Rolling bearing residual life prediction method and device, terminal, medium and product

Abstract

The invention discloses a method and device for predicting the residual life of a rolling bearing, a terminal, a storage medium and a computer program product, and the method comprises the steps: building a digital twinborn model based on a historical signal of the full life cycle data of the rolling bearing; performing simulation by using a digital twin model to obtain vibration amplitudes of the rolling bearing under different defect sizes, and taking the vibration amplitudes as simulation signals of the rolling bearing; training a BP neural network by using the simulation signal of the rolling bearing, and updating the digital twinborn model; establishing a physical information network, and training the physical information network by using the simulation signal of the rolling bearing; and based on the process of updating the digital twin model and the trained physical information network, predicting the residual life of the rolling bearing to obtain a prediction result of the residual life of the rolling bearing. According to the scheme, the rolling bearing digital twin model based on the dynamical model is established, the RUL prediction of the rolling bearing is performed by using the physical information network, and the accuracy of predicting the residual life is improved.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and specifically relates to a method, device, terminal, storage medium and computer program product for predicting the remaining life of a rolling bearing, and in particular to a method, device, terminal, storage medium and computer program product for predicting the remaining life of a rolling bearing based on a digital twin model and physical information. Background Art

[0002] Predicting the remaining useful life is a key element of prognostics and health management (PHM) technology, which is widely used in various mechanical equipment. As a key component of rotating mechanical systems, rolling bearings are used in complex and changing environments. Therefore, it is necessary to continuously monitor the condition of rolling bearings and accurately predict their remaining useful life (RUL) to ensure optimal performance and prevent unexpected failures.

[0003] The RUL prediction methods of rolling bearings mainly include data-driven methods, physical model methods and hybrid methods. Data-driven methods include statistical methods and machine learning methods. However, the remaining life prediction methods of rolling bearings based on deep learning in related schemes often rely too much on a large amount of experimental data to train the network, and ignore the objective physical laws during the training process, resulting in the prediction results after the network training are inconsistent with the physical laws, affecting the accuracy of the prediction results after the network training is completed. Among them, the actual physical laws, such as the actual remaining life of rolling bearings, are gradually decreasing, and it is impossible to increase; and the rolling bearing has defects during operation, and this defect will only gradually increase, and will not disappear or become smaller.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The purpose of the present invention is to provide a method, device, terminal, storage medium and computer program product for predicting the remaining life of a rolling bearing, so as to solve the problem that the remaining life prediction method of a rolling bearing based on deep learning in related schemes often relies too much on a large amount of experimental data to train the network, ignores the objective physical laws during the training process, and affects the accuracy of the prediction of the remaining life of the rolling bearing. The effect of improving the accuracy of the prediction of the remaining life of the rolling bearing is achieved by establishing a digital twin model of the rolling bearing based on a dynamic model, utilizing the construction of a physical information network prediction model and the RUL prediction of the rolling bearing.

[0006] The present invention provides a method for predicting the remaining life of a rolling bearing, comprising: collecting historical signals of the full life cycle data of the rolling bearing; establishing a digital twin model based on the historical signals of the full life cycle data of the rolling bearing; and using the digital twin model to simulate and obtain the vibration amplitude of the rolling bearing under different defect sizes as the simulation signal of the rolling bearing; using the simulation signal of the rolling bearing to train a BP neural network and update the digital twin model; establishing a physical information network and training the physical information network using the simulation signal of the rolling bearing; based on the process of updating the digital twin model and the trained physical information network, predicting the remaining life of the rolling bearing, and obtaining a prediction result of the remaining life of the rolling bearing.

[0007] In some embodiments, the full life cycle data of the rolling bearing includes: the mass of the outer ring and the inner ring of the rolling bearing, the vertical displacement of the outer ring and the inner ring of the rolling bearing, the equivalent stiffness of the outer ring and the inner ring of the rolling bearing, the equivalent damping of the outer ring and the inner ring of the rolling bearing, the force caused by the eccentricity of the rolling bearing, the speed of the inner ring of the rolling bearing, and the external load of the rolling bearing.

[0008] In some embodiments, a digital twin model is established based on the historical signals of the full life cycle data of the rolling bearing, including: determining the gap between the bearing rolling elements and the retaining cage of the rolling bearing and determining the roughness of the bearing surface of the rolling bearing based on the historical signals of the full life cycle data of the rolling bearing; based on the gap between the bearing rolling elements and the retaining cage of the rolling bearing and the roughness of the bearing surface of the rolling bearing, improving the four-degree-of-freedom dynamic model of the rolling bearing, and using the improved dynamic model as the digital twin model.

[0009] In some embodiments, the BP neural network is trained using the simulation signal of the rolling bearing, and the digital twin model is updated, including: training the BP neural network using the simulation signal of the rolling bearing, inputting the actual signal of the full life cycle data of the rolling bearing into the trained BP neural network to obtain an estimated value of the defect size of the rolling bearing; using the estimated value of the defect size of the rolling bearing to update the digital twin model, and repeating this cycle.

[0010] In some embodiments, a physical information network is established, and the physical information network is trained using the simulation signal of the rolling bearing; a physical information network including a fully connected layer, a Dropout layer, and a physical information layer is constructed; wherein the physical information layer is established by using a loss function including physical information, and the Dropout layer is used to achieve uncertainty estimation of the prediction results; the physical information network is trained using the simulation signal of the rolling bearing to obtain the trained physical information network.

[0011] In some embodiments, based on the process of updating the digital twin model and the trained physical information network, the remaining life of the rolling bearing is predicted to obtain a prediction result of the remaining life of the rolling bearing, including: determining an estimated value of the defect size of the rolling bearing obtained in the process of updating the digital twin model; inputting the estimated value of the defect size of the rolling bearing into the trained physical information network to obtain a predicted value of the remaining life of the rolling bearing; and using the predicted value of the remaining life of the rolling bearing within a set error range as the prediction result of the remaining life of the rolling bearing.

[0012] Matching the above method, the present invention provides, on the other hand, a device for predicting the remaining life of a rolling bearing, comprising: an acquisition unit, configured to collect historical signals of the full life cycle data of the rolling bearing; a control unit, configured to establish a digital twin model based on the historical signals of the full life cycle data of the rolling bearing; and use the digital twin model to simulate and obtain the vibration amplitude of the rolling bearing under different defect sizes as the simulation signal of the rolling bearing; the control unit is also configured to use the simulation signal of the rolling bearing to train a BP neural network and update the digital twin model; the control unit is also configured to establish a physical information network and train the physical information network using the simulation signal of the rolling bearing; the control unit is also configured to predict the remaining life of the rolling bearing based on the process of updating the digital twin model and the trained physical information network, and obtain a prediction result of the remaining life of the rolling bearing.

[0013] Matching the above-mentioned device, the present invention further provides a terminal on the other hand, including: the above-mentioned device for predicting the remaining life of a rolling bearing.

[0014] Matching the above method, the present invention further provides a computer program product on the other hand, including a computer program, which implements the steps of the above-mentioned method for predicting the remaining life of a rolling bearing when executed by a processor.

[0015] In accordance with the above method, the present invention provides a storage medium on another aspect, wherein the storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned method for predicting the remaining life of a rolling bearing.

[0016] Therefore, the scheme of the present invention constructs a high-fidelity digital twin model of a rolling bearing by establishing a digital twin model based on an improved dynamic model; uses a BP network to learn the relationship between the defect size and amplitude in the digital twin model of the rolling bearing, and uses the trained BP network to estimate the defect size during actual operation; builds a physical information network prediction model, which is established by using a loss function containing physical information, and the Dropout layer is used to realize the uncertainty estimation of the prediction result; uses the full life cycle data of the rolling bearing to update the digital twin model, and during the update process, the BP network estimates the defect size, and inputs the estimated defect size into the established RUL prediction network to obtain a prediction result that is more in line with physical laws; thereby, the accuracy of the remaining life prediction of the rolling bearing is improved through the establishment of a digital twin model of the rolling bearing based on the dynamic model, the digital twin model update method, the construction of the physical information network prediction model and the RUL prediction of the rolling bearing.

[0017] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0018] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic flow chart of an embodiment of a method for predicting the remaining life of a rolling bearing according to the present invention; Figure 2 A schematic diagram of a process for establishing a digital twin model in an embodiment of the method of the present invention; Figure 3 A schematic diagram of a process of updating the digital twin model in an embodiment of the method of the present invention; Figure 4 A schematic diagram of a flow chart of an embodiment of establishing a physical information network in the method of the present invention; Figure 5 A schematic diagram of a flow chart of an embodiment of predicting the remaining life of the rolling bearing in the method of the present invention; Figure 6 It is a structural schematic diagram of an embodiment of a device for predicting the remaining life of a rolling bearing of the present invention; Figure 7A schematic flow chart of a method for predicting the remaining life of a rolling bearing based on a digital twin model and physical information according to the present invention; Figure 8 It is a structural schematic diagram of the four-degree-of-freedom dynamics model; Fig. 9 Schematic diagrams of rolling element operation, wherein (a) is a schematic diagram of the relationship between the cage and the rolling element, (b) is a schematic diagram of a case where the defect size is larger, and (c) is a schematic diagram of a case where the defect size is smaller; Fig.10 A flowchart of the method for updating the digital twin model; Fig.11 It is a schematic diagram of the structure of the physical information network Dr-PINN; Fig.12 is a schematic diagram of the prediction results of the method of the present invention; Fig.13 It is a schematic diagram of the comparison results between the method of the present invention and the method in the related scheme; Fig.14 The BP network parameter setting table is Table 1; Fig.15 The bearing operation data parameter table is Table 2; Fig.16 The comparison result table with the related schemes is Table 3; Fig.17 The results are compared with other advanced methods in Table 4.

[0020] In conjunction with the accompanying drawings, the reference numerals in the embodiments of the present invention are as follows: 1-acquisition unit; 2-control unit. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] In related schemes, the rolling bearing remaining life prediction method based on deep learning often relies too much on a large amount of experimental data to train the network, ignoring the objective physical laws during the training process, which affects the accuracy of the prediction of the remaining life of the rolling bearing.

[0023] In addition, the deep learning technology in the relevant scheme is learned based on a large amount of data. However, in the process of collecting data, the data will be influenced by factors such as the environment, resulting in the knowledge learned by the deep learning method not being consistent with the actual rules, that is, the remaining life will increase. For example, Fig.13 In the example shown, for other deep learning methods, the remaining life span will show a significant decrease at the 200th minute, but then a rapid increase trend. However, in the actual process, the remaining life span will only gradually decrease and cannot increase.

[0024] Secondly, the deep learning method in the relevant scheme makes a point estimate of the remaining life prediction, ignoring the uncertainty of the prediction results. Most of the prediction results of the deep learning network in the relevant scheme are a specific value, but because the deep learning network is affected by various conditions during the learning process, such as initial values ​​and other factors, even for the same set of data, the test results after multiple training of the network will be different, resulting in different prediction results. If the prediction result only gives a specific value, it is impossible to judge the extent to which the network is affected by these uncertain factors, and it is impossible to accurately estimate the credibility of the prediction result.

[0025] It can be seen that the deep learning-based rolling bearing RUL prediction method in the relevant scheme relies on experimental data during the training process. However, the bearing is affected by many factors during the actual operation, resulting in the prediction results being inconsistent with the objective physical laws. Secondly, for point estimates in deep learning, uncertainty cannot be quantified.

[0026] Furthermore, given that statistical methods rely on expert experience and knowledge, machine learning methods often lack interpretability, and physical methods rely on the accuracy of physical models. Hybrid-driven methods can effectively reduce these problems and have made significant progress in recent years. Combining physical models with data-driven methods can fully utilize their respective advantages, mitigate their respective shortcomings, and produce more reliable and efficient prediction results.

[0027] Digital twin (DT) technology has attracted widespread attention due to its ability to create simulation models that are very similar to physical entities, thereby enabling effective status detection and fault diagnosis of physical equipment. DT is a comprehensive technology that integrates methods from physical modeling and simulation, data analysis and processing, and deep learning and related schemes. It has a wide range of application possibilities in location anomaly identification and RUL prediction.

[0028] Physics-Informed Neural Networks (PINN) integrate physical knowledge into neural networks and use it as constraints to make the results produced by neural networks consistent with physical principles. This has also been applied to RUL prediction.

[0029] In the prediction process, point estimates only provide a single estimate and do not provide information about the accuracy of the estimate. However, quantifying uncertainty is crucial for the assessment of the device state. A commonly used uncertainty method in neural networks is Bayesian neural networks. However, Bayesian networks face the problems of high model complexity and sensitivity to parameter changes, which makes them unsuitable for integration with PINNs. Therefore, a more concise and efficient method is needed to quantify uncertainty. The Dropout-based uncertainty quantification method has the significant advantage of not requiring the explicit definition of complex prior distributions. It can be directly applied to the neural network architecture in the relevant scheme without being restricted by the prior choice.

[0030] In order to solve the above problems, improve the kinetic model, and establish a more accurate kinetic model as the DT model. In RUL prediction, due to the lack of interpretability of deep learning methods and the limited accuracy of physical model methods alone, a method combining deep learning and physical model methods using PINN is adopted.

[0031] Therefore, the solution of the present invention proposes a method for predicting the remaining life of a rolling bearing, specifically a method for predicting the remaining life of a rolling bearing based on digital twins and physical information. Through the uncertainty quantification method, not only the value of the prediction result can be given, but also the confidence interval of the prediction result can be given, which solves the problem that the training result is inconsistent with the objective physical law, and quantifies the uncertainty of the remaining life, thereby improving the accuracy of the prediction of the remaining life of the rolling bearing. It can be seen that the solution of the present invention takes into account the uncertainty of the prediction result, while the related solutions ignore the uncertainty of the prediction result.

[0032] According to an embodiment of the present invention, a method for predicting the remaining life of a rolling bearing is provided. Figure 1 The flowchart of an embodiment of the method of the present invention is shown in FIG. The method for predicting the remaining life of a rolling bearing may include: steps S10 to S50.

[0033] In step S10, historical signals of the full life cycle data of the rolling bearing are collected.

[0034] Among them, the full life cycle data of the rolling bearing includes: the mass of the outer ring and the inner ring of the rolling bearing, the vertical displacement of the outer ring and the inner ring of the rolling bearing, the equivalent stiffness of the outer ring and the inner ring of the rolling bearing, the equivalent damping of the outer ring and the inner ring of the rolling bearing, the force caused by the bearing eccentricity of the rolling bearing, the speed of the inner ring of the rolling bearing, and the external load of the rolling bearing.

[0035] In step S20, a digital twin model is established based on the historical signal of the full life cycle data of the rolling bearing; and the vibration amplitude of the rolling bearing under different defect sizes is simulated using the digital twin model as the simulation signal of the rolling bearing.

[0036] In step S30, the BP neural network is trained using the simulation signal of the rolling bearing, and the digital twin model is updated.

[0037] In step S40, a physical information network is established, and the physical information network is trained using the simulation signal of the rolling bearing.

[0038] At step S50, based on the process of updating the digital twin model and the trained physical information network, the remaining life of the rolling bearing is predicted to obtain a prediction result of the remaining life of the rolling bearing.

[0039] The scheme of the present invention provides a rolling bearing RUL prediction scheme based on digital twin and physical information, improves the dynamic model of the rolling bearing, uses the dynamic model as a digital twin model, and provides a digital twin model update method; constructs a physical information network based on Dropout to predict the RUL of the rolling bearing, so that the prediction result is more in line with the objective physical law, and provides uncertainty quantification of the prediction result, thereby improving the accuracy of the prediction result. Thus, through the uncertainty quantification method, not only the value of the prediction result can be given, but also the confidence interval of the prediction result can be given, which solves the problem that the training result is inconsistent with the objective physical law, and quantifies the uncertainty of the remaining life, thereby improving the accuracy of the remaining life prediction of the rolling bearing.

[0040] In some embodiments, the specific process of establishing the digital twin model based on the historical signal of the full life cycle data of the rolling bearing in step S20 is described in the following exemplary embodiment.

[0041] Combine the following Figure 2 The flowchart of an embodiment of establishing a digital twin model in the method of the present invention is shown, which further illustrates the specific process of establishing the digital twin model in step S20, including: step S21 to step S22.

[0042] Step S21, determining the clearance between the rolling element and the cage of the rolling bearing based on the historical signal of the full life cycle data of the rolling bearing, and determining the roughness of the bearing surface of the rolling bearing; Step S22, based on the clearance between the rolling elements and the retaining cage of the rolling bearing and the roughness of the bearing surface of the rolling bearing, the four-degree-of-freedom dynamic model of the rolling bearing is improved, and the improved dynamic model is used as the digital twin model.

[0043] Figure 7 FIG. 1 is a flow chart of a method for predicting the remaining life of a rolling bearing based on a digital twin model and physical information according to the present invention. Figure 7 As shown, the method for predicting the remaining life of a rolling bearing based on a digital twin model and physical information includes: the first step, i.e., step S110, is to establish a digital twin model.

[0044] Figure 8 This is a schematic diagram of the structure of the four-degree-of-freedom dynamic model. Figure 8 The x and y in the figure are a set of unit vectors, x refers to the unit vector in the horizontal direction, and y refers to the unit vector in the vertical direction. The function of these two sets of unit vectors is to define the direction of other physical quantities with directions. For example, when the bearing load F acts vertically upward on the bearing, the direction of the vector is opposite to the y direction in the figure, and the value of F is negative. Figure 8 According to the structural diagram in the figure, a digital twin model is first established based on the improved dynamic model, and a high-fidelity digital twin model of the rolling bearing is constructed. This model takes into account the surface roughness of the bearing, the gap between the rolling element and the cage, and the uneven defective surface on the basis of the dynamic model of the relevant scheme. Figure 8 The following is a schematic diagram of the dynamic model of a four-degree-of-freedom bearing. It is assumed that the outer ring is fixed and the inner ring rotates. The inner ring and outer ring of the bearing have one degree of freedom in the horizontal direction and one degree of freedom in the vertical direction respectively. Formula (1) is the dynamic model of a four-degree-of-freedom rolling bearing. In the following formulas, formulas (1), (2), (5), (8) and (9) are existing formulas, and the rest are modified or self-defined formulas.

[0045] In the formula, and Respectively represent the mass of the outer and inner rings of the bearing; and Respectively represent the horizontal displacement of the outer ring and inner ring of the bearing; and Respectively represent the vertical displacement of the outer ring and inner ring of the bearing; and Respectively represent the equivalent stiffness of the outer ring and inner ring of the bearing; and c i Respectively represent the equivalent damping of the outer and inner rings of the bearing; Indicates the force caused by bearing eccentricity; Indicates the speed of the inner ring of the bearing; Indicates the external load of the bearing; represents the acceleration due to gravity, which is 9.8 . and F Hy They represent the contact forces generated by the rolling elements in the horizontal and vertical directions of the bearing, respectively, and are calculated by formula (2); (1); (2).

[0046] In the formula, Indicates the first Angular position of each rolling element; Indicates the number of rolling elements; Indicates The contact deformation produced by each rolling element; represents the equivalent stiffness; represents the Hayvis function; represents the deformation index, which is 1.5 for ball bearings; and The calculation method of is expressed by formula (3) and formula (4); (3); (4).

[0047] In calculation In the process, the solution of the present invention takes into account the influence of the gap between the bearing rolling elements and the cage. Fig. 9 The following are schematic diagrams of the rolling element operation, where (a) is a schematic diagram of the relationship between the cage and the rolling element, (b) is a schematic diagram of the case where the defect size is larger, and (c) is a schematic diagram of the case where the defect size is smaller. Fig. 9 As shown in (a), during operation, the rolling element will randomly appear within the range constrained by the cage. In formula (3), Indicates the angle corresponding to the gap between the rolling element and the cage; Indicates the initial angular position of the rolling body; rand indicates a random number between 0 and 1; represents the angular velocity of the rolling element center position, as shown in equation (5); (5).

[0048] In the formula, Indicates the rolling element diameter, Indicates the bearing pitch diameter; In calculation In the process, the influence of the bearing surface roughness is considered, and the roughness of the bearing is assigned a discrete random value, where the convex surface is positive and the concave surface is negative. Indicates the bearing surface roughness; Indicates the radial clearance of the bearing; Indicates the displacement caused by bearing defects; in conventional dynamic models, it is assumed that the defect surface is smooth, but in actual processes, the defect surface is rough. In order to establish a high-fidelity dynamic model, the solution of the present invention assumes that the defect surface is not smooth. According to the literature, the defect depth at the defect position conforms to the normal distribution. Therefore, in this study, Multiply it by a random number between -1 and 1 that follows a normal distribution. To represent the non-smooth defective surface. Fig. 9 As shown in (b) and (c), the solution of the present invention takes into account the situation when the defect surface is irregular and the bearing surface is not smooth. When the bearing defect size is small, the bearing rolling element will not contact the bottom of the defect. At this time, the displacement caused by the defect is calculated by formula (6). When the defect size increases, the bearing rolling element moves to the defect position and can contact the bottom of the defect. At this time, the displacement caused by the defect is calculated by formula (7).

[0049] like Fig. 9 As shown in the figure, when the outer ring fails, the displacement caused by the defect will occur when the rolling element runs to the defect position. When the defect size is small and the rolling element cannot contact the bottom of the defect, the displacement caused by the defect for: (6); In formula (6) .

[0050] When the defect size is large and the rolling element can contact the bottom of the defect, the displacement caused by the defect for: (7).

[0051] In the formula, mod means remainder operation, Indicates the initial angular position of the rolling element; Indicates the angle corresponding to the defect; Indicates the angle at which the rolling element contacts the bottom of the defect.

[0052] This step uses MATLAB to build a model, and uses MATLAB's ode45 command to solve the model. In the solution of the present invention, the actual running time of the bearing and the defect size estimation value output by the BP network in MATLAB are exported to Excel format, and imported into the Dr-PINN network built in Python to obtain the RUL prediction result of the rolling bearing.

[0053] In some embodiments, the specific process of using the simulation signal of the rolling bearing to train the BP neural network and update the digital twin model in step S30 is described in the following exemplary embodiment.

[0054] Combine the following Figure 3 The flowchart of an embodiment of updating the digital twin model in the method of the present invention is shown, which further illustrates the specific process of updating the digital twin model in step S30, including: step S31 to step S32.

[0055] Step S31, using the simulation signal of the rolling bearing to train the BP neural network, inputting the actual signal of the full life cycle data of the rolling bearing into the trained BP neural network to obtain an estimated value of the defect size of the rolling bearing.

[0056] Step S32, using the estimated value of the defect size of the rolling bearing, update the digital twin model and repeat this cycle.

[0057] like Figure 7 As shown, the method for predicting the remaining life of a rolling bearing based on a digital twin model and physical information also includes: a second step, i.e., step S120, a method for updating the digital twin model.

[0058] Fig.10 The flowchart of the digital twin model update method is shown in Figure 2. Fig.10 The BP network parameter settings are shown in Table 1. Fig.14 The BP network parameter setting table is Table 1.

[0059] In the digital twin model update method, the trained BP network is used to update the defect size of the model. The training data of the BP network is generated by the dynamic model, where different defect sizes are input to generate corresponding vibration signals and vibration amplitudes for each sampling period. The vibration amplitude of the simulation signal is used as input data, and the defect size is used as a label. The BP network learns the relationship between the defect size and the vibration amplitude in the dynamic model. During the model update process, the amplitude of the actual vibration signal is input into the BP network, and then the defect size is output. Table 1 illustrates the structure of the BP network.

[0060] This part uses MATLAB to model the BP network and update the dynamic model. The amplitude of the vibration signal and the defect size generated by the dynamic model in the previous step are output to the BP network for training. The actual vibration signal is input into the trained BP network to obtain an estimated value of the defect size. The full life cycle data of the rolling bearing is used to update the digital twin model. During the update process, the BP network estimates the defect size, and the estimated defect size is input into the established RUL prediction network to obtain a prediction result that is more in line with physical laws. Among them, the full life cycle data of the rolling bearing refers to the data of the entire cycle from the start of the rolling bearing's operation in a healthy state to the damage of the rolling bearing; in the scheme of the present invention, the data used is specifically the vibration signal of the rolling bearing.

[0061] In some implementations, the specific process of establishing a physical information network in step S40 and training the physical information network using the simulation signal of the rolling bearing is described in the following exemplary embodiments.

[0062] Combine the following Figure 4 The flowchart of an embodiment of establishing a physical information network in the method of the present invention further illustrates the specific process of establishing the physical information network in step S40, including: step S41 to step S42.

[0063] Step S41, building a physical information network including a fully connected layer, a Dropout layer, and a physical information layer; wherein the physical information layer is established by using a loss function containing physical information, and the Dropout layer is used to achieve uncertainty estimation of the prediction results.

[0064] Step S42: training the physical information network using the simulation signal of the rolling bearing to obtain the trained physical information network.

[0065] like Figure 7 As shown, the method for predicting the remaining life of a rolling bearing based on a digital twin model and physical information also includes: a third step, i.e., step S130, for establishing a physical information network model (Dropout-physical informed neural network, Dr-PINN) for predicting RUL.

[0066] Fig.11 Figure 1 is a schematic diagram of the physical information network Dr-PINN structure. Fig.11 As shown in Fig.11 As shown in Figure 2. A Neural Network (NN) is used to find the relationship between the input features and the remaining life. The physics of bearing degradation can be represented by a partial differential equation (PDE). Typically, a PDE is defined as: (8).

[0067] In the formula, u is the solution function in the differential equation, and the independent variable of the solution function is t and x , is a nonlinear operator. u about t The partial derivative of . An NN is used to approximate u in the formula, and the actual value and predicted value of the NN are used as one of the loss terms. At the same time, the automatic differentiation method inherent in the neural network is used to obtain the partial derivatives used to calculate (8). The actual value and true value of equation (8) are used as another loss of the PINN network, called physical loss. Finally, through network training, the output results of the network can be made consistent with the embedded physical rules. Therefore, the optimization goal of the network is: (9).

[0068] In the formula, the first term represents the loss of predicted value and true value, and the second term represents the physical loss. Represents the coefficient of physical loss. The meaning of this formula is that the optimization goal is to minimize the two losses.

[0069] Here is a more specific explanation of the embedded physics: The remaining useful life of a bearing is the time from now on that it can continue to operate reliably. Based on experience, there are three key physical facts about bearing defects and life: (1) Irreversibility of defect formation: The process of defect formation in bearings is irreversible. Defects such as cracks, dents or localized wear may occur due to various factors such as wear, fatigue or corrosion. Once these defects are formed, they are usually impossible to repair or restore to their original state. Over time, the size of these defects tends to increase due to stress concentration around the defects or further physical damage; (2) Life expectancy decreases monotonically: Given that defects in bearings are irreversible, their size can only increase, so the life of the bearing decreases monotonically over time. As the number of defects increases, the remaining useful life of the bearing continues to decrease; (3) Initial conditions: When the operating time of the bearing is 0 and the defect size is 0, the remaining service life of the bearing is considered to be 100%.

[0070] The three physical constraints obtained from the above three rules are as follows: (10); (11); (12).

[0071] In the formula, represents the network's prediction value of the rolling bearing RUL, T Indicates the bearing running time, L represents the defect size of the bearing. Equations (10), (11) and (12) are written as loss functions in the network as follows: (13); (14); (15).

[0072] In the formula, N represents the number of samples.

[0073] The theory of Dropout uncertainty quantification method in the network is as follows: In any deep nonlinear neural network, adding a Dropout layer before the weight layer is mathematically equivalent to an approximation of a probabilistic deep Gaussian process. In Monte Carlo Dropout, Dropout is kept constant during both training and testing phases, but during the testing phase, Dropout is only used to evaluate uncertainty. Specifically, during the testing phase, in order to evaluate the model uncertainty of the input data points, this can be achieved by forward propagating the model multiple times and observing the distribution of the outputs. During each forward propagation, Dropout is randomly applied, resulting in different outputs, i.e., the number of forward propagations on the network. Finally, by performing statistical analysis on these outputs, information about the uncertainty of the predictions can be obtained. The mean and variance of the output are expressed as (16) and (17), respectively: (16); (17).

[0074] In the formula, M represents the number of network forward propagation times, Represents the predicted value during each forward propagation process.

[0075] This part builds the Dr-PINN network model in Python.

[0076] In the scheme of the present invention, an improved rolling bearing dynamics model is used, which takes into account the clearance between the rolling element and the cage and the surface roughness of the bearing; based on the BP network bearing dynamics model update scheme, based on the physical information network Dr-PINN for uncertainty quantification, this network adds an uncertainty quantification mechanism on the basis of the ordinary PINN network, so that the network can output uncertainty quantification values. At the same time, the network combines the physical objective laws of the rolling bearing RUL, defines three physical rules as loss functions and embeds them into the network. Compared with other schemes, the scheme of the present invention uses a digital twin model and its update method to monitor the operating status of the bearing and estimate the defect size. The output value of the RUL prediction network Dr-PINN is more in line with the physical laws and has better interpretability than other deep learning networks. The accuracy of RUL prediction is improved by using the scheme of the present invention.

[0077] In the scheme of the present invention, in the first stage, the dynamic part of the digital twin model can be replaced by finite element analysis, etc., and simulation signals can also be generated; in the second stage, in the dynamic model update step, the BP network can be replaced by other networks. The substantial reason why the prediction results of the scheme of the present invention can be in line with the laws of physics is that the digital twin model is first established, and the results produced by the model are in line with the laws of physics and will not be affected by factors such as the environment; secondly, in the prediction model Dr-PINN, by embedding physical knowledge into the network in the form of a loss function, the network prediction results are constrained, so that the prediction results output by the network can be more in line with the laws of physics.

[0078] In some embodiments, in step S50, based on the process of updating the digital twin model and the trained physical information network, the remaining life of the rolling bearing is predicted, and the specific process of obtaining the prediction result of the remaining life of the rolling bearing is shown in the following exemplary description.

[0079] Combine the following Figure 5 The flowchart of an embodiment of predicting the remaining life of the rolling bearing in the method of the present invention further illustrates the specific process of predicting the remaining life of the rolling bearing in step S50, including: steps S51 to S53.

[0080] Step S51, determining an estimated value of the defect size of the rolling bearing obtained in the process of updating the digital twin model.

[0081] Step S52, inputting the estimated value of the defect size of the rolling bearing into the trained physical information network to obtain the predicted value of the remaining life of the rolling bearing.

[0082] Step S53, taking the predicted value of the remaining life of the rolling bearing within a set error range as the predicted result of the remaining life of the rolling bearing.

[0083] like Figure 7 As shown, the method for predicting the remaining life of a rolling bearing based on a digital twin model and physical information also includes: a fourth step, namely step S140, predicting the RUL of the bearing using a full life cycle data set.

[0084] The XJTU-SY bearing data set is used for verification. This data set tests the life cycle data of rolling bearings under three different operating conditions. The bearing model is LDK204, the sampling frequency is 25.6kHz, the sampling interval is 1s, and the sampling time is 1.28s each time. The parameters of the bearing are shown in Table 2. Fig.15 Table 2 is the bearing operation data parameter table.

[0085] Use Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Coefficient of Determination R 2 As evaluation indicators, the following is the calculation method of each evaluation indicator. The smaller the MAE and RMSE, the better. 2 Bigger is better; (18); (19); (20).

[0086] In the formula represents the predicted value, Indicates the actual value, N is the number of samples.

[0087] To verify the effectiveness of the model, some neural networks CNN, LSTM, TCN and RNN widely used in prediction are first used for comparison with the solution of the present invention. Table 3 is the comparison results. Fig.16 The comparison result table of the methods of the related schemes is Table 3. As can be seen from Table 3, the schemes of the present invention are superior to the methods in the related schemes. Fig.12 It can be seen that the solution of the present invention can not only provide a prediction result, but also provide a quantitative value of the uncertainty of the prediction result. Fig.12 is a schematic diagram of the prediction results of the method of the present invention, Fig.13 Schematic diagram of the comparison results between the method of the present invention and the method in the related scheme. Fig.13It can be seen that the method in the related scheme has a process in which the predicted value suddenly drops and then rises rapidly, which is inconsistent with the objective laws of physics. The method proposed in the scheme of the present invention greatly alleviates this phenomenon and is more in line with the laws of physics.

[0088] Table 3 is the comparison results of the proposed method Dr-PINN of the present invention and the methods of related solutions, MAE, RMSE and R 2 These three evaluation indicators are used to evaluate the accuracy of network prediction results. The smaller the MAE and RMSE, the better. 2 From Table 3, it can be seen that the three evaluation indicators of the method proposed by the present invention are better than those of the methods in the related schemes, indicating that the prediction results of the scheme of the present invention are more accurate. Fig.12 The black line in the figure is the actual value, the blue line is the predicted value, and the light blue part is the confidence interval of the prediction result, that is, the quantitative result of uncertainty. This result shows that the method proposed by the solution of the present invention can not only provide accurate prediction results but also provide uncertainty quantification of the prediction results, and at the same time, the prediction results do not show a significant increase. Fig.13 Schematic diagram of the comparison results between the method of the present invention and the method in the related scheme, Fig.13 In the figure, the blue line is the prediction result of the solution method of the present invention, and the light blue area is the uncertainty quantification result of the solution method of the present invention. The method in the related solution cannot provide uncertainty quantification results (no similar Fig.12 and Fig.13 The light blue area in the figure). In addition, the solution of the present invention is more in line with the laws of physics. For example, at the 200th minute, the methods in the related solutions all showed a situation where the remaining life dropped sharply and then rose rapidly, while the method proposed by the present invention did not have this problem. In addition, at the early stage of the fault, the prediction result of the solution of the present invention is closer to the true value, while the method in the related solution showed a phenomenon of slow rise, which is also inconsistent with the laws of physics.

[0089] To further verify the effectiveness of the present invention and compare it with other advanced methods, Table 4 shows the experimental results. Fig.16 The results are compared with other advanced methods in Table 4.

[0090] As can be seen from Table 4, the methods proposed in the present invention have good performance in the experiments. Although the performance of two groups is not as good as that of the related schemes, the overall difference is not large. Table 4 compares the scheme of the present invention with the advanced methods in other papers. In the figure, RMSE and MAE are evaluation indicators used to evaluate the accuracy of the prediction results. The smaller the index value, the better. It can be seen from Table 4 that among the 8 groups of experiments, 6 groups have the best results with the scheme of the present invention. Although the results of the remaining 2 groups are not the best, they are not much different from the minimum value of the index, which shows that the prediction results of the scheme of the present invention are the best overall.

[0091] The scheme of the present invention includes four steps: establishment of a rolling bearing digital twin model based on a dynamic model, a digital twin model update method, construction of a physical information network prediction model, and RUL prediction of rolling bearings. Among them, digital twin model establishment: by considering the gap between the bearing rolling element and the cage, and the roughness of the bearing surface, the four-degree-of-freedom dynamic model of the rolling bearing is improved, and the improved dynamic model is used as a digital twin model. Digital twin model update method: use the BP network to learn the relationship between the defect size and amplitude in the rolling bearing digital twin model, and use the trained BP network to estimate the defect size during actual operation; wherein the data for training the BP network comes from the simulation signal generated by the established digital twin model. Physical information network prediction model construction: the network is mainly composed of a fully connected layer, a Dropout layer, and a physical information layer. The physical information layer is established by using a loss function containing physical information. The Dropout layer is used to realize the uncertainty estimation of the prediction result. This layer is established based on the following three physical objective laws. RUL prediction of rolling bearings: Use the full life cycle data of rolling bearings to update the digital twin model. During the update process, the BP network estimates the defect size and inputs the estimated defect size into the established RUL prediction network to obtain a prediction result that is more in line with physical laws.

[0092] By adopting the technical solution of this embodiment, a high-fidelity digital twin model of a rolling bearing is constructed by establishing a digital twin model based on an improved dynamic model; the BP network is used to learn the relationship between the defect size and amplitude in the digital twin model of the rolling bearing, and the trained BP network is used to estimate the defect size during actual operation; a physical information network prediction model is built by using a loss function containing physical information, and the Dropout layer is used to realize the uncertainty estimation of the prediction result; the digital twin model is updated using the full life cycle data of the rolling bearing, and the BP network estimates the defect size during the update process, and the estimated defect size is input into the established RUL prediction network to obtain a prediction result that is more in line with physical laws; thus, the accuracy of the remaining life prediction of the rolling bearing is improved by establishing a digital twin model of the rolling bearing based on the dynamic model, the digital twin model updating method, the construction of the physical information network prediction model and the RUL prediction of the rolling bearing.

[0093] According to an embodiment of the present invention, a device for predicting the remaining life of a rolling bearing corresponding to the method for predicting the remaining life of a rolling bearing is also provided. Figure 6 The structural schematic diagram of an embodiment of the device of the present invention is shown. The device for predicting the remaining life of a rolling bearing may include: an acquisition unit 1 and a control unit 2.

[0094] The acquisition unit 1 is configured to collect historical signals of the full life cycle data of the rolling bearing. The specific functions and processing of the acquisition unit 1 refer to step S10.

[0095] The control unit 2 is configured to establish a digital twin model based on the historical signal of the full life cycle data of the rolling bearing; and use the digital twin model to simulate and obtain the vibration amplitude of the rolling bearing under different defect sizes as the simulation signal of the rolling bearing. The specific functions and processing of the control unit 2 refer to step S20.

[0096] The control unit 2 is further configured to train the BP neural network using the simulation signal of the rolling bearing and update the digital twin model. The specific functions and processing of the control unit 2 are also shown in step S30.

[0097] The control unit 2 is further configured to establish a physical information network and train the physical information network using the simulation signal of the rolling bearing. The specific functions and processing of the control unit 2 are also shown in step S40.

[0098] The control unit 2 is further configured to predict the remaining life of the rolling bearing based on the process of updating the digital twin model and the trained physical information network to obtain a prediction result of the remaining life of the rolling bearing. The specific functions and processing of the control unit 2 are also shown in step S50.

[0099] The solution of the present invention improves the dynamic model of the rolling bearing, uses the dynamic model as a digital twin model, and provides a digital twin model update method; constructs a physical information network based on Dropout to predict the RUL of the rolling bearing, so that the prediction result is more in line with the objective physical law, and provides uncertainty quantification of the prediction result, thereby improving the accuracy of the prediction result. Thus, through the uncertainty quantification method, not only the value of the prediction result can be given, but also the confidence interval of the prediction result can be given, which solves the problem that the training result is inconsistent with the objective physical law, and quantifies the uncertainty of the remaining life, thereby improving the accuracy of the remaining life prediction of the rolling bearing.

[0100] Since the processing and functions implemented by the device of this embodiment basically correspond to the embodiments, principles and examples of the aforementioned method, for the details not fully described in this embodiment, please refer to the relevant descriptions in the aforementioned embodiments, and no further elaboration will be made here.

[0101] According to an embodiment of the present invention, a terminal corresponding to the prediction device for the remaining life of a rolling bearing is also provided. The terminal may include: the prediction device for the remaining life of a rolling bearing described above.

[0102] Since the processing and functions implemented by the terminal of this embodiment basically correspond to the embodiments, principles and examples of the aforementioned devices, for the details not fully described in this embodiment, please refer to the relevant descriptions in the aforementioned embodiments, and no further elaboration will be made here.

[0103] According to an embodiment of the present invention, a computer program product corresponding to a method for predicting the remaining life of a rolling bearing is also provided, comprising a computer program which, when executed by a processor, implements the steps of the control method of the voltage detection device described above.

[0104] Since the processing and functions implemented by the product of this embodiment basically correspond to the embodiments, principles and examples of the aforementioned method, for the details not fully described in this embodiment, please refer to the relevant descriptions in the aforementioned embodiments, and no further elaboration will be made here.

[0105] According to an embodiment of the present invention, a storage medium corresponding to a method for predicting the remaining life of a rolling bearing is also provided, wherein the storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the method for predicting the remaining life of a rolling bearing described above.

[0106] Since the processing and functions implemented by the storage medium of this embodiment basically correspond to the embodiments, principles and examples of the aforementioned method, for the details not fully described in this embodiment, please refer to the relevant descriptions in the aforementioned embodiments, and no further elaboration will be made here.

[0107] In summary, it is easy for those skilled in the art to understand that, under the premise of no conflict, the above-mentioned advantageous methods can be freely combined and superimposed.

[0108] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the remaining life of a rolling bearing, characterized in that: include: Collecting historical signals of full life cycle data of the rolling bearing; A digital twin model is established based on the historical signal of the full life cycle data of the rolling bearing; and the vibration amplitude of the rolling bearing under different defect sizes is simulated by using the digital twin model as the simulation signal of the rolling bearing; Using the simulation signal of the rolling bearing to train the BP neural network and update the digital twin model; Establishing a physical information network, and training the physical information network using the simulation signal of the rolling bearing; Based on the process of updating the digital twin model and the trained physical information network, the remaining life of the rolling bearing is predicted to obtain a prediction result of the remaining life of the rolling bearing.

2. The method for predicting the remaining life of a rolling bearing according to claim 1, characterized in that: in, The full life cycle data of the rolling bearing includes: the mass of the outer ring and the inner ring of the rolling bearing, the vertical displacement of the outer ring and the inner ring of the rolling bearing, the equivalent stiffness of the outer ring and the inner ring of the rolling bearing, the equivalent damping of the outer ring and the inner ring of the rolling bearing, the force caused by the bearing eccentricity of the rolling bearing, the speed of the inner ring of the rolling bearing, and the external load of the rolling bearing.

3. The method for predicting the remaining life of a rolling bearing according to claim 1 or 2, characterized in that: Based on the historical signals of the full life cycle data of the rolling bearing, a digital twin model is established, including: Based on the historical signal of the full life cycle data of the rolling bearing, determining the clearance between the bearing rolling element and the cage of the rolling bearing, and determining the roughness of the bearing surface of the rolling bearing; Based on the clearance between the rolling elements and the retaining cage of the rolling bearing and the roughness of the bearing surface of the rolling bearing, the four-degree-of-freedom dynamic model of the rolling bearing is improved, and the improved dynamic model is used as a digital twin model.

4. The method for predicting the remaining life of a rolling bearing according to any one of claims 1 to 3, characterized in that: Using the simulation signal of the rolling bearing to train the BP neural network and update the digital twin model includes: Using the simulation signal of the rolling bearing to train the BP neural network, inputting the actual signal of the full life cycle data of the rolling bearing into the trained BP neural network to obtain an estimated value of the defect size of the rolling bearing; The digital twin model is updated using the estimated value of the defect size of the rolling bearing, and the cycle is repeated.

5. The method for predicting the remaining life of a rolling bearing according to any one of claims 1 to 4, characterized in that: Establishing a physical information network, and training the physical information network using the simulation signal of the rolling bearing; Building a physical information network including a fully connected layer, a Dropout layer, and a physical information layer; wherein the physical information layer is established by using a loss function including physical information, and the Dropout layer is used to achieve uncertainty estimation of the prediction results; The physical information network is trained using the simulation signal of the rolling bearing to obtain the trained physical information network.

6. The method for predicting the remaining life of a rolling bearing according to any one of claims 1 to 5, characterized in that: Based on the process of updating the digital twin model and the trained physical information network, the remaining life of the rolling bearing is predicted to obtain a prediction result of the remaining life of the rolling bearing, including: Determining an estimated value of a defect size of the rolling bearing obtained during the process of updating the digital twin model; Inputting the estimated value of the defect size of the rolling bearing into the trained physical information network to obtain the predicted value of the remaining life of the rolling bearing; The predicted value of the remaining life of the rolling bearing within the set error range is used as the prediction result of the remaining life of the rolling bearing.

7. A device for predicting the remaining life of a rolling bearing, characterized in that: include: An acquisition unit configured to collect historical signals of full life cycle data of the rolling bearing; A control unit is configured to establish a digital twin model based on historical signals of the full life cycle data of the rolling bearing; and use the digital twin model to simulate and obtain the vibration amplitude of the rolling bearing under different defect sizes as a simulation signal of the rolling bearing; The control unit is further configured to train a BP neural network using a simulation signal of the rolling bearing and update the digital twin model; The control unit is further configured to establish a physical information network and train the physical information network using the simulation signal of the rolling bearing; The control unit is also configured to predict the remaining life of the rolling bearing based on the process of updating the digital twin model and the trained physical information network to obtain a prediction result of the remaining life of the rolling bearing.

8. A terminal, characterized in that: include: The device for predicting the remaining life of a rolling bearing as claimed in claim 7.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the method for predicting the remaining life of a rolling bearing according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the control method of the voltage detection device according to any one of claims 1 to 6 are implemented.

Images