JAAN-based rolling bearing remaining service life prediction method

Abstract

The invention discloses a JAAN-based rolling bearing remaining service life prediction method. The method comprises the following steps of obtaining a data set of a bearing complete life cycle; calculating an RMS value from the complete life cycle data set to obtain a feature vector; normalizing the RMS value, taking m samples in the feature vectors as training samples, and taking the residual vectors as prediction samples; reconstructing a multivariable matrix; constructing a JAAN prediction model, wherein a plurality of JAAN memory units are stacked in the model; inputting the training sample into a JAAN prediction model to train the model; and inputting a test sample into the trained JAAN prediction model to analyze an estimation result. The method has the advantages that different information can be dynamically emphasized by embedding the attention mechanism and the JAAN memory unit, the prediction capacity of the network is improved, the prediction model is constructed based on the JAAN, network parameters are initialized for RUL estimation of the bearing, higher robustness is achieved, and the RUL value more accurate than that of an existing method can be provided.

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation and maintenance of mechanical equipment, and in particular to a method for predicting the remaining service life of a rolling bearing based on JAAN. Background Art

[0002] In the era of the Internet of Things and Industry 4.0, rolling bearings play a vital role in most industrial machinery. The remaining useful life (RUL) of a rolling bearing is a key indicator that determines whether the machinery is in good operating condition or in a faulty state. Currently, existing RUL estimation methods are mainly divided into two categories: model-based methods and data-driven methods. Model-based methods have inherent limitations. The degradation trends of complex machines often show nonlinearity and diversity, which makes it difficult to construct accurate physical models in practical engineering applications. Data-driven methods have achieved remarkable results in RUL estimation. End-to-end RUL prediction methods based on deep learning use raw monitoring signals as input to the network for prediction, reducing human interference and becoming a new approach to addressing the shortcomings of traditional methods. Among them, RNN and its improved methods such as LSTM, Bi-LSTM, and GDAU have been applied in the field of end-to-end RUL prediction of rolling bearings. Based on this, the JANET model was proposed by simplifying the GRU memory cell. However, this model does not weight different degradation features when predicting raw feature inputs. Different features have equal importance to the network prediction. When the rolling bearing performance degradation stage is different and the features have different effects on prediction, JANET has difficulty in automatically weighing the input features, and the prediction accuracy needs to be improved. Therefore, the attention mechanism is introduced into the JANET cell to design a new memory cell JAAN, which helps to improve the accuracy of rolling bearing remaining useful life prediction. Summary of the Invention

[0003] The purpose of the present invention is to provide a JAAN-based method for predicting the remaining service life of rolling bearings, which is accurate, efficient and robust in predicting the remaining service life of rolling bearings and can effectively improve the accuracy of the remaining service life prediction of rolling bearings.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] The JAAN-based method for predicting the remaining service life of rolling bearings includes the following steps:

[0006] S1. Obtain a data set of the complete life cycle of a bearing;

[0007] S2. Calculate the RMS value from the complete life cycle data set and obtain the feature vector X = [x1, x2, x3, ..., x n ] T ;

[0008] S3, normalize the RMS value and use the m samples in the feature vector as training samples T train =[y1,y2,y3,…,y m ] T , the remaining vector is used as the prediction sample Y test [y m+1 ,y m+2 ,y m+3 ,…,y n ] T ;

[0009] S4, reconstruct the multivariate matrix;

[0010] S5. Build a JAAN prediction model, where the model is stacked with multiple JAAN memory units.

[0011] S6. Inputting the training samples into the JAAN prediction model to train the model;

[0012] S7. Input the test sample into the trained JAAN prediction model to analyze the estimation results.

[0013] Furthermore, in step S4, Y train The multivariate matrix is ​​as follows:

[0014]

[0015] The first k feature vectors β are input into the model to predict β k+1 The value of

[0016] β k+1 =f(β1,β2,…,β k )

[0017] Y test The multivariate matrix is ​​as follows:

[0018]

[0019] Here, k represents the number of units in the input layer.

[0020] Furthermore, in step S5, the JAAN memory unit includes an update gate and an attention gate. The update gate is used to update the unit state at the current time, and the output of the update gate is used as the input of the attention gate. The JAAN memory unit update process is as follows:

[0021] f t =σ(W xf x t +W hf h t-1 +b f )

[0022]

[0023] A(x t ,h t-1 )=v T ⊙tanh(W xa x t +W ha h t-1 )

[0024]

[0025]

[0026]

[0027] h t-1 =c t

[0028] Among them, f t is the forget gate, x t is the input signal, A(x t ,h t-1 ) is the output of the attention gate, v T is the initialized matrix vector, W xf 、W hf 、W xc 、W hc 、W xa 、W ha 、 and is the weight matrix of the corresponding gate in the JAAN memory unit, b f 、b c 、 is the bias of the corresponding gate in the JAAN memory unit, α t is the probability distribution value of the attention gate, c t is the hidden state of the memory unit, c t-1 is the previous hidden state of the memory cell, is the candidate memory unit hidden state, is the candidate cell hidden state of the attention gate, h t-1 and h t are the hidden layer outputs of the previous moment and the current moment respectively.

[0029] Furthermore, in step S5, the JAAN prediction model is constructed, and the input signal X=[x1,x2,…,x k ] Input to the first layer of JAAN memory unit to extract the temporal features H (1) for:

[0030]

[0031] Where f(·) is the JAAN memory unit update process, Network parameters for the JAAN memory unit;

[0032] The temporal features H extracted by the n-th layer JAAN (n) for:

[0033]

[0034] The features extracted by the nth layer of JAAN are input into the fully connected layer, and the output of the network is:

[0035] Z=g(W o H (n) +b o )

[0036] Among them, g(W o H (n) +b o ) represents the activation function of the fully connected layer, W O and b o are the weight matrix and bias term corresponding to the fully connected layer respectively;

[0037] Finally, Z is input into the linear regression layer to calculate the predicted value The calculation process is:

[0038]

[0039] Among them, W Y Represents the network weights of the linear regression layer.

[0040] Furthermore, in step S6, the training process of the JAAN prediction model is implemented by the SGD optimization algorithm, and the mean square error function is used as the loss function:

[0041]

[0042] Where n is the length of the training sample. i and represent the true value and predicted value at time t respectively.

[0043] Furthermore, in step S7, the remaining service life of the bearing is:

[0044] RUL t =FLC-t

[0045] Among them, t is the running time of the rolling bearing, FLC is the full life cycle of the bearing, RUL t is the remaining life of the bearing after running to time t.

[0046] Furthermore, in step S7, the score and root mean square error are used to obtain the RUL prediction performance:

[0047]

[0048]

[0049] in, and y i They represent the predicted remaining useful life and the actual remaining useful life respectively, and m represents the length of the sample.

[0050] The present invention has the following advantages:

[0051] By embedding the attention mechanism, the JAAN memory unit can dynamically emphasize different information and improve the prediction ability of the network. Based on JAAN, a prediction model is built and the network parameters are initialized for RUL estimation of bearings, which has higher robustness and can provide more accurate RUL values ​​than existing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the process of the present invention.

[0053] Figure 2 This is a vibration signal diagram of the present invention running to the fault data set.

[0054] Figure 3 This is a diagram of the JAAN memory unit.

[0055] Figure 4 This is the RMS prediction diagram of the bearing of the present invention.

[0056] Figure 5 This is the RUL prediction diagram of the bearing of the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0058] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are also within the scope of protection of the present invention.

[0059] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other.

[0060] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0061] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, or are the orientations or positional relationships in which the inventive product is typically placed when in use, or are the orientations or positional relationships commonly understood by those skilled in the art. These terms are intended only to facilitate the description of the present invention and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0062] In the description of the present invention, it should also be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0063] refer to Figure 1-5 As shown, one embodiment of the present invention is:

[0064] The JAAN-based method for predicting the remaining service life of rolling bearings includes the following steps:

[0065] S1. Obtain a dataset of the complete life cycle of a bearing.

[0066] The dataset was collected from a PRONOSTIA experimental system. Two accelerometers and a radial force sensor were installed on the test bench to measure run-to-failure experiments. The vibration signal was repeated every 10 seconds with a sampling frequency of 25.6 kHz. Figure 2 The vibration profile of the bearing over its entire service life is shown. It can be seen that the dataset remains stable at the beginning, which is not useful for bearing RUL prediction. Therefore, the RMS values ​​of the bearings (1150 to 2725) were selected to verify the reliability of the proposed JAAN. The speed was 1800 RPM and the load was 4000 N.

[0067] S2. Calculate the RMS value from the complete life cycle data set and obtain the feature vector X = [x1, x2, x3, ..., x n ] T .

[0068] Specifically, degradation trends are extracted from the run-to-failure dataset to estimate future RUL.

[0069] S3, normalize the RMS value and use the m samples in the feature vector as training samples Y train [y1, y2, y3, …, y m ] T , the remaining vector is used as the prediction sample Y test [y m+1 ,y m+2 ,y m+3 ,…,y n ] T .

[0070] S4. Reconstruct the multivariate matrix.

[0071] To X train Normalize and get Y train =[y1,y2,y3,…,y m ] T Reconstruction, Y train The multivariate matrix is ​​as follows:

[0072]

[0073] The first k feature vectors β are input into the model to predict β k+1 The value of

[0074] β k+1 =f(β1,β2,…,β k )

[0075] Similarly, for X test Normalize and get Y test Reconstruction, Y test The multivariate matrix is ​​as follows:

[0076]

[0077] Here, k represents the number of units in the input layer.

[0078] S5. Build a JAAN prediction model, where the model is stacked with multiple JAAN memory units.

[0079] The JAAN memory unit includes an update gate and an attention gate. The update gate is used to update the unit state at the current time, and the output of the update gate is used as the input of the attention gate. The JAAN memory unit is as follows: Figure 3 As shown, the update process is as follows:

[0080] f t =σ(W xf x t +W hf h t-1 +b f )

[0081]

[0082] A(x t ,h t-1 )=v T ⊙tanh(W xa x t +W ha h t-1 )

[0083]

[0084]

[0085]

[0086] h t-1 =c t

[0087] Among them, f t is the forget gate, x t is the input signal, A(x t ,h t-1 ) is the output of the attention gate, v T is the initialized matrix vector, W xf 、W hf 、W xc 、W hc 、W xa 、W ha 、 and is the weight matrix of the corresponding gate in the JAAN memory unit, b f 、b c 、 is the bias of the corresponding gate in the JAAN memory unit, α t is the probability distribution value of the attention gate, c t is the hidden state of the memory unit, c t-1 is the previous hidden state of the memory cell, is the candidate memory unit hidden state, is the candidate cell hidden state of the attention gate, h t-1 and h t are the hidden layer outputs of the previous moment and the current moment respectively.

[0088] There are two gates in JAAN, including an update gate and an attention gate. The update gate is used to update the cell state at the current time. The output of the update gate is used as the input of the attention gate. The attention gate can help memory cells quickly discard redundant information and increase attention to important information, thereby improving the network estimation performance of the proposed method.

[0089] In constructing the JAAN model, the input signal X=[x1,x2,…,x k ] Input to the first layer of JAAN memory unit to extract the temporal features H (1) for:

[0090]

[0091] Where f(·) is the JAAN memory unit update process, Network parameters for the JAAN memory unit;

[0092] The temporal features H extracted by the n-th layer JAAN (n) for:

[0093]

[0094] The features extracted by the nth layer of JAAN are input into the fully connected layer, and the output of the network is:

[0095] Z=g(W o H (n) +b o )

[0096] Among them, g(W o H (n) +b o ) represents the activation function of the fully connected layer, W O and b o are the weight matrix and bias term corresponding to the fully connected layer respectively;

[0097] Finally, Z is input into the linear regression layer to calculate the predicted value The calculation process is:

[0098]

[0099] Among them, W Y Represents the network weights of the linear regression layer.

[0100] In the configuration of JAAN, the number of input units is selected as 100, the number of hidden units is selected as 80, the number of fully connected layer units is selected as 100, the number of output nodes is selected as 1, the learning rate is selected as 0.01, and the number of training times is selected as 200.

[0101] S6. Input the training samples into the JAAN prediction model to train the model.

[0102] The training process is implemented by the SGD optimization algorithm, and the mean square error function (MSE) is used as the loss function:

[0103]

[0104] Where n is the length of the training sample. i and represent the true value and predicted value at time t respectively.

[0105] S7. Input the test sample into the trained JAAN prediction model to analyze the estimation results.

[0106] Remaining bearing service life:

[0107] RUL t =FLC-t

[0108] Among them, t is the running time of the rolling bearing, FLC is the full life cycle of the bearing, RUL t is the remaining life of the bearing after running to time t.

[0109] To evaluate the prognostic ability, the score, root mean square error (RMSE) was used to obtain the RUL prediction performance, which was defined as follows:

[0110]

[0111]

[0112] in, and y i Represent the predicted remaining useful life and the actual remaining useful life respectively, and m represents the length of the sample

[0113] In this embodiment, a real industry experiment on short-term RMS prediction is conducted. In the experiment, the entire dataset is used as training samples, and JAAN is applied to predict the remaining samples. Figure 4 The predicted RMS values ​​of the bearing are shown. We can observe that as the deterioration of the bearing performance increases, the RMS value also increases. The training and predicted values ​​are very close to the actual values ​​in the short term, and their trends are similar. Therefore, the proposed method can effectively predict future RMS values.

[0114] By using the training samples as Importing to train the proposed JAAN, the corresponding RUL can be obtained. Figure 5 The RUL estimation results are shown. It can be observed that the predicted RUL values ​​obtained by JAAN gradually converge to the actual RUL values. This shows that JAAN's RUL estimation error is very small, especially during the later stages of operation. This is very helpful for maintenance in the actual engineering industry.

[0115] To verify the effectiveness of the proposed JAAN, we used GAU, JANET, DSCN, and MCNN for comparison. Two performance evaluation metrics were used, including Score and MAPE. The results are shown in the following table:

[0116]

[0117] As shown in the table above, the scores of the five methods are 184.73, 152.04, 165.67, 179.15, and 143.45, respectively. The method proposed in this paper achieves the highest score. The MAPE of JAAN is 12.42, GAU is 13.12, JANET is 15.44, DSCN is 13.66, and MCNN is 21.67. JAAN proposed in this paper significantly outperforms the other methods. This demonstrates that our method provides more accurate estimation results and is more suitable for practical RUL prediction of rolling bearings.

[0118] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the remaining service life of rolling bearings based on JAAN, characterized by: The following steps are involved: S1, obtain the data set of the complete life cycle of the bearing; S2. Calculate the RMS value from the complete life cycle data set to obtain the feature vector X = [x1, x2, x3, ..., x n ] T ; S3, normalize the RMS value and use the m samples in the feature vector as training samples Y train =[y1,y2,y3,…,y m ] T , the remaining vector is used as the prediction sample Y test [y m+1 ,y m+2 ,y m+3 , …, y n ] T ; S4, reconstruct the multivariate matrix; S5. Build a JAAN prediction model, where the model is stacked with multiple JAAN memory units; S6. Input the training samples into the JAAN prediction model to train the model; S7. Input the test sample into the trained JAAN prediction model to analyze the estimation results.

2. The method for predicting the remaining useful life of a rolling bearing based on JAAN according to claim 1, characterized in that: In the step S4, Y train The multivariate matrix is ​​as follows: The first k feature vectors β are input into the model to predict β k+1 The value of β k+1 ＝f(β1，β2，…，β k ) Y test The multivariate matrix is ​​as follows: Where k represents the number of units in the input layer.

3. The method for predicting the remaining useful life of a rolling bearing based on JAAN according to claim 1, characterized in that: In step S5, the JAAN memory unit includes an update gate and an attention gate. The update gate is used to update the unit state at the current time, and the output of the update gate is used as the input of the attention gate. The JAAN memory unit update process is as follows: f t =σ(W xf x t +W hf h t-1 +b f ) A(x t ,h t-1 )=v T ⊙tanh(W xa x t +W ha h t-1 ) h t-1 =c t Among them, f t is the forget gate, x t is the input signal, A(x t ,h t-1 ) is the output of the attention gate, v T is the initialized matrix vector, W xf , W hf , W xc , W hc , W xa , W ha , and is the weight matrix of the corresponding gate in the JAAN memory unit, b f 、b c , is the bias of the corresponding gate in the JAAN memory cell, α t is the probability distribution value of the attention gate, c t is the hidden state of the memory unit, c t-1 is the previous hidden state of the memory cell, is the candidate memory unit hidden state, is the candidate cell hidden state of the attention gate, h t-1 and h t are the hidden layer outputs of the previous moment and the current moment respectively.

4. The method for predicting the remaining useful life of a rolling bearing based on JAAN according to claim 3, characterized in that: In the step S5, the JAAN prediction model is constructed, and the input signal X=[x1, x2,…, x k ] is input into the first layer of JAAN memory unit and extracted into the temporal feature H (1) for: Where f(·) is the JAAN memory unit update process, is the network parameter of JAAN memory unit; the temporal feature H extracted by JAAN at the nth layer (n) for: The features extracted by the nth layer of JAAN are input into the fully connected layer, and the output of the network is: Z=g(W o H (n) +b o ) Among them, g(W o H (n) +b o ) represents the activation function of the fully connected layer, W O and b o They are the weight matrix and bias term corresponding to the fully connected layer respectively; Finally, Z is input into the linear regression layer to calculate the predicted value The calculation process is: Among them, W Y Represents the network weights of the linear regression layer.

5. The method for predicting the remaining useful life of a rolling bearing based on JAAN according to claim 1, characterized in that: In step S6, the training process of the JAAN prediction model is implemented by the SGD optimization algorithm, and the mean square error function is used as the loss function: Where n is the length of the training sample. i and Represent the true value and predicted value at time t respectively.

6. The method for predicting the remaining useful life of a rolling bearing based on JAAN according to claim 5, characterized in that: In step S7, the remaining service life of the bearing is: NUMBER t =FLC-t Among them, t is the running time of the rolling bearing, FLC is the full life cycle of the bearing, RUL t is the remaining life of the bearing after running to time t.

7. The method for predicting the remaining useful life of a rolling bearing based on JAAN according to claim 1, characterized in that: In step S7, the RUL prediction performance is obtained using the score and root mean square error: in, and i They represent the predicted remaining useful life and the actual remaining useful life respectively, and m represents the length of the sample.

Images