifier; the first channel is used to acquire temporal fault information from the operating data, the second channel is used to acquire spatial fault information from the operating data, and the classifier combines the outputs of the first and second channels to complete fault diagnosis.

[0007] Furthermore, the first channel includes a gated loop unit for acquiring timing information of the running data to enhance the interdependence between data.

[0008] Furthermore, the second channel includes a multi-scale wide kernel layer, which is used to extract multi-scale features from the running data through at least two wide kernel layers of different sizes, and then concatenate the extracted multi-scale features.

[0009] Furthermore, the second channel also includes a second wide kernel layer, which is used to process the input operating data. The processing includes constructing new signals to distinguish more fault information and high-frequency noise suppression processing. The processed operating data is then input to the multi-scale wide kernel layer so that the multi-scale wide kernel layer can perform multi-scale feature extraction on the processed operating data.

[0010] Furthermore, the second channel also includes a small kernel convolution module, which is used to reduce the number of neurons output by multi-scale wide kernel layers. The small kernel convolution module includes a small kernel convolution unit or at least two small kernel convolution units connected in sequence. The small kernel convolution unit includes a small kernel convolution layer and a pooling layer. The size of the convolution kernel in the small kernel convolution layer is smaller than the size of the convolution kernel in the wide kernel layer.

[0011] Furthermore, the first channel also includes a first wide kernel layer, which is used to process the input operating data. The processing includes constructing new signals to distinguish more fault information and high-frequency noise suppression processing, and then inputting the processed operating data to the gated loop unit so that the gated loop unit can obtain the time information of the processed operating data.

[0012] Furthermore, the number of the wide core layers of different sizes is three.

[0013] Furthermore, the specific formula for calculating the activation function used in the small kernel convolutional unit is as follows:

[0014]

[0015] Where α is a positive hyperparameter, e x is an exponential function, x is the input value, and ELU is the activation function.

[0016] To address the aforementioned technical problems, the present invention also provides a time-varying speed bearing fault diagnosis device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the time-varying speed bearing fault diagnosis method described above.

[0017] Its beneficial effects are as follows: This invention is an improved invention. It utilizes bearing operating data under different speed conditions to complete bearing fault diagnosis through a time-varying speed bearing fault diagnosis model. This operating data includes vibration data and speed pulse data. The time-varying speed bearing fault diagnosis model includes two channels and a classifier. The first channel is used to acquire temporal fault information from the operating data, and the second channel is used to acquire spatial fault information from the operating data. The classifier combines the outputs of the first and second channels to complete the fault diagnosis. The time-varying speed bearing fault diagnosis method of this invention comprehensively considers the temporal and spatial information of the input signal, enhances the interdependence between time-series data, and can simultaneously obtain spatial and temporal fault information of the input signal, thus improving the accuracy of fault diagnosis for time-varying speed bearings. Attached Figure Description

[0018] Figure 1 This is a flowchart of a time-varying speed bearing fault diagnosis method according to an embodiment of the present invention;

[0019] Figure 2-1 This is a schematic diagram of the multi-scale module convolution process according to an embodiment of the present invention;

[0020] Figure 2-2 This is a schematic diagram of the multi-scale module max pooling process in an embodiment of the present invention;

[0021] Figure 2-3 This is a schematic diagram of the multi-scale module average pooling process in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the GRU part of an embodiment of the present invention;

[0023] Figure 4 This is a t-SNE classification effect diagram of time-varying speed data of the present invention for three fault types under noise-free conditions;

[0024] Figure 5 This is a t-SNE classification effect diagram of time-varying speed data under strong noise in an embodiment of the present invention;

[0025] Figure 6-1This is a t-SNE classification result of an existing CNN model on three fault types under strong noise for time-varying speed data;

[0026] Figure 6-2 This is a t-SNE classification result of the existing WDCNN model on three fault types under strong noise for time-varying speed data;

[0027] Figure 6-3 This is a t-SNE classification effect of the existing GRU model on three fault types of time-varying speed data under strong noise.

[0028] Figure 6-4 This is a t-SNE classification result of the existing GRU-WDCNN model on three fault types in time-varying speed data under strong noise. Detailed Implementation

[0029] This invention acquires bearing operating data at a predicted time and inputs this data into a trained time-varying speed bearing fault diagnosis model to obtain fault type results. The time-varying speed bearing fault diagnosis model includes two channels and a classifier. The first channel acquires temporal fault information from the operating data, the second channel acquires spatial fault information from the operating data, and the classifier combines the outputs of the first and second channels to complete fault diagnosis. This invention's time-varying speed bearing fault diagnosis method comprehensively considers both temporal and spatial information of the input signal, enhancing the interdependence between time-series data. It can simultaneously obtain both spatial and temporal fault information of the input signal, solving the problem of inaccurate diagnostic results for time-varying speed bearings in existing bearing fault diagnosis methods.

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0031] Method Implementation Examples:

[0032] A method for diagnosing time-varying speed bearing faults according to an embodiment of the present invention, the main process of which is as follows:

[0033] Step 1: Construct a fault diagnosis model for time-varying speed bearings.

[0034] like Figure 1As shown, the model mainly consists of two channels and a classifier. The first channel includes a gated recurrent unit (GRU) for acquiring temporal fault information from the input data, preceded by a first wide kernel layer to suppress high-frequency noise. The second channel includes a multi-scale wide kernel layer for acquiring spatial fault information from the input data; preceded by this multi-scale wide kernel layer is a second wide kernel layer to suppress high-frequency noise, followed by a small kernel convolutional module. This small kernel convolutional module comprises three sequentially connected small kernel convolutional units, each consisting of a small kernel convolutional layer and a pooling layer. The classifier combines the outputs of the first and second channels to perform fault diagnosis. The first and second channels are connected in parallel. The structure, function, and connections between each module are described in detail below:

[0035] (1) First wide-core layer and gated recurrent unit (GRU):

[0036] The operational data obtained by the fault acquisition module is processed through a constructed 64*1*16 first wide-core layer. This processing includes constructing new signals to distinguish more fault information and high-frequency noise suppression. The processed operational data is then input into a gated recurrent unit (GRU). The wide-core layer provides a larger receptive field, allowing it to see more input data and better capture the overall characteristics of the input signal, thus facilitating the construction of a more global new signal. The GRU is used to acquire the temporal information of the operational data, comprehensively considering the temporal information of the input signal to enhance the interdependence between time-series data. The structure diagram of the GRU is shown below. Figure 3 As shown, its specific implementation method is as follows:

[0037] X t =Wconv(input) (1)

[0038] Z t =σ(W z X t +V z H t-1 +b z (2)

[0039] R t =σ(W r X t +V r H t-1 +b r (3)

[0040]

[0041] Among them, X t Z represents the input sequence of the data.t and R t These represent two gates in the GRU module: the update gate and the reset gate, H. t The current hidden state of the input sequence. H represents the hidden states of the candidate input sequence. t-1 W represents the hidden state of the input sequence at the previous time step. z W represents the weight matrix of the updated gate. r W represents the weight matrix of the reset gate. c V represents the weight matrix in the hidden state; z V represents the cyclic connection weight matrix of the update gate. r V represents the cyclic connection weight matrix of the update gate. c b represents the weight matrix of the hidden layer with circular connections; z b r denoted by , σ and tanh represent the activation functions, ⊙ represents the dot product, and W represents the weight matrix.

[0042] (2) Second wide core layer and multi-scale wide core layer:

[0043] First, a 64*1*16 second wide kernel layer is constructed to process the input running data. This processing includes constructing new signals to distinguish more fault information and high-frequency noise suppression. The wide kernel layer provides a larger receptive field, allowing it to see more input data and better capture the overall features of the input signal, thus facilitating the construction of a more global new signal. Next, a multi-layered, multi-scale wide kernel layer is constructed below this second wide kernel layer to extract multi-scale features from the running data processed by the second wide kernel layer, and these extracted multi-scale features are then concatenated. In this embodiment, the multi-scale wide kernel layer consists of three layers with kernel sizes of [100, 200, 300]. In other implementations, the number and size of the multi-scale wide kernel layer can be set according to actual needs.

[0044] (3) Mini-kernel convolution module:

[0045] A small-kernel convolutional module is placed after the multi-scale wide kernel layer. This small-kernel convolutional module includes three sequentially connected small-kernel convolutional units. Each small-kernel convolutional unit includes a small-kernel convolutional layer and a pooling layer, which is used to reduce the large number of neurons output by the multi-scale wide kernel layer. As another implementation, the number of small-kernel convolutional units can be adjusted according to actual needs.

[0046] The small kernel convolutional units have sizes of 8*1*8, 32*1*8, and 16*1*8. This not only reduces the number of neurons but also allows for the capture of more potential sensitive features. The specific implementation of the constructed multi-scale module is as follows:

[0047] Convolutional layers: such as Figure 2-1 The diagram shown illustrates the principle of multi-scale module convolution, assuming the input signal is x∈T. n and filter w∈T m The convolution process is described as follows:

[0048]

[0049] Where, x l b is the output feature of the l-th layer. l It is a bias term, z l Y is the linear activation vector of the l-th layer. l w refers to the feature vector after convolution. l The kernel size of the l-th layer is wl(i), where σ represents the kernel of the l-th layer; σ is the non-linear activation function; conv represents the convolution operation; same represents zero-padding; x l-1 The output features of layer l-1; (yl(t),…,yl(n-m+1))∈Tn-m+ 1 Indicates the convolution process, Represents the convolution process y l The calculation process of (t), x l-1 (t+i-1) is the signal of the (l-1)th layer.

[0050] Pooling layers: Typically located between two consecutive convolutional layers, their main function is to reduce the dimensionality of the output values ​​from the upper layer, maintaining a certain degree of consistency in the scale of output features while significantly reducing the amount of computational data. In this embodiment of the invention, max pooling or average pooling is used as the pooling layer, such as... Figure 2-2 , 2-3 As shown. It can be represented as:

[0051] x l =f(down(x) l-1 )+b l (7)

[0052] Where, x l-1 The previous layer's feature vector is referred to as "down", which means downsampling, and f represents the pooling operation.

[0053] Furthermore, to better extract effective features in noisy environments, a new activation function is introduced after each small convolutional layer and pooling layer. This activation function processes the input data, making the model more adaptable to changes in negative noise. The specific implementation is as follows:

[0054]

[0055] Where α is a positive hyperparameter, usually set to 1, ex is an exponential function, x is the input value, and ELU is the activation function.

[0056] (4) SoftMax classifier:

[0057] The SoftMax classifier performs fault diagnosis based on the outputs of the channels composed of the second wide kernel layer and the gated recurrent unit (GRU) and the multi-scale module channels, thus realizing time-varying speed bearing fault diagnosis and classification. Its specific implementation is as follows:

[0058]

[0059] Among them, z i Let be the output value of the i-th node, C be the number of output nodes, i.e., the number of categories. e is an exponential function.

[0060] Step 2: Obtain operating data of time-varying speed bearings with different fault types, including vibration data and speed pulse data; randomly divide these data into datasets for training and testing, and train and test the time-varying speed bearing fault diagnosis model constructed in Step 1.

[0061] In this embodiment, different fault types and speeds are addressed by replacing bearings with different fault types. Accelerometers and speed encoders are used to collect vibration data and speed pulse data, respectively. The multiple signals collected from these bearings with different fault types at different speeds constitute the required dataset. The different speeds include four states: continuously increasing speed (e.g., speed range 0-1500 r / min), continuously decreasing speed (e.g., speed range 1500-0 r / min), speed first increasing then decreasing (e.g., speed range 0-1500-0 r / min), and speed first decreasing then increasing (e.g., speed range 1500-0-1500 r / min). The different fault types used in this embodiment for training and testing the time-varying speed bearing fault diagnosis model include: normal state, inner ring fault, and outer ring fault.

[0062] The time-varying speed bearing fault diagnosis method of this invention is specifically designed for time-varying speeds. These time-varying speed signals are more complex, significantly increasing the difficulty of fault diagnosis. However, precisely because of this complexity, the method is more closely aligned with common working conditions in practice, improving the practicality and accuracy of fault diagnosis.

[0063] The vibration data from the acquired operational data was named Channel_1, and the velocity pulse data from the acquired operational data was named Channel_2. Training data was created using the acquired vibration and velocity pulse data, with a window length of 2048 data points and a step size of 64, to train the time-varying speed bearing fault diagnosis model.

[0064] In order to make full use of the data source, this embodiment introduces dual signal inputs, namely Channel_1 and Channel_2. This dual signal input design enables the network to independently extract fault features from vibration and velocity pulse signals, effectively improving the model's representation ability.

[0065] Step 3: Obtain the bearing's operating data within the predicted time period, and use the time-varying speed bearing fault diagnosis model obtained in Step 2 to diagnose the bearing fault.

[0066] like Figure 4 The image shows the t-SNE classification results of the method of the present invention for three fault types under noise-free conditions using the provided time-varying speed data. Each sample is visualized as a point, and fault types of the same type are represented by the same color. Figure 5 The figure shown is a t-SNE classification effect diagram of the method of the present invention for three fault types under a strong noise interference level of 6dB for the provided time-varying speed data.

[0067] like Figure 6-1 , 6-2 Figures 6-3 and 6-4 show the t-SNE classification results of time-varying speed data for three fault types under a strong noise level of 6dB, using conventional models, respectively. The model used in 6-1 is a CNN model, the model used in 6-2 is a WDCNN model, the model used in 6-3 is a GRU model, and the model used in 6-4 is a GRU-WDCNN model.

[0068] Compare Figure 4 , 5 As can be seen from 6-1, 6-2, 6-3, and 6-4, the method of the present invention targets the transmission data, and regardless of whether the transmission bearing is under noise interference or not, it can be seen that various faults are accurately separated, achieving accurate classification.

[0069] In summary, regardless of whether the variable speed bearing is under noise interference or not, the various faults diagnosed using the method of this invention are accurately separated. This demonstrates that the method of this invention can still meet practical application requirements in noisy environments and can effectively handle the fault diagnosis of bearings with time-varying speeds. Therefore, this invention considers both noise interference and time-varying speed conditions, making it more closely aligned with actual operating conditions and resulting in more accurate diagnostic results for bearings with time-varying speeds.

[0070] Device Example:

[0071] This invention discloses a time-varying speed bearing fault diagnosis device, comprising a memory, a processor, and an internal bus. The processor and memory communicate and interact with each other via the internal bus. The memory includes at least one software functional module stored in the memory. The processor executes various functional applications and data processing by running the computer program and module stored in the memory, thereby implementing the time-varying speed bearing fault diagnosis method described in the method embodiments of this invention. The principle, implementation process, and achievable effects of this method have been fully described in the method embodiments and will not be repeated here.

Claims

1. A method for diagnosing faults in a time-varying speed bearing, characterized in that, The method includes: acquiring bearing operating data for the predicted time period and inputting it into the trained time-varying speed bearing fault diagnosis model to obtain fault type results; the operating data includes vibration data and speed pulse data; The time-varying speed bearing fault diagnosis model includes two channels and a classifier; The first channel is used to acquire time-related fault information of the running data, including the first wide kernel layer and GRU; the first wide kernel layer is used to process the input running data and input it into the GRU; the GRU is used to acquire the time information of the running data to enhance the interdependence between data. The second channel is used to obtain spatial fault information of the running data, including a second wide kernel layer, a multi-scale wide kernel layer, and a small kernel convolution module; the second wide kernel layer is used to process the input running data and input it to the multi-scale wide kernel layer; the multi-scale wide kernel layer is used to extract multi-scale features from the running data through at least two wide kernel layers of different sizes and then concatenate them; the small kernel convolution module is used to reduce the number of neurons output by the multi-scale wide kernel layer, including a small kernel convolution unit or at least two small kernel convolution units connected in sequence, the small kernel convolution unit includes a small kernel convolution layer and a pooling layer, and the size of the convolution kernel in the small kernel convolution layer is smaller than the size of the convolution kernel in the wide kernel layer; The processing of both the first and second wide kernel layers includes constructing new signals to distinguish more fault information and suppressing high-frequency noise. The classifier is used to combine the outputs of the first and second channels to complete fault diagnosis.

2. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, The small kernel convolution module consists of three small kernel convolution units connected in sequence.

3. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, The pooling layer is either max pooling or average pooling.

4. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, The classifier is the SoftMax classifier.

5. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, Vibration data and velocity pulse data are collected by an accelerometer and a speed encoder, respectively.

6. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, Fault types include normal state, inner ring fault, and outer ring fault.

7. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, The number of wide core layers of different sizes is three.

8. The time-varying speed bearing fault diagnosis method according to claim 1, characterized in that, The specific formula for calculating the activation function used in the small kernel convolutional unit is as follows: in, It is a positive hyperparameter. It is an exponential function. is the input value, and ELU is the activation function.

9. A time-varying speed bearing fault diagnosis device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the time-varying speed bearing fault diagnosis method as described in claim 1.

10. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, The small kernel convolution module consists of three small kernel convolution units connected in sequence.

11. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, The pooling layer is either max pooling or average pooling.

12. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, The classifier is the SoftMax classifier.

13. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, The vibration data and velocity pulse data were collected by an accelerometer and a speed encoder, respectively.

14. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, Fault types include normal state, inner ring fault, and outer ring fault.

15. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, The number of wide core layers of different sizes is three.

16. The time-varying speed bearing fault diagnosis device according to claim 9, characterized in that, The specific formula for calculating the activation function used in the small kernel convolutional unit is as follows: in, It is a positive hyperparameter. It is an exponential function. is the input value, and ELU is the activation function.

Images