Bearing remaining service life prediction method, device, equipment, medium and product

Abstract

The invention discloses a method, a device and equipment for predicting the remaining service life of a bearing, a medium and a product. The method comprises the following steps: acquiring a first vibration signal sequence corresponding to a first rolling bearing at a current sampling moment; the first vibration signal sequence comprises a first vibration signal; the first vibration signal is used for indicating the amplitude and frequency characteristics of the first rolling bearing in the horizontal stress direction; determining at least one characteristic component corresponding to the first vibration signal sequence, and determining a second vibration signal sequence corresponding to the first rolling bearing according to the at least one characteristic component and the first vibration signal sequence; and processing the second vibration signal sequence according to a residual service life prediction model obtained by pre-training, and predicting to obtain a residual service life corresponding to the first rolling bearing. According to the technical scheme, the effect of precisely and accurately predicting the remaining service life of the rolling bearing based on the vibration signal of the rolling bearing in the horizontal stress direction through the neural network model is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing condition detection, and particularly to a method, device, equipment, medium and product for predicting the remaining service life of a bearing. Background Art

[0002] Rolling bearings are the most basic and widely used rotating components in mechanical equipment. By predicting the remaining service life of rolling bearings, bearing damage and defects can be detected as early as possible, and maintenance and repairs can be carried out to avoid equipment damage and safety accidents. Therefore, it is very important to predict the remaining service life of rolling bearings.

[0003] In the related art, predicting the remaining service life of rolling bearings is usually based on empirical formulas or simple statistical models. However, the prediction methods based on empirical formulas or simple statistical models are difficult to effectively capture the complex degradation laws in the signals, resulting in low prediction accuracy. Summary of the Invention

[0004] The present invention provides a method, device, equipment, medium and product for predicting the remaining service life of a bearing, so as to achieve the effect of accurately predicting the remaining service life of a rolling bearing based on the vibration signal of the rolling bearing in the horizontal force direction through a neural network model.

[0005] According to one aspect of the present invention, a method for predicting the remaining service life of a bearing is provided. The method includes:

[0006] Obtaining a first vibration signal sequence corresponding to a first rolling bearing at a current sampling moment; wherein, the first vibration signal sequence includes a first vibration signal corresponding to the current sampling moment and first vibration signals corresponding to at least one historical sampling moment before the current sampling moment; the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction at the corresponding sampling moment;

[0007] Determining at least one characteristic component corresponding to the first vibration signal sequence, and determining a second vibration signal sequence corresponding to the first rolling bearing according to the at least one characteristic component and the first vibration signal sequence;

[0008] Processing the second vibration signal sequence by using a pre-trained remaining service life prediction model, and predicting the remaining service life corresponding to the first rolling bearing;

[0009] Wherein, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model and a feature correlation determination sub-model.

[0010] According to another aspect of the present invention, a device for predicting the remaining service life of a bearing is provided. The device includes:

[0011] a vibration signal sequence acquisition module, configured to acquire a first vibration signal sequence corresponding to the first rolling bearing at a current sampling moment; wherein the first vibration signal sequence includes a first vibration signal corresponding to the current sampling moment and a first vibration signal corresponding to at least one historical sampling moment before the current sampling moment; the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction at the corresponding sampling moment;

[0012] a vibration signal sequence determination module, configured to determine at least one characteristic component corresponding to the first vibration signal sequence, and determine a second vibration signal sequence corresponding to the first rolling bearing based on the at least one characteristic component and the first vibration signal sequence;

[0013] A remaining service life prediction module is used to process the second vibration signal sequence according to a pre-trained remaining service life prediction model to predict the remaining service life corresponding to the first rolling bearing; wherein the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model and a feature correlation determination sub-model.

[0014] According to another aspect of the present invention, an electronic device is provided, comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the method for predicting the remaining service life of a bearing according to any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for predicting the remaining service life of a bearing according to any embodiment of the present invention when executed.

[0019] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the method for predicting the remaining service life of a bearing according to any embodiment of the present invention.

[0020] In the technical solution of the embodiment of the present invention, by obtaining a first vibration signal sequence corresponding to a first rolling bearing at the current sampling moment; since the first vibration signal sequence includes a first vibration signal corresponding to the current sampling moment and first vibration signals corresponding to at least one historical sampling moment before the current sampling moment, the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction, providing a rich data basis for subsequent analysis. The vibration signal in the horizontal force direction can accurately reflect the working state of the internal components of the bearing. By analyzing the vibration signal in this direction, potential fault hazards of the bearing can be detected in a timely manner, providing an important basis for fault diagnosis and prediction of the remaining service life; further, determining at least one characteristic component corresponding to the first vibration signal sequence, and determining a second vibration signal sequence corresponding to the first rolling bearing according to the at least one characteristic component and the first vibration signal sequence, can more prominently highlight the key features in the vibration signal, remove noise or irrelevant information, improve the quality and analyzability of the signal, and help to more accurately extract the features related to the bearing state; further, processing the second vibration signal sequence according to a pre-trained remaining service life prediction model to predict the remaining service life corresponding to the first rolling bearing; wherein, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model, solving the problem that the prediction method in the related technology is difficult to effectively capture the complex degradation law in the signal, resulting in low prediction accuracy, achieving the effect of accurately predicting the remaining service life of the rolling bearing based on the vibration signal of the rolling bearing in the horizontal force direction through a neural network model, and moreover, through the mutual cooperation of the time series feature extraction sub-model, the context feature extraction sub-model, and the feature correlation determination sub-model in the remaining service life prediction model, the time series information, context information, and correlation between features in the vibration signal sequence can be fully mined, thereby improving the prediction accuracy of the remaining service life, achieving the effect of accurately capturing the complex degradation law in the vibration signal based on the mutual cooperation between the sub-models in the neural network model to accurately predict the remaining service life.

[0021] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 is a flowchart of a method for predicting the remaining service life of a bearing provided according to an embodiment of the present invention;

[0024] Figure 2 is a flowchart of a method for predicting the remaining service life of a bearing provided according to an embodiment of the present invention;

[0025] Figure 3 is a model structure diagram of a timing feature extraction sub - model provided according to an embodiment of the present invention;

[0026] Figure 4 is a flowchart of a method for predicting the remaining service life of a bearing provided according to an embodiment of the present invention;

[0027] Figure 5 is a schematic structural diagram of a device for predicting the remaining service life of a bearing provided according to an embodiment of the present invention;

[0028] Figure 6 is a schematic structural diagram of an electronic device for implementing the method for predicting the remaining service life of a bearing according to an embodiment of the present invention. Detailed implementation manners

[0029] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above - mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non - exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or device.

[0031] Figure 1The flowchart of a method for predicting the remaining service life of a bearing provided by an embodiment of the present invention. This embodiment is applicable to the situation of predicting the remaining service life of a rolling bearing. This method can be executed by a device for predicting the remaining service life of a bearing, and this device for predicting the remaining service life of a bearing can be implemented in the form of hardware and / or software, and this device for predicting the remaining service life of a bearing can be configured in a terminal and / or a server. As Figure 1 shown, the method includes:

[0032] S110. Obtain a first vibration signal sequence corresponding to a first rolling bearing at the current sampling moment.

[0033] Among them, the first rolling bearing can be a rolling bearing that can be used and for which the remaining service life is to be predicted. That is to say, the first rolling bearing can be a rolling bearing that has not reached a failure state. The first rolling bearing can be a rolling bearing of any type and / or specification. Optionally, it includes deep groove ball bearings, cylindrical roller bearings, tapered roller bearings, self-aligning ball bearings, angular contact ball bearings, etc. The current sampling moment can be the sampling moment corresponding to the prediction of the remaining service life of the first rolling bearing. Generally, during the operation of the first rolling bearing, the state of the first rolling bearing can be detected, and sampling can be performed at regular time intervals. The moment of sampling can be used as the sampling moment. The first vibration signal sequence can be a sequence composed of a series of first vibration signals. The first vibration signal sequence includes the first vibration signal corresponding to the current sampling moment and the first vibration signals corresponding to at least one historical sampling moment before the current sampling moment. The interval between the current moment and the adjacent historical sampling moment before it is the preset sampling interval. The preset sampling interval can be 5 seconds, 10 seconds, 20 seconds, etc. The first vibration signal can be the vibration signal of the first rolling bearing in the horizontal force direction at the corresponding sampling moment. This vibration signal can be used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction at the corresponding sampling moment. The amplitude represents the intensity or magnitude of the first vibration signal and can reflect the severity of the vibration of the rolling bearing in the horizontal force direction. The larger the amplitude, the stronger the vibration. It can be understood that when the rolling bearing is operating, it can be subjected to loads in the horizontal direction and / or vertical direction. Furthermore, after signal sampling, vibration signals in the horizontal force direction and / or vibration signals in the vertical force direction can be obtained. Compared with the vibration signal in the vertical force direction, the vibration signal in the horizontal force direction can better show the degradation trend of the rolling bearing and can better reflect the operating state of the rolling bearing. By analyzing the vibration signal in the horizontal force direction, fault information such as wear, looseness, and fatigue of the rolling bearing can be obtained, and thus its remaining service life can be predicted. Moreover, the vibration signal in the horizontal force direction often has more obvious characteristics. Compared with other force directions, the vibration signal in the horizontal force direction may be more easily affected by internal defects of the bearing, and its characteristic parameters such as vibration amplitude and frequency will change significantly as the bearing deteriorates. Therefore, for predicting the remaining service life of the rolling bearing, using the vibration signal in the horizontal force direction will improve the prediction accuracy of the remaining service life.

[0034] In a specific implementation, a vibration sensor can be pre-installed at positions such as the outer shell of the rolling bearing or the bearing housing. Further, during the operation of the first rolling bearing, the vibration state of the first rolling bearing in the horizontal force direction can be detected by the vibration sensor, and the vibration signal in the horizontal force direction can be sampled according to a preset sampling interval. Further, a plurality of first vibration signals corresponding to different sampling moments can be obtained. Further, the first vibration signal corresponding to the current sampling moment and the first vibration signals corresponding to at least one historical sampling moment before the current moment can be acquired, and the acquired first vibration signals can be arranged in chronological order, and the arranged signals can be used as the first vibration signal sequence corresponding to the first rolling bearing at the current sampling moment.

[0035] S120. Determine at least one characteristic component corresponding to the first vibration signal sequence, and determine a second vibration signal sequence corresponding to the first rolling bearing according to the at least one characteristic component and the first vibration signal sequence.

[0036] Among them, the characteristic component can be a representative characteristic decomposed or extracted from the first vibration signal sequence, and this characteristic can reflect some essential attributes or characteristics of the first vibration signal sequence. In this embodiment, the characteristic component can characterize the bearing degradation characteristics of the first rolling bearing, that is, it can characterize the vibration characteristics of the first rolling bearing in an abnormal operating state. It should be noted that the characteristic component can be a sequence with the sampling moment as the independent variable. For example, assuming that the first vibration signal sequence is represented by f(t), the characteristic component can be represented by u k (t). The second vibration signal sequence is a vibration signal sequence obtained after reconstruction and optimization processing based on the characteristic component and the first vibration signal sequence. Compared with the first vibration signal sequence, the second vibration signal sequence can more prominently reflect the operating state information of the first rolling bearing, providing high-quality input data for the subsequent remaining service life prediction model and improving the prediction accuracy at the data level.

[0037] In this embodiment, the characteristic component can be obtained by signal decomposition of the first vibration signal sequence, or can be obtained by feature extraction from the first vibration signal sequence. Optionally, it can be a feature vector obtained by decomposing the first vibration signal sequence through a signal decomposition algorithm. Or, at least one characteristic component can also be obtained by extracting signals from the first vibration signal sequence through a signal extraction algorithm.

[0038] Optionally, determining at least one characteristic component corresponding to the first vibration signal sequence includes: decomposing the first vibration signal sequence according to a preset signal decomposition algorithm to obtain at least one characteristic component corresponding to the first vibration signal sequence.

[0039] Among them, the signal decomposition algorithm can be a method or rule for decomposing a signal to decompose the signal into multiple components with different characteristics. Optionally, the signal decomposition algorithm includes an empirical mode decomposition algorithm, an ensemble empirical mode decomposition algorithm, a wavelet decomposition algorithm, or a variational mode decomposition algorithm.

[0040] It should be noted that the first vibration signal sequence is constructed based on the sampled vibration signal, and will exhibit non-linear and non-stationary characteristics. Especially, the early wear signal is weak and not prominent. In this case, the signal decomposition algorithm can be used to decompose the first vibration signal sequence, accurately extract the subtle degradation characteristics therein, and also enhance the stability of the non-linear time series, converting the complex non-linear sequence into multiple stable linear sequences.

[0041] In a specific implementation, after obtaining the first vibration signal sequence, the variational mode decomposition algorithm can be used to perform multi-modal signal decomposition on the first vibration signal sequence, decomposing the first vibration signal sequence into multiple intrinsic mode function components with specific center frequencies and finite bandwidths, and the decomposed intrinsic mode function components can be used as the characteristic components corresponding to the first vibration signal sequence. It can be understood that the intrinsic mode function components decomposed by the variational mode decomposition algorithm correspond to the vibration characteristics in different frequency ranges of the first vibration signal sequence, and can effectively separate the mixed components such as the normal operation vibration, fault impact vibration, and environmental noise of the first rolling bearing, especially having good extraction ability for the early weak bearing wear, pitting and other degradation characteristics.

[0042] Exemplarily, the specific operation process of performing signal decomposition on the first vibration signal sequence using the variational mode decomposition algorithm can be implemented based on the following formula:

[0043] First, each intrinsic mode function component (i.e., characteristic component) is defined as a frequency modulation - amplitude modulation signal with different center frequencies and bandwidths, that is where, A k (t) is the instantaneous amplitude of the k-th intrinsic mode function component, is the instantaneous phase of the k-th intrinsic mode function component; a variational model is constructed, and the goal is to minimize the sum of the bandwidths of all intrinsic mode function components, while satisfying that the first vibration signal sequence f(t) is equal to the sum of all intrinsic mode function components, that is The constraint condition is where, {u k} is the set of intrinsic mode function components, {ω k} is the set of center frequencies, K is the number of intrinsic mode function components, δ(t) is the Dirac function, / represents the convolution operation, is the time derivative, f(t) represents the first vibration signal sequence, j represents the imaginary unit, ω kt represents the central frequency, and u k (t) represents the intrinsic mode function component;

[0044] After that, the Lagrange multiplier Z and the quadratic penalty factor α are introduced to transform the constrained variational problem into an unconstrained variational problem, and the augmented Lagrangian function is obtained:

[0045]

[0046] where <.,.> represents the inner product operation;

[0047] After that, the augmented Lagrangian function is solved by the alternating direction method of multipliers, and the function is minimized by alternately updating u k , ω k and Z;

[0048] Update u k : Fix ω k and Z, take the partial derivative of u k and set it to zero, and we get where represents the value of u k that minimizes the subsequent function L({u k}, {ω k}, Z); the update formula of can be obtained through Fourier transform and some mathematical operations as where is the Fourier transform of u k (t), and is the Fourier transform of f(t);

[0049] Update ω k : Fix u k and Z, take the partial derivative of ω k and set it to zero, and we get Through calculation, we can get

[0050] Update Z: According to the update rule of the alternating direction method of multipliers, where τ is a relatively small positive number used to control the update step size;

[0051] Based on the above formulas, iterative updates are performed. When the preset iteration stop condition is reached, the iteration is stopped, and the final K intrinsic mode function components are obtained. The preset iteration stop condition can be that the number of iterations reaches the preset number threshold, or it can also be that the average change amount of all intrinsic mode function components between two adjacent iterations is less than the preset change amount threshold, etc.

[0052] In this embodiment, after obtaining at least one feature component, for the at least one feature component, the correlation degree between the feature component and the first vibration signal sequence can be determined. Furthermore, signal reconstruction can be performed according to the determined correlation degree, and a second vibration signal sequence can be obtained.

[0053] Optionally, determining a second vibration signal sequence corresponding to the first rolling bearing according to the at least one feature component and the first vibration signal sequence includes: for the at least one feature component, determining the Pearson correlation coefficient between the feature component and the first vibration signal sequence; determining at least one target correlation coefficient greater than a preset correlation threshold from the at least one Pearson correlation coefficient, and using the feature component corresponding to the target correlation coefficient as an effective feature component; linearly superimposing the at least one effective feature component to obtain a second vibration signal sequence corresponding to the first rolling bearing.

[0054] Among them, the Pearson correlation coefficient can be a statistic for measuring the linear correlation degree between two variables. The value range of the Pearson correlation coefficient can be [-1, 1]. The closer the Pearson correlation coefficient is to 1, the higher the positive linear correlation degree between the two variables; the closer the Pearson correlation coefficient is to -1, the higher the negative linear correlation degree between the two variables; approaching 0 indicates a weaker linear correlation. The preset correlation threshold can be a preset value, which can be used to screen out the feature components with a strong correlation with the first vibration signal sequence. The preset correlation threshold can be any value within the range of [-1, 1]. Optionally, it is 0.3, 0.4, or 0.5, etc. The target correlation coefficient is the Pearson correlation coefficient whose absolute value is greater than the preset correlation threshold selected from the at least one determined Pearson correlation coefficient. It can be understood that in the field of signal processing, linear superposition refers to adding multiple effective feature components in a linear combination manner to obtain a new signal, and the obtained new signal is the second vibration signal sequence. Exemplarily, assuming there are n effective feature components, which are x1, x2, …, x n , then, performing linear superposition on these n effective feature components is to multiply these effective feature components by corresponding coefficients and then sum them to obtain a new signal y, and the mathematical expression is y = a1x1 + a2x2 + … + a n x n , where, a1, a2, …, a n are coefficients (in simple linear superposition, the coefficients are usually 1).

[0055] In a specific implementation, for at least one feature component, the feature component and the first vibration signal sequence can be substituted into the Pearson correlation coefficient calculation formula to obtain the Pearson correlation coefficient between the feature component and the first vibration signal sequence. Further, in the case of obtaining at least one Pearson correlation coefficient, the Pearson correlation coefficient can be compared with a preset correlation threshold to screen out the Pearson correlation coefficients whose absolute values are greater than the preset correlation threshold from the at least one Pearson correlation coefficient, and the screened Pearson correlation coefficients are used as target correlation coefficients, and the feature components corresponding to the target correlation coefficients are used as effective feature components. Further, the linear coefficients corresponding to each effective feature component can be determined, the effective feature components are multiplied by their corresponding linear coefficients, and the values obtained after multiplication are superimposed. Furthermore, the data obtained after addition can be used as the second vibration signal sequence corresponding to the first rolling bearing.

[0056] It should be noted that screening the Pearson correlation coefficient using the preset correlation threshold can retain the feature components highly correlated with the first vibration signal sequence or the fault index, and eliminate the low-frequency / high-frequency components dominated by noise or irrelevant. Through the screening of the Pearson correlation coefficient and the linear superposition processing, the finally obtained second vibration signal sequence not only retains the key vibration characteristics (such as the periodic impact signal corresponding to the fault frequency) in the degradation process of the first rolling bearing, but also suppresses the interference of noise and irrelevant modes, and the signal-to-noise ratio is significantly improved. Compared with the direct input of the first vibration signal sequence, the first vibration signal sequence after signal decomposition and Pearson correlation coefficient optimization can more clearly reflect the dynamic changes of the operating state of the first rolling bearing, provide high-quality input data for the subsequent remaining service life prediction model, and improve the feature learning efficiency and prediction accuracy of the remaining service life prediction model for the bearing remaining service life from the data level.

[0057] S130. Process the second vibration signal sequence according to the pre-trained remaining service life prediction model to predict the remaining service life corresponding to the first rolling bearing.

[0058] Among them, the remaining service life prediction model can be a neural network model capable of predicting the remaining service life of the first rolling bearing based on the input vibration signal sequence. The remaining service life prediction model can be obtained by training a pre-constructed deep learning model based on the full-cycle signal sequence of the sample rolling bearing and the corresponding actual service life of the sample rolling bearing. And the full-cycle signal sequence can be a signal sequence obtained after signal decomposition and signal reconstruction. The remaining service life prediction model can be a deep neural network model including multiple sub-models. The remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model. The remaining service life refers to the length of time that the first rolling bearing can continue to operate normally from the current sampling moment until it cannot work normally or reaches the specified failure standard due to various reasons (such as wear, fatigue, etc.).

[0059] Among them, the time series feature extraction sub-model can be a neural network model for extracting the time series features of the input data. The context feature extraction sub-model can be a neural network model for capturing and extracting the context information of the time series feature data. The feature correlation determination sub-model can be a neural network model that dynamically assigns different weights to each element in the sequence by calculating the correlation between each element and other elements in the sequence, so as to better capture the long-term dependencies in the sequence.

[0060] In a specific implementation, in the case of obtaining the second vibration signal sequence, the second vibration signal sequence can be input into the pre-trained remaining service life prediction model. Further, the second vibration signal sequence can be processed successively based on the time series feature extraction sub-model, the context feature extraction sub-model, and the feature correlation determination sub-model in the remaining service life prediction model, and the remaining service life corresponding to the first rolling bearing at the current sampling moment can be output.

[0061] The technical solution of the embodiment of the present invention obtains the first vibration signal sequence corresponding to the first rolling bearing at the current sampling moment. Since the first vibration signal sequence includes the first vibration signal corresponding to the current sampling moment and the first vibration signals corresponding to at least one historical sampling moment before the current sampling moment, the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction, providing a rich data basis for subsequent analysis. The vibration signal in the horizontal force direction can accurately reflect the working state of the internal components of the bearing. By analyzing the vibration signal in this direction, potential fault hazards of the bearing can be detected in a timely manner, providing an important basis for fault diagnosis and prediction of the remaining service life. Further, at least one characteristic component corresponding to the first vibration signal sequence is determined, and according to the at least one characteristic component and the first vibration signal sequence, a second vibration signal sequence corresponding to the first rolling bearing is determined, which can more prominently extract the key features in the vibration signal, remove noise or irrelevant information, improve the quality and analyzability of the signal, and help to more accurately extract the features related to the bearing state. Further, the remaining service life prediction model obtained by pre-training is used to process the second vibration signal sequence, and the remaining service life corresponding to the first rolling bearing is predicted. The remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model, which solves the problem that the prediction method in the related art is difficult to effectively capture the complex degradation law in the signal, resulting in low prediction accuracy, and realizes the effect of accurately predicting the remaining service life of the rolling bearing based on the vibration signal of the rolling bearing in the horizontal force direction through the neural network model. Moreover, through the cooperation of the time series feature extraction sub-model, the context feature extraction sub-model, and the feature correlation determination sub-model in the remaining service life prediction model, the time series information, context information, and correlation between features in the vibration signal sequence can be fully mined, thereby improving the prediction accuracy of the remaining service life, and realizing the efficiency of accurately capturing the complex degradation law in the vibration signal based on the cooperation between the sub-models in the neural network model to accurately predict the remaining service life.

[0062] Figure 2 FIG. is a flowchart of a method for predicting the remaining service life of a bearing provided by an embodiment of the present invention. On the basis of the above embodiment, the process of predicting the remaining service life based on the remaining service life prediction model is further refined. The specific implementation manner can refer to the technical solution of this embodiment. Among them, the same or similar technical terms as those in the above embodiment will not be described in detail here. As Figure 2 shown, the method includes:

[0063] S210. Obtain the first vibration signal sequence corresponding to the first rolling bearing at the current sampling moment.

[0064] S220. Determine at least one eigen-component corresponding to the first vibration signal sequence, and determine a second vibration signal sequence corresponding to the first rolling bearing according to the at least one eigen-component and the first vibration signal sequence.

[0065] S230. Perform time-series feature extraction on the second vibration signal sequence based on the time-series feature extraction sub-model to obtain a time-series feature matrix corresponding to the first rolling bearing.

[0066] Among them, the time-series feature extraction sub-model can be a neural network model with any model structure. In this embodiment, the time-series feature extraction sub-model includes a time-series convolutional network, and this time-series convolutional network can be composed of multiple extended causal convolution modules and ordinary convolution modules. By using extended causal convolution, the time-series convolutional network can capture long-term dependencies and avoid the problem of gradient disappearance. Extended causal convolution inserts an expansion factor between the elements of the convolution kernel, thereby expanding the receptive field, being able to consider a wider range of input information, and through causal constraints, only relying on the current and past time steps to avoid introducing future information. Through the skip connections and residual connections in the time-series convolutional network, the stability of the time-series convolutional network can be enhanced to ensure that there will be no gradient disappearance or gradient explosion during the training process.

[0067] In specific implementation, it can be combined with Figure 3 to illustrate the time-series feature extraction process. Figure 3 is a model structure diagram of the time-series feature extraction sub-model provided by an embodiment of the present invention. As Figure 3 shown, the time-series feature extraction sub-model can include two extended causal convolution modules and one ordinary convolution module. In the case of obtaining the second vibration signal sequence, the second vibration signal sequence can be input into the remaining service life prediction model. Further, the second vibration signal sequence is input into the time-series feature extraction sub-model, and the second vibration signal sequence is successively subjected to convolution processing, normalization processing, activation function application processing, and overfitting processing through the two extended causal convolution modules in the time-series feature extraction sub-model to obtain a first residual feature; and, the second vibration signal sequence is subjected to convolution processing through the ordinary convolution module in the time-series feature extraction sub-model to obtain a second residual feature. Further, the first residual feature and the second residual feature can be subjected to residual processing to obtain a time-series feature matrix corresponding to the first rolling bearing.

[0068] S240. Perform context feature extraction on the time-series feature matrix based on the context feature extraction sub-model to obtain a context feature matrix corresponding to the second vibration signal sequence.

[0069] Among them, the context feature extraction sub-model can be a neural network model with any model structure. In this embodiment, the context feature extraction sub-model includes a bidirectional long short-term memory network. The bidirectional long short-term memory network is a deep learning model for processing sequence data. This deep learning model combines the advantages of the long short-term memory network in processing sequence data and uses a bidirectional structure to utilize the forward and reverse information of the sequence simultaneously. Since there is a certain time delay in the change of the health state of the rolling bearing, the long short-term memory network may miss some important information. Through bidirectional learning, the bidirectional long short-term memory network can comprehensively understand the dependencies between different moments in the time series, improve the modeling ability for time series data, and enhance the prediction accuracy of the model.

[0070] In this embodiment, after obtaining the time series feature matrix, the time series feature matrix can be input into the context feature extraction sub-model. Further, based on the context feature extraction sub-model, forward features and backward features of the time series feature matrix can be extracted and fused, and a context feature matrix corresponding to the second vibration signal sequence can be obtained.

[0071] Optionally, the context feature extraction sub-model includes a forward long short-term memory network and a backward long short-term memory network; extracting context features of the time series feature matrix based on the context feature extraction sub-model to obtain a context feature matrix corresponding to the second vibration signal sequence includes: extracting forward features of the time series feature matrix based on the forward long short-term memory network to obtain a forward feature matrix; and extracting backward features of the time series feature matrix based on the backward long short-term memory network to obtain a backward feature matrix; fusing the forward feature matrix and the backward feature matrix to obtain a context feature matrix corresponding to the second vibration signal sequence.

[0072] Among them, the forward long short-term memory network is a neural network that processes data in the forward order of the time series. When processing the time series feature matrix through the forward long short-term memory network, starting from the starting position of the sequence, the feature information of each time step is sequentially input into the network, and the forward features in the data, that is, the feature change law as time progresses, are learned and extracted through the mechanism of the long short-term memory network. The backward long short-term memory network corresponds to the forward long short-term memory network and processes data in the reverse order of the time series. Starting from the end position of the sequence, the feature information of each time step is input into the network in reverse time order to learn and extract the backward features in the data, that is, the feature learning law of looking at the data from back to front.

[0073] In a specific implementation, after obtaining the temporal feature matrix, the temporal feature matrix can be input into the context feature extraction sub-model. Further, based on the forward long short-term memory network in the context feature extraction sub-model, forward feature extraction is performed on the temporal feature matrix, and the extracted forward feature information is used as the forward feature matrix. Also, based on the backward long short-term memory network in the context feature extraction sub-model, backward feature extraction can be performed on the temporal feature matrix, and the extracted backward feature information is used as the backward feature matrix. Further, the forward feature matrix and the backward feature matrix can be fused, and the fused feature information is used as the context feature matrix. Among them, the fusion method of the feature matrix can include splicing, addition, weighted summation, etc.

[0074] S250. Based on the feature correlation determination sub-model, feature weighting is performed on the context feature matrix to obtain a weighted feature matrix corresponding to the second vibration signal sequence.

[0075] Among them, the feature correlation determination sub-model can be a neural network model with any model structure. In this embodiment, the feature correlation determination sub-model is a neural network model constructed based on the self-attention mechanism. The self-attention mechanism can dynamically assign different weights to each element in the sequence by calculating the correlation between each element and other elements in the sequence, so as to better capture the long-term dependencies in the sequence. The self-attention scores are calculated through the self-attention mechanism in the feature correlation determination sub-model and weighted and summed to strengthen the model's attention to key time steps, improving the accuracy and robustness of the prediction. The feature vector processed by the self-attention mechanism contains the temporal information that the model pays the most attention to. Constructing the feature correlation determination sub-model based on the self-attention mechanism can enable the model to adaptively process long sequence data, and also enable the model to not only reduce the computational complexity, improve the training efficiency, but also significantly improve the accuracy and robustness.

[0076] In a specific implementation, the context feature matrix can be input into the feature correlation determination sub-model. Further, based on the self-attention mechanism in the feature correlation determination sub-model, the context feature matrix can be respectively generated into a query matrix, a key matrix, and a value matrix through three different linear transformations. Further, by calculating the dot product of the query matrix and the key matrix, an attention score matrix is obtained, and then the Softmax function is applied to the attention score matrix to convert the attention scores into an attention weight matrix. Then, based on the attention weight matrix, weighted summation is performed on the value matrix to obtain a weighted feature matrix.

[0077] S260. Based on the fully connected layer, the weighted feature matrix is processed to obtain the remaining service life corresponding to the first rolling bearing.

[0078] Among them, the fully connected layer is used to perform linear transformation and non-linear mapping on the input data to extract higher-level features and map the data to a specific output space.

[0079] In a specific implementation, after obtaining the weighted feature matrix, the weighted feature matrix can be input into the fully connected layer. Furthermore, the weighted feature matrix can be processed based on the fully connected layer, and the remaining service life corresponding to the first rolling bearing can be output.

[0080] The technical solution of the embodiment of the present invention extracts the time series features of the second vibration signal sequence through the time series feature extraction sub-model to obtain the time series feature matrix corresponding to the first rolling bearing; extracts the context features of the time series feature matrix through the context feature extraction sub-model to obtain the context feature matrix corresponding to the second vibration signal sequence; weights the context feature matrix based on the feature correlation determination sub-model to obtain the weighted feature matrix corresponding to the second vibration signal sequence; processes the weighted feature matrix based on the fully connected layer to obtain the remaining service life corresponding to the first rolling bearing, realizing the cooperation of the time series feature extraction sub-model, the context feature extraction sub-model and the feature correlation determination sub-model in the remaining service life prediction model, being able to fully exploit the time series information, context information and the correlation between features in the vibration signal sequence, thereby improving the prediction accuracy of the remaining service life, and realizing the efficiency of accurately capturing the complex degradation law in the vibration signal based on the cooperation between each sub-model in the neural network model to accurately predict the remaining service life.

[0081] Figure 4 is a flowchart of a method for predicting the remaining service life of a bearing provided by an embodiment of the present invention. On the basis of the above embodiment, before applying the remaining service life prediction model, training samples can be constructed first, and the remaining service life prediction model can be trained based on the training samples. Furthermore, the remaining service life of the rolling bearing can be predicted based on the remaining service life prediction. The specific implementation manner can refer to the technical solution of this embodiment. Among them, the same or similar technical terms as those in the above embodiment will not be described in detail here. As Figure 4 shown, the method includes:

[0082] S310. Obtain a plurality of training samples, where the training samples include the full-cycle signal sequence of the sample rolling bearing and the actual service life corresponding to the sample rolling bearing.

[0083] Among them, the training samples are the data sets used to train the model. The sample rolling bearing can be the rolling bearing used as the training sample. Each training sample corresponds to a specific sample rolling bearing. It should be noted that the multiple sample rolling bearings included in the multiple training samples can be rolling bearings under at least three different load and speed conditions. The full-cycle signal sequence can refer to the signal sequence obtained during the entire process from the start of use to the failure of the sample rolling bearing. This signal sequence can be obtained after signal decomposition and signal reconstruction of the collected signal sequence. The vibration signals included in the full-cycle signal sequence can characterize features such as the vibration amplitude and vibration frequency of the sample rolling bearing in the horizontal force direction. The actual service life can refer to the time length experienced by the sample rolling bearing from the start of use until it cannot work properly or reaches the specified failure standard due to various reasons (such as wear, fatigue, damage, etc.).

[0084] In a specific implementation, the initial full-cycle signal sequence and the actual service life of the rolling bearings under at least three different load and speed conditions can be obtained from a preset data set, and the obtained rolling bearings are used as sample rolling bearings. Further, for each sample rolling bearing, the initial full-cycle signal sequence of the sample rolling bearing can be decomposed according to a preset signal decomposition algorithm to obtain at least one characteristic component corresponding to the initial full-cycle signal sequence. Further, for at least one characteristic component, the Pearson correlation coefficient between the characteristic component and the initial full-cycle signal sequence is determined, and at least one target correlation coefficient whose absolute value is greater than a preset correlation threshold is determined from at least one Pearson correlation coefficient, and the characteristic component corresponding to the target correlation coefficient is used as the effective characteristic component. Then, at least one effective characteristic component is linearly superimposed to obtain the full-cycle signal sequence corresponding to the sample rolling bearing. Further, in the case of obtaining the full-cycle signal sequence corresponding to each sample rolling bearing, for each sample rolling bearing, a training sample can be constructed based on the full-cycle signal sequence and the actual service life corresponding to the sample rolling bearing. Furthermore, multiple training samples can be obtained.

[0085] S320. For multiple training samples, input the full-cycle signal sequences in the training samples into the model to be trained to obtain the predicted service life corresponding to the sample rolling bearing.

[0086] Among them, the model to be trained can be a neural network model whose model parameters are default values or initial values. In this embodiment, the model to be trained includes a time series feature extraction sub-model, a context feature extraction sub-model and a feature correlation determination sub-model. The context feature extraction sub-model is constructed based on a predetermined target hyperparameter. The target hyperparameter is a hyperparameter used to construct the context feature extraction sub-model to be trained. The hyperparameter is a parameter that needs to be determined before training the model, and the parameter will significantly affect the training effect of the model performance. Optionally, the target hyperparameter includes the number of hidden layers, learning rate, batch size, etc. It should be noted that in addition to the target hyperparameter, there are other hyperparameters used to construct the context feature extraction sub-model, which will not be repeated here in this embodiment.

[0087] It is understandable that increasing the hidden layer data can improve the expressiveness of the model, but it will also make training more difficult and time-consuming, and it is prone to gradient vanishing or exploding. Therefore, before training the model, you can first determine a better number of hidden layers. The learning rate controls the step size of the model parameter update. If the learning rate is too large, the model may skip the optimal solution and fail to converge. If the learning rate is set too small, the model training speed will be very slow. Therefore, before training the model, you can first determine a better learning rate. The batch size refers to the number of samples used in each iteration. A larger batch size can make the model training more stable, but it may increase memory requirements. A smaller batch size can increase the randomness of the model and help to escape the local optimal solution, but the training speed may be slower. Therefore, before training the model, you can first determine a better batch size.

[0088] Among them, the target hyperparameters are determined based on the hyperparameter optimization algorithm. Compared with the manual parameter adjustment method, determining the target hyperparameters based on the hyperparameter optimization algorithm can effectively improve the parameter adjustment efficiency and avoid falling into the local optimum; and the hyperparameters adjusted based on the hyperparameter optimization algorithm are hyperparameters that have an important impact on the prediction performance, and thus, it can be achieved on the basis of improving the parameter adjustment efficiency, and effectively improve the accuracy and stability of the prediction results. In this embodiment, the hyperparameter optimization algorithm used to determine the target hyperparameters can be any optimization algorithm that can determine the optimal hyperparameter combination to improve the model performance. Optionally, the hyperparameter optimization algorithm includes the crown porcupine optimization algorithm, the particle swarm optimization algorithm, the genetic algorithm, the simulated annealing algorithm, the firefly algorithm, and the like.

[0089] The predicted service life may be a predicted value obtained by predicting the remaining service life of the sample rolling bearing based on the full-cycle signal sequence by the to-be-trained model.

[0090] In a specific implementation, after obtaining a plurality of training samples, for the plurality of training samples, the full-cycle signal sequence corresponding to the sample rolling bearing in the training samples can be input into the model to be trained. Furthermore, the full-cycle signal sequence can be processed in sequence by the time-series feature extraction sub-model, the context feature extraction sub-model, and the feature correlation determination sub-model in the model to be trained, and the predicted service life corresponding to the sample rolling bearing can be output.

[0091] S330. Determine the loss value according to the actual service life and the predicted service life, and correct the model parameters in the model to be trained based on the loss value. Take the convergence of the loss function in the model to be trained as the training objective to obtain a remaining service life prediction model for predicting the remaining service life of the rolling bearing.

[0092] Among them, the loss value can be a numerical value representing the degree of difference between the model output and the true output. The model parameters refer to the learnable variables in the model. The model parameters will be continuously adjusted and updated during the training process to make the output of the model as close as possible to the true value, thereby improving the model performance. Optionally, the model parameters include weights and biases, etc. The loss function can be determined based on the loss value and is a function used to represent the degree of difference between the predicted output and the actual output. Optionally, the loss function includes a cross-entropy loss function, a logarithmic loss function, or a root mean square error loss function, etc.

[0093] In a specific implementation, when the predicted service life corresponding to the sample rolling bearing is obtained, the predicted service life can be compared with the actual service life in the training samples, and a loss value can be obtained. Further, when the loss value is obtained, the model parameters in the model to be trained can be corrected based on the loss value, and the corrected model can be used as the first model to be processed, and a backup of the first model to be processed can be made and the backup model can be saved. Further, the process of processing the training samples and determining the loss value can be repeatedly executed. Furthermore, the model parameters in the model to be processed can be corrected based on the obtained loss value, and the corrected model can be used as the second model to be processed, a backup of the second model to be processed can be made, and the backup model can be saved. Further, when it is detected that the loss function converges, such as the training error of the loss function is less than a preset error, or the error change trend tends to be stable, the iterative training can be stopped at this time, and the model obtained at this time can be used as the model to be verified. Further, the saved backup model can be retrieved, and the retrieved backup model can also be used as the model to be verified. Furthermore, at least one model to be verified can be obtained. Further, in order to evaluate the model performance of at least one model to be verified, test samples can be obtained. For at least one model to be verified, the model to be verified can be tested and verified based on the training samples, and verification results corresponding to the model to be verified can be obtained. Further, it can be determined whether the verification results of at least one model to be verified meet the preset conditions. Furthermore, the model to be verified whose verification results meet the preset conditions can be used as the remaining service life prediction model for predicting the remaining service life of the rolling bearing.

[0094] Among them, the model to be verified can be a model obtained by correcting the model parameters in the model to be trained according to the loss value. The number of models to be verified can be consistent with the number of training iterations (or the number of times of model parameter correction). The test samples can be part of the training samples or sample data reconstructed based on the construction method of the training samples. The preset conditions can include at least one of the mean-square error (MSE) reaching the first error threshold, the mean absolute error (MAE) reaching the second error threshold, and the mean absolute percentage error (MAPE) reaching the third error threshold. Among them, the mean-square error is usually used to measure the average squared difference between the predicted value and the true value. The mean absolute error is usually used to measure the average absolute difference between the predicted value and the true value. The mean absolute percentage error is usually used to measure the average absolute percentage by which the predicted value deviates from the true value.

[0095] In the technical solution of the embodiment of the present invention, by obtaining a plurality of training samples, where the training samples include the full-cycle signal sequences of the sample rolling bearings and the actual service life corresponding to the sample rolling bearings; further, for the plurality of training samples, input the full-cycle signal sequences in the training samples into the model to be trained to obtain the predicted service life corresponding to the sample rolling bearings; where the model to be trained includes a time-series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model, and the context feature extraction sub-model is constructed based on the pre-determined target hyperparameters; the target hyperparameters are determined based on the hyperparameter optimization algorithm; further, according to the actual service life and the predicted service life, determine the loss value, and correct the model parameters in the model to be trained based on the loss value, and take the convergence of the loss function in the model to be trained as the training target to obtain a remaining service life prediction model for predicting the remaining service life of the rolling bearing, realizing the training of a remaining service life prediction model that can accurately predict the remaining service life of the rolling bearing based on the constructed training samples, effectively improving the accuracy and robustness of the model, and effectively enhancing the prediction performance of the model.

[0096] It should be noted that before training the model to be trained, the hyperparameters of the model to be trained need to be set first. Since the context feature extraction sub-model in the model to be trained has a greater impact on the task of predicting the remaining service life of the bearing, the setting of its hyperparameters can effectively affect the model prediction performance of the remaining service life prediction model. Therefore, in order to improve the performance of the context feature extraction sub-model in the task of predicting the remaining service life of the bearing, a hyperparameter optimization algorithm can be used to determine the hyperparameters of the context feature extraction sub-model. For the hyperparameters of the other sub-models in the model to be trained except the context feature extraction sub-model, they can be optimized by using the hyperparameter optimization algorithm, or, it can also be determined by randomly determining the hyperparameters. This embodiment does not make specific limitations on this. Generally, if all the hyperparameters of the model to be trained are determined by the hyperparameter optimization algorithm, it may consume a large amount of computing resources and time costs. Therefore, in this embodiment, in order to be able to improve the prediction accuracy and stability of the model on the basis of improving the hyperparameter tuning efficiency, the hyperparameters of the context feature extraction sub-model in the model to be trained can be determined by the hyperparameter optimization algorithm, and the hyperparameters of the other sub-models can be determined by randomly determining the hyperparameters.

[0097] Optionally, the process of determining the target hyperparameters of the context feature extraction sub-model may be as follows: Based on the preset hyperparameter value range corresponding to the context feature extraction sub-model, determine at least one individual to be optimized corresponding to the context feature extraction sub-model; wherein, one individual to be optimized corresponds to a set of hyperparameters; for at least one individual to be optimized, based on the hyperparameters corresponding to the individual to be optimized, construct a model to be optimized, and based on the pre-acquired test samples and the model to be optimized, determine the fitness value corresponding to the individual to be optimized; wherein, the model to be optimized includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model; adjust the hyperparameters corresponding to at least one individual to be optimized according to the fitness value and the preset mutation probability to obtain the adjusted hyperparameters corresponding to the individual to be optimized; repeat the steps of constructing the model to be optimized, determining the fitness value, and hyperparameter adjustment until the preset iteration end condition is reached, to obtain the adjusted hyperparameters corresponding to at least one individual to be optimized and their corresponding fitness values; take the individual to be optimized corresponding to the minimum fitness value as the target optimization individual, and take the adjusted hyperparameters corresponding to the target optimization individual as the target hyperparameters corresponding to the context feature extraction sub-model, so as to construct the context feature extraction sub-model in the model to be trained based on the target hyperparameters.

[0098] Among them, the hyperparameter value range may include the value ranges of at least one hyperparameter to be optimized. Optionally, the hyperparameter value range includes the value range of the number of hidden layers, the value range of the learning rate, the value range of the batch size, etc. For example, the value range of the number of hidden layers can be set between 1 and 3. If the number of layers is too small, it is difficult to capture complex relationships. If the number of layers is too large, it will increase the computational amount and training difficulty. The value range of the learning rate can be set between 0.0001 and 0.1. If the learning rate is too small, the model convergence speed is slow. If it is too large, it may cause the model to not converge stably during training. The individual to be optimized represents a potential solution in the optimization search space. Generally, when solving the hyperparameter optimization problem based on the hyperparameter optimization algorithm, in the initialization stage, a group of individuals will be randomly generated, and each individual represents a set of possible hyperparameters. This group of randomly generated individuals can be used as the individuals to be optimized.

[0099] Among them, the model to be optimized can be a neural network model constructed based on the hyperparameters of the individual to be optimized and the hyperparameters of other pre-determined sub-models. The model to be optimized includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model. Among them, the context feature extraction sub-model is constructed based on the hyperparameters of the individual to be optimized. The time series feature extraction sub-model and the feature correlation determination sub-model are constructed based on the pre-determined hyperparameters.

[0100] Among them, the fitness value is a quantitative indicator used to measure the performance of an individual in an optimization problem. During the iteration process of the hyperparameter optimization algorithm, by comparing the fitness values of different individuals, it is determined which individuals are more likely to be retained, replicated, or mutated, etc., to guide the population to optimize towards a better solution space. Optionally, the fitness value can be determined based on the mean square error function or the cross-entropy function.

[0101] Among them, the preset iteration end condition can be a criterion used to determine whether the hyperparameter iteration process can end. Optionally, the preset iteration end conditions include that the number of iterations reaches the preset iteration number threshold, the fitness value converges, and the optimal solution that meets the requirements is determined, etc.

[0102] In a specific implementation, first, the hyperparameter value range corresponding to the context feature extraction submodel can be determined according to the preset context features, and at least one individual to be optimized is generated randomly or according to a certain rule. Each individual to be optimized corresponds to a different set of hyperparameter combinations. Further, for each individual to be optimized, a model is constructed according to the hyperparameters corresponding to the individual to be optimized, and the model is used as the context feature extraction submodel in the model to be optimized. In addition, other corresponding submodels are constructed according to the preset hyperparameters corresponding to other submodels. The other submodels constructed include the time series feature extraction submodel and the feature correlation determination submodel. The time series feature extraction submodel, the context feature extraction submodel, and the feature correlation determination submodel are spliced according to the preset model splicing order, and then the model to be optimized corresponding to the individual to be optimized can be obtained. Further, the pre-acquired test samples can be input into the model to be optimized. Through the calculation and prediction of the model, according to a certain evaluation criterion (such as the value of the loss function, the root mean square error, etc.), the fitness value corresponding to the individual to be optimized is determined. The fitness value reflects the performance of the model to be optimized under this hyperparameter combination on the test samples.

[0103] Further, at least one individual to be optimized is sorted in ascending order of the fitness value. Except for the individual to be optimized ranked first, other individuals to be optimized randomly select a learning object from the individuals to be optimized ranked higher with a certain probability, and adjust their corresponding hyperparameters according to the learning rate. At the same time, a mutation probability is set for each individual to be optimized, and the corresponding partial hyperparameters of the mutated individual to be optimized are mutated according to the rules. Then, after the adjustment and mutation operations, the adjusted hyperparameters corresponding to the individual to be optimized can be obtained.

[0104] Further, repeat the steps of constructing the model to be optimized, determining the fitness value, and adjusting the hyperparameters, and continuously optimize the hyperparameters. This iterative process will continue until the preset iteration end condition is met, such as reaching the maximum number of iterations, the fitness value converges to a certain degree, etc. When the iteration ends, the adjusted hyperparameters corresponding to at least one individual to be optimized and their corresponding fitness values are obtained. Among these results, select the individual to be optimized with the smallest fitness value as the target optimization individual, and determine the adjusted hyperparameters corresponding to the target optimization individual as the target hyperparameters corresponding to the context feature extraction submodel. Further, based on the obtained target hyperparameters, construct the context feature extraction submodel in the model to be trained, so that the submodel reaches a relatively optimal state in terms of hyperparameters, laying a foundation for the subsequent training and performance improvement of the entire model to be trained.

[0105] S340. Obtain the first vibration signal sequence corresponding to the first rolling bearing at the current sampling moment.

[0106] S350. Determine at least one feature component corresponding to the first vibration signal sequence, and determine the second vibration signal sequence corresponding to the first rolling bearing according to the at least one feature component and the first vibration signal sequence.

[0107] S360. Process the second vibration signal sequence according to the pre-trained remaining service life prediction model, and predict the remaining service life corresponding to the first rolling bearing.

[0108] Figure 5 It is a schematic structural diagram of a device for predicting the remaining service life of a bearing provided by an embodiment of the present invention. As Figure 5 shown, the device includes: a vibration signal sequence acquisition module 410, a vibration signal sequence determination module 420, and a remaining service life prediction module 430.

[0109] Among them, the vibration signal sequence acquisition module 410 is configured to acquire a first vibration signal sequence corresponding to the first rolling bearing at the current sampling moment; wherein, the first vibration signal sequence includes the first vibration signal corresponding to the current sampling moment and the first vibration signals corresponding to at least one historical sampling moment before the current sampling moment; the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction at the corresponding sampling moment; the vibration signal sequence determination module 420 is configured to determine at least one eigencomponent corresponding to the first vibration signal sequence, and determine a second vibration signal sequence corresponding to the first rolling bearing according to the at least one eigencomponent and the first vibration signal sequence; the remaining service life prediction module 430 is configured to process the second vibration signal sequence according to a pre-trained remaining service life prediction model, and predict the remaining service life corresponding to the first rolling bearing; wherein, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model.

[0110] The technical solution of the embodiment of the present invention obtains the first vibration signal sequence corresponding to the first rolling bearing at the current sampling moment; since the first vibration signal sequence includes the first vibration signal corresponding to the current sampling moment and the first vibration signals corresponding to at least one historical sampling moment before the current sampling moment, the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction, providing a rich data basis for subsequent analysis. The vibration signal in the horizontal force direction can accurately reflect the working state of the internal components of the bearing. By analyzing the vibration signal in this direction, potential fault hazards of the bearing can be detected in a timely manner, providing an important basis for fault diagnosis and prediction of the remaining service life; further, determining at least one characteristic component corresponding to the first vibration signal sequence, and determining the second vibration signal sequence corresponding to the first rolling bearing according to the at least one characteristic component and the first vibration signal sequence can more prominently highlight the key features in the vibration signal, remove noise or irrelevant information, improve the quality and analyzability of the signal, and help to more accurately extract the features related to the bearing state; further, processing the second vibration signal sequence according to the pre-trained remaining service life prediction model to predict the remaining service life corresponding to the first rolling bearing; wherein, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model, which solves the problem that the prediction method in the related technology is difficult to effectively capture the complex degradation law in the signal, resulting in low prediction accuracy, and realizes the effect of accurately predicting the remaining service life of the rolling bearing based on the vibration signal of the rolling bearing in the horizontal force direction through the neural network model. Moreover, through the cooperation of the time series feature extraction sub-model, the context feature extraction sub-model, and the feature correlation determination sub-model in the remaining service life prediction model, the time series information, context information, and correlation between features in the vibration signal sequence can be fully mined, thereby improving the prediction accuracy of the remaining service life, and realizing the efficiency of accurately predicting the remaining service life by precisely capturing the complex degradation law in the vibration signal based on the cooperation between the sub-models in the neural network model.

[0111] Optionally, the vibration signal sequence determination module 420 includes: a signal decomposition unit. The signal decomposition unit is configured to decompose the first vibration signal sequence according to a preset signal decomposition algorithm to obtain at least one characteristic component corresponding to the first vibration signal sequence.

[0112] Optionally, the vibration signal sequence determination module 420 includes: a correlation coefficient determination unit, a feature component determination unit, and a vibration signal sequence determination unit. Among them, the correlation coefficient determination unit is configured to determine the Pearson correlation coefficient between the feature component and the first vibration signal sequence for at least one feature component; the feature component determination unit is configured to determine at least one target correlation coefficient whose absolute value is greater than a preset correlation threshold from at least one Pearson correlation coefficient, and use the feature component corresponding to the target correlation coefficient as an effective feature component; the vibration signal sequence determination unit is configured to linearly superimpose at least one effective feature component to obtain a second vibration signal sequence corresponding to the first rolling bearing.

[0113] Optionally, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, a feature correlation determination sub-model, and a fully connected layer; the remaining service life prediction module 430 includes: a time series feature matrix determination unit, a context feature matrix determination unit, a weighted feature matrix determination unit, and a remaining service life determination unit. Among them, the time series feature matrix determination unit is configured to perform time series feature extraction on the second vibration signal sequence based on the time series feature extraction sub-model to obtain a time series feature matrix corresponding to the first rolling bearing; the context feature matrix determination unit is configured to perform context feature extraction on the time series feature matrix based on the context feature extraction sub-model to obtain a context feature matrix corresponding to the second vibration signal sequence; the weighted feature matrix determination unit is configured to perform feature weighting on the context feature matrix based on the feature correlation determination sub-model to obtain a weighted feature matrix corresponding to the second vibration signal sequence; the remaining service life determination unit is configured to process the weighted feature matrix based on the fully connected layer to obtain the remaining service life corresponding to the first rolling bearing.

[0114] Optionally, the context feature extraction sub-model includes a forward long short-term memory network and a backward long short-term memory network; the context feature matrix determination unit includes: a forward feature matrix determination sub-unit, a backward feature matrix determination sub-unit, and a context feature matrix determination sub-unit. Among them, the forward feature matrix determination sub-unit is configured to perform forward feature extraction on the time series feature matrix based on the forward long short-term memory network to obtain a forward feature matrix; and, the backward feature matrix determination sub-unit is configured to perform backward feature extraction on the time series feature matrix based on the backward long short-term memory network to obtain a backward feature matrix; the context feature matrix determination sub-unit is configured to fuse the forward feature matrix and the backward feature matrix to obtain a context feature matrix corresponding to the second vibration signal sequence.

[0115] Optionally, the device further includes: a training sample acquisition module, a predicted service life determination module, and a model parameter correction module. Among them, the training sample acquisition module is used to acquire a plurality of training samples, where the training samples include the full-cycle signal sequences of the sample rolling bearings and the corresponding actual service life of the sample rolling bearings; the predicted service life determination module is used to, for the plurality of training samples, input the full-cycle signal sequences in the training samples into the model to be trained, and obtain the predicted service life corresponding to the sample rolling bearings; among them, the model to be trained includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model, and the context feature extraction sub-model is constructed based on the predetermined target hyperparameters; the target hyperparameters are determined based on the hyperparameter optimization algorithm; the model parameter correction module is used to determine the loss value according to the actual service life and the predicted service life, and correct the model parameters in the model to be trained based on the loss value, and take the convergence of the loss function in the model to be trained as the training target, so as to obtain a remaining service life prediction model for predicting the remaining service life of the rolling bearings.

[0116] Optionally, the device further includes: an individual determination module, a model construction module, a hyperparameter adjustment module, a hyperparameter determination module, and a target hyperparameter determination module. Among them, the individual determination module is used to determine at least one individual to be optimized corresponding to the context feature extraction sub-model according to the preset hyperparameter value range corresponding to the context feature extraction sub-model; where one individual to be optimized corresponds to a set of hyperparameters; the model construction module is used to, for at least one individual to be optimized, construct an optimized model based on the hyperparameters corresponding to the individual to be optimized, and determine the fitness value corresponding to the individual to be optimized based on the pre-acquired test samples and the optimized model; where the optimized model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model; the hyperparameter adjustment module is used to adjust the hyperparameters corresponding to at least one individual to be optimized according to the fitness value and the preset mutation probability, and obtain the adjusted hyperparameters corresponding to the individual to be optimized; the hyperparameter determination module is used to repeat the steps of constructing the optimized model, determining the fitness value, and adjusting the hyperparameters until the preset iteration end condition is reached, and obtain the adjusted hyperparameters corresponding to at least one individual to be optimized and their corresponding fitness values; the target hyperparameter determination module is used to use the individual to be optimized corresponding to the minimum fitness value as the target optimization individual, and use the adjusted hyperparameters corresponding to the target optimization individual as the target hyperparameters corresponding to the context feature extraction sub-model in the model to be trained, so as to construct the context feature extraction sub-model in the model to be trained based on the target hyperparameters.

[0117] The bearing remaining service life prediction device provided by the embodiments of the present invention can execute the bearing remaining service life prediction method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0118] Figure 6 FIG. 1 shows a schematic structural diagram of an electronic device 10 that can be used to implement an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as, for example, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, personal digital processors, cellular telephones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0119] As Figure 6 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. The memory stores a computer program executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0120] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0121] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the bearing remaining useful life prediction method.

[0122] In some embodiments, the method for predicting the remaining service life of a bearing can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for predicting the remaining service life of a bearing described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the method for predicting the remaining service life of a bearing by any other suitable means (e.g., by means of firmware).

[0123] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), system-on-a-chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0124] The computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowchart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0125] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0126] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic, speech, or tactile input).

[0127] The systems and techniques described herein can be implemented in a computing system that includes backend components (such as, for example, a data server), or a computing system that includes middleware components (such as, for example, an application server), or a computing system that includes frontend components (such as, for example, a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (such as, for example, a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a target blockchain network, and the Internet.

[0128] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0129] In particular, according to an embodiment of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication unit 19, or installed from the storage unit 18, or installed from the ROM 12. When the computer program is executed by the processor 11, the above-mentioned functions defined in the method of the embodiment of the present invention are executed.

[0130] It should be understood that various forms of the flow shown above can be used, steps can be reordered, added or deleted. For example, the steps described in the present invention can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0131] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the remaining service life of a bearing, characterized in that, Including: Obtain a first vibration signal sequence corresponding to a first rolling bearing at a current sampling moment; wherein, the first vibration signal sequence includes a first vibration signal corresponding to the current sampling moment and first vibration signals corresponding to at least one historical sampling moment before the current sampling moment; the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction at the corresponding sampling moment; Determine at least one characteristic component corresponding to the first vibration signal sequence, and determine a second vibration signal sequence corresponding to the first rolling bearing according to at least one of the characteristic components and the first vibration signal sequence; Process the second vibration signal sequence according to a pre-trained remaining service life prediction model, and predict the remaining service life corresponding to the first rolling bearing; Wherein, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model.

2. The method for predicting the remaining service life of a bearing according to claim 1, characterized in that, The determining at least one characteristic component corresponding to the first vibration signal sequence includes: Perform signal decomposition on the first vibration signal sequence according to a preset signal decomposition algorithm to obtain at least one characteristic component corresponding to the first vibration signal sequence.

3. The method for predicting the remaining service life of a bearing according to claim 1, wherein The determining a second vibration signal sequence corresponding to the first rolling bearing according to at least one of the characteristic components and the first vibration signal sequence includes: For at least one of the characteristic components, determine the Pearson correlation coefficient between the characteristic component and the first vibration signal sequence; Determine at least one target correlation coefficient whose absolute value is greater than a preset correlation threshold from at least one of the Pearson correlation coefficients, and use the characteristic component corresponding to the target correlation coefficient as an effective characteristic component; Linearly superimpose at least one of the effective characteristic components to obtain a second vibration signal sequence corresponding to the first rolling bearing.

4. The method for predicting the remaining service life of a bearing according to claim 1, characterized in that, The remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, a feature correlation determination sub-model, and a fully connected layer; the processing the second vibration signal sequence according to a pre-trained remaining service life prediction model to predict the remaining service life corresponding to the first rolling bearing includes: Perform time series feature extraction on the second vibration signal sequence based on the time series feature extraction sub-model to obtain a time series feature matrix corresponding to the first rolling bearing; Perform context feature extraction on the time series feature matrix based on the context feature extraction sub-model to obtain a context feature matrix corresponding to the second vibration signal sequence; Perform feature weighting on the context feature matrix based on the feature correlation determination sub-model to obtain a weighted feature matrix corresponding to the second vibration signal sequence; Process the weighted feature matrix based on the fully connected layer to obtain the remaining service life corresponding to the first rolling bearing.

5. The method for predicting the remaining service life of a bearing according to claim 4, characterized in that, The context feature extraction sub-model includes a forward long short-term memory network and a backward long short-term memory network; the context feature extraction of the time series feature matrix based on the context feature extraction sub-model to obtain a context feature matrix corresponding to the second vibration signal sequence includes: Performing forward feature extraction on the time series feature matrix based on the forward long short-term memory network to obtain a forward feature matrix; and, Performing backward feature extraction on the time series feature matrix based on the backward long short-term memory network to obtain a backward feature matrix; Fusing the forward feature matrix and the backward feature matrix to obtain a context feature matrix corresponding to the second vibration signal sequence.

6. The method for predicting the remaining service life of a bearing according to claim 1, wherein It further includes: Obtaining a plurality of training samples, where the training samples include the full-cycle signal sequence of the sample rolling bearing and the actual service life corresponding to the sample rolling bearing; For the plurality of training samples, inputting the full-cycle signal sequence in the training samples into the model to be trained to obtain a predicted service life corresponding to the sample rolling bearing; where the model to be trained includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model, and the context feature extraction sub-model is constructed based on pre-determined target hyperparameters; the target hyperparameters are determined based on a hyperparameter optimization algorithm; Determining a loss value according to the actual service life and the predicted service life, and correcting the model parameters in the model to be trained based on the loss value, and taking the convergence of the loss function in the model to be trained as the training target to obtain a remaining service life prediction model for predicting the remaining service life of the rolling bearing.

7. The method for predicting the remaining service life of a bearing according to claim 5, wherein, It further includes: Determining at least one individual to be optimized corresponding to the context feature extraction sub-model according to a preset hyperparameter value range corresponding to the context feature extraction sub-model; where one individual to be optimized corresponds to a set of hyperparameters; For at least one individual to be optimized, constructing an optimized model based on the hyperparameters corresponding to the individual to be optimized, and determining a fitness value corresponding to the individual to be optimized based on the pre-obtained test samples and the optimized model; where the optimized model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model; Adjusting the hyperparameters corresponding to at least one individual to be optimized according to the fitness value and a preset mutation probability to obtain adjusted hyperparameters corresponding to the individual to be optimized; Repeating the steps of constructing the optimized model, determining the fitness value, and hyperparameter adjustment until a preset iteration end condition is reached to obtain the adjusted hyperparameters corresponding to at least one individual to be optimized and their corresponding fitness values; Taking the individual to be optimized corresponding to the minimum fitness value as the target optimized individual, and taking the adjusted hyperparameters corresponding to the target optimized individual as the target hyperparameters corresponding to the context feature extraction sub-model in the model to be trained, so as to construct the context feature extraction sub-model in the model to be trained based on the target hyperparameters.

8. A bearing remaining service life prediction device, characterized in that, It includes: A vibration signal sequence acquisition module, configured to acquire a first vibration signal sequence corresponding to a first rolling bearing at a current sampling moment; wherein, the first vibration signal sequence includes a first vibration signal corresponding to the current sampling moment and first vibration signals corresponding to at least one historical sampling moment before the current sampling moment; the first vibration signal is used to indicate the amplitude and frequency characteristics of the first rolling bearing in the horizontal force direction at the corresponding sampling moment; A vibration signal sequence determination module, configured to determine at least one characteristic component corresponding to the first vibration signal sequence, and determine a second vibration signal sequence corresponding to the first rolling bearing according to the at least one characteristic component and the first vibration signal sequence; A remaining service life prediction module, configured to process the second vibration signal sequence according to a pre-trained remaining service life prediction model, and predict a remaining service life corresponding to the first rolling bearing; wherein, the remaining service life prediction model includes a time series feature extraction sub-model, a context feature extraction sub-model, and a feature correlation determination sub-model.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute the bearing remaining service life prediction method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions are used to implement the bearing remaining service life prediction method according to any one of claims 1-7 when executed by a processor.

Images