Method and system for predicting residual service life of motor bearing by using health index

Abstract

The invention discloses a motor bearing remaining service life prediction method and system using health indexes, and belongs to the technical field of automation. According to the method, time domain, frequency domain and time-frequency domain characteristic extraction is carried out on a vibration signal sequence, and sensitive characteristics of a degradation process can be accurately reflected through monotonicity and time correlation screening. And performing principal component analysis and dimension reduction on the screened features, and selecting the maximum principal component as a health index. And calculating an initial degradation point of the health index and dividing the data set. When residual service life prediction is carried out, an input signal is decomposed into a trend part and a residual error part, different models are used for feature extraction, and finally a residual service life prediction result is output. Through the steps of health index construction, signal decomposition, feature extraction, residual service life prediction and the like, the degradation stage can be accurately divided, meanwhile, accurate feature extraction and service life prediction can be carried out on data of the degradation stage, and the service life prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of automation technology, and in particular relates to a method and system for predicting the remaining service life of a motor bearing using health indicators. Background Art

[0002] China's "Smart Manufacturing 2035" initiative and the rapid development of big data and artificial intelligence are placing higher demands on the safety of motor bearings. As a core technology for fault prediction and health management, rolling bearing life prediction has become a key measure to ensure the normal operation of motors.

[0003] Although there are many bearing remaining service life prediction methods based on physical models and deep learning, the existing methods still have the following problems: (1) Lack of health indicator construction, that is, the existing methods fail to fully consider that the degradation of bearings is a gradual process. From initial normal operation to final failure, they will experience a stable wear period and a degradation period. When the bearing is first put into operation, it is in a stable wear period. Although the bearing will have a certain degree of wear and friction at this time, it has little impact on the overall operation. As time goes by, the damage to the bearing will gradually expand, and even abnormal conditions such as noise and vibration will occur, which indicates that the bearing has entered the degradation period. This affects the accuracy of the remaining service life results when predicting the remaining service life of the bearing. (2) The feature extraction lacks completeness, that is, the feature extraction effect of the deep learning model is often not optimal. The vibration signal contains the process reflecting the long-term degradation trend and the local features of the short-term fluctuation. The usual feature extraction does not extract them separately, resulting in insufficient feature extraction, which affects the final remaining service life prediction results. (3) Traditional feature extraction models such as convolutional neural networks and time convolutional networks are relatively simple and cannot fully extract the features of the vibration signal degradation process, which affects the final remaining service life results. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems existing in the prior art and to provide a method and system for predicting the remaining service life of motor bearings using health indicators.

[0005] In order to achieve the above-mentioned object of the invention, the present invention specifically adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a method for predicting the remaining useful life of a motor bearing using a health indicator, comprising the following steps:

[0007] S1: Obtain the vibration signal sequence of the motor bearing collected by the sensor and perform feature extraction processing on it to obtain time domain features, frequency domain features and time-frequency domain features;

[0008] S2: Calculate the monotonicity index and time correlation index of each time domain feature, frequency domain feature, and time-frequency domain feature, calculate the final comprehensive index based on the monotonicity index and time correlation index, and take the feature whose final comprehensive index is greater than the preset comprehensive index threshold as the sensitive feature;

[0009] S3: Based on principal component analysis, multiple sensitive features are merged into principal components, and the largest principal component is selected as the health indicator;

[0010] S4: Use the mean and standard deviation of the health indicator to design the indicator threshold, and take the time point corresponding to the first time the health indicator exceeds the indicator threshold as the starting degradation point;

[0011] S5: forming a data set from the vibration signal sequence, dividing the data set using a starting degradation point, considering the remaining service life label before the starting degradation point as 1, and the remaining service life label after the starting degradation point gradually decreasing to 0;

[0012] S6: Training a remaining service life prediction model on the divided data set, inputting the vibration signal sequence to be predicted into the trained remaining service life prediction model, obtaining a remaining service life prediction result, and completing the remaining service life prediction of the motor bearing;

[0013] In the remaining useful life prediction model, the input vibration signal sequence is first decomposed to obtain trend terms and residual terms. A deep learning model is used to extract features from the trend terms and residual terms respectively, and trend features and local features are obtained accordingly. The trend features and local features are concatenated as comprehensive features. After the comprehensive features are flattened, a fully connected layer is used to predict the remaining useful life.

[0014] Based on the above solution, each step can be implemented in the following preferred specific manner.

[0015] As a preferred embodiment of the first aspect, the feature extraction process of step S1 is specifically as follows:

[0016] S11: Extract time domain features from the vibration signal sequence and obtain 12 time domain features, namely F1-mean, F2-average amplitude, F3-standard deviation, F4-root mean square, F5-peak, F6-skewness, F7-kurtosis, F8-peak factor, F9-shape factor, F10-pulse factor, F11-variance, and F12-margin factor;

[0017] S12: The vibration signal sequence is converted into a frequency domain vibration signal by fast Fourier transform, and six frequency domain features are extracted, namely F13-center of gravity frequency, F14-spectral entropy, F15-frequency standard deviation, F16-spectral skewness, F17-spectral kurtosis, and F18-frequency domain amplitude;

[0018] S13: Wavelet packet decomposition is used to perform a three-layer decomposition of the vibration signal sequence to obtain 8 nodes. The energy transformation of the 8 nodes over time is extracted as the final time-frequency domain features.

[0019] As a preferred embodiment of the first aspect, the monotonicity index and the time correlation index are calculated as follows:

[0020]

[0021] Where: f i represents the time domain features, frequency domain features or time-frequency domain features calculated in S1; t i Indicates time; cov(f i ,t i ) represents f i With t i The covariance between represents f i The standard deviation of Indicates t i The standard deviation of C i represents f i Time correlation index of Mon i (f i ) is f i Monotonicity index; Sum(d / df i >0) indicates f i The number of monotonically increasing data in the sum (d / df i <0) indicates f i The number of monotonically decreasing data in f i The data length.

[0022] As a preferred embodiment of the first aspect, the average of the monotonicity index and the time correlation index is used as the final comprehensive index α:

[0023]

[0024] As a preferred embodiment of the first aspect, in step S2, the average of the monotonicity index and the time correlation index is used as the final comprehensive index.

[0025] As a preferred embodiment of the first aspect, the specific process of step S3 is as follows:

[0026] S31: Perform Z-score standardization on each sensitive feature so that the mean of each sensitive feature is 0 and the standard deviation is 1, thereby obtaining the processed sensitive feature;

[0027] S32: Calculate the covariance matrix of the processed sensitive features to reflect the linear correlation between the various sensitive features:

[0028]

[0029] Where: Represents the feature matrix composed of all processed sensitive features; n represents the feature matrix The number of samples in ; R represents the calculated covariance matrix; the superscript T represents the transpose.

[0030] S33: Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalue λ corresponding to the processed sensitive feature i and the eigenvector v i :

[0031] Rv i =λ i v i

[0032] S34: Sort the eigenvectors according to the size of the eigenvalues, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components;

[0033] S35: Project the data on the principal components to map the high-order data to a low-dimensional space, and select the largest principal component as the health indicator:

[0034]

[0035] Where: Y represents the feature matrix after dimensionality reduction.

[0036] As a preferred embodiment of the first aspect, in step S4, the sum of three times the standard deviation of the health indicator and its average value is used as the indicator threshold.

[0037] As a preferred embodiment of the first aspect, the specific process of step S6 is as follows:

[0038] S61: performing a learnable convolution kernel decomposition on the input vibration signal sequence to obtain a trend term representing the long-term degradation of the signal, and subtracting the vibration signal sequence from the trend term to obtain a residual term representing the local change;

[0039] S62: Use the attention temporal convolutional network to extract features from the trend item to obtain trend features, and use the multi-scale convolutional neural network to extract features from the residual item to obtain local features;

[0040] S63: Combine trend features and local features to obtain comprehensive features;

[0041] S64: Flatten the comprehensive features and use a fully connected layer to predict the remaining useful life to obtain a remaining useful life prediction result.

[0042] In a second aspect, the present invention provides a system for predicting the remaining useful life of a motor bearing using a health indicator, comprising:

[0043] A feature extraction module is used to obtain the vibration signal sequence of the motor bearing collected by the sensor and perform feature extraction processing on it to obtain time domain features, frequency domain features and time-frequency domain features;

[0044] The sensitive feature acquisition module is used to calculate the monotonicity index and time correlation index of each time domain feature, frequency domain feature, and time-frequency domain feature, calculate the final comprehensive index based on the monotonicity index and time correlation index, and take the feature whose final comprehensive index is greater than the preset comprehensive index threshold as the sensitive feature;

[0045] A health indicator generation module is used to combine multiple sensitive features into principal components using principal component analysis and select the largest principal component as the health indicator;

[0046] A starting degradation point determination module is used to design an indicator threshold using the average value and standard deviation of the health indicator, and to use the time point corresponding to the first time the health indicator exceeds the indicator threshold as the starting degradation point;

[0047] a data set partitioning module, configured to form a data set from the vibration signal sequence, partition the data set using a starting degradation point, and consider the remaining service life label before the starting degradation point to be 1, and the remaining service life label after the starting degradation point to gradually decrease to 0;

[0048] The prediction module is used to train the remaining service life prediction model on the divided data set, input the vibration signal sequence to be predicted into the trained remaining service life prediction model, obtain the remaining service life prediction result, and complete the remaining service life prediction of the motor bearing;

[0049] The specific processing process in the remaining service life prediction model is as follows: first, the input vibration signal sequence is decomposed to obtain trend terms and residual terms. The deep learning model is used to extract features of the trend terms and residual terms respectively, and the trend features and local features are obtained accordingly. The trend features and local features are spliced ​​as comprehensive features. After the comprehensive features are flattened, a fully connected layer is used to predict the remaining service life.

[0050] In a third aspect, the present invention provides a computer program product comprising a computer program / instruction, which, when executed by a processor, can implement a method for predicting the remaining useful life of a motor bearing using health indicators as described in any one of the schemes of the first aspect above.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting the remaining useful life of a motor bearing using health indicators as described in any of the schemes of the first aspect above is implemented.

[0052] In a fifth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0053] The memory is used to store computer programs;

[0054] The processor is used to implement the method for predicting the remaining service life of a motor bearing using a health indicator as described in any of the solutions of the first aspect when executing the computer program.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] (1) The present invention proposes a method for predicting the remaining service life of motor bearings using health indicators. The method uses health indicators to divide the stages of bearing degradation, and at the same time decomposes the vibration signal sequence of the motor bearing to obtain trend terms and residual terms, and uses different models to extract features respectively, thereby achieving accurate remaining service life prediction.

[0057] (2) The present invention processes the vibration signal sequence of the motor bearing by using feature extraction, sensitive feature screening, feature dimensionality reduction and determination of the starting degradation point to obtain a health indicator that can reflect the degradation trend of the bearing, and thereby determines the starting degradation point and divides the degradation stage.

[0058] (3) The present invention decomposes the vibration signal sequence of the motor bearing into a trend term and a residual term by introducing a learnable convolution kernel decomposition method. The learnable convolution kernel can update the weights through the backpropagation process during training, thereby capturing more complex trend patterns.

[0059] (4) The present invention innovatively proposes to use ATTCN and MSCNN models to extract the features of trend terms and residual terms respectively, and finally combine them and output the remaining useful life through a fully connected layer to achieve sufficient feature extraction and accurate remaining useful life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the overall architecture of the method of the present invention;

[0061] Figure 2 Schematic diagram of the vibration signal sequence of the motor bearing in the PHM2012 dataset of this embodiment;

[0062] Figure 3 This is a graph showing the health index results calculated based on the vibration signal sequence of the motor bearing in this embodiment;

[0063] Figure 4 This is a schematic diagram of the remaining useful life prediction results of this embodiment. DETAILED DESCRIPTION

[0064] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in the various embodiments of the present invention can be combined accordingly without conflicting with each other.

[0065] To address the problems of existing motor bearing RSU prediction methods, such as a lack of health indicator construction and incomplete feature extraction, the present invention proposes a motor bearing RSU prediction method using health indicators. This method achieves accurate prediction of the bearing's RSU by constructing health indicators, decomposing the input signal, and extracting features from trend and residual terms. Specifically, the present invention first extracts features from the vibration signal sequence of the motor bearing, collected by a sensor, in the time, frequency, and time-frequency domains. It then uses monotonicity and temporal correlation to select sensitive features that accurately reflect the degradation process. Furthermore, the filtered sensitive features are subjected to principal component analysis for dimensionality reduction. The largest principal component is selected as the health indicator. The starting degradation point of the health indicator is calculated and the dataset is partitioned. The RSU label before the starting degradation point is considered 1, while the RSU label after the starting degradation point gradually decreases to 0. Furthermore, when performing RSU prediction, the vibration signal sequence is decomposed into a trend component and a residual component. Feature extraction is performed using an attention temporal convolutional network and a multi-scale convolutional neural network, respectively. Finally, the RSU is output through a fully connected layer. This process realizes the linkage between the stage division of the bearing degradation process and the remaining service life prediction, which not only eliminates the interference in the stable stage, but also significantly improves the effect of signal feature extraction.

[0066] like Figure 1 As shown, in a preferred implementation of the present invention, the above-mentioned method for predicting the remaining service life of a motor bearing using a health indicator includes the following steps S1 to S6. The specific implementation process is described in detail below.

[0067] 1. Raw data acquisition and feature extraction

[0068] S1: Obtain the vibration signal sequence of the motor bearing collected by the sensor and perform feature extraction processing on it to obtain time domain features, frequency domain features and time-frequency domain features.

[0069] It should be noted that, in the present invention, the feature extraction process of step S1 is specifically as follows:

[0070] S11: For the vibration signal sequence X=[x1(Ns ),x2(N s ),…,x i (N s )] to extract time domain features and obtain 12 time domain features, namely F1-mean, F2-average amplitude, F3-standard deviation, F4-root mean square, F5-peak, F6-skewness, F7-kurtosis, F8-peak factor, F9-shape factor, F10-pulse factor, F11-variance, F12-margin factor; Among them, x1(N s ),x2(N s ),…,x i (N s ) represent N s is the number of data points in a single time window; x1,x2,…,x i are the 1st, 2nd,…,ith samples.

[0071] S12: The vibration signal sequence is converted into a frequency domain vibration signal by fast Fourier transform (FFT), and six frequency domain features are extracted, namely F13-center of gravity frequency, F14-spectral entropy, F15-frequency standard deviation, F16-spectral skewness, F17-spectral kurtosis, and F18-frequency domain amplitude;

[0072] S13: Wavelet packet decomposition is used to perform a three-layer decomposition of the vibration signal sequence to obtain 8 nodes. The energy transformation of the 8 nodes over time is extracted as the final time-frequency domain features (F19-F26).

[0073] 2. Construction of Health Indicators

[0074] 1) Feature screening

[0075] S2: Calculate the monotonicity index and time correlation index of each time domain feature, frequency domain feature, and time-frequency domain feature, calculate the final comprehensive index based on the monotonicity index and time correlation index, and set the final comprehensive index greater than the preset comprehensive index threshold. features as sensitive features.

[0076] It should be noted that in step S2 of the present invention, the monotonicity and time correlation of the features are calculated to screen out sensitive features that can accurately reflect degradation. The calculation method of the above monotonicity index and time correlation index is specifically as follows:

[0077]

[0078] Where: f i represents the time domain features, frequency domain features or time-frequency domain features calculated in S1; t i Indicates time; cov(f i ,t i ) represents fi The covariance between t and represents f i The standard deviation of represents the standard deviation of t; C i represents f i Time correlation index of Mon i (f i ) is f i Monotonicity index; Sum(d / df i >0) indicates f i The number of monotonically increasing data in the sum (d / df i <0) indicates f i The number of monotonically decreasing data in f i The data length.

[0079] After calculating the monotonicity index and the time correlation index, the final comprehensive index can be calculated. In this embodiment, the average of the monotonicity index and the time correlation index is used as the final comprehensive index α:

[0080]

[0081] Then, the calculated final comprehensive index is compared with a manually set comprehensive index threshold, and features with a value greater than the comprehensive index threshold are selected and retained. In this embodiment, the comprehensive index threshold is set to 0.6.

[0082] 2) Construction of Health Indicators (HI)

[0083] S3: Based on principal component analysis, multiple sensitive features are merged into principal components, and the largest principal component PCA1 is selected as the health indicator.

[0084] It should be noted that, in the present invention, in step S2, a principal component analysis method is used to merge multiple sensitive features into fewer principal components, thereby reducing the dimension of the data while retaining as much original information as possible.

[0085] It should be noted that the specific process of step S3 of the present invention is as follows:

[0086] S31: Perform data standardization on each sensitive feature. Specifically, Z-score standardization is used here to convert each sensitive feature into a standard normal distribution, that is, the mean is 0 and the standard deviation is 1, so as to avoid the influence of the dimensional differences of different features on the results of principal component analysis, and finally obtain the processed sensitive features.

[0087] Z-score standardization can be performed as follows:

[0088]

[0089] Where: μ i and σ i x i The mean and standard deviation of x′ i Represents x i Results after data standardization.

[0090] S32: Calculate the covariance matrix of the processed sensitive features to reflect the linear correlation between the various sensitive features:

[0091]

[0092] Where: Represents the feature matrix composed of all processed sensitive features; n represents the feature matrix The number of samples in ; R represents the calculated covariance matrix; the superscript T represents the transpose.

[0093] S33: Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalue λ corresponding to the processed sensitive feature i and the eigenvector v i :

[0094] Rv i =λ i v i

[0095] S34: Sort the eigenvectors according to the size of the eigenvalues, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal component V k =[v1,v2,…,v k ];

[0096] S35: Perform data projection on the principal components to map high-order data to a low-dimensional space, and select the largest principal component as a health indicator.

[0097]

[0098] Where: Y represents the feature matrix after dimensionality reduction.

[0099] It should be noted that, in this embodiment, the contribution of each principal component to the total variance in the vibration signal sequence is analyzed. The higher the variance explanation rate, the more important information the principal component covers. Therefore, the largest principal component is selected as the health indicator HI.

[0100] S4: Use the mean μ and standard deviation σ of the health indicator HI to design the indicator threshold T h , in the health indicator, the time point corresponding to the first time the indicator threshold is exceeded is taken as the starting degradation point.

[0101] It should be noted that in step S4 of the present invention, the sum of three times the standard deviation of the health index and its average value is used as the index threshold. The calculation method of the above index threshold is specifically as follows:

[0102] T h =μ+3σ

[0103] In this embodiment, T h =1338.

[0104] S5: The vibration signal sequence is used to form a data set, and the data set is divided using a starting degradation point, and the remaining service life label before the starting degradation point is regarded as 1, and the remaining service life label after the starting degradation point gradually decreases to 0.

[0105] 3. Vibration signal decomposition and remaining service life prediction

[0106] S6: Training a remaining service life prediction model on the divided data set, inputting the vibration signal sequence to be predicted into the trained remaining service life prediction model, obtaining a remaining service life prediction result, and completing the remaining service life prediction of the motor bearing;

[0107] In the remaining useful life prediction model, the input vibration signal sequence is first decomposed to obtain trend terms and residual terms. A deep learning model is used to extract features from the trend terms and residual terms respectively, and trend features and local features are obtained accordingly. The trend features and local features are concatenated as comprehensive features. After the comprehensive features are flattened, a fully connected layer is used to predict the remaining useful life.

[0108] It should be noted that the specific process of step S6 of the present invention is as follows:

[0109] S61: First, the input vibration signal sequence is decomposed for subsequent feature extraction. Specifically, the input vibration signal sequence is subjected to a learnable convolutional kernel decomposition (Learnable Convolutional Kernel Decomposition) to obtain the trend term X representing the long-term degradation of the signal. Trend , subtract the vibration signal sequence from the trend term to obtain the residual term X representing the local change Residual :

[0110] X Trend =LCKD(Padding(X))

[0111] X Residual =XX Trend

[0112] Where: LCKD represents the learnable convolution kernel decomposition operation; Padding represents the padding operation.

[0113] S62: Based on the trend term and residual term obtained above, two different deep learning models are used to achieve sufficient feature extraction, as follows:

[0114] (1) The trend item is extracted using the attention temporal convolutional network ATTCN(·) to obtain the trend feature F1:

[0115] F1=ATTCN(X Trend )

[0116] (2) The residual term is subjected to feature extraction using a multi-scale convolutional neural network ASCNN(·) to obtain the local feature F2:

[0117] F2=MSCNN(X Residual )

[0118] S63: Combine trend features and local features to obtain comprehensive features F:

[0119] F=Concat[F1,F2]

[0120] Where: Concat represents the concatenation operation.

[0121] S64: Flatten the comprehensive features and use the fully connected layer to predict the remaining useful life to obtain the remaining useful life prediction result y:

[0122] y=FC(flatten(F1+F2))

[0123] Where: flatten represents the flattening operation; FC represents the fully connected layer.

[0124] In order to better demonstrate the specific implementation and technical effects of the present invention, a method for predicting the remaining service life of a motor bearing using health indicators shown in steps S1 to S6 in the above preferred implementation method is applied to a specific example.

[0125] Example

[0126] The specific implementation process of the method for predicting the remaining service life of a motor bearing using a health indicator adopted in this embodiment is as described above and will not be repeated here.

[0127] This example is based on the PHM2012 dataset, which contains 2803 samples, each of which contains 2560 data points. Figure 2 In this embodiment, when performing RSU prediction, the LCKD convolution kernel size is 25, the loss function is MSELoss, the optimizer is Adam, the learning rate is 0.0001, the drop rate is 0.1, the training rounds are 100, and the batch size is 32. Figure 3The health indicators and degradation stages of this embodiment are: Figure 4 The final remaining useful life prediction result is shown in FIG. It can be seen that the method of the present invention exhibits good remaining useful life prediction performance.

[0128] It should also be noted that the method for predicting the remaining useful life of a motor bearing using a health indicator in the above embodiment can essentially be executed by a computer program or module. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a system for predicting the remaining useful life of a motor bearing using a health indicator, corresponding to the method for predicting the remaining useful life of a motor bearing using a health indicator provided in the above embodiment, comprising:

[0129] A feature extraction module is used to obtain the vibration signal sequence of the motor bearing collected by the sensor and perform feature extraction processing on it to obtain time domain features, frequency domain features and time-frequency domain features;

[0130] The sensitive feature acquisition module is used to calculate the monotonicity index and time correlation index of each time domain feature, frequency domain feature, and time-frequency domain feature, calculate the final comprehensive index based on the monotonicity index and time correlation index, and take the feature whose final comprehensive index is greater than the preset comprehensive index threshold as the sensitive feature;

[0131] A health indicator generation module is used to combine multiple sensitive features into principal components using principal component analysis and select the largest principal component as the health indicator;

[0132] A starting degradation point determination module is used to design an indicator threshold using the average value and standard deviation of the health indicator, and to use the time point corresponding to the first time the health indicator exceeds the indicator threshold as the starting degradation point;

[0133] a data set partitioning module, configured to form a data set from the vibration signal sequence, partition the data set using a starting degradation point, and consider the remaining service life label before the starting degradation point to be 1, and the remaining service life label after the starting degradation point to gradually decrease to 0;

[0134] The prediction module is used to train the remaining service life prediction model on the divided data set, input the vibration signal sequence to be predicted into the trained remaining service life prediction model, obtain the remaining service life prediction result, and complete the remaining service life prediction of the motor bearing;

[0135] The specific processing process in the remaining service life prediction model is as follows: first, the input vibration signal sequence is decomposed to obtain trend terms and residual terms. The deep learning model is used to extract features of the trend terms and residual terms respectively, and the trend features and local features are obtained accordingly. The trend features and local features are spliced ​​as comprehensive features. After the comprehensive features are flattened, a fully connected layer is used to predict the remaining service life.

[0136] It is understood that the method for predicting the remaining useful life of a motor bearing using a health indicator described in S1 to S6 above can essentially be implemented via a computer program. Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer program product corresponding to the method for predicting the remaining useful life of a motor bearing using a health indicator provided in the above embodiment. The computer program product includes a computer program / instructions. When executed by a processor, the computer program / instructions implement the method for predicting the remaining useful life of a motor bearing using a health indicator as described in the above embodiment.

[0137] Similarly, based on the same inventive concept, another preferred embodiment of the present invention further provides a computer electronic device corresponding to the method for predicting the remaining service life of a motor bearing using a health indicator provided in the above embodiment, which includes a memory and a processor;

[0138] The memory is used to store computer programs;

[0139] The processor is configured to implement the method for predicting the remaining service life of a motor bearing using a health indicator in the above embodiment when executing the computer program.

[0140] Furthermore, the logic instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention.

[0141] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the method for predicting the remaining service life of a motor bearing using health indicators provided in the above embodiment, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the method for predicting the remaining service life of a motor bearing using health indicators provided in the above embodiment can be implemented.

[0142] It is understood that the storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage medium may be any medium capable of storing program code, such as a USB flash drive, a mobile hard drive, a magnetic disk, or an optical disk.

[0143] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0144] It should also be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division. In actual implementation, there may be other division methods, for example, multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0145] The embodiment described above is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Persons skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent substitution or equivalent transformation falls within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining service life of a motor bearing using a health indicator, characterized in that: The following steps are involved: S1: Obtain the vibration signal sequence of the motor bearing collected by the sensor and perform feature extraction processing on it to obtain time domain features, frequency domain features and time-frequency domain features; S2: Calculate the monotonicity index and time correlation index of each time domain feature, frequency domain feature, and time-frequency domain feature, calculate the final comprehensive index based on the monotonicity index and time correlation index, and take the feature whose final comprehensive index is greater than the preset comprehensive index threshold as the sensitive feature; S3: Based on principal component analysis, multiple sensitive features are merged into principal components, and the largest principal component is selected as the health indicator; S4: Use the mean and standard deviation of the health indicator to design the indicator threshold, and take the time point corresponding to the first time the health indicator exceeds the indicator threshold as the starting degradation point; S5: forming a data set from the vibration signal sequence, dividing the data set using a starting degradation point, considering the remaining service life label before the starting degradation point as 1, and the remaining service life label after the starting degradation point gradually decreasing to 0; S6: Training a remaining service life prediction model on the divided data set, inputting the vibration signal sequence to be predicted into the trained remaining service life prediction model, obtaining a remaining service life prediction result, and completing the remaining service life prediction of the motor bearing; In the remaining useful life prediction model, the input vibration signal sequence is first decomposed to obtain trend terms and residual terms. A deep learning model is used to extract features from the trend terms and residual terms respectively, and trend features and local features are obtained accordingly. The trend features and local features are concatenated as comprehensive features. After the comprehensive features are flattened, a fully connected layer is used to predict the remaining useful life.

2. The method for predicting the remaining useful life of a motor bearing using a health indicator according to claim 1, wherein: The feature extraction process of step S1 is specifically as follows: S11: Extract time domain features from the vibration signal sequence and obtain 12 time domain features, namely F1-mean, F2-average amplitude, F3-standard deviation, F4-root mean square, F5-peak, F6-skewness, F7-kurtosis, F8-peak factor, F9-shape factor, F10-pulse factor, F11-variance, and F12-margin factor; S12: The vibration signal sequence is converted into a frequency domain vibration signal by fast Fourier transform, and six frequency domain features are extracted, namely F13-center of gravity frequency, F14-spectral entropy, F15-frequency standard deviation, F16-spectral skewness, F17-spectral kurtosis, and F18-frequency domain amplitude; S13: Wavelet packet decomposition is used to perform a three-layer decomposition of the vibration signal sequence to obtain 8 nodes. The energy transformation of the 8 nodes over time is extracted as the final time-frequency domain features.

3. The method for predicting the remaining useful life of a motor bearing using a health indicator according to claim 1, wherein: In step S2, the average of the monotonicity index and the time correlation index is taken as the final comprehensive index.

4. The method for predicting the remaining useful life of a motor bearing using a health indicator according to claim 1, wherein: The specific process of step S3 is: S31: Perform Z-score standardization on each sensitive feature so that the mean of each sensitive feature is 0 and the standard deviation is 1, thereby obtaining the processed sensitive feature; S32: Calculate the covariance matrix of the processed sensitive features to reflect the linear correlation between the various sensitive features; S33: Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and eigenvectors corresponding to the processed sensitive features; S34: Sort the eigenvectors according to the size of the eigenvalues, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components; S35: Perform data projection on the principal components to map high-order data to a low-dimensional space, and select the largest principal component as a health indicator.

5. The method for predicting the remaining useful life of a motor bearing using a health indicator according to claim 1, wherein: In step S4, the sum of three times the standard deviation of the health indicator and its average value is used as the indicator threshold.

6. The method for predicting the remaining useful life of a motor bearing using a health indicator according to claim 1, wherein: In step S6, the remaining useful life prediction model, the specific processing process is as follows: S61: performing a learnable convolution kernel decomposition on the input vibration signal sequence to obtain a trend term representing the long-term degradation of the signal, and subtracting the vibration signal sequence from the trend term to obtain a residual term representing the local change; S62: Use the attention temporal convolutional network to extract features from the trend item to obtain trend features, and use the multi-scale convolutional neural network to extract features from the residual item to obtain local features; S63: Combine trend features and local features to obtain comprehensive features; S64: Flatten the comprehensive features and use a fully connected layer to predict the remaining useful life to obtain a remaining useful life prediction result.

7. A system for predicting the remaining useful life of a motor bearing using health indicators, characterized in that: include: A feature extraction module is used to obtain the vibration signal sequence of the motor bearing collected by the sensor and perform feature extraction processing on it to obtain time domain features, frequency domain features and time-frequency domain features; The sensitive feature acquisition module is used to calculate the monotonicity index and time correlation index of each time domain feature, frequency domain feature, and time-frequency domain feature, calculate the final comprehensive index based on the monotonicity index and time correlation index, and take the feature whose final comprehensive index is greater than the preset comprehensive index threshold as the sensitive feature; A health indicator generation module is used to combine multiple sensitive features into principal components using principal component analysis and select the largest principal component as the health indicator; A starting degradation point determination module is used to design an indicator threshold using the average value and standard deviation of the health indicator, and to use the time point corresponding to the first time the health indicator exceeds the indicator threshold as the starting degradation point; a data set partitioning module, configured to form a data set from the vibration signal sequence, partition the data set using a starting degradation point, and consider the remaining service life label before the starting degradation point to be 1, and the remaining service life label after the starting degradation point to gradually decrease to 0; The prediction module is used to train the remaining service life prediction model on the divided data set, input the vibration signal sequence to be predicted into the trained remaining service life prediction model, obtain the remaining service life prediction result, and complete the remaining service life prediction of the motor bearing; The specific processing process in the remaining service life prediction model is as follows: first, the input vibration signal sequence is decomposed to obtain trend terms and residual terms. The deep learning model is used to extract features of the trend terms and residual terms respectively, and the trend features and local features are obtained accordingly. The trend features and local features are spliced ​​as comprehensive features. After the comprehensive features are flattened, a fully connected layer is used to predict the remaining service life.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, it can implement the method for predicting the remaining service life of a motor bearing using a health indicator as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the method for predicting the remaining service life of a motor bearing using a health indicator as described in any one of claims 1 to 6 is implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the method for predicting the remaining useful life of a motor bearing using a health indicator as described in any one of claims 1 to 6 when executing the computer program.

Images