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

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

A method and system for predicting the remaining life of rolling bearings based on bidirectional GRU

technical field

The invention relates to the technical field of state evaluation of rolling bearings, in particular to a method and system for predicting the remaining life of rolling bearings based on a bidirectional GRU.

Background technique

The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

With the rapid development of smart sensing, wireless communication, and computer technologies, decision making (DM) is developing toward being intelligent, robust, and adaptive. Remaining useful life (RUL) prediction, as one of the most critical technologies in decision-making, can predict the failure time in advance, which is convenient for maintenance engineers to conduct qualitative risk analysis and formulate corresponding maintenance strategies, so as to avoid catastrophic situations and better ensure the mechanical equipment. reliability and safety.

Rolling bearings are one of the most common but important parts in mechanical equipment, and their health will directly affect the safety, reliability and availability of mechanical equipment. At present, the RUL prediction methods of rolling bearings are mainly divided into two categories: model-based methods and data-driven methods. Among them, the model-based method infers the future trend of health status by establishing a physical model, which requires a lot of prior knowledge and experience about the research object, resulting in poor generalization ability. The data-driven method mainly makes health predictions by modeling historical data, without the need for mathematical models or expert experience of the research object. Therefore, in recent years, data-driven methods have been widely used in remaining life prediction.

Acquisition of time correlation of degradation trend and health prediction are the key steps of data-driven residual life prediction method. The current mainstream method is based on the construction of degradation index, combined with models such as support vector regression machine, artificial neural network and deep neural network to carry out residual life expectancy. However, the method of constructing degradation indicators based on artificially extracted features relies heavily on empirical knowledge, and most of the current remaining life prediction models can only capture the time correlation of data in a single time direction (ie, forward or backward). Accuracy needs to be improved.

Contents of the invention

In order to solve the shortcomings of the existing technology, the present invention provides a method and system for predicting the remaining life of rolling bearings based on bidirectional GRU, which automatically extracts the degradation trend from the original state signal of the bearing, and effectively captures the hidden long-term correlation between time series signals , to achieve accurate prediction of bearing remaining life.

In order to achieve the above object, the present invention adopts the following technical solutions:

The first aspect of the present invention provides a method for predicting the remaining life of a rolling bearing based on a bidirectional GRU.

A method for predicting the remaining life of a rolling bearing based on a two-way GRU, including the following processes:

Obtain vibration signals of rolling bearings;

According to the obtained vibration signal and the preset convolutional neural network model, the estimated value of the degradation index of the rolling bearing is obtained;

According to the influence of the monotonicity, trendiness and robustness of degradation indicators on the prediction results of remaining life, different weights are given to the evaluation indicators, and the evaluation value of degradation indicators integrating trendiness, monotonicity and robustness is obtained.

According to the estimated value of the degradation index and the preset first BiGRU model, the predicted value of the degradation index is obtained;

According to the estimated value of the degradation index and the preset second BiGRU model, the predicted value of the remaining service life is obtained;

As an optional implementation, in the training of the preset convolutional neural network model, the degradation index acquisition method adopted is:

Based on the AVMD-KPCA degradation index extraction method, the bearing signal is adaptively decomposed to obtain K narrowband intrinsic mode component signals and calculate their intrinsic energy. The obtained narrowband intrinsic mode component signals are obtained through the KPCA kernel principal component analysis algorithm. Intrinsic energy is converted into a degradation indicator.

As an optional implementation manner, the intrinsic energies of the K narrowband intrinsic mode component signals are dimensionally reduced and compressed by KPCA, and the kernel function of KPCA is a Gaussian kernel function.

As an optional implementation manner, the first principal component extracted by KPCA is used as the estimated value of the degradation index.

As an optional implementation, the training of the convolutional neural network includes:

The original vibration signal X∈R ^p×q of the rolling bearing is used as the input for training the convolutional neural network model. The input matrix is convoluted by M convolution kernels with dimensions a ₁ ×b _1. Using the ReLU activation function, the volume The dimension of the product layer is (pa ₁ +1) × (qb ₁ +1), and the output feature map of the convolutional layer is sub-sampled in the pooling layer.

As an optional implementation, the training data is converted into multiple training sample vectors through a sliding time window.

As an optional implementation manner, a grid search method is used to optimize the number of hidden layers and units in each hidden layer of the first BiGRU model and the second BiGRU model.

The second aspect of the present invention provides a system for predicting the remaining life of a rolling bearing based on a bidirectional GRU.

A rolling bearing residual life prediction system based on bidirectional GRU, including:

The data acquisition module is configured to: acquire the vibration signal of the rolling bearing;

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

The state evaluation module is configured to: obtain the evaluation value of the degradation index that integrates the trend, monotonicity, and robustness according to the influence of the monotonicity, trendiness, and robustness of the degradation index on the prediction result of the remaining life.

The degradation index prediction module is configured to: obtain the predicted value of the degradation index according to the estimated value of the degradation index and the preset first BiGRU model;

The remaining service life prediction module is configured to: obtain the remaining service life prediction value according to the estimated value of the degradation index and the preset second BiGRU model;

The third aspect of the present invention provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, the steps in the method for predicting the remaining life of a rolling bearing based on a bidirectional GRU as described in the first aspect of the present invention are implemented .

The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored on the memory and operable on the processor. Steps in the bidirectional GRU-based rolling bearing residual life prediction method.

Compared with prior art, the beneficial effect of the present invention is:

1. The method and system for predicting the remaining life of a rolling bearing based on bidirectional GRU according to the present invention obtains the predicted value of the degradation index according to the estimated value of the degradation index and the preset first BiGRU model; and obtains the predicted value of the degradation index according to the estimated value of the degradation index and the preset second BiGRU model , to obtain the predicted value of the remaining service life; the more accurate estimation of the degradation index of the rolling bearing and the more accurate prediction of the remaining service life are realized.

2. In the method and system for predicting the remaining life of rolling bearings based on two-way GRU according to the present invention, in the training of the preset convolutional neural network model, first use the degradation index extraction method based on AVMD-KPCA to adaptively decompose the bearing signal, K narrowband natural mode component signals are obtained and their intrinsic energies are calculated, and the intrinsic energy of the obtained narrowband natural mode component signals is converted into a degradation index by KPCA kernel principal component analysis algorithm. Then, the bearing vibration data is used as the input of the convolutional neural network model, and the degradation index is used as the target output to obtain the convolutional neural network model for online extraction of the degradation index. The subjectivity of degradation index acquisition is overcome, and the accuracy of degradation index prediction is improved.

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

Description of drawings

The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention.

FIG. 1 is a schematic structural diagram of a method for predicting the remaining life of a rolling bearing based on a bidirectional GRU provided in Embodiment 1 of the present invention.

FIG. 2 is a schematic diagram of the framework of the CNN-based DI estimation method provided by Embodiment 1 of the present invention.

FIG. 3 is a schematic diagram of sliding time window processing provided by Embodiment 1 of the present invention.

Fig. 4 is a structural diagram of the BiGRU model provided by Embodiment 1 of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present 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 should be noted that the terminology used here is only for describing specific embodiments, and is not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural, and it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they mean There are features, steps, operations, means, components and/or combinations thereof.

In the case of no conflict, the embodiments and the features in the embodiments of the present invention can be combined with each other.

Example 1:

Embodiment 1 of the present invention provides a method for predicting the remaining life of rolling bearings based on bidirectional GRU. First, through the fully integrated empirical mode decomposition of adaptive noise and dynamic principal component analysis (AVMD-KPCA), new The nonlinear DI is more in line with the real law of bearing degradation to describe the degradation process; then, by learning and capturing the mapping relationship between the original vibration signal and the training rolling bearing DI, a CNN model for DI estimation is constructed, and its representative features are used The CNN model is automatically extracted from the original vibration signal, eliminating the need for manual feature extraction and selection; due to its generalization and robustness, the model can be transferred to other bearings with similar operating conditions without changing the model hyperparameters ; Once DI was estimated, a BiGRU model was built for lifetime prediction, including future DI and RUL predictions.

Specifically, the following processes are included:

S1: Data collection. It is collected by the multi-channel online detection and analysis system DH5972N on the rotating machinery fault simulation platform HFZZ-II.

S2: DI extraction. Using AVMD-KPCA to extract DI from the collected vibration signal;

S3: DI estimation. Build a CNN model, capture the representative features hidden in the original vibration signal, and automatically estimate the DI of the online rolling bearing;

S4: DI prediction, the BiGRU model is constructed, and the estimated DI of the online rolling bearing is sent to the trained DI prediction model for future DI prediction.

S5: Life prediction, a BiGRU model is constructed, and the estimated DI of the online rolling bearing is sent into the trained life prediction model for RUL prediction.

In S1, the dataset is acquired from a simulation platform consisting of a series of components that generate 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 working conditions. The shaft speed is controlled by the motor speed controller. The two accelerometers are used to collect vibration signals with a sampling frequency of 20kHz in the horizontal and vertical directions respectively. The sampling period is set to 1 minute, and each sampling is 1 second. Only the horizontal vibration signal is used in this paper because it contains more information useful for the health degradation of rolling bearings. The dataset includes 5 SDK6205 bearings, which were tested under three operating conditions. For each case, the first three rolling bearings were used as training bearings and the rest as test bearings.

In S2, the DI extraction method based on AVMD-KPCA is used to adaptively decompose the bearing signal to obtain K narrow-band intrinsic mode component signals, which are converted into degradation indicators DI by KPCA kernel principal component analysis algorithm.

The detailed steps of VMD are as follows:

First, for each modal signal u _k (t), its analytic signal is obtained by Hilbert transform, and its one-sided spectrum is calculated:

并加入相应的信号中，从而将该信号调制的固定的基频带上，即：Second, estimate a center frequency for each resolved signal

And add it to the corresponding signal, so that the signal is modulated on a fixed baseband, that is:

Finally, by solving the Gaussian smoothness of the modulated signal, that is, the L2 norm of its gradient, the frequency band of the signal is estimated. The variational problem can be expressed as:

所有信号模态加和。In the formula, {u _k }:＝{u ₁ ,u ₂ ,…u _K } represent all modes of signal u _k (t), {ω _k }:＝{ω ₁ ,ω ₂ ,…ω _K } represent The center frequency of the signal, K is the total number of signal modes,

All signal modes are summed.

In order to transform the band constraint variational problem of formula (3) into the unconstrained variational problem, introduce quadratic penalty factor α and Lagrangian multiplier λ; 1, formula (3) can transform augmented Lagrangian expression:

The above variational problem is solved by using the multiplication operator alternation algorithm. The specific algorithm flow is as follows:

λ¹,n＝0(1) Initialization

λ ¹ ,n=0

(2) For k=1:1:K, u _k is updated by solving the next optimization problem.

Using the Parseval/Plancherel Fourier equidistant transform, the above problem is converted to the frequency domain:

Replace ω in formula (6) with ω-ω _k , then the above formula can be transformed into:

Using the Hermitian symmetry of the real signal in the process of reconstructing the fidelity term, the above formula can be rewritten as the integral form of the non-negative frequency interval, as follows:

Solving this quadratic optimization problem gives:

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

Using the same processing method as S2, the optimization problem represented by the above formula is converted to the frequency domain for processing, namely:

Solving the above formula can get:

则终止迭代，否则返回步骤(2)；(4) For a given error discriminant e, if

Then terminate the iteration, otherwise return to 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 realize the adaptive selection of the penalty factor α. SPC can assess the degree of modal aliasing and is used to quantify "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, the Fourier transform of each decomposition mode is calculated, and then its SPC is obtained according to equation (14). Then choose the one closest to the SPC average, and choose its corresponding penalty factor α value as the best value.

Through the above algorithm, the vibration signal can be decomposed into K narrowband intrinsic mode component signals and their intrinsic energy can be calculated. Then, send K sequences composed of inherent energy to KPCA for dimensionality reduction and compression. The kernel function used is a Gaussian kernel function, and the formula is as follows:

In this paper, the first principal component extracted by KPCA is selected as DI to describe the degradation process.

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

The original vibration signal X ∈ R ^{p × q} of the rolling bearing is used as the input to train the CNN model, while the DI value is used as the target output. The input matrix is convolved by M convolution kernels with dimensions a ₁ ×b ₁ .

Using the ReLU activation function, the dimension is (pa ₁ +1) × (qb ₁ +1), and the output feature map of the convolutional layer is sub-sampled in the following pooling layer. Then there are several convolutional and pooling layers to capture representative features from the input raw vibration signal. Then, the fully connected layer acts as a regression layer, generating the predicted output (DI labels). After the training of the CNN model is completed, the online vibration signal of the test bearing is input into the trained CNN model. The CNN model directly captures representative features from raw vibration signals and can obtain estimated DI values.

And according to the influence of monotonicity, trendiness and robustness of health indicators on the prediction results of remaining life, different weights are given to the evaluation indicators, and the evaluation criteria of degradation indicators integrating trendiness, monotonicity and robustness are established.

In the formula, ω _i , i=1, 2, 3 are the evaluation index weights, Y(t _k ) is the degradation index, T(t _k ) is the time vector, V _corr (Y(t _k ),T(t _k )) is the trend value, which is used to represent the trend of health indicators, V _mon (Y(t _k )) is a monotonic value, used to represent the increase or decrease of health indicators, V _rob (Y(t _k )) is the robustness The stickiness value reflects the tolerance of health indicators to outliers.

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

x _t represents the input data, y _t is the output of the GRU, h _t represents the output of the GRU unit, r is the reset gate, z is the update gate, r and z together control how to get from the previous hidden state h _t-1 through calculation The new hidden state h _t , σ represents the sigmoid activation function, and W _z is the update gate weight.

The most basic unit of the BiGRU model is composed of a GRU unit for forward propagation and a GRU unit for backward propagation. In a unidirectional GRU network, state information is always output from front to back. In the remaining life prediction problem, the output information at the current moment can be related to the state information at the previous moment and the state information at the next moment.

和反向的隐含状态的输出/>

三个部分一起决定。BiGRU's current hidden state information is based on the current input x _t , the hidden state forward at time t-1

and the output of the inverted hidden state />

The three parts are decided together.

The model consists of two steps, a training step and a testing step.

In the training step, the DI values are used to construct the training dataset. In order to improve the information content of the input data and the prediction accuracy of the model, the construction of the training data set adopts the sliding time window technology. A fixed-length time window is used for continuous sampling, and the time window slides one measurement unit at a time to continuously acquire new samples until the end of the life cycle.

X_t＝[d_t,d_t+1,…,d_t+w]是第i个训练样本向量，其中d_t表示训练滚动轴承在t 时刻的归一化DI值，w是时间窗口的长度。y_t＝d_t+w+1是对应的标签。通过输入训练数据，可以通过最小化均方误差(MSE)函数来训练BiGRU，表示为：By using a sliding time window processing technique, the training dataset can be formed as

X _t =[d _t ,d _t+1 ,…,d _t+w ] is the i-th training sample vector, where d _t represents the normalized DI value of the training rolling bearing at time t, and w is the length of the time window. y _t =d _t+w+1 is the corresponding label. By inputting training data, BiGRU can be trained by minimizing the mean square error (MSE) function, expressed as:

是实际和预测的标签。T表示训练样本的总数。为了训练过程的快速收敛，BiGRU模型配备了Adam优化器，该优化器已被证实在预测问题中是有效的。Among them, y _t and

are the actual and predicted labels. T represents the total number of training samples. For fast convergence of the training process, the BiGRU model is equipped with the Adam optimizer, which has been proven to be effective in prediction problems.

In the test step, the DI value of the online test rolling bearing is directly input into the BiGRU model, which can predict the future DI.

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

During training, a training dataset is constructed using DI values and corresponding RUL values. The sliding time window slides step by step to obtain the input sample I _t , where I _t = [d _t ,d _t+1 ,…d _t+l-1 ] is the t-th input vector, and dt represents the normalization of the training rolling bearing at time t DI value, using the normalized RUL value o _t as the output of the BiGRU model.

Finally, the training set of the BiGRU model for RUL prediction can be expressed as

Based on the grid search technique, the hyperparameters of the BiGRU model for RUL prediction, such as the number of hidden layers and the units in each hidden layer, are optimized.

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

Fig. 1 is a schematic diagram of the whole process of the present invention, the horizontal vibration signal of the bearing is analyzed by adaptive variational mode decomposition algorithm (AVMD) and KPCA kernel principal component, the vibration signal is extracted and converted into the degradation index DI, and then the CNN network is used for To extract DI, the original vibration signal X∈R ^p×q of the rolling bearing is used as the input of the training CNN model, and the DI value is used as the target output, which excludes the artificial interference when extracting DI; finally, two BiGRU networks are trained, for DI and RUL for prediction.

Figure 2 is a structural diagram of the CNN neural network. A CNN model was constructed to test the DI estimation of rolling bearings. Due to the generality and robustness of the proposed CNN model, the hyperparameters of the CNN model are the same in the three cases. The model consists of 7 layers, including two convolutional layers, two max pooling layers and three fully connected layers (F1, F2 and F3). Raw vibration signals for training rolling bearings are used as input to the CNN model. In each training sample, 32,400 data points are taken from each original sample, forming a matrix of size 180×180. During training, the MSE function is used as the loss function of CNN. After 200 epochs, the optimal model parameters can be obtained through the Adam optimizer.

Figure 3 is the processing of DI data when training bidirectional GRU. The input form of the GRU model is (batch_size, time_steps, feature_nums), where batch_size refers to the number of samples processed in batches during model training, time_steps is the time series step size, and feature_nums is the feature dimension. In order to meet the GRU input requirements, time window sliding is performed on the original multi-dimensional sensor sequence to construct training samples. The length of the time window is time window, which represents the time step of the GRU model, and slides forward one time unit each time along the time direction, so a single training sample is a one-dimensional tensor of the length of the time window, and there will be overlapping. For BiGRU training to predict DI, the value corresponding to a moment after the time window is used as the label of the sample. For BiGRU trained to predict RUL, the normalized RUL value o _t is used as the output of the BiGRU model. The performance of BiGRU is different for different window size values. When the window size is 5, BiGRU can bear the best performance. Therefore, in this case, the window size of BiGRU is set to 5.

Figure 4 is a structural diagram of BiGRU. After obtaining the estimated DI by the CNN model, a BiGRU model was constructed to predict the future DI of the rolling bearing online test. The number H of hidden layers and the units K in each hidden layer are two hyperparameters that control the architecture and topology of BiGRU models, which have a key impact on model performance. In this paper, the grid search technique is used to optimize these two important hyperparameters. When the H and K values are different, the performance of BiGRU changes. Apparently, BiGRU achieves a good predictive function when it consists of 3 hidden layers and 100 neurons in each layer. Furthermore, we can see an overall upward trend in training time as the values of H and K increase. This is because larger values of H and K mean that more parameters included in the BiGRU model need to be optimized.

Example 2:

Embodiment 2 of the present invention provides a two-way GRU-based system for predicting the remaining life of rolling bearings, including:

The data acquisition module is configured to: acquire the vibration signal of the rolling bearing;

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

The degradation index prediction module is configured to: obtain the predicted value of the degradation index according to the estimated value of the degradation index and the preset first BiGRU model;

The remaining service life prediction module is configured to: obtain the remaining service life prediction value according to the estimated value of the degradation index and the preset second BiGRU model;

The state evaluation module is configured to: evaluate the state of the rolling bearing according to the obtained predicted value of the degradation index and the predicted value of the remaining service life.

The working method of the system is the same as the method for predicting the remaining life of the rolling bearing based on the bidirectional GRU provided in Embodiment 1, and will not be repeated here.

Example 3:

Embodiment 3 of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the steps in the method for predicting the remaining life of a rolling bearing based on bidirectional GRU as described in Embodiment 1 of the present invention are implemented. .

Example 4:

Embodiment 4 of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and operable on the processor. When the processor executes the program, the implementation as described in Embodiment 1 of the present invention Steps in the bidirectional GRU-based rolling bearing residual life prediction method.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Accordingly, the present invention can 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 disk storage and optical storage, etc.) 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 should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented through computer programs to instruct related hardware, and the programs can be stored in a computer-readable storage medium. During execution, it may include the processes of the embodiments of the above-mentioned methods. Wherein, the storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM) or a random access memory (Random Access Memory, RAM) and the like.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (9)

1. A rolling bearing residual life prediction method based on bidirectional GRU, characterized in that: Including the following process: Obtain vibration signals of rolling bearings; According to the obtained vibration signal and the preset convolutional neural network model, the estimated value of the degradation index of the rolling bearing is obtained; in the training of the preset convolutional neural network model, the method of obtaining the degradation index is as follows: Based on the AVMD-KPCA degradation index extraction method, the bearing signal is adaptively decomposed to obtain K narrow-band intrinsic mode component signals and calculate the intrinsic energy. The intrinsic energy of the narrow-band intrinsic mode component signals is obtained through the KPCA kernel principal component analysis algorithm. The energy is converted into a degradation index, and a new index called spectral cross-correlation (SPC) is proposed as a selection condition to realize the adaptive selection of the penalty factor α. SPC evaluates the degree of modal aliasing and is used to quantify Overlap, the indicator SPC is as follows:

where F( ) is the Fourier transform of a specific signal or time series; For each penalty factor α, calculate the Fourier transform of each decomposition mode, then obtain its SPC according to the above equation, then select the one closest to the SPC average value, and select its corresponding penalty factor α value as the best value; According to the estimated value of the degradation index and the preset first BiGRU model, the predicted value of the degradation index is obtained; According to the estimated value of the degradation index and the preset second BiGRU model, the predicted value of the remaining service life is obtained; Specifically: Use the original vibration data of the training bearing and the degradation index constructed by AVMD+KPCA to train the CNN network model; Input the vibration data of the test bearing into the trained CNN model to obtain the degradation index of the test bearing; After the obtained degradation index is processed by a sliding window with a window size of 5, it is input to the first trained BiGRU network model to predict its future degradation index; The status evaluation of rolling bearings is carried out according to the obtained degradation index prediction value and remaining service life prediction value. 2. the rolling bearing residual life prediction method based on bidirectional GRU as claimed in claim 1, is characterized in that: The intrinsic energy of the K narrowband intrinsic mode component signals is reduced and compressed by KPCA, and the kernel function of KPCA is a Gaussian kernel function. 3. the rolling bearing residual life prediction method based on bidirectional GRU as claimed in claim 1, is characterized in that: The first principal component extracted by KPCA is used as the estimated value of the degradation index. 4. the rolling bearing residual life prediction method based on bidirectional GRU as claimed in claim 1, is characterized in that: In the training of convolutional neural network, including: The original vibration signal X∈R ^p×q of the rolling bearing is used as the input for training the convolutional neural network model. The input matrix is convoluted by M convolution kernels with dimensions a ₁ ×b _1. Using the ReLU activation function, the volume The dimension of the product layer is (pa ₁ +1) × (qb ₁ +1), and the output feature map of the convolutional layer is sub-sampled in the pooling layer. 5. the rolling bearing residual life prediction method based on bidirectional GRU as claimed in claim 1, is characterized in that: Transform the training data into multiple training sample vectors via a sliding time window. 6. the rolling bearing residual life prediction method based on bidirectional GRU as claimed in claim 1, is characterized in that: The number of hidden layers and the units in each hidden layer of the first BiGRU model and the second BiGRU model were optimized using a grid search method. 7. A rolling bearing residual life prediction system based on bidirectional GRU, characterized in that: include: The data acquisition module is configured to: acquire the vibration signal of the rolling bearing; The degradation index estimation module is configured to: obtain the estimated value of the degradation index of the rolling bearing according to the obtained vibration signal and the preset convolutional neural network model; in the training of the preset convolutional neural network model, the method of obtaining the degradation index is: Based on the AVMD-KPCA degradation index extraction method, the bearing signal is adaptively decomposed to obtain K narrow-band intrinsic mode component signals and calculate the intrinsic energy. The intrinsic energy of the narrow-band intrinsic mode component signals is obtained through the KPCA kernel principal component analysis algorithm. The energy is converted into a degradation index, and a new index called spectral cross-correlation (SPC) is proposed as a selection condition to realize the adaptive selection of the penalty factor α. SPC evaluates the degree of modal aliasing and is used to quantify Overlap, the indicator SPC is as follows:

where F( ) is the Fourier transform of a specific signal or time series; For each penalty factor α, calculate the Fourier transform of each decomposition mode, then obtain its SPC according to the above equation, then select the one closest to the SPC average value, and select its corresponding penalty factor α value as the best value; The degradation index prediction module is configured to: obtain the predicted value of the degradation index according to the estimated value of the degradation index and the preset first BiGRU model; The remaining service life prediction module is configured to: obtain the remaining service life prediction value according to the estimated value of the degradation index and the preset second BiGRU model; specifically: Use the original vibration data of the training bearing and the degradation index constructed by AVMD+KPCA to train the CNN network model; Input the vibration data of the test bearing into the trained CNN model to obtain the degradation index of the test bearing; After the obtained degradation index is processed by a sliding window with a window size of 5, it is input to the first trained BiGRU network model to predict its future degradation index; The state evaluation module is configured to: evaluate the state of the rolling bearing according to the obtained predicted value of the degradation index and the predicted value of the remaining service life. 8. A computer-readable storage medium on which a program is stored, wherein the program is implemented when the program is executed by a processor in the method for predicting the remaining life of a rolling bearing based on a bidirectional GRU according to any one of claims 1-6. A step of. 9. An electronic device comprising a memory, a processor, and a program stored in the memory and operable on the processor, wherein the processor implements any one of claims 1-6 when executing the program Steps in the method for predicting the remaining life of a rolling bearing based on a two-way GRU.

Images