Rolling bearing remaining life prediction method based on deep learning and particle filtering

Abstract

The present application belongs to the field of rolling bearing reliability assessment, and particularly relates to a method for predicting the remaining life of rolling bearings based on deep learning and particle filtering. The present application realizes the adaptive construction of rolling bearing degradation indicators, realizes the adaptive construction of degradation indicators by building an efficient deep learning model, builds the model only based on normal data, and predicts the evolution of abnormal states, which is of great significance for the state assessment and remaining life prediction of rolling bearings under the condition of missing fault samples in actual engineering; realizes the adaptive division of the evolution stage of rolling bearing faults, and adaptively divides the evolution stage by unsupervised cluster learning, which is more scientific and accurate than the traditional artificial fuzzy definition method; realizes the prediction of the remaining life of rolling bearings based on particle filtering, and based on the extracted bearing degradation characteristics, uses a particle filtering method that is more suitable for nonlinear and non-Gaussian signal prediction problems to achieve accurate prediction of the remaining life of rolling bearings.

Description

Technical Field

[0001] The present application relates to the field of rolling bearing reliability assessment, and in particular to a rolling bearing remaining life prediction method based on deep learning and particle filtering. Background Art

[0002] Evaluating the reliability of rolling bearings, detecting the extent of performance degradation, and accurately predicting their remaining useful life can provide a basis for preventive maintenance decisions and improve aircraft engine safety. The construction of rolling bearing health indicators plays a crucial role in life prediction. Selecting appropriate indicators can accurately reflect bearing degradation trends and significantly improve the performance of the remaining useful life prediction model. The quality of the extracted features directly determines the accuracy of the prediction results.

[0003] However, the health characteristic indicators currently used in research are still essentially traditional manual features based on time or frequency domains. These require a certain level of professional skills and knowledge, and can only reflect certain one-sided characteristics of bearing vibration signals. They have certain limitations in intelligently extracting and fully reflecting bearing degradation trends. For example, it is difficult to extract degradation indicators that reflect the evolution of bearing status from complex rolling bearing vibration signals; it is difficult to automatically classify the stages of bearing fault evolution; and it is difficult to predict nonlinear and non-Gaussian rolling bearing signals.

[0004] Therefore, it is desired to have a technical solution to overcome or at least alleviate at least one of the above-mentioned deficiencies of the prior art. Summary of the Invention

[0005] The purpose of this application is to provide a rolling bearing remaining life prediction method based on deep learning and particle filtering to solve at least one problem existing in the prior art.

[0006] The technical solution of this application is:

[0007] A method for predicting the remaining life of rolling bearings based on deep learning and particle filtering, comprising:

[0008] Step 1: Build a deep convolutional neural network model;

[0009] Step 2: Acquire the full-life vibration data of the rolling bearing, construct a rolling bearing evolution feature extractor based on a deep convolutional neural network, extract the full-life vibration data of the rolling bearing in the normal stage from the full-life vibration data of the rolling bearing by the rolling bearing evolution feature extractor, and train the deep convolutional neural network model based on the full-life vibration data of the rolling bearing in the normal stage;

[0010] Step 3: Inputting the full-life vibration data of the rolling bearing into the trained deep convolutional neural network model, and obtaining the rolling bearing degradation index based on the characteristic distance between the normal sample and the fault sample;

[0011] Step 4: Perform k-means clustering on the rolling bearing degradation index, divide the rolling bearing evolution stage into a normal stage, a degradation stage, and a failure stage, and mark the eigenvalue corresponding to the starting point of the degradation stage as the degradation threshold, and mark the eigenvalue corresponding to the starting point of the failure stage as the failure threshold;

[0012] Step 5: Fit the rolling bearing degradation index before degradation by least squares fitting, and use the particle filtering method to gradually predict the characteristic value from the beginning of degradation. When the characteristic value is greater than the failure threshold, the prediction algorithm is terminated, and the remaining life prediction curve of the rolling bearing from the beginning of degradation to complete failure is obtained.

[0013] In at least one embodiment of the present application, in step one, when constructing the deep convolutional neural network model, the best combination of the deep convolutional neural network is achieved by gradually deepening the convolution-pooling-activation layers.

[0014] In at least one embodiment of the present application, in step one, when constructing the deep convolutional neural network model, the network structure and optimal network parameters are obtained by examining the loss accuracy and training time of the network on the validation set.

[0015] In at least one embodiment of the present application, the method further includes step six, verifying the correctness and effectiveness of the rolling bearing remaining life prediction method based on deep learning and particle filtering through vibration monitoring data of the rolling bearing full life fatigue test.

[0016] The invention has at least the following beneficial technical effects:

[0017] The rolling bearing remaining life prediction method based on deep learning and particle filtering in this application adaptively extracts bearing degradation indicators, avoids the one-sidedness of traditional degradation indicators, and improves the accuracy of bearing evolution state identification; adaptively divides bearing evolution states, avoiding the subjectivity of traditional manual division methods; and uses particle filtering method to realize the prediction of bearing remaining life, greatly improving the accuracy and adaptability of bearing life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a method for predicting the remaining life of a rolling bearing based on deep learning and particle filtering, according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the implementation of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below in conjunction with the drawings in the embodiments of this application. In the drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of this application, not all of the embodiments. The embodiments described below with reference to the drawings are exemplary and are intended to be used to explain this application, and should not be understood as limitations on this application. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The embodiments of this application are described in detail below in conjunction with the drawings.

[0020] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be understood as limiting the scope of protection of this application.

[0021] The following is combined with Figure 1 This application is described in further detail.

[0022] This application provides a method for predicting the remaining life of a rolling bearing based on deep learning and particle filtering, comprising the following steps:

[0023] Step 1: Build a deep convolutional neural network model;

[0024] Step 2: Obtain the full-life vibration data of the rolling bearing, construct a rolling bearing evolution feature extractor based on a deep convolutional neural network, extract the full-life vibration data of the rolling bearing in the normal stage from the full-life vibration data of the rolling bearing through the rolling bearing evolution feature extractor, and train a deep convolutional neural network model based on the full-life vibration data of the rolling bearing in the normal stage;

[0025] Step 3: Input the full-life vibration data of the rolling bearing into the trained deep convolutional neural network model, and obtain the rolling bearing degradation index based on the characteristic distance between normal samples and fault samples;

[0026] Step 4: Perform k-means clustering on the rolling bearing degradation indicators, divide the rolling bearing evolution stage into normal stage, degradation stage and failure stage, and mark the eigenvalue corresponding to the starting point of the degradation stage as the degradation threshold, and mark the eigenvalue corresponding to the starting point of the failure stage as the failure threshold;

[0027] Step 5. Fit the rolling bearing degradation index before degradation through least squares fitting, and use the particle filtering method to gradually predict the characteristic value from the beginning of degradation. When the characteristic value is greater than the failure threshold, the prediction algorithm is terminated, and the remaining life prediction curve of the rolling bearing from the beginning of degradation to complete failure is obtained.

[0028] The present invention discloses a method for predicting the remaining life of a rolling bearing based on deep learning and particle filtering. First, a deep convolutional neural network model is established. The optimal combination of the deep convolutional neural network is determined by gradually deepening the convolution-pooling-activation layer, and the network structure and optimal network parameters are obtained by examining the network's loss accuracy and training time on the validation set. Then, a rolling bearing evolution feature extractor based on the deep convolutional neural network is established, and only normal stage data is trained. The model is trained based on the vibration data of the rolling bearing in the normal stage. The full life evolution data is input into the trained model, and the rolling bearing degradation index is obtained based on the characteristic distance between the normal sample and the fault sample. The rolling bearing evolution stage is automatically divided into normal, degradation and failure stages by k-means clustering, and the characteristic values ​​corresponding to the starting points of the degradation stage and the failure stage are marked as degradation threshold and failure threshold respectively. A four-parameter exponential model is constructed, and the data before degradation are fitted using least squares fitting. The particle filtering method is used to gradually predict the characteristic values ​​from the beginning of degradation and compare them with the failure threshold. When the characteristic value is greater than the failure threshold, the prediction algorithm is terminated, and a prediction curve for the remaining life of the rolling bearing from the beginning of degradation to complete failure is obtained.

[0029] The present method for predicting the remaining life of rolling bearings based on deep learning and particle filtering was also validated using three sets of vibration monitoring data from full-life fatigue tests of rolling bearings. The method was also validated using three sets of vibration data from full-life evolution of rolling bearings obtained from three full-life fatigue tests. The data included an inner race failure in the HRB 6206#1 bearing, a mixed inner race and ball bearing failure in the HRB 6206#1 bearing, and an outer race failure in the ZA 2115 bearing.

[0030] The rolling bearing remaining life prediction method based on deep learning and particle filtering in the present application realizes the adaptive construction of rolling bearing degradation indicators. The adaptive construction of degradation indicators is achieved by building an efficient deep learning model. The model is built only based on normal data, and the evolution of abnormal conditions is predicted. It is of great significance for rolling bearing status assessment and remaining life prediction under the condition of missing fault samples in actual engineering projects; it realizes the adaptive division of rolling bearing fault evolution stages, and adaptively divides the evolution stages through unsupervised poly-learning, which is more scientific and accurate than the traditional artificial fuzzy definition method; it realizes the remaining life prediction of rolling bearings based on particle filtering, and based on the extracted bearing degradation characteristics, uses a particle filtering method that is more suitable for nonlinear and non-Gaussian signal prediction problems to achieve accurate prediction of the remaining life of rolling bearings.

[0031] The present invention provides a method for predicting the remaining life of rolling bearings based on deep learning and particle filtering, which has the following beneficial effects:

[0032] (1) Adaptive extraction of bearing degradation indicators avoids the one-sidedness of traditional degradation indicators and improves the accuracy of bearing evolution state identification;

[0033] (2) Adaptive division of bearing evolution states, avoiding the subjectivity of traditional manual division methods;

[0034] (3) The particle filtering method is used to predict the remaining life of the bearing, which greatly improves the accuracy and adaptability of the bearing life prediction.

[0035] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for predicting the remaining life of rolling bearings based on deep learning and particle filtering, characterized in that: include: Step 1: Build a deep convolutional neural network model; Step 2: Acquire the full-life vibration data of the rolling bearing, construct a rolling bearing evolution feature extractor based on a deep convolutional neural network, extract the full-life vibration data of the rolling bearing in the normal stage from the full-life vibration data of the rolling bearing by the rolling bearing evolution feature extractor, and train the deep convolutional neural network model based on the full-life vibration data of the rolling bearing in the normal stage; Step 3: Inputting the full-life vibration data of the rolling bearing into the trained deep convolutional neural network model, and obtaining the rolling bearing degradation index based on the characteristic distance between the normal sample and the fault sample; Step 4: Perform k-means clustering on the rolling bearing degradation index, divide the rolling bearing evolution stage into a normal stage, a degradation stage, and a failure stage, and mark the eigenvalue corresponding to the starting point of the degradation stage as the degradation threshold, and mark the eigenvalue corresponding to the starting point of the failure stage as the failure threshold; Step 5: Fit the rolling bearing degradation index before degradation by least squares fitting, and use the particle filter method to gradually predict the characteristic value from the beginning of degradation. When the characteristic value is greater than the failure threshold, the prediction algorithm is terminated, and the remaining life prediction curve of the rolling bearing from the beginning of degradation to complete failure is obtained; Step 6: Verify the correctness and effectiveness of the rolling bearing remaining life prediction method based on deep learning and particle filtering through the rolling bearing full life fatigue test vibration monitoring data.

2. The method for predicting the remaining life of a rolling bearing based on deep learning and particle filtering according to claim 1, characterized in that: In step 1, when constructing the deep convolutional neural network model, the best combination of deep convolutional neural networks is to gradually deepen the convolution-pooling-activation layers.

3. The method for predicting the remaining life of a rolling bearing based on deep learning and particle filtering according to claim 2, characterized in that: In step 1, when constructing the deep convolutional neural network model, the network structure and optimal network parameters are obtained by examining the network's loss accuracy and training time on the validation set.

Images