CN117370742A - Bearing residual life prediction method under data loss - Google Patents

Bearing residual life prediction method under data loss Download PDF

Info

Publication number
CN117370742A
CN117370742A CN202311414891.0A CN202311414891A CN117370742A CN 117370742 A CN117370742 A CN 117370742A CN 202311414891 A CN202311414891 A CN 202311414891A CN 117370742 A CN117370742 A CN 117370742A
Authority
CN
China
Prior art keywords
data
bearing
missing
interpolation
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311414891.0A
Other languages
Chinese (zh)
Inventor
邱明
刘静涛
李军星
刘志卫
董艳方
杨传猛
李迎春
杜辉
庞晓旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Henan University of Science and Technology
Original Assignee
Henan University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Henan University of Science and Technology filed Critical Henan University of Science and Technology
Priority to CN202311414891.0A priority Critical patent/CN117370742A/en
Publication of CN117370742A publication Critical patent/CN117370742A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0475—Generative networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2123/00—Data types
    • G06F2123/02—Data types in the time domain, e.g. time-series data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing

Abstract

The invention belongs to the technical field of bearing residual life prediction, and particularly relates to a method for predicting the residual life of a bearing under the condition of data loss. Obtaining vibration signal data of bearing missing to be detected, inputting the vibration signal data to a trained interpolation network model to perform interpolation filling processing on missing parts in the vibration signal data, wherein the interpolation network model is used for learning a data missing mechanism of bearing missing sample data; and then inputting the processed result into a trained residual life prediction model to obtain a residual life prediction result of the bearing to be detected. Compared with the method for directly predicting the residual life of the bearing with missing data, the method has higher accuracy, and can effectively solve the problem of degradation information loss caused by data missing.

Description

Bearing residual life prediction method under data loss
Technical Field
The invention belongs to the technical field of bearing residual life prediction, and particularly relates to a method for predicting the residual life of a bearing under the condition of data loss.
Background
Rolling bearings are also increasingly required for their safety as extremely important components in mechanical equipment. About 30% of the mechanical failures are caused by rolling bearings, so that development of the residual life prediction (RUL) work of the bearings is also necessary.
Currently, the algorithms for predicting the remaining life of a bearing are roughly classified into three categories: a method based on a physical model, a method based on statistics, a method based on data driving. The method based on the physical model utilizes Paris crack propagation model, forman crack propagation law and the like to describe the degradation process of the bearing by analyzing the failure mechanism of the bearing or summarizing experience, and then carries out the residual life prediction of the bearing. The statistical method is to build Weibull distribution models by analyzing the life distribution rule of the bearing and combining PCA and other methods, and solve relevant parameters by parameter estimation methods, so as to obtain a life distribution model and predict the life. The data driving-based method does not need to build a physical model or a statistical model, but obtains the relation between the bearing degradation state and the residual life through a neural network such as CNN and RNN and other models from the existing monitoring data, so as to predict the residual life of the rolling bearing. However, in actual engineering, due to the limitation of the acquisition equipment, the data acquisition data are lost due to various reasons such as external interference or data missing reading, so that an incomplete data set is formed, and the incomplete data set causes the loss of the degradation information of the bearing, so that the accurate prediction of the residual life of the bearing cannot be realized by adopting any one of the methods.
Disclosure of Invention
The invention aims to provide a method for predicting the residual life of a bearing under the condition of data loss, which is used for solving the problem that the accurate prediction of the residual life of the bearing can not be realized under the condition of data loss.
In order to solve the technical problems, the invention provides a method for predicting the residual life of a bearing under the condition of data loss, which is used for acquiring vibration signal data of the bearing to be detected, inputting the vibration signal data into a trained interpolation network model to perform interpolation filling processing on a missing part in the vibration signal data, wherein the interpolation network model is used for learning a data loss mechanism of bearing missing sample data; and then inputting the processed result into a trained residual life prediction model to obtain a residual life prediction result of the bearing to be detected.
The beneficial effects of the technical scheme are as follows: in consideration of data loss, the invention constructs an interpolation network model which is used for fully learning a data loss mechanism of bearing loss sample data, carrying out interpolation filling on the lost bearing data, and further predicting the residual life of the bearing by using a residual life prediction model after interpolation. Compared with a method for directly predicting the residual life of the bearing with missing data, the method has higher accuracy and can effectively solve the problem of degradation information loss caused by data missing.
Further, the interpolation network model is an RFGAIN network model obtained by improving the GAIN network model, and the RFGAIN network model comprises a generator and a discriminator; the generator is used for obtaining an interpolation matrix by calculation according to the input data and a mask matrix used for representing distribution of the input data and the missing part of the input data, and inputting the input data and the interpolation matrix into the discriminator; the discriminator is used for comparing whether the generated value of the non-missing part in the input data is the same as the true value data or not: if the parameters are different, the parameters of the generator and the discriminator are optimized through back propagation. The beneficial effects of the technical scheme are as follows: because the probability distribution of the vibration signal of the bearing at a certain moment is simpler, the role of the prompt matrix in the traditional GAIN network is not important, the RFGAIN network obtained by improving the GAIN network discards the prompt matrix, and a generator and a discriminator are reserved, so that the model is simpler, and the calculation processing efficiency is improved.
Further, the calculation process of generating the interpolation matrix in the generator is as follows:
wherein G () represents a generator; x is an input data matrix; m is a mask matrix, the undelayed mark is 1, and the undelayed mark is 0;is a distribution matrix corresponding to the input data matrix; z is noise; />Represents the inner product of Croke; />Is an interpolation matrix.
The beneficial effects of the technical scheme are as follows: the structure of the generator is simpler.
Further, the activation functions used in the generator are a Tanh activation function and a Relu activation function.
The beneficial effects of the technical scheme are as follows: the optimizer may converge faster using the Tanh activation function and the Relu activation function.
Further, the loss function used in training the interpolation network model includes:
wherein L (D) represents the loss value of the discriminator, and L (G) represents the loss value of the generator; e (E) X,M []Representing the error averaging of the true value in the arbiter,representing averaging the error of the interpolation matrix in the arbiter; m is a mask matrix, the undelayed mark is 1, and the undelayed mark is 0; d () represents a arbiter; />Is an interpolation matrix; alpha represents a super parameter; />Representing the sign of the multiplication of the corresponding position element.
The beneficial effects of the technical scheme are as follows: when the loss function is designed, the generator loss and the discriminator loss are comprehensively considered, and the prediction precision of the interpolation network model is ensured.
Further, the process of training the remaining life prediction model includes:
acquiring full-life vibration signals of a plurality of bearings as original samples, and carrying out deletion treatment on part of the original samples to obtain deletion samples;
obtaining a complete sample after interpolation filling of the missing sample by using the trained interpolation network model;
and taking the sample data which is not subjected to the missing processing and the corresponding prediction label as training data, taking the complement sample and the corresponding prediction label as test data, and correspondingly training and testing the constructed residual life prediction model.
The beneficial effects of the technical scheme are as follows: the complete data set is used as a training set to better realize life prediction, and then the residual life prediction model is tested by using the complement sample and the corresponding label to carry out fine adjustment on the residual life prediction model, so that the prediction precision of the complete sample can be improved, and the data after interpolation can be well completed for life prediction.
Further, the remaining life prediction result is a health index, and the calculation formula of the health index is:
wherein HI represents a health index value at the time t; t represents the duration of the full life cycle of the bearing.
The beneficial effects of the technical scheme are as follows: the index can accurately reflect the residual life of the bearing, and is simple and convenient to calculate.
Further, the residual life prediction model is a TCN network model.
The beneficial effects of the technical scheme are as follows: the TCN network model integrates causal convolution, expansion convolution and residual connection, has flexible receptive fields, can customize corresponding receptive fields according to task requirements, and ensures the accuracy of prediction results.
Further, the deletion process is a random deletion process.
The beneficial effects of the technical scheme are as follows: the random deletion processing mode is adopted instead of a single deletion processing mode, so that the diversity of the deletion situation can be improved, and the interpolation filling capacity of the interpolation network model for different deletion situations can be improved.
Drawings
FIG. 1 is a flow chart of a method of predicting bearing remaining life in the absence of data in accordance with an embodiment of the present invention;
FIG. 2 (a) is a graph showing the comparison of the error of the interpolation data RMSE at different loss rates according to the embodiment of the invention;
FIG. 2 (b) is a graph showing MAE error comparison of interpolation data at different loss rates according to an embodiment of the present invention;
FIG. 3 (a) is a diagram of the predicted results of the raw data;
FIG. 3 (b) is a graph of the predicted outcome of missing interpolated data;
fig. 3 (c) is a diagram of the result of prediction of missing non-interpolated data.
Detailed Description
The main conception of the invention is as follows: in consideration of data loss, the invention constructs an interpolation network model which is used for fully learning a data loss mechanism of bearing loss sample data, carrying out interpolation filling on the lost bearing data, and further predicting the residual life of the bearing by using a residual life prediction model after interpolation. Compared with a method for directly predicting the residual life of the bearing with missing data, the method has higher accuracy and can effectively solve the problem of degradation information loss caused by data missing. The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent.
An embodiment of a method for predicting remaining life of a bearing in the absence of data:
the flowchart of the method for predicting the remaining life of the bearing under the data loss introduced in this embodiment is shown in fig. 1, and specifically includes the following steps:
and step one, carrying out a full life cycle test on a plurality of rolling bearings with the same model under the same test condition until the bearings completely fail, and obtaining full life vibration signals of each bearing as sample data.
In this example, the sampling frequency during the test was set to 25KHz and a vibration signal of 0.1s was acquired every 10 s. These parameters can be adjusted according to the actual situation.
And step two, carrying out deletion treatment on part of the sample data obtained in the step one to obtain a missing sample, and marking the missing part of the missing sample.
In this embodiment, the deletion processing method is various, and in order to increase the sample deletion diversity, a random deletion processing method is used instead of a fixed deletion processing method. Moreover, here, the missing portion of the missing sample is marked to obtain a mask matrix, and standardized processing is performed to eliminate the dimension influence, and the specific mask matrix is set as follows:
and thirdly, constructing and training an interpolation network model for carrying out interpolation filling on the missing sample, and obtaining a full sample after interpolation filling on the missing sample by utilizing the trained interpolation network model.
In this embodiment, the interpolation network model selects an RFGAIN network model obtained by improving the GAIN network model. Correspondingly, according to the missing samples and the mask matrix obtained in the second step, the missing samples and the mask matrix are sent to the RFGAIN network as input, and interpolation filling processing of the missing samples is achieved through initializing network parameters and a preset loss function, so that the complement samples are obtained.
The GAIN network presents a hint matrix H, the generator generates a plurality of probability distributions of data, and the presence constraint generator of the hint matrix H can only generate one probability distribution. However, the probability distribution of the vibration signal at a certain moment of the bearing is not complex, and the role of the cue matrix H is not important. Furthermore, the GAIN network also requires a discriminator to identify whether the generated hint matrix is close to 1, which also increases the computation time. Therefore, the improved RFGAIN network discards the prompt matrix H, changes the neural network from the original three layers into two layers, and uses the full connection layer to make the model simpler. In the selection of the activation function, RFGAIN selects Tanh and Relu as the activation function, unlike sigmoid selected by the GAIN network, and the derivative of Tanh is greater than sigmoid, so that the optimizer can converge more quickly.
The RFGAIN framework structure mainly comprises a generator (G) and a discriminator (D), data samples which are distributed in the same way as real data are generated through the countermeasure between the generator and the discriminator, and finally the interpolation of the data is completed.
The inputs to the generator (G) mainly comprise: the missing data matrix X (i.e., the input data matrix of RFGAIN), the random noise Z, and the mask matrix M. The incomplete data matrix consists of measurement data and NAN data, the mask matrix consists of 0 and 1, the distribution of the measurement data and missing data is represented by the mask matrix, the value of the mask matrix corresponding to the measurement data is 1, and the value of the mask matrix corresponding to the missing data is 0. Firstly, obtaining a distribution matrix of bearing vibration data through a formula (2)Then generating an interpolation matrix +.>The data is input into a generator, the distribution matrix is output through a full connection layer, a full connection layer formula of the generator is shown as a formula (4), the formula (4) is an expansion expression of a neural network of the formula (2), and x i Shown are specific elements of formula (2) after the operation in the G () brackets. Next, the missing data matrix X, the generated interpolation matrix +.>The data is input into a discriminator (D), and if the discriminator judges false, the generator continues to generate false data matrixes by comparing whether the generated value of the non-missing part is the same as the original data or not, and the data matrixes are optimized through back propagation until the discriminator judges true. The neural network of the discriminator is also output through the full connection layer, and the full connection layer formula of the discriminator is shown as formula (5).
Wherein G () represents a generator;represents the inner product of Croke; w (W) ii Representing a weight matrix; f () represents an activation function, including Relu and Tanh; />Representing an output of the arbiter;
the network model is optimized by a loss function, in the RFGAIN framework, the objective of the arbiter D is to maximize the probability of predicting the M matrix, and the objective of the generator G is to minimize the probability of predicting the M matrix, so the loss is determined by both the arbiter and the generator. The specific loss function is as follows:
wherein L (D) represents the loss value of the discriminator, and L (G) represents the loss value of the generator; e (E) X,M []Representing the error averaging of the true value in the arbiter,representing averaging the error of the interpolation matrix in the arbiter; m is a mask matrix; d () represents a arbiter; />Is an interpolation matrix; alpha represents a super parameter and is set when a model is built; />Representing the sign of the multiplication of the corresponding position element.
And fourthly, constructing a health index of the whole life cycle of the bearing as a prediction label.
Specifically, let the time of the bearing full life cycle be T, and the Health Index (HI) of the training sample at time T be:
by the calculation of the formula, the service life label of the bearing is distributed to be 0-100%, the service life label of the bearing is 100% when the bearing starts to operate, and the service life label of the bearing is 0 when the bearing fails.
And fifthly, constructing a residual life prediction model, and training the network weight parameters in the constructed residual life prediction model by taking the sample data which is not subjected to the deletion processing in the second step and the corresponding prediction labels as training data (namely a training set).
In this embodiment, the remaining life prediction model selects a time convolution network (TCN network), which is a network for solving the time series problem, and on a conventional one-dimensional convolution neural network, causal convolution (causal convolutions), dilation convolution (dilated convolution) and residual connection (residual connections) are fused.
The input of TCN is time sequence data shown in formula (9), the filter is shown in formula (10), and the final hidden layer is shown in formula (11).
F＝(f 1 ,f 2 ,...,f k ) (10)
Wherein d is the expansion coefficient; s-d.i is the history data in the input sequence; k is a filter coefficient; f (f) j Representing a filter function; y(s) represents the output of the hidden layer.
When constructing the TCN model, three residual modules are selected to be stacked, with expansion factors (12,3,1), (6,3,2), (4, 3, 4), respectively. And taking other life cycle data except random missing processing as a training set, selecting Adam as an optimizer, selecting MSE as a loss function, and enabling the health index HI output by the model to be more attached to the actual health index HI through back propagation of training weight parameters.
And step six, taking the complement sample obtained in the step three as test data (namely, a test set) so as to realize the test of the residual life prediction model.
Specifically, in order to better demonstrate the interpolation and prediction effects of the method of the present invention, RMSE, MAE and an improved Score function are introduced as evaluation indexes, and the specific formulas are as follows:
q represents the number of missing data in the evaluation index as an interpolation model and the number of samples in the evaluation index as a life prediction. The smaller the two indices represent the better the model.
Wherein y is i The actual service life of the bearing;is a life prediction value; e (E) i Error for the ith sample; n is the total number of samples; m is the percentage of the early life of the bearing; omega 1 And omega 2 Weights for early and late life, respectively; s represents the Score function value, and the larger the value of S, the higher the Score, and the higher the prediction accuracy. In this embodiment, m=n/2, ω 1 ＝0.35，ω 2 ＝0.65。
And step seven, for the bearing to be tested, which lacks part of vibration signal data, obtaining the vibration signal data of the bearing to be tested, inputting the vibration signal data into the trained interpolation network model obtained in the step three for interpolation filling treatment, and inputting the treated result into the trained and tested residual life prediction model obtained in the step six to obtain the health index of the bearing to be tested, so as to realize the residual life prediction of the bearing to be tested.
It should be noted that the bearing to be measured is preferably the same type and working condition as the bearing of the training set, otherwise, the predicted result is affected.
The validity and correctness of the present invention is verified in connection with the examples below, with the dataset employing PHM2012 challenge data set provided by IEEE. In this embodiment, the data of Bearing1_3 under the working condition is selected to be artificially and randomly deleted, and the deletion rates are respectively set to be 10%, 20%, 30% and 40%. Inputting the missing data into an interpolation model, and comparing the RFGAIN model with interpolation results of a GAIN model and a random forest model (MissForest) under the condition that other conditions are unchanged. RMSE and MAE were chosen as evaluation indices, see formulas (12) and (13). The comparison results are shown in FIG. 2 (a) to FIG. 2 (b) and Table 1. And predicting the Bearing1_3 vibration signal under the 20% loss rate, and inputting other Bearing data sets under the working condition as training sets into the TCN model. The remaining life prediction results of the original vibration signal of bearing1_3, the non-interpolated vibration signal and the interpolated vibration signal are compared with each other. The comparison results are shown in FIGS. 3 (a) to 3 (c). In addition to RMSE and MAE as evaluation indices, an improved Score function was used. See formulas (14) - (16). The comparison of the service life prediction evaluation index results is shown in table 2.
TABLE 1 evaluation index of interpolation results
TABLE 2 evaluation index of predicted results
Compared with other methods, the method has more excellent performance in interpolation of the bearing data, and the method for predicting the life of the missing data of the bearing has higher accuracy in prediction of the life of the missing data of the bearing, so that the method can effectively solve the problem of degradation information loss caused by data missing.
In summary, the interpolation network model in the invention can fully learn the data deletion mechanism of the missing data sample of the bearing, performs interpolation filling on the missing bearing data by utilizing the distribution information of the missing data sample, combines the TCN prediction model to obtain the residual life prediction result of the missing data bearing, effectively combines the advantages of the two methods, and solves the technical problem that the residual life of the bearing cannot be accurately predicted due to the missing of the bearing data in the prior art.

Claims (9)

1. The method is characterized in that vibration signal data of bearing missing to be detected is obtained and is input into a trained interpolation network model to carry out interpolation filling processing on missing parts in the vibration signal data, and the interpolation network model is used for learning a data missing mechanism of bearing missing sample data; and then inputting the processed result into a trained residual life prediction model to obtain a residual life prediction result of the bearing to be detected.
2. The method for predicting the residual life of a bearing in the absence of data according to claim 1, wherein the interpolation network model is an RFGAIN network model obtained by improving a GAIN network model, and the RFGAIN network model includes a generator and a discriminator; the generator is used for obtaining an interpolation matrix by calculation according to the input data and a mask matrix used for representing distribution of the input data and the missing part of the input data, and inputting the input data and the interpolation matrix into the discriminator; the discriminator is used for comparing whether the generated value of the non-missing part in the input data is the same as the true value data or not: if the parameters are different, the parameters of the generator and the discriminator are optimized through back propagation.
3. The method for predicting remaining life of a bearing in the absence of data according to claim 2, wherein the calculation process of generating the interpolation matrix in the generator is:
wherein G () represents a generator; x is an input data matrix; m is a mask matrix, the undelayed mark is 1, and the undelayed mark is 0;is a distribution matrix corresponding to the input data matrix; z is noise; />Represents the inner product of Croke; />Is an interpolation matrix.
4. A method of predicting bearing residual life in the absence of data as claimed in claim 3 wherein the activation functions used in the generator include a Tanh activation function and a Relu activation function.
5. The method for predicting remaining life of a bearing in the absence of data as recited in claim 2, wherein the loss function used in training the interpolation network model comprises:
wherein L (D) represents the loss value of the discriminator, and L (G) represents the loss value of the generator; e (E) X,M []Representing the error averaging of the true value in the arbiter,representing averaging the error of the interpolation matrix in the arbiter; m is a mask matrix, the undelayed mark is 1, and the undelayed mark is 0; d () represents a arbiter; />Is an interpolation matrix; alpha represents a super parameter; the symbol by which the corresponding position element is multiplied is indicated.
6. The method for predicting remaining life of a bearing in the absence of data of claim 1, wherein training the remaining life prediction model comprises:
acquiring full-life vibration signals of a plurality of bearings as original samples, and carrying out deletion treatment on part of the original samples to obtain deletion samples;
obtaining a complete sample after interpolation filling of the missing sample by using the trained interpolation network model;
and taking the sample data which is not subjected to the missing processing and the corresponding prediction label as training data, taking the complement sample and the corresponding prediction label as test data, and correspondingly training and testing the constructed residual life prediction model.
7. The method for predicting the residual life of a bearing in the absence of data according to claim 1, wherein the residual life prediction result is a health index, and the calculation formula of the health index is:
wherein HI represents a health index value at the time t; t represents the duration of the full life cycle of the bearing.
8. The method for predicting remaining life of a bearing in the absence of data of claim 1, wherein the remaining life prediction model is a TCN network model.
9. The method for predicting remaining life of a bearing in the absence of data according to any one of claims 1 to 8, wherein the missing processing is random missing processing.
CN202311414891.0A 2023-10-27 2023-10-27 Bearing residual life prediction method under data loss Pending CN117370742A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311414891.0A CN117370742A (en) 2023-10-27 2023-10-27 Bearing residual life prediction method under data loss

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311414891.0A CN117370742A (en) 2023-10-27 2023-10-27 Bearing residual life prediction method under data loss

Publications (1)

Publication Number Publication Date
CN117370742A true CN117370742A (en) 2024-01-09

Family

ID=89403789

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311414891.0A Pending CN117370742A (en) 2023-10-27 2023-10-27 Bearing residual life prediction method under data loss

Country Status (1)

Country Link
CN (1) CN117370742A (en)

Similar Documents

Publication Publication Date Title
CN107832219B (en) Construction method of software fault prediction technology based on static analysis and neural network
CN108960303B (en) Unmanned aerial vehicle flight data anomaly detection method based on LSTM
CN110610035A (en) Rolling bearing residual life prediction method based on GRU neural network
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
CN111079836A (en) Process data fault classification method based on pseudo label method and weak supervised learning
CN112508243B (en) Training method and device for multi-fault prediction network model of power information system
CN113723007B (en) Equipment residual life prediction method based on DRSN and sparrow search optimization
CN113642754A (en) Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network
CN107832789B (en) Feature weighting K nearest neighbor fault diagnosis method based on average influence value data transformation
CN112414715B (en) Bearing fault diagnosis method based on mixed feature and improved gray level symbiosis algorithm
CN112217668B (en) Self-adaptive network fault diagnosis method based on comparison model
CN112214880B (en) Multi-stage maintainability assessment method based on virtual-real fusion
CN114707712A (en) Method for predicting requirement of generator set spare parts
CN114817034A (en) Test quality analysis method and device and electronic equipment
CN114169091A (en) Method for establishing prediction model of residual life of engineering mechanical part and prediction method
CN116994044A (en) Construction method of image anomaly detection model based on mask multi-mode generation countermeasure network
CN117370742A (en) Bearing residual life prediction method under data loss
CN111079348B (en) Method and device for detecting slowly-varying signal
CN116304587A (en) Rolling bearing degradation trend prediction method based on CAE and AGRU
CN114818281A (en) Section inversion method and system based on convolutional neural network
CN112307918A (en) Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network
CN111105074A (en) Fault prediction method based on improved deep belief learning
CN113343787B (en) Deep learning-based medium-level assessment method suitable for map contrast scene
CN112203311A (en) Network element abnormity diagnosis method, device, equipment and computer storage medium
CN116610484B (en) Model training method, fault prediction method, system, equipment and medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination