Rolling bearing life prediction method and device, electronic equipment and vehicle

Abstract

The invention provides a rolling bearing service life prediction method and device, electronic equipment and a vehicle, and the method comprises the steps: obtaining a current operation moment, a full-life-cycle health degradation index curve and a failure time threshold value of a tested rolling bearing, the full life cycle health degradation index curve comprises a slow degradation stage and an accelerated degradation stage of the test rolling bearing; a degradation model is regressed through a staged nonlinear random degradation coefficient, a degradation state value corresponding to the current operation moment of the tested rolling bearing is determined, and different functions are set for the model in a slow degradation stage and an accelerated degradation stage; through the full life cycle health degradation index curve, obtaining a target duration of the test rolling bearing from the current operation moment to the moment of the failure time threshold; the particle filter algorithm is combined with the degradation state value and the target duration to determine the probability density distribution prediction result and the residual duration prediction result of the residual life of the rolling bearing, and the accuracy of predicting the life of the rolling bearing is improved.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a rolling bearing life prediction method, device, electronic equipment and vehicle. Background Art

[0002] Rolling bearings are widely used in various rotating machinery and equipment. Their primary function is to connect two components that rotate relative to each other, or allow relative rotation, to achieve efficient and stable support. As the "joints" of mechanical systems, predicting the remaining useful life of rolling bearings is crucial to ensuring the normal and safe operation of rotating machinery and equipment.

[0003] Existing methods extract degradation features from vibration signals or multi-sensor data through adversarial learning or trend memory attention mechanism to construct health indicators, and then use degradation models and Kalman filtering to study the health indicators, capture the degradation laws of rolling bearings and predict the life of rolling bearings.

[0004] However, there are multiple degradation stages when rolling bearings degrade. Currently, a single degradation model is used, which leads to the neglect of the adaptability between the degradation characteristics of different degradation stages and the degradation model when predicting the life of rolling bearings. In addition, Kalman filtering is suitable for state estimation of linear systems, and it is difficult to perform effective state estimation for the nonlinear characteristics presented by rolling bearings during degradation, which leads to low accuracy in predicting the remaining life. Summary of the Invention

[0005] In view of this, the present invention aims to propose a rolling bearing life prediction method, device, electronic device, and vehicle to address the problems that current rolling bearing life prediction ignores the compatibility between degradation characteristics at different degradation stages and degradation models, and that it is difficult to effectively estimate the nonlinear characteristics exhibited by rolling bearings during degradation, resulting in low accuracy in prediction results. The specific technical solutions are as follows: According to a first aspect of the present invention, a method for predicting the life of a rolling bearing is provided, the method comprising: Obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, wherein the life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing; Determining the degradation state value corresponding to the test rolling bearing at the current operating moment by a staged nonlinear random degradation coefficient regression degradation model, wherein the staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and in the accelerated degradation stage; Obtaining the target duration of the test rolling bearing from the current operating moment to the moment at which the failure time threshold is reached through the full life cycle health degradation index curve; A particle filter algorithm is used in combination with the degradation state value and the target duration to determine a probability density distribution prediction result and a remaining duration prediction result of the remaining life of the test rolling bearing.

[0006] Optionally, before obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, the following steps may be further included: Obtain the vibration signal of the entire life cycle of the training rolling bearing; Extracting time domain, frequency domain and time-frequency domain features from the full life cycle vibration signal, and screening multi-dimensional sensitive degradation features from the time domain, frequency domain and time-frequency domain features; Inputting the multi-dimensional sensitive degradation features into a convolution module with a residual structure, and outputting a query volume carrying spatial dimension degradation information; Input the multi-dimensional sensitive degradation features into a pre-set autoencoder, and output a key value carrying time dimension degradation information; Matching the query amount with the key value to obtain a matching score; Training the hidden state vector of the final hidden layer included in the autoencoder using the matching score; Determining the hidden state vector of the trained final hidden layer as the low-dimensional embedding vector of the training rolling bearing; Obtaining a normal state vector of the rolling bearing; Determining an offset of the low-dimensional embedding vector using the normal state vector; Normalizing the offset to obtain a full life cycle health degradation index of the training rolling bearing; The target life cycle health degradation index curve of the training rolling bearing is determined by the life cycle health degradation index.

[0007] Optionally, the step of obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing further includes: Get the current running time of the test rolling bearing; Obtaining a health degradation index curve of a portion of the life cycle of the test rolling bearing running up to the current running moment; Obtaining similarities between the partial life cycle health degradation index curve and target full life cycle health degradation index curves of different categories of training rolling bearings; Determining the target category corresponding to the test rolling bearing in different categories of training rolling bearings through the similarity; Determining a full life cycle health degradation index curve of the test rolling bearing through the target category; The failure time threshold of the test rolling bearing is obtained through the health degradation index curve of the test rolling bearing throughout its life cycle.

[0008] Optionally, after obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, the method further includes: Determining the multiphase degradation change point of the test rolling bearing through the full life cycle health degradation index curve of the test rolling bearing; The slow degradation stage and the accelerated degradation stage of the test rolling bearing are determined by the multi-phase degradation change points.

[0009] Optionally, the staged nonlinear random degradation coefficient regression degradation model is obtained by the following formula:

[0010] in, yes The degraded state value at the moment, yes The degraded state value at the moment, is the inter-unit variability coefficient in the model that describes the differences between components of the same unit, is a non-decreasing function describing the degradation characteristics of rolling bearings, is the process noise that conforms to the Gaussian distribution, is The observed value measured at the moment, is the measurement noise that conforms to the Gaussian distribution. m is used to describe the degradation stage of the rolling bearing. PD means that the rolling bearing is in the slow degradation stage, and RD means that the rolling bearing is in the accelerated degradation stage. When m=PD, , when m=RD, , is the regression coefficient of the slow degradation stage, is the regression coefficient of the accelerated degradation phase, means time, means time.

[0011] Optionally, before using a particle filter algorithm in combination with the degradation state value and the target duration to determine the probability density distribution prediction result and the remaining duration prediction result of the tested rolling bearing, the method further includes: Obtaining model parameters to be calculated in the staged nonlinear random degradation coefficient regression degradation model, wherein the model parameters to be calculated include a mean value of an inter-unit variability coefficient, a variance of an inter-unit variability coefficient, a regression coefficient, a process noise variance, and a measurement noise variance; Obtaining available observation values ​​and estimated implicit state values ​​of the test rolling bearing at the current operating moment; A likelihood function is generated based on the available observation values ​​and the estimated implicit state value, and the current actual values ​​of the mean of the inter-unit variability coefficient, the variance of the inter-unit variability coefficient, the regression coefficient, the process noise variance and the measurement noise variance are determined using the likelihood function and the Bayesian criterion.

[0012] Optionally, the method of using a particle filter algorithm in combination with the degradation state value and the target duration to determine a probability density distribution prediction result and a remaining duration prediction result of the remaining life of the test rolling bearing further includes: Determine the normalized weight of the test rolling bearing by using a particle filter algorithm in combination with the degradation state value and the current actual value of the model parameter to be calculated; Determining a probability density distribution prediction result of the remaining life of the test rolling bearing by using the normalized weight; The remaining duration prediction result of the remaining life of the test rolling bearing is determined by the normalized weight and the target duration.

[0013] According to a second aspect of the present invention, a rolling bearing life prediction device is provided, the device comprising: A first acquisition module is used to obtain the current operating time, the life cycle health degradation index curve and the failure time threshold of the test rolling bearing, wherein the life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing; A first determination module is configured to determine a degradation state value corresponding to the test rolling bearing at a current operating moment by using a staged nonlinear random degradation coefficient regression degradation model, wherein the staged nonlinear random degradation coefficient regression degradation model is pre-set with different functions in the slow degradation stage and in the accelerated degradation stage; A second acquisition module is configured to obtain, through the full life cycle health degradation index curve, a target duration for the test rolling bearing to reach a failure time threshold from a current operating moment; The second determination module is used to use a particle filtering algorithm in combination with the degradation state value and the target duration to determine the probability density distribution prediction result and the remaining duration prediction result of the remaining life of the test rolling bearing.

[0014] According to another aspect of the present invention, there is provided an electronic device, comprising: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the rolling bearing life prediction method as described above.

[0015] According to another aspect of the present invention, a vehicle is provided, comprising: the above-mentioned rolling bearing life prediction device.

[0016] The rolling bearing life prediction method provided by the present invention obtains the current operating time of the test rolling bearing, the full life cycle health degradation index curve and the failure time threshold. The full life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing. The above information can be used to determine the current degradation state of the test rolling bearing. The degradation state value corresponding to the test rolling bearing at the current operating time is determined by a staged nonlinear random degradation coefficient regression degradation model. The staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and the accelerated degradation stage. By setting different functions for different degradation stages, the adaptability requirements between the degradation characteristics of different degradation stages and the degradation model are met. The full life cycle health degradation index curve is used to obtain the target duration from the current operating time of the test rolling bearing to the time when the failure time threshold is located. The particle filter algorithm is used to combine the degradation state value and the target duration to determine the probability density distribution prediction result and the remaining duration prediction result of the remaining life of the test rolling bearing. The particle filter algorithm is used for life prediction, which can effectively estimate the nonlinear characteristics presented during rolling bearing degradation and improve the accuracy of the remaining life prediction.

[0017] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings: Figure 1 This is a flowchart of the steps of a rolling bearing life prediction method provided by the present invention; Figure 2 yes Figure 1 The flowchart of step 101 in the rolling bearing life prediction method provided by the present invention is shown; Figure 3 yes Figure 1 The schematic diagram shown is a similarity clustering-based method for predicting the life of a rolling bearing provided by the present invention; Figure 4 yes Figure 1 The diagram shown is a schematic diagram of the degradation stage division in a rolling bearing life prediction method provided by the present invention; Figure 5 yes Figure 1 The diagram shown is a schematic diagram of prediction results output by different models in a rolling bearing life prediction method provided by the present invention; Figure 6 This is a structural schematic diagram of a rolling bearing life prediction device provided by the present invention; Figure 7 It is a structural schematic diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] To make the purpose, technical solutions and advantages of the present invention clearer, each embodiment of the present invention will be described in detail below with reference to the accompanying drawings. However, it will be understood by those skilled in the art that in each embodiment of the present invention, many technical details are provided to enable the reader to better understand the present invention. However, even without these technical details and the various changes and modifications according to the following embodiments, the technical solutions claimed in the present invention can be implemented. The division of the following embodiments is for convenience of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with each other and referenced to each other under the premise that there is no contradiction.

[0020] Existing methods for constructing bearing health indicators include automatically constructing an integrated health indicator by coupling multimodal samples of vibration signals, extracting degradation features through an adversarial autoencoder model and estimating the health indicator value using Bayes' theorem in probability space, effectively fusing historical health information and trend information through a gated recursive unit based on trend memory attention to estimate the health indicator of the rolling bearing, and designing an unsupervised deep autoencoder to fuse multiple nonlinear sensor data and estimate the health indicator through a residual-based similarity model. After constructing the health indicator, the rolling bearing life is predicted using physical model methods, data-driven methods, and hybrid methods. Physical model-based methods generally establish physical mathematical models based on mechanical failure principles or statistical models based on empirical knowledge to describe the degradation process of the system. Data-driven methods can be based on mathematical statistics knowledge to determine the statistical characteristics of the life distribution of the mechanical system from monitoring data and then predict the remaining life of the mechanical system.

[0021] However, the above-mentioned method for constructing health indicators does not adequately model the spatiotemporal correlation of the degradation process. In the methods for predicting the life of rolling bearings, most of them use a single degradation model to model different degradation stages, ignoring the adaptability between the degradation characteristics of different degradation stages and the random degradation model. In addition, current research on the degradation stage of bearings mostly uses state estimation methods such as Kalman filtering that are suitable for linear systems. It is difficult to effectively handle the nonlinear characteristics presented. Based on the above problems, the present invention proposes a method for predicting the life of rolling bearings. Figure 1 , shows a flowchart of the steps of a rolling bearing life prediction method provided by the present invention, the method may include: Step 101: Obtain the current operating time, life cycle health degradation index curve, and failure time threshold of the test rolling bearing.

[0022] Before obtaining the lifecycle health degradation index curve for a test rolling bearing, the present invention first requires obtaining target lifecycle health degradation index curves for several different types of training rolling bearings. Then, after determining the corresponding category of training rolling bearings for the test rolling bearing, the target lifecycle health degradation index curve for the corresponding training rolling bearing is used as the lifecycle health degradation index curve for the test rolling bearing.

[0023] Among them, when determining the target full life cycle health degradation index curve of the training rolling bearing, first, for any type of training rolling bearing, first obtain n full life cycle vibration signals of the training rolling bearing from health to failure, which can be expressed as X = [x1, x2, …, xn], where x1 refers to the full life cycle vibration signal 1, x2 refers to the full life cycle vibration signal 2, xn refers to the full life cycle vibration signal n, and n is the total number of signals (also known as length). After performing time domain, frequency domain and time-frequency domain analysis on the full life cycle vibration signal, the time domain, frequency domain and time-frequency domain features are obtained. The maximum correlation minimum redundancy algorithm is used to screen out several multidimensional sensitive degradation features with significant degradation characteristics in the corresponding analysis domains (i.e., time domain, frequency domain and time-frequency domain), which are expressed as F = [ f 1, f 2, f 3, … , f t ] TTo determine whether a multidimensional sensitive degradation feature has significant degradation characteristics, the features can be ranked by feature importance evaluation (such as random forest or ReliefF algorithm) or degradation indicators (such as monotonicity, robustness, and correlation), and then the top-ranked features (e.g., the top 30) are used as multidimensional sensitive degradation features. t is the total number of multidimensional sensitive degradation features, each of which contains m degradation features. f1 refers to multidimensional sensitive degradation feature 1 composed of the state values ​​of the m degradation features at the first moment, f2 refers to multidimensional sensitive degradation feature 2 composed of the state values ​​of the m degradation features at the second moment, and ft refers to multidimensional sensitive degradation feature t composed of the state values ​​of the m degradation features at the tth moment. The m degradation features are pre-calibrated according to business needs and can be adjusted according to different business needs and actual needs. The present invention does not impose specific limitations on this. The multi-dimensional sensitive degradation features are input into the convolution module with residual structure in batches according to the time window length w in the time dimension, and the spatial dimension degradation information is extracted as the query amount of the self-attention mechanism ( ), which provides a theoretical basis for the subsequent calculation of the correlation between the current moment information and information at other moments. The specific calculation process formula is as follows:

[0024] in, is a multidimensional sensitive degradation feature, is the output feature of the convolution module with residual structure, is the output feature of the linear variation layer 1, It is the output feature of the linear change layer 2, also known as the query volume carrying spatial dimension degradation information. is the transformation function of the convolutional layer and residual connection contained in the convolutional module with residual structure, It is a linear change layer 1 used for dimensional conversion. is the linear change layer 2 for dimensionality conversion, t is the number of samples of the multidimensional sensitive degradation feature, m is the feature dimension of the multidimensional sensitive degradation feature, and h is the dimension of the output feature.

[0025] At the same time, the multi-dimensional sensitive degradation features are input into the pre-set TSCA-Bi-GRU autoencoder. The encoder encoding stage will extract the time dimension degradation information and generate a key value carrying the time dimension degradation information to match the query volume. The specific calculation process of the key value is as follows:

[0026] in, is a multidimensional sensitive degradation feature, Refers to the conversion function of the encoding part in the autoencoder, is the hidden state vector of the final hidden layer contained in the autoencoder, It is the vector of the hidden state output of the autoencoder at each moment, also known as the key value that carries the time dimension degradation information. is the final hidden state vector of the forward propagation recurrent neural network, is the final hidden state vector of the back-propagated recurrent neural network.

[0027] The query amount and key value are input into the self-attention layer for matching. The formula is as follows:

[0028] in, is the self-attention function, score is the attention score matrix, It is the vector of the hidden state output of the autoencoder at each moment, also known as the key value that carries the time dimension degradation information. It is the output feature of the linear change layer 2, also known as the query volume carrying spatial dimension degradation information. is the value matrix in the self-attention layer, is the output vector of the self-attention layer, is the vector of the hidden state output of the decoder at each moment, is the hidden state vector output by the final hidden layer of the decoder.

[0029] Through Output after decoding ,Will After inputting the linear prediction layer, the decoded output is finally obtained . Calculate the input value of the model and output value The difference between and is used to obtain the reconstruction error of the encoder at each time step. The calculation formula of the reconstruction error is as follows:

[0030] in, is the 1-norm operator of a vector, It is a multidimensional sensitive degradation feature at the i-th moment (each moment corresponds to a time step), yes The corresponding value of the decoded output at the i-th moment, t is the number of samples, and E is the reconstruction error of the encoder at each time step.

[0031] The present invention trains the autoencoder by reconstructing the error value and adjusting the When the reconstruction error converges to a certain degree (customizable), the training of the autoencoder is completed. It can be used as a low-dimensional embedding vector for training rolling bearings and is expressed as: .

[0032] Multidimensional sensitive degradation features F = [ f 1, f 2, f 3, … , f t ] T , each are composed of the state values ​​of m degenerate features at time i, for example, f 1= [ d 1, d 2, d 3, …, d m ], set a fixed window of size w that slides along the time direction on F, and after the sliding window processing, a window matrix consisting of t-w+1 windows will be generated , then the i-th window can be expressed as The spatiotemporal fusion of multi-dimensional sensitive degradation features from the autoencoder calculation can be transformed into a univariate time series carrying the bearing degradation state. In the early stages of operation, the health of the bearing has not yet begun to deteriorate. Therefore, the first few (for example, the first three) low-dimensional embedding vectors are considered to be in a healthy state, and the subsequent low-dimensional embedding vectors are used to calculate the degree of deviation from the normal state vector. The formula is as follows:

[0033] in, is the offset of the multidimensional sensitive degradation feature at the tth moment, N is the number of samples, is the normal state vector, is the low-dimensional embedding vector at time t, is the 2-norm operator of a vector.

[0034] By Normalized to the interval [0, 1] to obtain a one-dimensional health degradation index, the formula is as follows:

[0035] in, is the maximum value that deviates from the normal state during operation, is the minimum deviation from the normal state during operation, is the offset of the multidimensional sensitive degradation feature at the tth moment, is the health degradation indicator of the training rolling bearing at the tth moment.

[0036] By obtaining the health degradation index of the training rolling bearing at each moment, the health degradation index of the training rolling bearing over its entire life cycle can be obtained. The target health degradation index curve of the training rolling bearing over its entire life cycle can be generated by using the health degradation index of the training rolling bearing over its entire life cycle. Therefore, the steps for generating the target health degradation index curve of the training rolling bearing over its entire life cycle include: Obtain the vibration signal of the entire life cycle of the training rolling bearing; Extract time domain, frequency domain and time-frequency domain features from the full life cycle vibration signal, and screen out multi-dimensional sensitive degradation features from the time domain, frequency domain and time-frequency domain features; The multi-dimensional sensitive degradation features are input into the convolution module with a residual structure, and the query volume carrying the spatial dimension degradation information is output; Input the multi-dimensional sensitive degradation features into the pre-set autoencoder and output the key value carrying the time dimension degradation information; Match the query volume and key value to obtain the matching score; The hidden state vector of the final hidden layer of the autoencoder is trained using the matching score; The hidden state vector of the trained final hidden layer is determined as the low-dimensional embedding vector of the training rolling bearing; Obtaining a normal state vector of the rolling bearing; Determine the offset of the low-dimensional embedding vector through the normal state vector; The offset is normalized to obtain the health degradation index of the training rolling bearing throughout its life cycle; The target life cycle health degradation index curve of the training rolling bearing is determined through the life cycle health degradation index.

[0037] The above steps fuse multidimensional sensitive features in the time domain, frequency domain, and time-frequency domain, combine the residual convolution module and autoencoder to capture degradation information in the spatial and temporal dimensions respectively, and use the attention mechanism to match the correlation between the two. Finally, a low-dimensional embedding vector is generated and the health state offset is quantified. The resulting health degradation indicator can not only capture the overall degradation trend in the temporal dimension, but also capture important degradation difference information in the spatial dimension, thereby improving the robustness of the degradation trend representation.

[0038] In the present invention, step 101 is as follows: Figure 2 As shown: Step 1011, obtaining the current operating time of the test rolling bearing.

[0039] Step 1012: Obtain a health degradation index curve of a portion of the life cycle of the test rolling bearing running up to the current running moment.

[0040] Step 1013: Obtain the similarity between the partial life cycle health degradation index curve and the target full life cycle health degradation index curve of different categories of training rolling bearings.

[0041] Step 1014 , determining the target category corresponding to the test rolling bearing among the training rolling bearings of different categories through similarity.

[0042] Step 1015: Determine the full life cycle health degradation index curve of the test rolling bearing through the target category.

[0043] Step 1016: Obtain the failure time threshold of the tested rolling bearing by testing the full life cycle health degradation index curve of the rolling bearing.

[0044] Among them, the test of the life cycle health degradation index of the rolling bearing running to the current running time is also achieved by referring to the steps of training the bearing to generate the life cycle health degradation index, and the present invention will not repeat it here. The running time is not recorded according to the timestamp, but is recorded from 0 according to the working time of the rolling bearing. The running time can be used to determine the elapsed running time. For example, if the running time is the 5th minute, the elapsed running time is 5 minutes. However, the running time is a time point, and the running time is a time period.

[0045] When calculating similarity, the present invention uses a longest common subsequence based on adaptive control parameters to measure the similarity of the health degradation indicators of the training rolling bearing and the test rolling bearing. The calculation formula is as follows:

[0046] in, is the health degradation index of the test rolling bearing with a length of m, is the health degradation index of the training rolling bearing with a length of n throughout its life cycle, Express and The length of the longest common subsequence after similarity evaluation; is the health degradation index of the test rolling bearing with a length of m-1, is the health degradation index of the training rolling bearing with a length of n-1 throughout its life cycle, Express and The longest common subsequence length after similarity evaluation, max(.) represents the function of taking the maximum value, Represents the threshold parameter, which is used to control the allowable error of the similarity between two points and eliminate the influence of noise on similarity matching; Represents the weight parameter, which is used to control the weight of the common subsequence at different degradation levels, taking into account the nonlinear relationship between similarity and lifespan during the matching process. Based on the calculated longest common subsequence length, calculate

[0047] in, represents the function that takes the minimum value, Indicates the similarity between the test rolling bearing and the training rolling bearing.

[0048] The present invention collects different categories of training rolling bearings, obtains their corresponding full life cycle health degradation indicators, and generates a target full life cycle health degradation indicator curve for the training rolling bearings. Then, the similarity between the test rolling bearings and the training rolling bearings of different categories is calculated. Based on the similarity analysis results, the rolling bearings with higher similarity have more similar degradation trajectories, and they are clustered, such as Figure 3 As shown in the figure, after clustering, three clusters are obtained, namely, category cluster 1, category cluster 2, and category cluster 3. Based on the clustering results, the target life cycle health degradation index curve of the training rolling bearing of the same category can be used as the life cycle health degradation index curve of the test rolling bearing.

[0049] The present invention can determine the failure time threshold of the tested rolling bearing based on the obtained lifecycle health degradation index curve. When determining the failure time threshold, a sliding window of length d is first set. The sliding window is used to divide the obtained lifecycle health degradation index curve into multiple different window intervals. A linear regression model is used to fit the one-dimensional health degradation index under different sliding windows. The linear regression model is:

[0050] Among them, x represents the time, y represents the health degradation index, w and b represent the fitting parameters of the model, and w is used to describe the gradient of the dependent variable relative to the independent variable. and They represent the health degradation index at the i-th moment and the i-th moment under a certain sliding window interval, and d represents the sliding window length.

[0051] Given any interval , the fitted w<0 indicates that the bearing is in a healthy state at the current moment, and w≥0 indicates that the bearing begins to degrade. Compare, if it is less than the given gradient threshold , then continue to determine the gradient of the next moment until a certain window length The gradient under is greater than the given gradient threshold , output the y value at this time as the failure time threshold.

[0052] The above steps cluster the test rolling bearings and training rolling bearings by calculating similarity, so that the full life cycle curve of the test rolling bearing can be estimated by the full life cycle curve of the training bearing of known category, and then the failure time threshold of the test rolling bearing can be predicted, thereby achieving an accurate assessment of the health status of the test rolling bearing and facilitating the subsequent prediction of the remaining rolling bearing life.

[0053] The life cycle health degradation index curve of the present invention includes a slow degradation stage and an accelerated degradation stage of the rolling bearing, wherein the steps of determining the slow degradation stage and the accelerated degradation stage include: By testing the health degradation index curve of the rolling bearing throughout its life cycle, the multi-phase degradation change point of the tested rolling bearing is determined; The slow degradation stage and accelerated degradation stage of the tested rolling bearing are determined by multi-phase degradation change points.

[0054] Among them, when determining the multi-phase degradation change point, it is necessary to first determine the performance degradation angle and angle change rate of the health degradation indicator, determine the degradation change point set based on the angle change rate, and further screen the degradation change point set to determine the multi-phase degradation change point. When calculating the performance degradation angle of the health degradation indicator, first set a sliding window of length l. Using the sliding window, the obtained full life cycle health degradation indicator curve is divided into multiple different window intervals. Then, the performance degradation angle of the health degradation indicator in each window is calculated. The formula is as follows:

[0055] in, Represents the i-th window on the life cycle health degradation index curve of the tested rolling bearing and The angle between the line connecting the two points and the horizontal line is defined as the performance degradation angle; l is the sliding window length; arctan is the inverse tangent function, and are the coordinates of points at different moments in the health degradation index curve of the entire life cycle, , is the change in the horizontal and vertical coordinates of the two points.

[0056] The angular change rate of the performance degradation angle is calculated as follows:

[0057] in, is the horizontal coordinate of the life cycle health degradation index curve at the kth moment, yes and The change between the horizontal coordinates, is the angular rate of change of the performance degradation angle.

[0058] Filter the local maximum of the performance degradation angle change rate and output the degradation change point set of the whole life cycle health degradation index curve, which is the degradation change point set. Then set a sliding window of length T, and use the sliding window to divide the obtained whole life cycle health degradation index curve into multiple different window intervals. Then calculate the slowness evolution rate of the i-th change point in the degradation change point set, extending forward for q consecutive windows, and extending backward for q consecutive windows. The calculation formula of the slowness evolution rate is as follows:

[0059] in, is the horizontal coordinate of the health degradation index of the i-th change point in the degradation change point set, is the horizontal coordinate of the health degradation index of the i+1th change point in the degradation change point set, is the unit time, T is the length of the time window, is the slowness evolution rate of the i-th change point, is the slowness corresponding to the i-th change point, is the slowness corresponding to the i-1th change point.

[0060] According to the degradation pattern of multi-phase degraded rolling bearings, since the slowness evolution rate will increase significantly at the stage change point, it is set to 0.1 times the maximum slowness evolution rate according to experimental results, and the number of extended windows before and after the candidate inflection point is set to 5. Determine the slowness change rate of the i-th change point Is it a maximum value and is the q consecutive slowness change rates at the change point less than a given threshold? If so, the i-th change point is determined to be a multi-phase degradation change point.

[0061] Normally, there are two or one multiphase degradation change points in the life cycle health degradation index curve of a rolling bearing. If there are two multiphase degradation change points, the time between the two multiphase degradation change points is determined as the slow degradation stage of the test rolling bearing, and the time after the second multiphase degradation change point is determined as the accelerated degradation stage of the test rolling bearing. The time when the failure time threshold is located is in the slow degradation stage. Figure 4 If there is only one multiphase degradation change point, the time before the multiphase degradation change point is determined as the slow degradation stage of the test rolling bearing, and the time after the multiphase degradation change point is determined as the accelerated degradation stage of the test rolling bearing.

[0062] The slow degradation stage and accelerated degradation stage of the tested rolling bearing are determined by multi-phase degradation change points, which facilitates the subsequent creation of a staged nonlinear random degradation coefficient regression degradation model. Different functions can be set for different degradation stages to describe the changing trends of different degradation stages of the bearing over time.

[0063] Step 102 : Determine the degradation state value corresponding to the test rolling bearing at the current operating moment by regressing the degradation model using a staged nonlinear random degradation coefficient.

[0064] In the staged nonlinear random degradation coefficient regression degradation model, the present invention sets different functions for the slow degradation stage and the accelerated degradation stage. The formula is as follows: (twenty one) in, yes The degraded state value at the moment, yes The degraded state value at the moment, is the inter-unit variability coefficient in the model that describes the differences between components of the same unit, is a non-decreasing function describing the degradation characteristics of rolling bearings, is the process noise that conforms to the Gaussian distribution, is The observed value measured at the moment, is the measurement noise that conforms to the Gaussian distribution. m is used to describe the degradation stage of the rolling bearing. PD means that the rolling bearing is in the slow degradation stage, and RD means that the rolling bearing is in the accelerated degradation stage. When m=PD, , when m=RD, , is the regression coefficient of the slow degradation stage, is the regression coefficient of the accelerated degradation phase, means time, means Moment. And Satisfies normal distribution ~N( , ), satisfy ~ N(0, ), satisfy ~ N(0, ). It can be seen that the power function is set in the slow degradation stage, and the exponential function is set in the accelerated degradation stage.

[0065] After obtaining the staged nonlinear random degradation coefficient regression degradation model, the present invention needs to calculate the unknown model parameters therein. The model parameters to be calculated include the mean value of the inter-unit variability coefficient , variance of the coefficient of variability between units , regression coefficient , process noise variance and measurement noise variance When calculating the model parameters, we need to use the observed values and the estimated implicit state value Calculation is carried out, and from formula (21), we can know that , due to measurement noise Much smaller than the observed value By itself, the measurement noise can usually be neglected. Therefore, the state value Can be approximated as observed value ,make ,in Represents the transpose of the vector, and the observation value is , obeys the multivariate normal distribution, and the likelihood function of the observed value is as follows:

[0066] Among them, the process noise variance and measurement noise variance Variance of coefficient of variability between units Has the following relationship:

[0067] For formula (22), and The first-order partial derivative of and Partial differential equations:

[0068] Let the above two partial differential equations (24) and (25) be 0, then we can obtain and The analytical expression of :

[0069] Substituting formulas (26) and (27) into formula (22), the only remaining model parameter to be calculated for the likelihood function is the regression coefficient , process noise variance and measurement noise variance ,

[0070] By multi-dimensional optimization, formula (28) is maximized and the regression coefficient is obtained. , process noise variance and measurement noise variance Substituting the estimated value into formula (26) and (27) yields the maximum likelihood estimate of and The initial value of .

[0071] The observed values ​​for the above settings are , when acquiring new observations ( S 1:k ), the degradation model parameters can be updated according to the Bayesian criterion. The following formula can be obtained from the Bayesian criterion:

[0072] Through the properties of the conjugate prior, we can get the expression of the updated posterior distribution parameters corresponding to the new observation value at the i-th moment:

[0073] Among them, when the observation data is in the slow degradation stage, When the observation data is in the accelerated degradation stage, , the updated and Substitute into formula (22) and obtain according to multidimensional search 、 and . So far, the update of model parameters has been realized.

[0074] The meanings of the parameters appearing in formulas (22) to (30) are as follows: It's about Likelihood function, M is the observed value length, is the variance of the coefficient of variability between units, yes, is the process noise variance, is the measurement noise variance, is the diagonal matrix in the maximum likelihood function, is the diagonal matrix in the maximum likelihood function, is the observation value from time 1 to M, is the mean of the coefficient of variability between units, are the model parameters to be calculated, yes The amount of change, is the i-th moment, is the i-1th moment, is the observation value at the i-th moment, is the observation value at the i-1th moment. When the observation data is in the slow degradation stage, When the observation data is in the accelerated degradation stage, .

[0075] According to the above content, the calculation steps of the model parameters to be calculated in the staged nonlinear random degradation coefficient regression degradation model include: Obtaining the model parameters to be calculated in the staged nonlinear random degradation coefficient regression degradation model, the model parameters to be calculated including the mean value of the inter-unit variability coefficient, the variance of the inter-unit variability coefficient, the regression coefficient, the process noise variance, and the measurement noise variance; Obtain available observation values ​​and estimated implicit state values ​​of the test rolling bearing at the current operating moment; A likelihood function is generated based on the available observations and the estimated latent state values, and the current actual values ​​of the mean of the inter-unit variability coefficient, the variance of the inter-unit variability coefficient, the regression coefficient, the process noise variance, and the measurement noise variance are determined using the likelihood function and the Bayesian criterion.

[0076] The above steps combine the real-time observation data and implicit state estimation of the test bearings, and use the likelihood function and Bayesian method to accurately calculate the key degradation parameters (model parameters to be calculated), thereby realizing dynamic modeling and parameter optimization of the bearing degradation process, and improving the accuracy and reliability of subsequent remaining rolling bearing life prediction.

[0077] Step 103 , obtaining the target duration for the test rolling bearing to reach the failure time threshold from the current operating moment through the full life cycle health degradation index curve.

[0078] When calculating the target duration, the present invention first determines the time required to reach the failure time threshold using the full life cycle health degradation index curve, and then determines the operating time of the test rolling bearing using the current operating time of the test rolling bearing. The target duration is determined by combining the time required to reach the failure time threshold and the operating time. The formula for the target duration is:

[0079] in, yes The moment when the degradation state of the rolling bearing reaches the failure threshold for the first time The corresponding length of time, represents the failure threshold, yes ,..., The estimated degradation state, yes The degradation state predicted at any moment in the future is predicted by the conversion function of the following formula: in, is the time interval between adjacent degenerate states and j∈N={0, 1, 2, …}, yes At the moment, when m=PD, , when m=RD, , is the regression coefficient of the slow degradation stage, is the regression coefficient of the accelerated degradation phase, yes The mean of the coefficient of variability between units at each moment, yes The moment-to-moment process noise variance.

[0080] Step 104 : Using a particle filter algorithm in combination with the degradation state value and the target duration, determine a probability density distribution prediction result and a remaining duration prediction result of the test rolling bearing remaining life.

[0081] After determining the current actual values ​​of the model parameters to be calculated in the staged nonlinear random degradation coefficient regression degradation model (the current actual values ​​are dynamically updated based on the available observation values ​​and estimated implicit state values ​​at different times), the present invention estimates the degradation of the test rolling bearing based on the particle filtering algorithm, including estimating the probability density distribution of the remaining life and the remaining time. The implementation steps include: The particle filter algorithm is used to combine the degradation state value and the current actual value of the model parameter to be calculated to determine the normalized weight of the test rolling bearing; Determine the probability density distribution prediction result of the remaining life of the test rolling bearing by normalizing the weight; The remaining time prediction result of the remaining life of the test rolling bearing is determined by the normalized weight and the target time.

[0082] Among them, when calculating the normalized weight based on the particle filter algorithm, it is necessary to first calculate the current running time of the test rolling bearing ( Time) to determine the previous running time ( The degradation state value ( x k-1 ) at the previous running time, the posterior probability density distribution is as follows:

[0083] in, p ( x k-1 | s 1:k-1 ) is the degenerate state in The posterior probability density distribution at time , p ( x k | x k-1 ) is the state transition probability density function, p ( x k | s 1:k-1 ) indicates the degenerate state in Prior probability distribution of time.

[0084] After determining the prior probability distribution, the prior probability density distribution is updated according to the Bayesian criterion to obtain the degraded state in The posterior probability density distribution at the moment is as follows:

[0085] in, p ( x k | s 1:k ) is the degenerate state in The posterior probability density distribution at time , p ( S k | x k ) is the likelihood function, p ( x k | s 1:k-1 ) indicates the degenerate state in The prior probability distribution of time, p ( S k | s 1:k ) is the marginal likelihood function. It is difficult to obtain an analytical solution for the posterior probability density distribution under nonlinear tasks. Therefore, the sequential importance sampling algorithm is used to obtain the importance distribution function q ( x k | x k-1， s 1:k ) to sample a set of particles { x i k} i=1: N To approximate the posterior probability density distribution shown in formula (35). At this time, the normalized weight corresponding to each particle must be determined first, as shown in the following formula:

[0086] in, is the likelihood function of the ith particle, is the state transition probability density function of the i-th particle, is the importance distribution function of the i-th particle.

[0087] Then the importance distribution function q ( x k | x k-1， s1:k ) is set as the state transition probability density function p ( x k | x k-1 ),Right now , the weight formula can be obtained as follows:

[0088] in, is the likelihood function of the ith particle, is the number of particles in The weight of the moment, is the number of particles in The weight of the moment, is the measurement noise variance, yes The degraded state value at the moment, is Observed value at time.

[0089] According to formula (38), the bearing health status is approximately obtained The posterior probability distribution at time t is as follows:

[0090] in, is the number of particles in The weight of the moment, p ( x k | s 1:k ) is the degenerate state in The posterior probability density distribution at time , yes The degraded state value at the moment, is the number of particles in The degraded state value at the moment, N is the number of particle samples.

[0091] It should be understood that particle degradation may occur during the prediction and update process of the particle filter algorithm. When particles degenerate, they need to be resampled. By introducing the effective sample size to determine whether particles have degenerated, the formula is as follows:

[0092] in, is the effective sample size, is the number of particles in The weight of the moment, N is the number of particle samples. If it appears during the update The value is less than a pre-set threshold In the case of , N particles need to be resampled and the particle weights are reset to 1 / N. Set to 2N / 3. With continuous iteration and update, particles with higher weights are retained. After the weights are determined, the weights can be determined based on the weights. The estimated value of the bearing degradation state at time is as follows:

[0093] in, is the number of particles in The weight of the moment, N is the number of particle samples, is the number of particles in The degraded state value at the moment, yes Estimated value of bearing degradation state at time t.

[0094] Based on the above content, the present invention uses the particle filter algorithm and the target duration to predict the life of the test rolling bearing. The probability density distribution prediction result of the remaining life at the moment is obtained by the following formula:

[0095] The remaining lifespan prediction result at the moment is obtained by the following formula:

[0096] Among them, the formulas (41) and (42) are yes The moment when the degradation state of the rolling bearing reaches the failure threshold for the first time The corresponding target duration, is the number of particles in The moment when the degradation state of the rolling bearing reaches the failure threshold for the first time The corresponding target duration, is the number of particles in The weight of the moment, N is the number of particle samples, p ( l k | s 1:k ) is the degenerate state in The probability density distribution prediction result of the remaining life of the test rolling bearing at the time (current running time) is: is in a degenerate state The remaining time prediction result of the remaining life of the test rolling bearing at the current running time.

[0097] The above steps effectively solve the uncertainty problem in the nonlinear degradation process, realize the upgrade of life prediction from point estimation to probabilistic evaluation, and provide a more robust decision-making basis for predictive maintenance.

[0098] The rolling bearing life prediction is performed by the staged nonlinear random degradation coefficient regression degradation model of the present invention. Compared with the single power function degradation model and the single exponential function degradation model, the prediction result is more consistent with the actual situation, such as Figure 5 shown.

[0099] The rolling bearing life prediction method provided by the present invention obtains the current operating time of the test rolling bearing, the full life cycle health degradation index curve and the failure time threshold. The full life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing. The above information can be used to determine the current degradation state of the test rolling bearing. The degradation state value corresponding to the test rolling bearing at the current operating time is determined by a staged nonlinear random degradation coefficient regression degradation model. The staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and the accelerated degradation stage. By setting different functions for different degradation stages, the adaptability requirements between the degradation characteristics of different degradation stages and the degradation model are met. The full life cycle health degradation index curve is used to obtain the target duration from the current operating time of the test rolling bearing to the time when the failure time threshold is located. The particle filter algorithm is used to combine the degradation state value and the target duration to determine the probability density distribution prediction result and the remaining duration prediction result of the remaining life of the test rolling bearing. The particle filter algorithm is used for life prediction, which can effectively estimate the nonlinear characteristics presented during rolling bearing degradation and improve the accuracy of the remaining life prediction.

[0100] Reference Figure 6 , shows a schematic structural diagram of a rolling bearing life prediction device provided by the present invention, the device comprising: The first acquisition module 201 is used to obtain the current operating time, the life cycle health degradation index curve and the failure time threshold of the test rolling bearing. The life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing.

[0101] The first determination module 202 is used to determine the degradation state value corresponding to the test rolling bearing at the current operating moment through a staged nonlinear random degradation coefficient regression degradation model. The staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and the accelerated degradation stage.

[0102] The second acquisition module 203 is used to obtain the target time length from the current operating moment to the moment when the test rolling bearing reaches the failure time threshold through the full life cycle health degradation index curve.

[0103] The second determination module 204 is configured to determine a probability density distribution prediction result and a remaining life prediction result of the test rolling bearing by using a particle filter algorithm in combination with the degradation state value and the target life.

[0104] Optionally, the rolling bearing life prediction device further includes: The third acquisition module is used to obtain the vibration signal of the entire life cycle of the training rolling bearing.

[0105] The feature screening module is used to extract time domain, frequency domain and time-frequency domain features from the full life cycle vibration signal, and to screen out multi-dimensional sensitive degradation features from the time domain, frequency domain and time-frequency domain features.

[0106] The first input-output module is used to input the multi-dimensional sensitive degradation features into a convolution module with a residual structure, and output the query volume carrying spatial dimension degradation information.

[0107] The second input-output module is used to input the multi-dimensional sensitive degradation features into a pre-set autoencoder and output a key value carrying the time dimension degradation information.

[0108] The matching module is used to match the query volume and key value to obtain the matching score.

[0109] The training module is used to train the hidden state vector of the final hidden layer included in the autoencoder through the matching score.

[0110] The third determination module is used to determine the hidden state vector of the trained final hidden layer as the low-dimensional embedding vector of the training rolling bearing.

[0111] The third acquisition module is used to acquire a normal state vector of the rolling bearing.

[0112] The fourth determining module is configured to determine an offset of the low-dimensional embedding vector using the normal state vector.

[0113] The normalization processing module is used to normalize the offset to obtain the health degradation index of the training rolling bearing throughout its life cycle.

[0114] The fifth determination module is used to determine the target life cycle health degradation index curve of the training rolling bearing through the life cycle health degradation index.

[0115] Optionally, the first obtaining module 201 includes: The first acquisition submodule is used to acquire the current running time of the test rolling bearing.

[0116] The second acquisition submodule is used to obtain a health degradation index curve of a portion of the life cycle of the test rolling bearing running to the current running moment.

[0117] The third acquisition submodule is used to obtain the similarity between the partial life cycle health degradation index curve and the target full life cycle health degradation index curve of different categories of training rolling bearings.

[0118] The first determination submodule is used to determine the target category corresponding to the test rolling bearing in different categories of training rolling bearings through similarity.

[0119] The second determination submodule is used to determine the full life cycle health degradation index curve of the test rolling bearing through the target category.

[0120] The fourth acquisition submodule is used to obtain the failure time threshold of the test rolling bearing by testing the health degradation index curve of the rolling bearing throughout its life cycle.

[0121] Optionally, the rolling bearing life prediction device further includes: The sixth determination module is used to determine the multiphase degradation change point of the test rolling bearing by testing the health degradation index curve of the rolling bearing throughout its life cycle.

[0122] The seventh determination module is used to determine the slow degradation stage and the accelerated degradation stage of the test rolling bearing through multi-phase degradation change points.

[0123] Optionally, the staged nonlinear random degradation coefficient regression degradation model is obtained by the following formula:

[0124] in, yes The degraded state value at the moment, yes The degraded state value at the moment, is the inter-unit variability coefficient in the model that describes the differences between components of the same unit, is a non-decreasing function describing the degradation characteristics of rolling bearings, is the process noise that conforms to the Gaussian distribution, is The observed value measured at the moment, is the measurement noise that conforms to the Gaussian distribution. m is used to describe the degradation stage of the rolling bearing. PD means that the rolling bearing is in the slow degradation stage, and RD means that the rolling bearing is in the accelerated degradation stage. When m=PD, , when m=RD, , is the regression coefficient of the slow degradation stage, is the regression coefficient of the accelerated degradation phase, means time, means time.

[0125] Optionally, the rolling bearing life prediction device further includes: The fourth acquisition module is used to obtain the model parameters to be calculated in the staged nonlinear random degradation coefficient regression degradation model, and the model parameters to be calculated include the mean of the inter-unit variability coefficient, the variance of the inter-unit variability coefficient, the regression coefficient, the process noise variance and the measurement noise variance.

[0126] The fifth acquisition module is used to obtain the available observation values ​​and estimated implicit state values ​​of the test rolling bearing at the current operating moment.

[0127] The eighth determination module is used to generate a likelihood function based on the available observation values ​​and the estimated implicit state value, and use the likelihood function and the Bayesian criterion to determine the current actual values ​​of the mean of the inter-unit variability coefficient, the variance of the inter-unit variability coefficient, the regression coefficient, the process noise variance and the measurement noise variance.

[0128] Optionally, the second determining module 204 includes: The third determination submodule is used to determine the normalized weight of the test rolling bearing by using a particle filter algorithm in combination with the degradation state value and the current actual value of the model parameter to be calculated.

[0129] The fourth determination submodule is used to determine the probability density distribution prediction result of the remaining life of the test rolling bearing through normalized weights.

[0130] The fifth determination submodule is used to determine the remaining time prediction result of the remaining life of the test rolling bearing through the normalized weight and the target time.

[0131] The rolling bearing life prediction method provided by the present invention obtains the current operating time of the test rolling bearing, the full life cycle health degradation index curve and the failure time threshold. The full life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing. The above information can be used to determine the current degradation state of the test rolling bearing. The degradation state value corresponding to the test rolling bearing at the current operating time is determined by a staged nonlinear random degradation coefficient regression degradation model. The staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and the accelerated degradation stage. By setting different functions for different degradation stages, the adaptability requirements between the degradation characteristics of different degradation stages and the degradation model are met. The full life cycle health degradation index curve is used to obtain the target duration from the current operating time of the test rolling bearing to the time when the failure time threshold is located. The particle filter algorithm is used to combine the degradation state value and the target duration to determine the probability density distribution prediction result and the remaining duration prediction result of the remaining life of the test rolling bearing. The particle filter algorithm is used for life prediction, which can effectively estimate the nonlinear characteristics presented during rolling bearing degradation and improve the accuracy of the remaining life prediction.

[0132] Reference Figure 7 , the present invention also provides an electronic device, such as Figure 7 As shown, it includes a processor 301, a communication interface 302, a memory 303 and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304. Processor 301, memory 303 for storing processor-executable instructions; The processor 301 is configured to execute the instructions to implement the rolling bearing life prediction method as described above: Obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, wherein the life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing; Determining the degradation state value corresponding to the test rolling bearing at the current operating moment by a staged nonlinear random degradation coefficient regression degradation model, wherein the staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and in the accelerated degradation stage; Obtaining the target duration of the test rolling bearing from the current operating moment to the moment at which the failure time threshold is reached through the full life cycle health degradation index curve; A particle filter algorithm is used in combination with the degradation state value and the target duration to determine a probability density distribution prediction result and a remaining duration prediction result of the remaining life of the test rolling bearing.

[0133] The communication bus mentioned in the terminal above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into address buses, data buses, control buses, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0134] The communication interface is used for communication between the above terminal and other devices.

[0135] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0136] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0137] In another embodiment provided by the present invention, a vehicle is further provided, which may specifically include: the above-mentioned rolling bearing life prediction device.

[0138] In the above embodiments, all or part of the embodiments can be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state drive (SSD)).

[0139] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0140] Each embodiment in this specification is described in a related manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For related parts, refer to the description of the method embodiment.

[0141] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. A method for predicting the life of a rolling bearing, characterized in that: The method comprises: Obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, wherein the life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing; Determining the degradation state value corresponding to the test rolling bearing at the current operating moment by a staged nonlinear random degradation coefficient regression degradation model, wherein the staged nonlinear random degradation coefficient regression degradation model pre-sets different functions in the slow degradation stage and in the accelerated degradation stage; Obtaining the target duration of the test rolling bearing from the current operating moment to the moment at which the failure time threshold is reached through the full life cycle health degradation index curve; A particle filter algorithm is used in combination with the degradation state value and the target duration to determine a probability density distribution prediction result and a remaining duration prediction result of the remaining life of the test rolling bearing.

2. The method according to claim 1, characterized in that Before obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, the following steps are also included: Obtain the vibration signal of the entire life cycle of the training rolling bearing; Extracting time domain, frequency domain and time-frequency domain features from the full life cycle vibration signal, and screening multi-dimensional sensitive degradation features from the time domain, frequency domain and time-frequency domain features; Inputting the multi-dimensional sensitive degradation features into a convolution module with a residual structure, and outputting a query volume carrying spatial dimension degradation information; Input the multi-dimensional sensitive degradation features into a pre-set autoencoder, and output a key value carrying time dimension degradation information; Matching the query amount with the key value to obtain a matching score; Training the hidden state vector of the final hidden layer included in the autoencoder using the matching score; Determining the hidden state vector of the trained final hidden layer as the low-dimensional embedding vector of the training rolling bearing; Obtaining a normal state vector of the rolling bearing; Determining an offset of the low-dimensional embedding vector using the normal state vector; Normalizing the offset to obtain a full life cycle health degradation index of the training rolling bearing; The target life cycle health degradation index curve of the training rolling bearing is determined by the life cycle health degradation index.

3. The method according to claim 2, characterized in that The step of obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing further includes: Get the current running time of the test rolling bearing; Obtaining a health degradation index curve of a portion of the life cycle of the test rolling bearing running up to the current running moment; Obtaining similarities between the partial life cycle health degradation index curve and target full life cycle health degradation index curves of different categories of training rolling bearings; Determining the target category corresponding to the test rolling bearing in different categories of training rolling bearings through the similarity; Determining a full life cycle health degradation index curve of the test rolling bearing through the target category; The failure time threshold of the test rolling bearing is obtained through the health degradation index curve of the test rolling bearing throughout its life cycle.

4. The method according to claim 1, wherein After obtaining the current operating time, the life cycle health degradation index curve, and the failure time threshold of the test rolling bearing, the method further includes: Determining the multiphase degradation change point of the test rolling bearing through the full life cycle health degradation index curve of the test rolling bearing; The slow degradation stage and the accelerated degradation stage of the test rolling bearing are determined by the multi-phase degradation change points.

5. The method according to claim 1, wherein The staged nonlinear random degradation coefficient regression degradation model is obtained by the following formula: in, yes The degraded state value at the moment, yes The degraded state value at the moment, is the inter-unit variability coefficient in the model that describes the differences between components of the same unit, is a non-decreasing function describing the degradation characteristics of rolling bearings, is the process noise that conforms to the Gaussian distribution, is The observed value measured at the moment, is the measurement noise that conforms to the Gaussian distribution. m is used to describe the degradation stage of the rolling bearing. PD means that the rolling bearing is in the slow degradation stage, and RD means that the rolling bearing is in the accelerated degradation stage. When m=PD, , when m=RD, , is the regression coefficient of the slow degradation stage, is the regression coefficient of the accelerated degradation phase, means time, means time.

6. The method according to claim 1, wherein Before determining the probability density distribution prediction result and the remaining life prediction result of the tested rolling bearing by combining the degradation state value and the target life using a particle filter algorithm, the method further includes: Obtaining model parameters to be calculated in the staged nonlinear random degradation coefficient regression degradation model, wherein the model parameters to be calculated include a mean value of an inter-unit variability coefficient, a variance of an inter-unit variability coefficient, a regression coefficient, a process noise variance, and a measurement noise variance; Obtaining available observation values ​​and estimated implicit state values ​​of the test rolling bearing at the current operating moment; A likelihood function is generated based on the available observation values ​​and the estimated implicit state value, and the current actual values ​​of the mean of the inter-unit variability coefficient, the variance of the inter-unit variability coefficient, the regression coefficient, the process noise variance and the measurement noise variance are determined using the likelihood function and the Bayesian criterion.

7. The method according to claim 6, characterized in that The method of using a particle filter algorithm in combination with the degradation state value and the target duration to determine a probability density distribution prediction result and a remaining duration prediction result of the remaining life of the test rolling bearing further includes: Determine the normalized weight of the test rolling bearing by using a particle filter algorithm in combination with the degradation state value and the current actual value of the model parameter to be calculated; Determining a probability density distribution prediction result of the remaining life of the test rolling bearing by using the normalized weight; The remaining duration prediction result of the remaining life of the test rolling bearing is determined by the normalized weight and the target duration.

8. A rolling bearing life prediction device, characterized in that: The device comprises: A first acquisition module is used to obtain the current operating time, the life cycle health degradation index curve and the failure time threshold of the test rolling bearing, wherein the life cycle health degradation index curve includes the slow degradation stage and the accelerated degradation stage of the test rolling bearing; A first determination module is configured to determine a degradation state value corresponding to the test rolling bearing at a current operating moment by using a staged nonlinear random degradation coefficient regression degradation model, wherein the staged nonlinear random degradation coefficient regression degradation model is pre-set with different functions in the slow degradation stage and in the accelerated degradation stage; A second acquisition module is configured to obtain, through the full life cycle health degradation index curve, a target duration for the test rolling bearing to reach a failure time threshold from a current operating moment; The second determination module is used to use a particle filtering algorithm in combination with the degradation state value and the target duration to determine the probability density distribution prediction result and the remaining duration prediction result of the remaining life of the test rolling bearing.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to execute the instructions to implement the rolling bearing life prediction method according to any one of claims 1 to 7.

10. A vehicle, characterized in that: include: The rolling bearing life prediction device according to claim 8.

Images