Method for predicting residual life of rolling bearing

Abstract

The invention discloses a rolling bearing residual life prediction method, which comprises the following steps of: firstly, dividing data detected by different mechanical equipment under different working conditions into source domain data and target domain data, and then constructing a training set and a verification set; carrying out batch standardization processing on the data; extracting time domain features and frequency domain features of the original vibration signals by using a deep convolutional neural network, and splicing the time domain features and the frequency domain features to form a replaceable feature set; outputting a full connection layer by using the deep time sequence characteristics, and comparing the deep time sequence characteristics with the RUL tag; and finally, repeatedly learning data features of different working conditions by using a large amount of source domain labeled data, calibrating corresponding values of the features and the RUL, constructing a mapping relationship between the features and the RUL, and then adding a Dropout layer, thereby constructing a model and checking in a source domain test set. The method can greatly reduce the consumption of manpower and material resources caused by the re-calibration of the label by the data, better excavates the internal degradation trend of the bearing, and effectively improves the prediction precision of the remaining service life.

Description

Technical Field

[0001] The present invention belongs to the field of mechanical component life prediction, and particularly relates to a method for predicting the remaining life of rolling bearings. Background Technique

[0002] With the rapid development of modern industry, the high integration of mechanical equipment has put forward higher requirements for the safety of the system. As one of the key basic components of mechanical equipment, according to relevant statistical data, more than 30% of the failures of rotating mechanical equipment are caused by bearing failures. Predictive and Health Management (PHM) of machines plays a crucial role in maintaining the stable operation of mechanical equipment. The degradation of mechanical equipment bearings is related to various factors, such as: operating mode, workload, material defects, and operating temperature. In addition, the long degradation period and random failure of bearings make the prediction of the remaining service life of bearings a challenging task. In view of the current problems such as different bearing operating conditions, less effective data, no data labels, and low prediction accuracy, a method for predicting the remaining life of rolling bearings is proposed, which can more effectively predict the remaining bearing life under different equipment and operating conditions. Summary of the Invention

[0003] In view of this, the present invention provides a method for predicting the remaining life of rolling bearings, which can solve the problem of low prediction accuracy in the prior art due to different bearing operating conditions, less effective data, and no data labels.

[0004] The object of the present invention can be achieved by the following technical solutions:

[0005] A method for predicting the remaining life of rolling bearings includes the following steps:

[0006] Step 1: Obtain the data detected under different operating conditions of different mechanical equipment, divide the data into source domain data and target domain data, and then construct a training set and a validation set based on the source domain data and the target domain data;

[0007] Step 2: Perform batch normalization processing on the source domain data and the target domain data;

[0008] Step 3: Use a deep convolutional neural network DCNN to extract the time-domain features and frequency-domain features of the original vibration signal, and splice the time-domain and frequency-domain features together to form a replaceable feature set;

[0009] Step 4: Utilize the deep time-series features, output a fully connected layer, and compare the deep time-series features with the RUL label;

[0010] Step 5: Repeatedly learn the data characteristics under different working conditions using a large amount of labeled data in the source domain, calibrate the corresponding values of the features and RUL, construct the mapping relationship between the features and RUL, and then add a Dropout layer to construct the source domain model and test it on the source domain test set;

[0011] Step 6: Use the source domain model and the unlabeled data set in the target domain to analyze the parameter characteristics of the source domain data and the target domain data, learn similar features, and minimize the distribution difference between the target domain and the source domain;

[0012] Step 7: Conduct transfer training based on the source domain model, update the parameters of the model, and train a transfer prediction model;

[0013] Step 8: Use the transfer prediction model to predict and verify the RUL of the rolling bearing for the signals under different working conditions of different devices.

[0014] Further, the source domain data and the target domain data described in step 1 include: acquiring the monitoring data under different working conditions of a certain mechanical equipment to construct the source domain, and dividing it into a training set and a validation set; constructing the target domain with the monitoring data under different working conditions of the same mechanical equipment or different mechanical equipment, and dividing it into a training set and a validation set.

[0015] Further, the training set and the validation set include: the training set of the source domain data has RUL labels, and the test set does not have RUL labels. The training set and the test set of the target domain data do not have RUL labels.

[0016] Further, the batch normalization process described in step 2 includes: mapping the source domain data and the target domain data into the range of 0 to 1 to ensure the stability of the data.

[0017] Further, the vibration original signals extracted by the deep convolutional neural network DCNN described in step 3 include: 11 time-domain indexes of vibration signals such as mean value, skewness, and kurtosis factor, and 12 frequency-domain indexes of vibration signals such as amplitude variance, amplitude average value, and spectral bandwidth. A total of 23 feature indexes are used to construct an optional feature set; the first 15 features with high scores are selected as the input of the RUL prediction model through a screening equation for predicting the remaining service life of the bearing; the screened time-domain indexes and frequency-domain indexes are standardized.

[0018] Further, in step 3, the deep convolutional neural network DCNN can gradually learn high-level semantic features from low-level edge and texture features through multiple layers of convolution and pooling operations. The expression of the convolution operation process is as follows:

[0019]

[0020] In the formula: is the weight of the i-th convolution kernel in the l-th layer; is the bias of the i-th convolutional kernel in the l-th layer; X l (j) is the j-th local area in the l-th layer; is the input of the j-th neuron in the operation result of the i-th convolutional kernel in the (l + 1)-th layer;

[0021] After the convolution operation, the ReLU function is used as the activation function of the model to accelerate the convergence speed of the model and prevent overfitting; the expression of the ReLU function is as follows:

[0022] f(x) = max{0, x}

[0023] In the deep convolutional neural network DCNN, a pooling layer is added after the convolutional layer to reduce the computational burden. There are two common pooling operations, namely average pooling and max pooling. The expressions of the two pooling operations are as follows;

[0024] v l(i,j) = max H(j-1)-1＜n＜jH {x l(i,n)}

[0025]

[0026] In the formula: l(i, n) is the activation value of the n-th neuron in the i-th channel in the l-th layer; H is the size of the sliding window; ν l(i,j) is the result after pooling of the j-th neuron in the current l-th layer;

[0027] The deep convolutional neural network DCNN is used to extract features from the original signal and a max pooling layer is adopted to improve the feature extraction speed. Then, the time-domain and frequency-domain features are concatenated to form an alternative feature set.

[0028] Furthermore, in step 7, a transfer learning bearing life prediction model is established. Transfer learning is carried out on the DCNN model, and the transfer prediction model is pre-trained with a large number of source domain datasets to ensure that the model learns as many data features under different working conditions as possible. At the same time, transfer learning is used to extract similar features of vibration signals under different working conditions and different devices to reduce the difference between the source domain and the target domain. Based on the DCNN model, a transfer model is re-trained by combining the learned similar features for the remaining life prediction in the target domain.

[0029] Furthermore, in step 8, after the bearing remaining life prediction is completed, its source domain test set and target domain test set need to be verified. For this, this method uses two indicators, the mean absolute error MAE and the root mean square error RMSE, to evaluate the prediction effect. The formulas are as follows:

[0030]

[0031]

[0032] Where: m is the number of prediction points; i is the serial number of the prediction point, and y i Actual value; Is the predicted value;

[0033] The value range of MAE is [0, +∞), the smaller the better. The smaller the RMSE value, the higher the accuracy.

[0034] Compared with the prior art, the beneficial effects of the rolling bearing remaining life prediction method provided by the present invention are as follows:

[0035] First, according to the data detected under different working conditions of different mechanical equipment, it is divided into source domain data and target domain data, and then a training set and a validation set are constructed based on the source domain data and the target domain data; secondly, the input data is batch-normalized; thirdly, a deep convolutional neural network (DCNN) is used to extract the time-domain features and frequency-domain features of the original vibration signal and splice the time-domain and frequency-domain features together to form an alternative feature set; thirdly, using deep temporal features, the output fully connected layer is used to compare the deep temporal features with the RUL label; finally, a large amount of source domain labeled data is used to repeatedly learn the data features of different working conditions, calibrate the corresponding values of the features and RUL, construct the mapping relationship between the features and RUL, and then add a Dropout layer to construct the model and test it on the source domain test set; the prediction model of the rolling bearing remaining life constructed by the DCNN network and transfer learning can effectively solve problems such as different bearing operating conditions, few effective data, unlabeled data, and low prediction accuracy.

[0036] Using the source domain model and the target domain unlabeled data set, analyze the parameters and features of the source domain data and the target data, learn similar features, and minimize the distribution difference between the target domain and the source domain; then perform transfer training based on the source domain model, update the parameters of the model, and train a transfer model; finally, use the transfer prediction model to predict and verify the RUL of the rolling bearing for signals of different equipment under different working conditions. Compared with the existing methods, this method can greatly reduce the consumption of manpower and material resources caused by re-calibrating the labels of data, better excavate the internal degradation trend of the bearing, and effectively improve the prediction accuracy of the remaining service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] By referring to the drawings and reading the following detailed description, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become easy to understand. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0038] Figure 1 Is the prediction flow chart of the present invention.

[0039] Figure 2 This is the DNCC network structure diagram of the present invention. Specific embodiments

[0040] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0041] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art to which the present invention belongs. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, the elements defined by the statement "including..." do not exclude the existence of additional identical elements in the process, method, article or device including the said elements.

[0042] As Figure 1 shown, it is a flowchart of a method for predicting the remaining life of a rolling bearing provided by the present invention. This method includes the following steps:

[0043] Step 1: According to the data detected under different working conditions of different mechanical equipment, divide it into source domain data and target domain data, and then construct a training set and a validation set according to the source domain data and the target domain data;

[0044] Step 2: Perform batch normalization processing on the input data;

[0045] Step 3: Use a deep convolutional neural network (DCNN) to extract the time-domain features and frequency-domain features of the original vibration signal and splice the time-domain and frequency-domain features together to form an alternative feature set;

[0046] Step 4: Utilize deep temporal features, output a fully connected layer, and compare the deep temporal features with the RUL label;

[0047] Step 5: Use a large amount of source domain labeled data to repeatedly learn the data features under different working conditions, calibrate the corresponding values of the features and the RUL, construct the mapping relationship between the features and the RUL, and then add a Dropout layer to construct a model and test it on the source domain test set;

[0048] Step 6: Using the source domain model and the target domain unlabeled data set, analyze the parameters and features of the source domain data and the target data, learn similar features, and minimize the distribution difference between the target domain and the source domain;

[0049] Step 7: Based on the source domain model, perform transfer training, update the parameters of the model, and train a transfer model;

[0050] Step 8: Through the transfer prediction model, predict the RUL of the rolling bearing for signals under different working conditions of different devices.

[0051] The following is a detailed description of the specific implementation manners of the above steps:

[0052] In Step 1, for the acquisition of this data, an acceleration sensor is used to collect vibration signals in the horizontal and vertical directions respectively. The sampling frequency of the data is 25.6 kHz, the data is recorded every 10 s, the acquisition duration is 0.1 s, and the vibration data collected each time is summarized; in the experiment, the cross-validation method is used for data division. According to different test data, it is divided into test data and training data, and then superimposed and input into the model for training.

[0053] In Step 2, the input data is normalized to map the data into the range of 0 to 1 to ensure the stability of the data. This method uses the Z-score normalization criterion to process the selected features. The formula of the Z-score normalization criterion is as follows:

[0054] Z = (X - μ) / σ

[0055] (In the formula: Z is the Z-score value, X is the original data, μ is the average value of the original data, and σ is the standard deviation of the original data)

[0056] In Step 3, as Figure 2 shown, Figure 2 shows the DNCC network structure diagram. The DCNN network model can gradually learn high-level semantic features from low-level edge, texture and other features through multiple layers of convolution and pooling operations. The expression of the convolution operation process is as follows:

[0057]

[0058] (In the formula: is the weight of the i-th convolution kernel in the l-th layer; is the bias of the i-th convolution kernel in the l-th layer; X l (j) is the j-th local area in the l-th layer; is the input of the j-th neuron in the operation result of the i-th convolution kernel in the (l + 1)-th layer)

[0059] After convolution operation, the ReLU function is usually used as the activation function of the model to accelerate the convergence speed of the model and prevent overfitting. The expression of the ReLU function is as follows:

[0060] f(x) = max{0, x}

[0061] In the DCNN network structure, a pooling layer is usually added after the convolutional layer to reduce the computational burden. There are two common pooling operations, namely average pooling and max pooling. The expressions of the two pooling operations are as follows;

[0062] v l(i,j) = max H(j-1)-1＜n＜jH {x l(i,n)}

[0063]

[0064] (where: l(i, n) is the activation value of the nth neuron in the ith channel of the lth layer; H is the size of the sliding window; v l (i,j) is the result after pooling of the jth neuron in the current lth layer)

[0065] Use the DCNN network to extract features from the original signal and adopt the max pooling layer to improve the feature extraction speed, and then concatenate the time-domain and frequency-domain features to form an alternative feature set.

[0066] In step 7, a transfer learning bearing life prediction model is established, transfer learning is carried out on the DCNN model, and the transfer prediction model is pre-trained through a large number of source domain data sets to ensure that the model learns as many data features under different working conditions as possible. At the same time, transfer learning is used to extract similar features of vibration signals under different working conditions and different devices to reduce the differences between the source domain and the target domain. Based on the DCNN model, a transfer model is re-trained by combining the similar features learned by transfer learning for the remaining life prediction in the target domain.

[0067] In step 8, after the bearing remaining life prediction is completed, its source domain test set and target domain test set need to be verified. For this, this method uses two indicators, mean absolute error (MAE) and root mean square error (RMSE), to evaluate the prediction effect. The formulas are as follows:

[0068]

[0069] (where: m is the number of prediction points; i is the serial number of the prediction point, y i is the actual value; is the predicted value)

[0070] The value range of MAE is [0, +∞), the smaller the better, and the smaller the RMSE value, the higher the accuracy.

[0071] In the present invention, first, according to the data detected under different working conditions of different mechanical equipment, it is divided into source domain data and target domain data, and then a training set and a validation set are constructed based on the source domain data and the target domain data; second, the input data is batch-normalized; third, a deep convolutional neural network (DCNN) is used to extract the time-domain features and frequency-domain features of the original vibration signal and splice the time-domain and frequency-domain features together to form an alternative feature set; fourth, using deep temporal features, the output of the fully connected layer is obtained, and the deep temporal features are compared with the RUL label; finally, a large amount of labeled data in the source domain is used to repeatedly learn the data features under different working conditions, calibrate the corresponding values of the features and RUL, construct the mapping relationship between the features and RUL, and then add a Dropout layer to construct a model and test it on the source domain test set; the prediction model of the remaining life of the rolling bearing constructed by the DCNN network and transfer learning can effectively solve the problems such as different bearing operating conditions, few effective data, unlabeled data, and low prediction accuracy.

[0072] Using the source domain model and the unlabeled data set in the target domain, analyze the parameters and features of the source domain data and the target data, learn the similar features, and minimize the distribution difference between the target domain and the source domain; then perform transfer training based on the source domain model, update the parameters of the model, and train a transfer model; finally, use the transfer prediction model to predict and verify the RUL of the rolling bearing for signals of different equipment under different working conditions. Compared with the existing methods, this method can greatly reduce the consumption of manpower and material resources caused by re-calibrating the labels of the data, better explore the internal degradation trend of the bearing, and effectively improve the prediction accuracy of the remaining service life.

[0073] Experimental process and verification: In order to verify the effectiveness of the rolling bearing remaining life prediction method, the following experimental process and test scheme are designed. The experimental process is divided into three parts: data acquisition, model training and verification, and result analysis.

[0074] 1. Data acquisition includes experimental equipment conditions and data recording: (1) Experimental equipment conditions: Build a corresponding bearing degradation test platform, select experimental bearing models (such as: 6007, 6008, 6009, etc.), arrange vibration acceleration with high sampling frequency in the radial direction of the bearing (horizontal and vertical directions), and at the same time set multiple groups of load and speed combinations to simulate the bearing degradation process under different working conditions; (2) Data recording: Record data once at a set time, collect the required number of vibration signals each time, and the obtained vibration signals include the normal state data in the initial stage of operation and the full life cycle data until the failure. The data is divided into source domain data (degraded signals with labels) and target domain data (unlabeled signals).

[0075] 2. Model Training and Validation: (1) Feature extraction includes time-domain features, frequency-domain features, and feature screening; (2) Model training: Use a deep convolutional neural network (DCNN) to perform feature learning on vibration signals, and set the network structure: including convolutional layers, each layer uses the ReLU activation function, the convolutional kernel size, the stride is [stride value], the pooling layer uses max pooling, and set the parameters of the Dropout layer to prevent overfitting; (3) Transfer learning: First, train an initial model based on the source domain data, use the unlabeled target domain data for transfer learning, extract the similar features between the source domain and the target domain, reduce the distribution difference, and update the parameters of the transfer model to obtain a transfer prediction model; (4) Validation test: Perform remaining useful life (RUL) predictions on the source domain test set and the target domain test set respectively, and use the calculated mean absolute error (MAE) and root mean square error (RMSE) for evaluation.

[0076] 3. Analysis of Test Results: (1) By comparing with traditional machine learning methods, the advantages of this method are reflected in that when under different working conditions, the MAE and RMSE values of the transfer learning model are better than those of traditional methods, especially in the test of unlabeled target domain data; the high-order semantic features extracted by the deep convolutional neural network can effectively reflect the degradation trend of the bearing and have stronger robustness; at the same time, the transfer learning method has stronger data adaptability between different working conditions and devices, and the feature differences between the source domain and the target domain can be effectively reduced, and finally the prediction accuracy can be improved.

[0077] As mentioned above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for predicting the remaining life of a rolling bearing, characterized in that, Including the following steps: Step 1: Obtain the data detected under different working conditions of different mechanical equipment, divide the data into source domain data and target domain data, and then construct a training set and a validation set based on the source domain data and the target domain data; Step 2: Perform batch normalization on the source domain data and the target domain data; Step 3: Use a deep convolutional neural network DCNN to extract the time-domain features and frequency-domain features of the original vibration signal, and splice the time-domain and frequency-domain features together to form an alternative feature set; Step 4: Utilize the deep temporal features, output a fully connected layer, and compare the deep temporal features with the RUL label; Step 5: Use a large amount of labeled data in the source domain to repeatedly learn the data features under different working conditions, calibrate the corresponding values of the features and RUL, construct the mapping relationship between the features and RUL, and then add a Dropout layer to construct the source domain model and test it on the source domain test set; Step 6: Utilize the source domain model and the target domain unlabeled data set to analyze the parameter features of the source domain data and the target domain data, learn similar features, and minimize the distribution difference between the target domain and the source domain; Step 7: Conduct transfer training based on the source domain model, update the parameters of the model, and train a transfer prediction model; Step 8: Predict and verify the RUL of the rolling bearing for the signals of different equipment under different working conditions through the transfer prediction model.

2. The method for predicting the remaining life of a rolling bearing according to claim 1, wherein, The source domain data and the target domain data described in Step 1 include: Obtain the monitoring data of a certain mechanical equipment under different working conditions to construct the source domain, and divide the training set and the validation set; The monitoring data of the same mechanical equipment or different mechanical equipment under different working conditions are used to construct the target domain, and the training set and the validation set are divided.

3. A method for predicting the remaining life of a rolling bearing according to claim 2, characterized in that, The training set and the validation set include: The training set of the source domain data has RUL labels, and the test set does not have RUL labels. The training set and the test set of the target domain data do not have RUL labels.

4. A method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that, The batch normalization process described in Step 2 includes: Map the source domain data and the target domain data to the range of 0 to 1 to ensure the stability of the data.

5. A method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that The original vibration signal extracted by the deep convolutional neural network DCNN described in Step 3 includes: 11 time-domain indicators of the vibration signal including mean, skewness, and kurtosis factor, 12 frequency-domain indicators of the vibration signal including amplitude variance, amplitude average, and spectral bandwidth, and a total of 23 feature indicators are used to construct an optional feature set; The first 15 features with high scores are selected as the input of the RUL prediction model through a screening equation for predicting the remaining service life of the bearing; The screened time-domain indicators and frequency-domain indicators are normalized.

6. A method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that In Step 3, the deep convolutional neural network DCNN can gradually learn high-level semantic features from low-level edge and texture features through multiple layers of convolution and pooling operations. The expression of the convolution operation process is as follows: Wherein: is the weight of the i-th convolution kernel in the l-th layer; is the bias of the i-th convolution kernel in the l-th layer; X l (j) is the j-th local area in the l-th layer; is the input of the j-th neuron in the operation result of the i-th convolution kernel in the (l + 1)-th layer; After the convolution operation, the ReLU function is used as the activation function of the model to accelerate the convergence speed of the model and prevent overfitting; The expression of the ReLU function is as follows: f(x) = max{0, x} In the deep convolutional neural network (DCNN), a pooling layer is added after the convolutional layer to reduce the computational burden. There are two common pooling operations, namely average pooling and max pooling. The expressions for the two pooling operations are as follows; v l(i,j) = max H(j-1)-1＜n＜jH {x l(i,n)} where: l(i,n) is the activation value of the nth neuron in the ith channel of the lth layer; H is the size of the sliding window; ν l(i,j) is the result after pooling of j neurons in the current l-th layer; The deep convolutional neural network (DCNN) is used to extract features from the original signal, and a max pooling layer is adopted to improve the feature extraction speed. Subsequently, the time-domain and frequency-domain features are concatenated to form an alternative feature set.

7. A method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that In step 7, a transfer learning bearing life prediction model is established. Transfer learning is performed on the DCNN model, and the transfer prediction model is pre-trained with a large number of source domain datasets to ensure that the model learns as many data features under different working conditions as possible. At the same time, transfer learning is used to extract similar features of vibration signals under different working conditions and different devices to reduce the differences between the source domain and the target domain; based on the DCNN model, a transfer model is re-trained by combining the similar features learned through transfer learning for the remaining life prediction in the target domain.

8. A method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that, In step 8, after the bearing remaining life prediction is completed, the source domain test set and the target domain test set need to be verified. For this, two metrics, namely the mean absolute error (MAE) and the root mean square error (RMSE), are used in this method to evaluate the prediction effect. The formulas are as follows: Where: m is the number of prediction points; i is the serial number of the prediction point, and y i Actual value; Is the predicted value; The value range of MAE is [0, +∞), and the smaller the better. The smaller the RMSE value, the higher the accuracy.

Images