A method for predicting remaining useful life of bearings considering the diffusion phenomenon of service degradation

Abstract

The present invention belongs to the technical field of mechanical equipment fault diagnosis and maintenance, and provides a bearing remaining service life prediction method considering the service degradation diffusion phenomenon, the method uses an acceleration sensor to pick up the rolling bearing accelerated life test data of each measuring point; reorganizes the normalized data into a two-dimensional matrix and divides it into a training data set and a test data set, and creates labels; constructs a pre-trained conditional time diffusion network prediction model, uses a pre-trained conditional encoder to extract features from the training data set to obtain a conditional vector; uses the diffusion model and the conditional vector to obtain a prediction sequence, and then applies the Lowess method to smooth the prediction sequence, and combines the least squares method to obtain the bearing remaining service life value; and simultaneously uses a differential loss function and an optimization algorithm to optimize the training to construct a pre-trained conditional time diffusion network prediction model to complete the bearing remaining service life prediction. The present invention can accurately predict the remaining service life of the bearing.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical equipment fault diagnosis and maintenance, and in particular relates to a method for predicting the remaining service life of a bearing taking into account the service degradation diffusion phenomenon. Background Art

[0002] In the current industrial context of the coordinated development of the Internet of Things, big data and intelligent manufacturing, the prediction of the service fatigue life of mechanical equipment can provide an important reference standard for the realization of predictive maintenance, intelligent decision-making and optimized operation of mechanical systems. As the core component of the mechanical equipment system, the normal operation of rolling bearings is directly related to the safe operation and reliable service of the entire mechanical equipment system. By accurately predicting the remaining service life of the bearings, maintenance or replacement can be performed before the bearings are about to fail, thereby avoiding shutdowns and safety accidents of the mechanical equipment system caused by bearing failures and achieving precise maintenance of the system. Therefore, research on the prediction of the remaining service life of bearings has become a hot issue in the field of fault diagnosis and predictive maintenance of mechanical equipment.

[0003] However, the long-term service monitoring data of bearings has nonlinear characteristics such as high dimension, large scale, strong multi-source fault coupling and time-varying parameters, and even lacks full-life sample data. The parameters of traditional prediction models (such as physical-based prediction models and data-driven prediction models) cannot be adaptively adjusted as the equipment operating conditions and monitoring data distribution characteristics change, resulting in the stability and prediction accuracy of the prediction model cannot meet the real-time and rapid response requirements of the project. Summary of the invention

[0004] The purpose of an embodiment of the present invention is to provide a method for predicting the remaining service life of a bearing taking into account the diffusion phenomenon of service degradation, aiming to solve the technical problem that the parameters of the traditional prediction model cannot be adaptively adjusted as the equipment operating conditions and the distribution characteristics of the monitoring data change, resulting in the stability and prediction accuracy of the prediction model failing to meet the real-time and rapid response requirements of the engineering. The present invention can achieve high-precision prediction of the remaining service life of the bearing and provide a theoretical basis for fault diagnosis and predictive maintenance of mechanical equipment systems.

[0005] To achieve the above object, the present invention provides the following technical solutions.

[0006] An embodiment of the present invention provides a method for predicting the remaining service life of a bearing taking into account the service degradation diffusion phenomenon, comprising:

[0007] Use acceleration sensors to pick up rolling bearing accelerated life test data at each vibration measurement point of rotating mechanical equipment;

[0008] The Min-Max normalization method is used to normalize the accelerated life test data to be predicted, and the normalized data is reorganized into a two-dimensional matrix;

[0009] Divide the reorganized two-dimensional matrix data into training data set and test data set, and create labels;

[0010] Construct a pre-trained conditional time diffusion network prediction model, including a pre-trained conditional encoder and a diffusion model; use the pre-trained conditional encoder to extract features from the training data set to obtain a conditional vector h;

[0011] The diffusion model and conditional vector h are used to obtain the prediction sequence, and then the Lowess method is used to smooth the prediction sequence, and the remaining service life of the bearing is obtained by combining the least squares method.

[0012] The differential loss function and optimization algorithm are used to optimize the training and construct a pre-trained conditional time diffusion network prediction model;

[0013] The remaining service life of the bearing is predicted based on the trained conditional time diffusion network prediction model, and the prediction result of the remaining service life of the bearing is obtained by using Min-Max denormalization.

[0014] Based on the prediction results of the remaining service life of the bearings, predictive maintenance is performed on mechanical system equipment.

[0015] Furthermore, the step of dividing the reorganized two-dimensional matrix data into a training data set and a test data set, and creating labels, includes:

[0016] Normalize the RUL value label of the bearing and express it as:

[0017] ;

[0018] In the formula, t represents the sampling time of the vibration signal; T represents the failure threshold of the bearing fault; FTP represents the first prediction time, and RUL (t) represents the digital label corresponding to the current sampling time t;

[0019] Convert RUL(t) from a numerical value to a sequence of length L that decreases linearly to 0. The sequence is represented as:

[0020] ;

[0021] Where X represents the sequence after the remaining service life of the bearing is converted into a sequence; L represents the length of the sequence vector; k represents the position index of the sequence, k=1,2,…,L; Represents the sequence value under the k-th position index.

[0022] Furthermore, the step of constructing a pre-trained conditional time diffusion network prediction model, including a pre-trained conditional encoder and a diffusion model; and extracting features from a training data set using the pre-trained conditional encoder to obtain a conditional vector h, includes:

[0023] Build a conditional encoder of the pre-trained conditional time diffusion network prediction model. The conditional encoder consists of the first convolutional layer of ResNet34 and the first three residual blocks, a flattening layer, a one-dimensional convolutional layer, a temporal convolutional network TCN, and a linear layer.

[0024] The reorganized two-dimensional matrix data is processed using a conditional encoder, and the conditional vector h is input into the diffusion model;

[0025] The signal matrix is ​​flattened by the frozen ResNet34 flattening layer and input into a one-dimensional convolution module with a convolution kernel size of 1 for dimensionality reduction;

[0026] The reduced-dimensional data is processed through the time convolution network TCN and the linear layer to obtain the conditional vector;

[0027] According to the constructed diffusion model, we get:

[0028] ;

[0029] Among them, α is the noise intensity; h is the conditional vector; z is Gaussian white noise, ; Represents a diffusion model with the independent variable being the learning weight θ.

[0030] Furthermore, the step of optimizing the training and constructing the pre-trained conditional time diffusion network prediction model by using the differential loss function and the optimization algorithm includes:

[0031] The degree of difference between the predicted result and the target value is expressed as:

[0032] ;

[0033] In the formula, is the target value set; is the i-th prediction result, is the jth prediction result; is the ith actual value; is the jth actual value, i=1,2,…,n;

[0034] The construction is based on the differential Huber loss function, expressed as:

[0035] ;

[0036] In the formula, and β is a hyperparameter, It is used to adjust the impact of outliers on the loss function, and β is used to alleviate the disappearance of gradients; represents the mean of the target values, Represents the mean of the prediction results;

[0037] The forward process of the conditional generative diffusion model is expressed as:

[0038] ;

[0039] In the formula, represents the noisy data at time step t in the forward process; is a parameter that controls the noise intensity; Indicates that the noise data Original data The contribution part; Parameter representing the noise intensity; I is the unit matrix;

[0040] In the reverse process of the conditional diffusion model, for a given time step t and noise vector And condition h, the learnable transfer kernel function is expressed as:

[0041] ;

[0042] In the formula, It means that under the given condition h, the sequence The conditional probability distribution of ; represents the learnable parameters of the network; is the noise distribution at time step T; represents the standard normal distribution, ; Indicates the state of the given current step t , under condition h and time step t, the conditional probability distribution of the state at time step t-1 is obtained; Indicates that the state of the current time step And condition h performs splicing operation function;

[0043] The training process is to minimize the negative log-likelihood function Optimization parameters ,in, represents the i-th sample in the target sequence; is the conditional vector corresponding to the i-th sample.

[0044] Compared with the prior art, the bearing remaining service life prediction method of the present invention has the following beneficial effects:

[0045] First, the present invention takes into account the statistical distribution, characteristic indicators, frequency, etc. of the data during the service degradation of the bearing, along with material failure, changes in operating conditions, or other sudden abnormal phenomena, and uses the pre-trained conditional time diffusion network prediction to accurately predict the remaining service life of the bearing;

[0046] Second, compared with the classical physics-based prediction model and the data-driven prediction model, the present invention combines the advantages of the ResNet34 network, the temporal convolutional network and the diffusion model. It can effectively capture long-distance sequence dependencies without increasing the number of parameters or computational complexity. The training process is not affected by the gradient vanishing and explosion problems, has strong parallel processing capabilities, high prediction accuracy, and fast algorithm operation speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0048] Figure 1 A system architecture block diagram of a method for predicting the remaining useful life of a bearing taking into account the diffusion phenomenon of service degradation according to the present invention;

[0049] Figure 2 for Figure 1 Schematic diagram of data preprocessing provided in;

[0050] Figure 3 for Figure 1 Schematic diagram of the pre-trained conditional encoder provided in;

[0051] Figure 4 for Figure 1 Schematic diagram of the diffusion model provided in;

[0052] Figure 5 for Figure 1 Schematic diagram of the prediction results provided in;

[0053] Figure 6 It is a schematic diagram of the implementation flow of the method for predicting the remaining service life of a bearing according to the present invention;

[0054] Figure 7 The bearing full life cycle vibration acceleration signal of the embodiment of the present invention;

[0055] Figure 8 It is the basic residual block diagram of 1DU-Net improved by the present invention;

[0056] Fig. 9 This is a schematic diagram of the denoising process results of the pre-trained conditional time diffusion network prediction model of the present invention when the step size is 0;

[0057] Fig.10 This is a schematic diagram of the denoising process results of the pre-trained conditional time diffusion network prediction model of the present invention when the step length is 249;

[0058] Fig.11This is a schematic diagram of the denoising process results of the pre-trained conditional time diffusion network prediction model of the present invention when the step length is 499;

[0059] Fig.12 This is a schematic diagram of the denoising process results of the pre-trained conditional time diffusion network prediction model of the present invention when the step length is 749;

[0060] Fig.13 This is a schematic diagram of the denoising process results of the pre-trained conditional time diffusion network prediction model of the present invention when the step length is 999;

[0061] Fig.14 This is the prediction result of the remaining service life of the bearing using the prediction model proposed in the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] At present, the long-term service monitoring data of bearings of mechanical equipment has nonlinear characteristics such as high dimension, large scale, strong coupling of multi-source faults and time-varying parameters, and even lacks full-life sample data. The parameters of traditional prediction models (such as physical-based prediction models and data-driven prediction models) cannot be adaptively adjusted as the equipment operating conditions and the distribution characteristics of monitoring data change, resulting in the stability and prediction accuracy of the prediction model failing to meet the requirements of engineering real-time and rapid response. In order to overcome the above-mentioned problems of poor stability and low prediction accuracy of the remaining life prediction of bearings, especially the interference of statistical distribution, characteristic indicators, frequency, etc. of data in the process of equipment service degradation with material failure, changes in operating conditions or other sudden abnormal phenomena, the present invention proposes a method for predicting the remaining service life of bearings that takes into account the diffusion phenomenon of service degradation, which can achieve high-precision prediction of the remaining life of bearings and provide a theoretical basis for fault diagnosis and predictive maintenance of mechanical equipment systems.

[0064] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0065] like Figure 1-Figure 6 As shown, in an embodiment of the present invention, a method for predicting the remaining service life of a bearing taking into account the service degradation diffusion phenomenon is provided, comprising the following steps:

[0066] S1. Use acceleration sensors to pick up rolling bearing accelerated life test data at each vibration measurement point of rotating mechanical equipment. In this step, an existing public data set for accelerated bearing life test (PRONOSTIA: an experimental platform for bearings accelerated life test, IEEE International Conference on Prognostics and Health Management PHM2012, IEEE, Denver, CO, USA, 2012) is used. Rotating bearings 2-6 are the research objects. The sampling frequency of the sensor in the experiment is 25.6 kHz. The experimental working conditions of the bearings are a rotation speed of 1650 rpm and a radial force of 4200 N. Samples are collected every 10 seconds. The sample length in one collection is 2560 sample points, that is, 0.1 s of data.

[0067] Figure 7 This is the vibration acceleration signal of bearing 2-6 throughout its life cycle. It can be seen that the amplitude of the vibration acceleration signal increases with time from the beginning of operation to the complete destruction of bearing 2-6.

[0068] S2, using the Min-Max normalization (i.e., deviation normalization) method to normalize the accelerated life test data to be predicted, and reorganizing the normalized data into a two-dimensional matrix;

[0069] S3, dividing the reorganized two-dimensional matrix data into a training data set and a test data set, and creating labels;

[0070] Among the data from the first prediction time to failure, the first 75% of the running data was augmented using the sliding window sampling method, and the entire data set was used as a test set for model training;

[0071] S4, construct a pre-trained conditional time diffusion network prediction model, including a pre-trained conditional encoder and a diffusion model; use the pre-trained conditional encoder to extract features from the training data set to obtain a conditional vector h;

[0072] In step S4 of the present invention, Figure 8As shown, the conditional encoder consists of the first convolution layer of ResNet34 and the first three residual blocks, a flattening layer, a one-dimensional convolution layer, a temporal convolutional network (TCN), and a linear layer. The signal matrix is ​​flattened by the frozen ResNet34 through the flattening layer, and input into the one-dimensional convolution module with a convolution kernel size of 1 for dimensionality reduction. The temporal convolutional network (TCN) then captures the long-term dependency in the time dimension by stacking multiple dilated causal convolutional layers and residual connections. Finally, the linear layer completes the linear transformation to obtain the conditional vector and inputs it into the diffusion model. Furthermore, the stacked network of the diffusion model uses 2DU-Net. In order to adapt to the distribution modeling of sequence data, the present invention changes the stacked network to a 1DU-Net model suitable for sequence modeling. The improved 1DU-Net basic residual block diagram, the number of downsampling channels of the U-Net network in the embodiment is 64, 128, 256, 512.

[0073] S5. Use the diffusion model and conditional vector h to obtain the prediction sequence, then apply the Lowess method to smooth the prediction sequence, and combine the least squares method to obtain the remaining useful life (RUL) value of the bearing;

[0074] Figure 9-13 This is a graph of the denoising process results of the pre-trained conditional time diffusion network prediction model at different step sizes. It can be seen that in the denoising process, at different denoising time steps, the predicted value of the model gradually converges to the true value from the initial distribution, and the denoising process gradually approaches the true value from the initial Gaussian distribution; in the middle stage, the overall trend of the data gradually emerges; in the later stage of denoising, the predicted value quickly converges to the true value; overall, the observed results are consistent with the denoising theoretical process, which proves the stability and interpretability of the diffusion model used in the present invention.

[0075] S6, using differential loss function and optimization algorithm to optimize training and build a pre-trained conditional time diffusion network prediction model; the present invention uses Lowess method to smooth the generated results, combines the least squares method to fit a straight line, and according to the formula , the RUL value is derived from the slope of the straight line;

[0076] S7, predicting the remaining service life of the bearing according to the trained conditional time diffusion network prediction model, and obtaining the prediction result of the remaining service life of the bearing by using Min-Max denormalization;

[0077] Fig.14 This is the prediction result of the remaining service life of the bearing. It can be seen that the prediction curve trend of the method proposed in the present invention is completely consistent with the actual degradation line of the bearing, and the prediction point fluctuates above and below the actual degradation line.

[0078] S8. Perform predictive maintenance on mechanical system equipment based on the prediction results of the remaining service life of the bearing.

[0079] Furthermore, the step of dividing the reorganized two-dimensional matrix data into a training data set and a test data set, and creating labels, includes:

[0080] Normalize the RUL value label of the bearing and express it as:

[0081] ;

[0082] In the formula, t represents the sampling time of the vibration signal; T represents the failure threshold of the bearing fault; FTP represents the first prediction time, and RUL (t) represents the digital label corresponding to the current sampling time t;

[0083] Convert RUL(t) from a numerical value to a sequence of length L that decreases linearly to 0. The sequence is represented as:

[0084] ;

[0085] Where X represents the sequence after the remaining service life of the bearing is converted into a sequence; L represents the length of the sequence vector; k represents the position index of the sequence, k=1,2,…,L; Represents the sequence value under the k-th position index.

[0086] Furthermore, the step of constructing a pre-trained conditional time diffusion network prediction model, including a pre-trained conditional encoder and a diffusion model; and extracting features from a training data set using the pre-trained conditional encoder to obtain a conditional vector h, includes:

[0087] Build a conditional encoder of the pre-trained conditional time diffusion network prediction model. The conditional encoder consists of the first convolutional layer of ResNet34 and the first three residual blocks, a flattening layer, a one-dimensional convolutional layer, a temporal convolutional network TCN, and a linear layer.

[0088] The reorganized two-dimensional matrix data is processed using a conditional encoder, and the conditional vector h is input into the diffusion model;

[0089] The signal matrix is ​​flattened by the frozen ResNet34 flattening layer and input into a one-dimensional convolution module with a convolution kernel size of 1 for dimensionality reduction;

[0090] The reduced-dimensional data is processed through the time convolution network TCN and the linear layer to obtain the conditional vector;

[0091] According to the constructed diffusion model, we get:

[0092] ;

[0093] Among them, α is the noise intensity; h is the conditional vector; z is Gaussian white noise, ; Represents a diffusion model with the independent variable being the learning weight θ.

[0094] Furthermore, the step of optimizing the training and constructing the pre-trained conditional time diffusion network prediction model by using the differential loss function and the optimization algorithm includes:

[0095] The degree of difference between the predicted result and the target value is expressed as:

[0096] ;

[0097] In the formula, is the target value set; is the i-th prediction result, is the jth prediction result; is the ith actual value; is the jth actual value, i=1,2,…,n;

[0098] The construction is based on the differential Huber loss function, expressed as:

[0099] ;

[0100] In the formula, and β is a hyperparameter, It is used to adjust the impact of outliers on the loss function, and β is used to alleviate the disappearance of gradients; represents the mean of the target values, Represents the mean of the prediction results;

[0101] The forward process of the conditional generative diffusion model is expressed as:

[0102] ;

[0103] In the formula, represents the noisy data at time step t in the forward process; is a parameter that controls the noise intensity; Indicates that the noise data Original data The contribution part; Parameter representing the noise intensity; I is the unit matrix;

[0104] In the reverse process of the conditional diffusion model, for a given time step t and noise vector And condition h, the learnable transfer kernel function is expressed as:

[0105] ;

[0106] In the formula, It means that under the given condition h, the sequence The conditional probability distribution of ; represents the learnable parameters of the network; is the noise distribution at time step T; represents the standard normal distribution, ; Indicates the state of the given current step t , under condition h and time step t, the conditional probability distribution of the state at time step t-1 is obtained; Indicates that the state of the current time step And condition h performs splicing operation function;

[0107] The training process is to minimize the negative log-likelihood function Optimization parameters ,in, represents the i-th sample in the target sequence; is the conditional vector corresponding to the i-th sample.

[0108] In summary, the present invention takes into account the statistical distribution, characteristic indicators, frequency, etc. of data during the service degradation of the bearing, along with material failure, operating condition changes, or other sudden abnormal phenomena, and uses the pre-trained conditional time diffusion network prediction to accurately predict the remaining service life of the bearing;

[0109] Compared with the classical physical-based prediction model and the data-driven prediction model, the present invention combines the advantages of the ResNet34 network, the temporal convolutional network and the diffusion model. It can effectively capture long-distance sequence dependencies without increasing the number of parameters or the computational complexity. The training process is not affected by the gradient vanishing and explosion problems, has strong parallel processing capabilities, high prediction accuracy, and fast algorithm operation speed.

[0110] Accordingly, an embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a method for predicting the remaining service life of a bearing taking into account the service degradation diffusion phenomenon described in the above-mentioned embodiment of the invention.

[0111] Accordingly, an embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a method for predicting the remaining useful life of a bearing that takes into account the service degradation diffusion phenomenon as described in the above-mentioned embodiment of the invention.

[0112] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0113] Those skilled in the art can clearly understand that for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0114] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0115] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the device, and various interfaces and lines are used to connect various parts of the entire device.

[0116] The memory can be used to store the computer program, and the processor realizes various functions of the device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0117] The storage medium is a storage medium, and the computer program is stored in the storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0118] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining useful life of a bearing considering the diffusion phenomenon of service degradation, characterized in that: The following steps are involved: Use acceleration sensors to pick up rolling bearing accelerated life test data at each vibration measurement point of rotating mechanical equipment; The Min-Max normalization method is used to normalize the accelerated life test data to be predicted, and the normalized data is reorganized into a two-dimensional matrix; Divide the reorganized two-dimensional matrix data into training data set and test data set, and create labels; Construct a pre-trained conditional time diffusion network prediction model, including a pre-trained conditional encoder and a diffusion model; use the pre-trained conditional encoder to extract features from the training data set to obtain a conditional vector h; The diffusion model and conditional vector h are used to obtain the prediction sequence, and then the Lowess method is used to smooth the prediction sequence, and the remaining service life of the bearing is obtained by combining the least squares method. The differential loss function and optimization algorithm are used to optimize the training and construct a pre-trained conditional time diffusion network prediction model; The remaining service life of the bearing is predicted based on the trained conditional time diffusion network prediction model, and the prediction result of the remaining service life of the bearing is obtained by using Min-Max denormalization. Based on the prediction results of the remaining service life of the bearings, predictive maintenance is performed on mechanical system equipment.

2. A method for predicting the remaining useful life of a bearing taking into account the diffusion phenomenon of service degradation according to claim 1, characterized in that: The steps of dividing the reorganized two-dimensional matrix data into a training data set and a test data set and creating labels include: Normalize the RUL value label of the bearing and express it as: ； Where, t represents the vibration signal sampling time; T represents the bearing fault failure threshold; FTP represents the first prediction time, and RUL (t) represents the digital label corresponding to the current sampling time t; Convert RUL(t) from a numerical value to a sequence of length L that decreases linearly to 0. The sequence is represented as: ； Where X represents the sequence after the remaining service life of the bearing is converted into a sequence; L represents the length of the sequence vector; k represents the position index of the sequence, k=1,2,…,L; Represents the sequence value under the k-th position index.

3. A method for predicting the remaining useful life of a bearing taking into account the diffusion phenomenon of service degradation according to claim 2, characterized in that: The step of constructing a pre-trained conditional time diffusion network prediction model, including a pre-trained conditional encoder and a diffusion model; and extracting features from a training data set using the pre-trained conditional encoder to obtain a conditional vector h, includes: Build a conditional encoder of the pre-trained conditional time diffusion network prediction model. The conditional encoder consists of the first convolutional layer of ResNet34 and the first three residual blocks, a flattening layer, a one-dimensional convolutional layer, a temporal convolutional network TCN, and a linear layer. The reorganized two-dimensional matrix data is processed using a conditional encoder, and the conditional vector h is input into the diffusion model; The signal matrix is ​​flattened by the frozen ResNet34 flattening layer and input into a one-dimensional convolution module with a convolution kernel size of 1 for dimensionality reduction; The reduced-dimensional data is processed through the time convolution network TCN and the linear layer to obtain the conditional vector; According to the constructed diffusion model, we get: ； Among them, α is the noise intensity; h is the conditional vector; z is Gaussian white noise, ; Represents a diffusion model with the independent variable being the learning weight θ.

4. A method for predicting the remaining useful life of a bearing taking into account the diffusion phenomenon of service degradation according to claim 3, characterized in that: The step of optimizing the training and constructing the pre-trained conditional time diffusion network prediction model by using the differential loss function and the optimization algorithm includes: The degree of difference between the predicted result and the target value is expressed as: ； In the formula, is the target value set; is the i-th prediction result, is the jth prediction result; is the ith actual value; is the jth actual value, i=1,2,…,n; The construction is based on the differential Huber loss function, expressed as: ； In the formula, and β is a hyperparameter, It is used to adjust the impact of outliers on the loss function, and β is used to alleviate the disappearance of gradients; represents the mean of the target values, Represents the mean of the prediction results; The forward process of the conditional generative diffusion model is expressed as: ； In the formula, represents the noisy data at time step t in the forward process; is a parameter that controls the noise intensity; Indicates that the noise data Original data The contribution part; Parameter representing the noise intensity; I is the unit matrix; In the reverse process of the conditional diffusion model, for a given time step t and noise vector And condition h, the learnable transfer kernel function is expressed as: ； In the formula, It means that under the given condition h, the sequence The conditional probability distribution of ; represents the learnable parameters of the network; is the noise distribution at time step T; represents the standard normal distribution, ; Indicates the state of the given current step t , under condition h and time step t, the conditional probability distribution of the state at time step t-1 is obtained; Indicates that the state of the current time step And condition h performs splicing operation function; The training process is to minimize the negative log-likelihood function Optimization parameters ,in, represents the i-th sample in the target sequence; is the conditional vector corresponding to the i-th sample.

Images