CN114720129B

CN114720129B - Rolling bearing residual life prediction method and system based on bidirectional GRU

Info

Publication number: CN114720129B
Application number: CN202210300479.5A
Authority: CN
Inventors: 张法业; 闫星宇; 姜明顺; 隋青美; 张雷; 贾磊
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2022-03-25
Filing date: 2022-03-25
Publication date: 2023-06-27
Anticipated expiration: 2042-03-25
Also published as: CN114720129A

Abstract

The invention provides a rolling bearing residual life prediction method and a system based on bidirectional GRU, which acquire a vibration signal of a rolling bearing; obtaining a degradation index estimated value of the rolling bearing according to the obtained vibration signal and a preset convolutional neural network model; obtaining a degradation index predicted value according to the degradation index estimated value and a preset first BiGRU model; obtaining a residual service life predicted value according to the degradation index estimated value and a preset second BiGRU model; carrying out state evaluation of the rolling bearing according to the obtained degradation index predicted value and the residual service life predicted value; according to the method, the degradation trend is automatically extracted from the original state signals of the bearing, the hidden long-term correlation between time sequence signals is effectively captured, and the accurate prediction of the residual life of the bearing is realized.

Description

Rolling bearing residual life prediction method and system based on bidirectional GRU

Technical Field

The invention relates to the technical field of rolling bearing state evaluation, in particular to a rolling bearing residual life prediction method and system based on bidirectional GRU.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

With the rapid development of smart sensing, wireless communication and computer technology, decision (DM) is evolving towards intelligence, robustness and adaptation. The residual service life (RUL) prediction is used as one of the most critical technologies in decision making, the fault time can be predicted in advance, a maintenance engineer can conveniently conduct qualitative risk analysis, and a corresponding maintenance strategy is formulated, so that the disastrous situation is avoided, and the reliability and the safety of mechanical equipment are better ensured.

Rolling bearings are one of the most common but important components in mechanical devices, the health of which will directly affect the safety, reliability and usability of the mechanical device. At present, rolling bearing RUL prediction methods are mainly divided into two types, namely a model-based method and a data driving method. Among them, model-based methods infer future trends in health by building physical models, requiring a great deal of prior knowledge and experience about the study subject, resulting in poor generalization ability. The data-driven method is mainly used for carrying out health prediction by modeling historical data, and a mathematical model or expert experience of a study object is not needed. Therefore, in recent years, the data-driven method is widely used for residual life prediction.

The degradation trend time correlation acquisition and the health prediction are key steps of a data-driven residual life prediction method, and the current mainstream method is to combine a support vector regression machine, an artificial neural network, a deep neural network and other models to perform residual life prediction on the basis of constructing degradation indexes. However, the degradation index construction method based on the manually extracted features relies heavily on empirical knowledge, and the residual life prediction models currently studied in most cases can only capture the time correlation of data in a single time direction (i.e., forward or backward), and the residual life prediction accuracy is to be improved.

Disclosure of Invention

In order to solve the defects in the prior art, the invention provides a rolling bearing residual life prediction method and a rolling bearing residual life prediction system based on bidirectional GRU, which automatically extract degradation trend from original state signals of the bearing, effectively capture hidden long-term correlation between time sequence signals and realize accurate prediction of the residual life of the bearing.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

the first aspect of the invention provides a rolling bearing residual life prediction method based on bidirectional GRU.

A rolling bearing residual life prediction method based on bidirectional GRU comprises the following steps:

acquiring a vibration signal of a rolling bearing;

obtaining a degradation index estimated value of the rolling bearing according to the obtained vibration signal and a preset convolutional neural network model;

and according to the influence of monotonicity, trend and robustness of the degradation indexes on the residual life prediction result, giving different weights to the evaluation indexes, and obtaining the degradation index evaluation values integrating the monotonicity, the monotonicity and the robustness.

Obtaining a degradation index predicted value according to the degradation index estimated value and a preset first BiGRU model;

obtaining a residual service life predicted value according to the degradation index estimated value and a preset second BiGRU model;

as an optional implementation manner, in training of the preset convolutional neural network model, the adopted degradation index acquisition mode is as follows:

the degradation index extraction method based on AVMD-KPCA carries out self-adaptive decomposition on the bearing signals to obtain K narrow-band inherent mode component signals and calculates the inherent energy of the K narrow-band inherent mode component signals, and the inherent energy of the obtained narrow-band inherent mode component signals is converted into degradation indexes through a KPCA kernel principal component analysis algorithm.

As an alternative embodiment, the intrinsic energy of the K narrowband intrinsic mode component signals is subjected to dimension reduction compression by using KPCA, and the kernel function of the KPCA is a gaussian kernel function.

As an alternative embodiment, the first principal component extracted by KPCA is taken as the degradation indicator estimation value.

As an alternative embodiment, the training of the convolutional neural network includes:

original vibration signal X epsilon R of rolling bearing ^p×q Used as input for training convolutional neural network model, the input matrix is composed of dimension a ₁ ×b ₁ Is convolved with a ReLU activation function, the dimension of the convolution layer being (p-a ₁ +1)×(q-b ₁ +1), the output feature map of the convolutional layer is subsampled in the pooling layer.

As an alternative embodiment, the training data is converted into a plurality of training sample vectors by sliding a time window.

As an alternative embodiment, the number of hidden layers and the cells in each hidden layer of the first biglu model and the second biglu model are optimized using a grid search method.

The second aspect of the invention provides a rolling bearing residual life prediction system based on bidirectional GRU.

A bi-directional GRU-based rolling bearing residual life prediction system, comprising:

a data acquisition module configured to: acquiring a vibration signal of a rolling bearing;

a degradation indicator estimation module configured to: obtaining a degradation index estimated value of the rolling bearing according to the obtained vibration signal and a preset convolutional neural network model;

a state evaluation module configured to: and obtaining the degradation index evaluation value integrating the monotonicity, the monotonicity and the robustness according to the influence of the monotonicity, the trend and the robustness of the degradation index on the residual life prediction result.

A degradation indicator prediction module configured to: obtaining a degradation index predicted value according to the degradation index estimated value and a preset first BiGRU model;

a remaining life prediction module configured to: obtaining a residual service life predicted value according to the degradation index estimated value and a preset second BiGRU model;

a third aspect of the present invention provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the steps in the rolling bearing residual life prediction method based on bidirectional GRUs according to the first aspect of the present invention.

A fourth aspect of the invention provides an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in the method for predicting the residual life of a rolling bearing based on bidirectional GRUs according to the first aspect of the invention when executing the program.

Compared with the prior art, the invention has the beneficial effects that:

1. according to the method and the system for predicting the residual life of the rolling bearing based on the bidirectional GRU, the predicted value of the degradation index is obtained according to the estimated value of the degradation index and the preset first BiGRU model; obtaining a residual service life predicted value according to the degradation index estimated value and a preset second BiGRU model; a more accurate estimation of the degradation indicator of the rolling bearing and a more accurate prediction of the remaining service life are achieved.

2. In the method and the system for predicting the residual life of the rolling bearing based on the bidirectional GRU, in the training of a preset convolutional neural network model, a degradation index extraction method based on AVMD-KPCA is firstly used for carrying out self-adaptive decomposition on a bearing signal, K narrow-band inherent mode component signals are obtained, the inherent energy of the narrow-band inherent mode component signals is calculated, and the inherent energy of the narrow-band inherent mode component signals is converted into a degradation index through a KPCA kernel principal component analysis algorithm. And then, using the bearing vibration data as input of the convolutional neural network model, and using the degradation index as target output to obtain the convolutional neural network model for extracting the degradation index on line. The subjectivity of degradation index acquisition is overcome, and the accuracy of degradation index prediction is improved.

Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.

Fig. 1 is a schematic structural diagram of a rolling bearing residual life prediction method based on bidirectional GRU according to embodiment 1 of the present invention.

Fig. 2 is a schematic diagram of a framework of a CNN-based DI estimation method according to embodiment 1 of the present invention.

Fig. 3 is a schematic diagram of a sliding time window processing according to embodiment 1 of the present invention.

Fig. 4 is a diagram of a biglu model structure according to embodiment 1 of the present invention.

Detailed Description

The invention will be further described with reference to the drawings and examples.

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.

Embodiments of the invention and features of the embodiments may be combined with each other without conflict.

Example 1:

the embodiment 1 of the invention provides a rolling bearing residual life prediction method based on bidirectional GRU, firstly, extracting a new nonlinear DI for training a bearing through fully integrated empirical mode decomposition of self-adaptive noise and dynamic principal component analysis (AVMD-KPCA), and describing a degradation process more in accordance with the real rule of bearing degradation; then, a CNN model for DI estimation is constructed by learning and capturing the mapping relation between the original vibration signal and the training rolling bearing DI, and the representative characteristic of the CNN model is automatically extracted from the original vibration signal by using the CNN model without manually extracting and selecting the characteristic; because of its generalization and robustness, the model can be transferred to other bearings with similar operating conditions without changing the model hyper-parameters; once DI is estimated, a biglu model is built for life prediction, including future DI and RUL predictions.

Specifically, the method comprises the following steps:

s1: and (5) data acquisition. The method is obtained by collecting a multi-channel online detection analysis system DH5972N on a rotating machinery fault simulation platform HFZZ-II.

S2: DI extraction. Extracting DI from the collected vibration signals by adopting AVMD-KPCA;

s3: and (5) DI estimation. Establishing a CNN model, capturing representative features hidden in an original vibration signal, and automatically estimating DI of the online rolling bearing;

s4: and (3) performing DI prediction, namely constructing a BiGRU model, and sending estimated DI of the online rolling bearing into a trained DI prediction model to perform future DI prediction.

S5: and (3) predicting the service life, constructing a BiGRU model, and sending the estimated DI of the online rolling bearing into the trained service life prediction model to perform RUL prediction.

In S1, a data set is obtained from an analog platform, which is composed of a series of components that produce vibration signals from operation to failure under different operating conditions. The radial force generated by the hydraulic loader is applied to the bearing under test to simulate different operating conditions. The shaft speed is controlled by a motor speed controller. The two accelerometers are used for collecting vibration signals with 20kHz sampling frequency in the horizontal direction and the vertical direction respectively. The sampling period was set to 1 minute, 1 second per sampling. Only a horizontal vibration signal is used here, as it contains more information useful for health degradation of the rolling bearing. The dataset included 5 SDK6205 bearings, which were tested under three operating conditions. For each case, the first three rolling bearings were used as training bearings, the rest as test bearings.

In S2, a DI extraction method based on AVMD-KPCA is adopted to carry out self-adaptive decomposition on the bearing signals, K narrow-band inherent mode component signals are obtained, and the K narrow-band inherent mode component signals are converted into degradation indexes DI through a KPCA kernel principal component analysis algorithm.

The specific steps of the VMD are described in detail as follows:

first, for each modal signal u _k (t) obtaining its resolved signal by hilbert transform and calculating its single-sided spectrum:

second, a center frequency is estimated for each of the resolved signals

And added to the corresponding signal, thereby modulating the signal on a fixed baseband, i.e.:

finally, the frequency band of the signal is estimated by solving the Gaussian smoothness of the modulated signal, i.e., the L2 norm of its gradient. The variation problem can be expressed as:

in the formula, { u _k }:＝{u ₁ ,u ₂ ,…u _K The signal u _k All modes of (t) { ω _k }:＝{ω ₁ ,ω ₂ ,…ω _K Represents the center frequency of the signal, K is the total number of signal modes,

all signal modalities add.

To convert the constrained variation problem of equation (3) to an unconstrained variation problem, introducing a quadratic penalty factor α and a lagrangian multiplier λ; i, formula (3) can transform the Index expression:

the above variational problem is solved by adopting a multiplication operator alternating algorithm, and the specific algorithm flow is as follows:

(1) Initialization of

λ ¹ ,n＝0

(2) For k=1:1:k, u is updated by solving the lower optimization problem _k 。

The above problem is transformed into the frequency domain using a Parseval/Planchrel Fourier equidistant transform:

by omega-omega _k Instead of ω in formula (6), the above formula can be converted into:

the Hermitian symmetry of the real signal in the process of reconstructing the fidelity term can be used for rewriting the above-mentioned form into an integral form of a non-negative frequency interval, and the method is as follows:

solving this quadratic optimization problem can be achieved:

(3) For k=1:1:k, ω is updated by solving the following optimization problem _k ：

The same processing manner as S2 is adopted, and the optimization problem represented by the above formula is converted into the frequency domain for processing, namely:

solving the above formula can obtain:

(4) For a given error criterion e, if

Terminating the iteration, otherwise returning to the step (2);

(5) The intrinsic energy E (t) of each narrowband intrinsic mode component signal is calculated as:

and a new index called spectral cross correlation (SPC) is proposed as a selection condition to achieve an adaptive selection of penalty factor a. SPC can evaluate the degree of modal aliasing for quantifying "overlap". The index SPC is as follows:

where F (·) is the Fourier transform of a particular signal or time series.

For each penalty factor α, a Fourier transform is computed for each decomposition mode, and then its SPC is taken according to equation (14). The value of the penalty factor alpha closest to the SPC average is then selected as the best value.

Through the algorithm, the vibration signal can be decomposed into K narrow-band natural mode component signals, and the natural energy of the K narrow-band natural mode component signals can be calculated. Then, K sequences consisting of inherent energy are sent to KPCA for dimension reduction compression, and the adopted kernel function is Gaussian kernel function, and the formula is as follows:

the first principal component of KPCA extraction is chosen herein as DI to describe the degradation process.

In S3, at the convolution layer, the input is convolved with a set of learnable kernels to obtain a new feature map, as follows:

original vibration signal X epsilon R of rolling bearing ^p×q As input to train the CNN model, while DI values are used as target outputs. The input matrix is formed by a dimension a ₁ ×b ₁ Is convolved with M convolution kernels.

Using ReLU activation function, we get the dimension (p-a ₁ +1)×(q-b ₁ +1), the output characteristics map of the convolutional layer is subsampled in the following pooling layer. Then several convolution layers and pooling layers capture representative features from the incoming raw vibration signal. The fully connected layers then act as regression layers, generating a prediction output (DI label). After the CNN model is trained, an online vibration signal of the test bearing is input into the trained CNN model. The CNN model captures representative features directly from the raw vibration signal and can obtain estimated DI values.

And according to the influence of monotonicity, trend and robustness of the health index on the residual life prediction result, different weights are given to the evaluation index, and a degradation index evaluation criterion integrating the monotonicity, the monotonicity and the robustness is established.

Wherein omega is _i I=1, 2,3 is the evaluation index weight, Y (t _k ) As an index of degradation, T (T _k ) As a time vector, V _corr (Y(t _k ),T(t _k ) Trend value for representing trend of health index, V _mon (Y(t _k ) For monotone value, for representing the increasing or decreasing variation of health index, V _rob (Y(t _k ) A robustness value, reflects the tolerance of the health indicator to outliers.

In S4, the GRU model contains 2 gate structures, reset gate and update gate, as follows:

x _t representing input data, y _t For the output of GRU, h _t Representing the output of the GRU unit, r is the reset gate, z is the update gate, r and z together control how the previous hidden state h is _t-1 Obtaining new hidden state h through calculation _t Sigma represents a sigmoid activation function, W _z Is to update the gate weights.

The most basic unit of the biglu model consists of a forward propagating GRU unit and a backward propagating GRU unit. In a unidirectional propagating GRU network, state information is always output from front to back. In the remaining life prediction problem, the output information at the present time may be associated with both the state information at the previous time and the state information at the subsequent time.

The current implicit state information of BiGRU is represented by the current input x _t Implicit state forward at time t-1

And the output of the implicit state in the opposite direction +.>

The three parts are decided together.

The model comprises two steps, a training step and a testing step.

In the training step, the DI values are used to construct a training dataset. In order to improve the information quantity of input data and the prediction accuracy of a model, a sliding time window technology is adopted for the construction of a training data set. And continuously sampling by adopting a time window with a fixed length, sliding one measurement unit each time by the time window, and continuously acquiring a new sample until the life cycle is finished.

By using a sliding time window processing technique, the training data set can be formed as

X _t ＝[d _t ,d _t+1 ,…,d _t+w ]Is the ith training sample vector, where d _t The normalized DI value at time t, w, representing the training rolling bearing, is the length of the time window. y is _t ＝d _t+w+1 Is the corresponding tag. By inputting training data, biGRU can be trained by minimizing a Mean Square Error (MSE) function, expressed as:

wherein y is _t And

are actual and predicted tags. T represents the total number of training samples. For rapid convergence of the training process, the biglu model is equipped with Adam optimizers that have proven to be effective in predicting problems.

In the test step, the DI value of the online test rolling bearing is directly input into the BiGRU model, so that future DI can be predicted.

In S5, another BiGRU model is established for realizing RUL prediction of rolling bearing online detection. The structure of the biglu model for RUL prediction is very similar to that of the biglu model for future DI prediction.

During training, a training data set is constructed using DI values and corresponding RUL values. The sliding time window is slid step by step to obtain the input samples I _t Wherein I _t ＝[d _t ,d _t+1 ,…d _t+l-1 ]Is the t input vector, dt represents the standardized DI value of the training rolling bearing at the t moment, and the normalized RUL value o is adopted _t As output of the biglu model.

Finally, the training set of biglu models for RUL prediction can be expressed as

Based on the grid search technique, the hyper-parameters of the biglu model for RUL prediction, such as the number of hidden layers and the cells in each hidden layer, are optimized.

During the test, the estimated DI values are input into the trained BiGRU model and the corresponding RUL values can be predicted.

FIG. 1 is a schematic diagram of the whole flow of the invention, wherein the horizontal vibration signal of the bearing is extracted and converted into a degradation index DI through an Adaptive Variation Modal Decomposition (AVMD) algorithm and KPCA kernel principal component analysis, and then a CNN network is trained to extract DI, so that the original vibration signal X epsilon R of the rolling bearing ^p×q Used as input for training the CNN model, while DI values are used as target outputs, eliminating artifacts when extracting DI; and finally training two BiGRU networks to predict DI and RUL.

Fig. 2 is a structural diagram of a CNN neural network. A CNN model was constructed for DI estimation of the test rolling bearing. Since the proposed CNN model is generic and robust, the hyper-parameters of the CNN model are identical in three cases. The model consists of 7 layers, including two convolution layers and two max pooling layers and three fully connected layers (F1, F2 and F3). The raw vibration signal of the training rolling bearing is used as input to the CNN model. In each training sample 32,400 data points were taken from each original sample to form a matrix of size 180 x 180. During training, the MSE function is used as a loss function of the CNN. After 200 epochs, optimal model parameters can be obtained by Adam optimizer.

Fig. 3 is a process of DI data in training a bi-directional GRU. The GRU model input is in the form of (batch_size, time_steps, feature_nums), where batch_size refers to the number of samples batched during model training, time_steps is the time sequence step, and feature_nums is the feature dimension. In order to meet the GRU input requirement, the original multidimensional sensor sequence is subjected to time window sliding, and a training sample is constructed. The time window length is time window, which represents the time step of the GRU model, and each time the GRU model slides forward along the time direction by one time unit, so that a single training sample is a one-dimensional tensor of one time window length, and two adjacent samples are overlapped. For BiGRU of training prediction DI, the value corresponding to the moment after the time window is used as the label of the sample. For BiGRU to train and predict RUL, normalized RUL value o is adopted _t As output of the biglu model. The performance of bigrus varies for different window size values. The biglu can achieve the best performance on the bearing when the window size is 5. Therefore, in this case, the window size of biglu is set to 5.

Fig. 4 is a diagram of the biglu structure. After obtaining the estimated DI by the CNN model, a biglu model is constructed to predict future DI for online testing of the rolling bearing. The number of hidden layers H and the cells K in each hidden layer are two hyper-parameters that control the architecture and topology of the biglu model, which has a key impact on model performance. These two important hyper-parameters are optimized herein using grid search techniques. When the H and K values are different, the performance of the BiGRU will change. Obviously, biglu achieves a good prediction function when it consists of 3 hidden layers and 100 neurons per layer. Furthermore, as the H and K values increase, we can see an overall upward trend in training time. This is because larger values of H and K mean that more parameters contained in the biglu model need to be optimized.

Example 2:

the embodiment 2 of the invention provides a rolling bearing residual life prediction system based on bidirectional GRU, which comprises:

a state evaluation module configured to: and carrying out state evaluation of the rolling bearing according to the obtained degradation index predicted value and the residual service life predicted value.

The working method of the system is the same as the rolling bearing residual life prediction method based on the bidirectional GRU provided in embodiment 1, and will not be described here again.

Example 3:

embodiment 3 of the present invention provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the steps in the rolling bearing remaining life prediction method based on bidirectional GRU according to embodiment 1 of the present invention.

Example 4:

embodiment 4 of the present invention provides an electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, where the processor implements the steps in the rolling bearing remaining life prediction method based on bidirectional GRU according to embodiment 1 of the present invention when executing the program.

It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), or the like.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A rolling bearing residual life prediction method based on bidirectional GRU is characterized in that:

the method comprises the following steps:

acquiring a vibration signal of a rolling bearing;

obtaining a degradation index estimated value of the rolling bearing according to the obtained vibration signal and a preset convolutional neural network model; in the training of the preset convolutional neural network model, the adopted degradation index acquisition mode is as follows:

the degradation index extraction method based on AVMD-KPCA carries out self-adaptive decomposition on bearing signals to obtain K narrow-band inherent mode component signals and calculates inherent energy, the inherent energy of the obtained narrow-band inherent mode component signals is converted into degradation indexes through a KPCA kernel principal component analysis algorithm, a new index called spectrum cross correlation degree (SPC) is provided, the self-adaptive selection of penalty factor alpha is realized as a selection condition, the SPC evaluates the degree of modal aliasing and is used for quantifying overlapping, and the index SPC adopts the following equation:

where F (·) is the Fourier transform of the particular signal or time series;

for each penalty factor alpha, calculating the Fourier transform of each decomposition mode, then obtaining the SPC according to the equation, then selecting the value of the penalty factor alpha closest to the SPC average value as the optimal value;

the method comprises the following steps:

training a CNN network model by using the training bearing original vibration data and the degradation index constructed by AVMD+KPCA;

inputting the vibration data of the test bearing into the trained CNN model to obtain a degradation index of the test bearing;

the obtained degradation index is processed through a sliding window with the window size of 5 and then is input into a trained first BiGRU network model, and future degradation indexes of the first BiGRU network model are predicted;

and carrying out state evaluation of the rolling bearing according to the obtained degradation index predicted value and the residual service life predicted value.

2. The method for predicting the remaining life of a rolling bearing based on bidirectional GRU according to claim 1, wherein:

and carrying out dimension reduction compression on the inherent energy of the K narrow-band inherent mode component signals by using KPCA, wherein the kernel function of the KPCA is Gaussian kernel function.

3. The method for predicting the remaining life of a rolling bearing based on bidirectional GRU according to claim 1, wherein:

and taking the first principal component extracted by KPCA as a degradation index estimated value.

4. The method for predicting the remaining life of a rolling bearing based on bidirectional GRU according to claim 1, wherein:

the training of the convolutional neural network comprises the following steps:

5. The method for predicting the remaining life of a rolling bearing based on bidirectional GRU according to claim 1, wherein:

the training data is converted into a plurality of training sample vectors by sliding a time window.

6. The method for predicting the remaining life of a rolling bearing based on bidirectional GRU according to claim 1, wherein:

the number of hidden layers and cells in each hidden layer of the first biglu model and the second biglu model are optimized using a mesh search method.

7. A rolling bearing residual life prediction system based on bidirectional GRU is characterized in that:

comprising the following steps:

a degradation indicator estimation module configured to: obtaining a degradation index estimated value of the rolling bearing according to the obtained vibration signal and a preset convolutional neural network model; in the training of the preset convolutional neural network model, the adopted degradation index acquisition mode is as follows:

where F (·) is the Fourier transform of the particular signal or time series;

a remaining life prediction module configured to: obtaining a residual service life predicted value according to the degradation index estimated value and a preset second BiGRU model; the method comprises the following steps:

8. A computer readable storage medium having stored thereon a program, which when executed by a processor, implements the steps of the bidirectional GRU-based rolling bearing residual life prediction method of any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor performs the steps in the method for predicting the residual life of a rolling bearing based on a bi-directional GRU as claimed in any one of claims 1-6 when the program is executed.