Bearing Causal Digital Twin Modeling Method and Device for Vertical Rotor Test Bench
By combining machine learning and causal relationship models in a synergistic approach on a vertical rotor test bench, a causal digital twin model of bearings is constructed, which solves the problems of insufficient accuracy and reliability in the existing technology, realizes more accurate condition assessment and fault diagnosis, and supports causal analysis and counterfactual experiments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI INSTITUTE OF APPLIED PHYSICS CHINESE ACADEMY OF SCIENCES
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-26
AI Technical Summary
Existing digital twin modeling methods cannot simultaneously achieve data-driven fitting capabilities and mechanism-driven causal explanatory power in bearing condition monitoring and fault diagnosis of vertical rotor test benches, resulting in insufficient accuracy and reliability.
By employing a collaborative approach combining machine learning and causal relationship models, a causal digital twin model of bearings on a vertical rotor test bench is constructed. By receiving vibration data in real time, a causal parameter prediction model and a causal relationship model are built to achieve accurate assessment of bearing condition and fault diagnosis.
It improves the accuracy and reliability of bearing condition assessment and fault diagnosis, enables causal analysis and counterfactual experiments, and provides effective support for fault prediction and root cause analysis.
Smart Images
Figure CN122087984A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital twin technology, and more specifically to a method and apparatus for causal digital twin modeling of bearings on a vertical rotor test bench. Background Technology
[0002] Digital twins provide an important tool for equipment condition monitoring, fault diagnosis, and optimization. By inputting real-time data from the physical system into a digital model for analysis, digital twins enable remote monitoring, predictive maintenance, and virtual commissioning of the physical system.
[0003] Currently, there are two main approaches to digital twin modeling: data-based modeling and physics-based modeling. Physics-based modeling, such as finite element method and CFD (computational fluid dynamics), offers high accuracy but requires significant computational resources. Data-based modeling offers better real-time performance but cannot trace the underlying operational mechanisms of the equipment. Summary of the Invention
[0004] The purpose of this invention is to provide a bearing causal digital twin modeling method and device for a vertical rotor test bench. By combining machine learning models and causal relationship models, it takes into account both the fitting ability of data-driven models and the causal interpretability of mechanism-driven models, thereby improving the accuracy and reliability of bearing condition assessment and fault diagnosis.
[0005] To achieve the above objectives, the present invention provides a bearing causal digital twin modeling method for a vertical rotor test bench, wherein the vertical rotor test bench includes a test bearing and a sample bearing, and the method includes:
[0006] Vibration data of the test bearing and the auxiliary bearing are received in real time at a preset receiving frequency.
[0007] The characteristic data of the accompanying bearing and the test bearing at each receiving moment are determined based on the vibration data of the accompanying bearing and the test bearing at that receiving moment.
[0008] A bearing causal digital twin model is constructed, which includes an interconnected causal parameter prediction model and a causal relationship model. The causal parameter prediction model is used to predict the causal parameters at each reception time based on the feature data of the test bearing and the companion bearing at that reception time. The causal relationship model is used to predict the feature data of the test bearing and the companion bearing at that reception time based on the causal parameters at that reception time. The causal parameters include the causal effect of the companion bearing on the test bearing and the causal effect of the test bearing on the companion bearing.
[0009] Acquire the characteristic data of the test bearing and the auxiliary bearing at N consecutive reception times, where the N consecutive reception times include the current reception time and the N-1 reception times before the current reception time;
[0010] The bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, so as to obtain a trained bearing causal digital twin model.
[0011] Optionally, the vibration data of the test bearing and the auxiliary bearing received at each receiving moment includes the vibration acceleration of multiple sampling points of the test bearing and the vibration acceleration of multiple sampling points of the test bearing obtained by sampling the vibration acceleration of the test bearing and the test bearing at a preset sampling frequency during the time between the receiving moment and the previous receiving moment.
[0012] Optionally, characteristic data of the accompanying bearing and the test bearing at each receiving moment are determined based on the vibration data of the accompanying bearing and the test bearing at that receiving moment, specifically including:
[0013] The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as the characteristic data of the test bearing at that receiving moment.
[0014] The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as characteristic data of the test bearing at that receiving moment.
[0015] Optionally, the bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, to obtain a trained bearing causal digital twin model, specifically including:
[0016] For each of the N consecutive receiving times, the feature data of the test bearing and the accompanying bearing at that receiving time are used as the first sample of that receiving time; the feature data of the test bearing and the accompanying bearing at that receiving time and the k receiving times before that receiving time are used as the second sample of that receiving time.
[0017] The first sample at each receiving time is input into the causal parameter prediction model so that the causal parameter prediction model outputs the predicted value of the causal parameter at that receiving time.
[0018] The predicted values of the causal parameters at each receiving time and the second sample at that receiving time are input into the causal relationship model so that the causal relationship model outputs the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving time.
[0019] The loss for each receiving moment is constructed based on the first sample at each receiving moment and the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving moment;
[0020] With the goal of minimizing the loss at each receiving time, the parameters of the causal parameter prediction model and the causal relationship model are updated to obtain a trained bearing causal digital twin model.
[0021] Optionally, the method further includes:
[0022] The characteristic data of the test bearing and the auxiliary bearing at the current receiving time are input into the trained bearing causal digital twin model to obtain the causal parameters at the current receiving time.
[0023] Another aspect of the present invention provides a bearing causal digital twin modeling device for a vertical rotor test bench, the vertical rotor test bench including a test bearing and a sample bearing, the device comprising:
[0024] The receiving module is used to receive vibration data of the test bearing and the auxiliary bearing in real time at a preset receiving frequency.
[0025] The determination module is used to determine the characteristic data of the test bearing and the monitoring bearing at each receiving time based on the vibration data of the test bearing and the monitoring bearing at each receiving time.
[0026] The module is used to construct a bearing causal digital twin model. The bearing causal digital twin model includes an interconnected causal parameter prediction model and a causal relationship model. The causal parameter prediction model is used to predict the causal parameters at each reception time based on the feature data of the test bearing and the companion bearing at that reception time. The causal relationship model is used to predict the feature data of the test bearing and the companion bearing at that reception time based on the causal parameters at that reception time. The causal parameters include the causal effect of the companion bearing on the test bearing and the causal effect of the test bearing on the companion bearing.
[0027] The acquisition module is used to acquire the characteristic data of the test bearing and the auxiliary bearing at N consecutive reception times. The N consecutive reception times include the current reception time and the N-1 reception times before the current reception time.
[0028] The training module is used to train the bearing causal digital twin model using feature data of the test bearing and the auxiliary bearing at N consecutive receiving times, so as to obtain a trained bearing causal digital twin model.
[0029] Optionally, the vibration data of the test bearing and the auxiliary bearing received at each receiving moment includes the vibration acceleration of multiple sampling points of the test bearing and the vibration acceleration of multiple sampling points of the test bearing obtained by sampling the vibration acceleration of the test bearing and the test bearing at a preset sampling frequency during the time between the receiving moment and the previous receiving moment.
[0030] Optionally, characteristic data of the accompanying bearing and the test bearing at each receiving moment are determined based on the vibration data of the accompanying bearing and the test bearing at that receiving moment, specifically including:
[0031] The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as the characteristic data of the test bearing at that receiving moment.
[0032] The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as characteristic data of the test bearing at that receiving moment.
[0033] Optionally, the bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, to obtain a trained bearing causal digital twin model, specifically including:
[0034] For each of the N consecutive receiving times, the feature data of the test bearing and the accompanying bearing at that receiving time are used as the first sample of that receiving time; the feature data of the test bearing and the accompanying bearing at that receiving time and the k receiving times before that receiving time are used as the second sample of that receiving time.
[0035] The first sample at each receiving time is input into the causal parameter prediction model so that the causal parameter prediction model outputs the predicted value of the causal parameter at that receiving time.
[0036] The predicted values of the causal parameters at each receiving time and the second sample at that receiving time are input into the causal relationship model so that the causal relationship model outputs the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving time.
[0037] The loss for each receiving moment is constructed based on the first sample at each receiving moment and the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving moment;
[0038] With the goal of minimizing the loss at each receiving time, the parameters of the causal parameter prediction model and the causal relationship model are updated to obtain a trained bearing causal digital twin model.
[0039] Optionally, the device further includes:
[0040] The prediction module is used to input the feature data of the test bearing and the companion bearing at the current receiving time into the trained bearing causal digital twin model to obtain the causal parameters at the current receiving time. Attached Figure Description
[0041] Figure 1A flowchart of a bearing causal digital twin modeling method for a vertical rotor test bench according to an embodiment of the present invention;
[0042] Figure 2 This is a schematic diagram of a cause-effect graph structure according to an embodiment of the present invention;
[0043] Figure 3 This is a structural block diagram of a bearing causal digital twin modeling device for a vertical rotor test bench according to an embodiment of the present invention. Detailed Implementation
[0044] The preferred embodiments of the present invention are given below with reference to the accompanying drawings and described in detail.
[0045] like Figure 1 As shown, this embodiment of the invention provides a causal digital twin modeling method for bearings on a vertical rotor test bench. The vertical rotor test bench includes a test bearing and a sample bearing. The modeling method includes the following steps:
[0046] S100: Receives vibration data from the test bearing and the auxiliary bearing in real time at a preset receiving frequency.
[0047] In some embodiments, the vibration data of the test bearing and the auxiliary bearing received at each receiving moment includes the vibration acceleration of multiple sampling points of the test bearing and the vibration acceleration of multiple sampling points of the test bearing obtained by sampling the vibration acceleration of the test bearing and the test bearing at a preset sampling frequency during the time between the receiving moment and the previous receiving moment; wherein, the sampling frequency is greater than the receiving frequency. Specifically, the vibration acceleration of the test bearing and the test bearing can be measured by two sensors respectively, and then the vibration acceleration of the test bearing and the test bearing can be sampled respectively. Since the sampling frequency is greater than the receiving frequency, there are multiple vibration accelerations of the test bearing and the test bearing in each receiving cycle. During each reception, multiple vibration accelerations of the test bearing and the test bearing during the time between the receiving moment and the previous receiving moment will be received.
[0048] S200: Determine the characteristic data of the test bearing and the auxiliary bearing at each receiving time based on the vibration data of the test bearing and the auxiliary bearing at each receiving time.
[0049] In some embodiments, step S200 specifically includes:
[0050] The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as the characteristic data of the test bearing at that receiving moment.
[0051] The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as characteristic data of the test bearing at that receiving moment.
[0052] S300: Construct a bearing causal digital twin model. The bearing causal digital twin model includes an interconnected causal parameter prediction model and a causal relationship model. The causal parameter prediction model is used to predict the causal parameters at each reception time based on the feature data of the test bearing and the companion bearing at each reception time. The causal relationship model is used to predict the feature data of the test bearing and the companion bearing at that reception time based on the causal parameters at that reception time. The causal parameters include the causal effect of the companion bearing on the test bearing and the causal effect of the test bearing on the companion bearing.
[0053] In some embodiments, the causal parameter prediction model can be various neural network models. The causal relationship model can be an SVAR (Structured Vector Autoregression) model. In this model, two causal graph structures are constructed, with the test bearing B1 and the experimental bearing B2 as the result nodes. Each causal graph structure consists of two nodes and one edge, as shown below. Figure 2 As shown.
[0054] S400: Acquire the characteristic data of the test bearing and the auxiliary bearing at N consecutive reception times. The N consecutive reception times include the current reception time and the N-1 reception times before the current reception time.
[0055] N can be set to any value as needed, for example, it can be set to 180~600.
[0056] S500: The bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, so as to obtain a trained bearing causal digital twin model.
[0057] The feature data of the accompanying bearings and test bearings at N consecutive reception times are used as inputs to the causal parameter prediction model, serving as training samples for the model. The causal parameter prediction model can predict the causal parameters for each reception time based on the feature data of the accompanying bearings and test bearings at each reception time. Then, the causal parameters for each reception time are input into the causal relationship model. At the same time, the feature data of the accompanying bearings and test bearings at the current reception time and the previous k reception times are also input into the causal relationship model. The causal relationship model predicts the feature data of the accompanying bearings and test bearings at the current reception time based on the causal parameters at each reception time and the feature data of the accompanying bearings and test bearings at the current reception time and the previous k reception times; where k is the time lag parameter.
[0058] Step S500 specifically includes:
[0059] S510: For each of the N consecutive receiving times, use the feature data of the test bearing and the auxiliary bearing at that receiving time as the first sample of that receiving time; obtain the feature data of the test bearing and the auxiliary bearing of the k receiving times before that receiving time as the second sample of that receiving time.
[0060] S520: Input the first sample at each receiving time into the causal parameter prediction model so that the causal parameter prediction model outputs the predicted value of the causal parameter at that receiving time;
[0061] S530: Input the predicted value of the causal parameter at each receiving time and the second sample at that receiving time into the causal relationship model so that the causal relationship model outputs the predicted value of the feature data of the test bearing and the auxiliary bearing at that receiving time.
[0062] S540: Construct the loss for each receiving moment based on the first sample at each receiving moment and the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving moment;
[0063] S550: With the goal of minimizing the loss at each receiving time, the parameters of the causal parameter prediction model and the causal relationship model are updated to obtain a trained bearing causal digital twin model.
[0064] For example, assuming the first sample at time t is x(t), then the second sample at time t includes x(t-τ), where τ = 0~k. After x(t) is input into the causal parameter prediction model, the causal parameter prediction model will perform the following calculations:
[0065]
[0066]
[0067] in, f is the hidden layer output, and f is the activation function (e.g., the Sigmoid function). and For gradient, For bias, Let be the causal parameters at time t.
[0068] When the causal relationship model is the SVAR model, the calculation is as follows:
[0069]
[0070] in, Let be the predicted values of the characteristic data of the test bearing and the auxiliary bearing at time t. The parameters used to simulate the perturbation of the stochastic process are independent and non-Gaussian.
[0071] The loss E is as follows:
[0072]
[0073] With the objective of minimizing E, the parameters of the causal relationship model and the causal parameter prediction model are updated using the backpropagation algorithm. The update strategy is as follows:
[0074]
[0075] Where lr is the learning rate and ω is the weight.
[0076] During training, the feature data of the test bearing and the feature data of the sample bearing are input simultaneously. The causal parameter prediction model estimates the causal parameters related to the two bearings at the same time. The causal relationship model predicts the feature data of the test bearing and the sample bearing at the current receiving time based on the causal parameters and the feature data of the test bearing at k+1 receiving times.
[0077] After the bearing causal digital twin model is trained, the feature data of the test bearing and the companion bearing at the current receiving moment can be input into the trained bearing causal digital twin model to obtain the causal parameters at the current receiving moment. These causal parameters not only quantify the operational influence between bearings but also enable hypothesis analysis and counterfactual experiments. Once the causal digital twin model is built, hypothesis analysis experiments can be conducted. For example, by analyzing and modifying the causal parameters, the operating state of the bearing before failure can be simulated. Therefore, by understanding the causal parameters, the possible future failure modes of the bearing system can be predicted. On the other hand, counterfactual experiments can also be conducted, such as the impact of the absence of the companion bearing on the test bearing. For more complex equipment systems, causal digital twin models will demonstrate even broader application prospects and value.
[0078] The bearing causal digital twin modeling method of the vertical rotor test bench of this invention combines machine learning model and causal relationship model. It constructs a bearing causal digital twin model through causal mapping network. Through structural adaptation, training optimization and online learning, it realizes real-time estimation of causal parameters. It can simulate the running state and perform causal analysis. It can also be applied to hypothesis analysis and counterfactual experiments, providing effective assistance for bearing condition assessment, fault diagnosis and root cause analysis.
[0079] like Figure 3 As shown, this embodiment of the invention also provides a bearing causal digital twin modeling device for a vertical rotor test bench, which includes a receiving module 10, a determining module 20, a constructing module 30, an acquiring module 40, and a training module 50.
[0080] The receiving module 10 is used to receive vibration data of the test bearing and the auxiliary bearing in real time at a preset receiving frequency.
[0081] The determination module 20 is used to determine the characteristic data of the test bearing and the auxiliary bearing at each receiving time based on the vibration data of the test bearing and the auxiliary bearing at each receiving time.
[0082] The construction module 30 is used to construct a bearing causal digital twin model. The bearing causal digital twin model includes an interconnected causal parameter prediction model and a causal relationship model. The causal parameter prediction model is used to predict the causal parameters at each reception time based on the feature data of the test bearing and the companion bearing at each reception time. The causal relationship model is used to predict the feature data of the test bearing and the companion bearing at that reception time based on the causal parameters at that reception time. The causal parameters include the causal effect of the companion bearing on the test bearing and the causal effect of the test bearing on the companion bearing.
[0083] The acquisition module 40 is used to acquire the characteristic data of the test bearing and the auxiliary bearing at N consecutive reception times. The N consecutive reception times include the current reception time and the N-1 reception times before the current reception time.
[0084] The training module 50 is used to train the bearing causal digital twin model using feature data of the test bearing and the companion bearing at N consecutive receiving times, so as to obtain a trained bearing causal digital twin model.
[0085] The receiving module 10, determining module 20, constructing module 30, acquiring module 40, and training module 50 are the functional modules corresponding to steps S100-S500 in the method embodiment, respectively. Their specific implementation methods can be found in the description in the method embodiment, and will not be repeated here.
[0086] In some embodiments, the apparatus may further include:
[0087] The prediction module is used to input the feature data of the test bearing and the companion bearing at the current receiving time into the trained bearing causal digital twin model to obtain the causal parameters at the current receiving time.
[0088] The bearing causal digital twin modeling device of the vertical rotor test bench of this invention combines machine learning model and causal relationship model, and constructs bearing causal digital twin model through causal mapping network. Through structural adaptation, training optimization and online learning, it realizes real-time estimation of causal parameters. It can simulate the running state and perform causal analysis. It can also be applied to hypothesis analysis and counterfactual experiments, providing effective assistance for bearing condition assessment, fault diagnosis and root cause analysis.
[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. That is, all simple and equivalent changes and modifications made based on the claims and description of this invention fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.
Claims
1. A bearing causal digital twin modeling method for a vertical rotor test bench, characterized in that, The vertical rotor test bench includes a test bearing and a test bearing, and the method includes: Vibration data of the test bearing and the auxiliary bearing are received in real time at a preset receiving frequency. The characteristic data of the accompanying bearing and the test bearing at each receiving moment are determined based on the vibration data of the accompanying bearing and the test bearing at that receiving moment. A bearing causal digital twin model is constructed, which includes an interconnected causal parameter prediction model and a causal relationship model. The causal parameter prediction model is used to predict the causal parameters at each reception time based on the feature data of the test bearing and the companion bearing at that reception time. The causal relationship model is used to predict the feature data of the test bearing and the companion bearing at that reception time based on the causal parameters at that reception time. The causal parameters include the causal effect of the companion bearing on the test bearing and the causal effect of the test bearing on the companion bearing. Acquire the characteristic data of the test bearing and the auxiliary bearing at N consecutive reception times, where the N consecutive reception times include the current reception time and the N-1 reception times before the current reception time; The bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, so as to obtain a trained bearing causal digital twin model.
2. The bearing causal digital twin modeling method for a vertical rotor test bench according to claim 1, characterized in that, The vibration data of the test bearing and the sample bearing received at each receiving moment includes the vibration acceleration of multiple sampling points of the test bearing and the vibration acceleration of multiple sampling points of the sample bearing obtained by sampling the vibration acceleration of the test bearing and the sample bearing at a preset sampling frequency during the time between the receiving moment and the previous receiving moment.
3. The bearing causal digital twin modeling method for a vertical rotor test bench according to claim 2, characterized in that, Based on the vibration data of the test bearing and the monitoring bearing at each receiving moment, the characteristic data of the test bearing and the monitoring bearing at that receiving moment are determined, specifically including: The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as the characteristic data of the test bearing at that receiving moment. The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as characteristic data of the test bearing at that receiving moment.
4. The bearing causal digital twin modeling method for a vertical rotor test bench according to claim 2, characterized in that, The bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, resulting in a well-trained bearing causal digital twin model. Specifically, this includes: For each of the N consecutive receiving times, the feature data of the test bearing and the accompanying bearing at that receiving time are used as the first sample of that receiving time; the feature data of the test bearing and the accompanying bearing at that receiving time and the k receiving times before that receiving time are used as the second sample of that receiving time. The first sample at each receiving time is input into the causal parameter prediction model so that the causal parameter prediction model outputs the predicted value of the causal parameter at that receiving time. The predicted values of the causal parameters at each receiving time and the second sample at that receiving time are input into the causal relationship model so that the causal relationship model outputs the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving time. The loss for each receiving moment is constructed based on the first sample at each receiving moment and the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving moment; With the goal of minimizing the loss at each receiving time, the parameters of the causal parameter prediction model and the causal relationship model are updated to obtain a trained bearing causal digital twin model.
5. The bearing causal digital twin modeling method for a vertical rotor test bench according to claim 2, characterized in that, Also includes: The characteristic data of the test bearing and the auxiliary bearing at the current receiving time are input into the trained bearing causal digital twin model to obtain the causal parameters at the current receiving time.
6. A bearing causal digital twin modeling device for a vertical rotor test bench, characterized in that, The vertical rotor test bench includes a test bearing and a test bearing, and the device includes: The receiving module is used to receive vibration data of the test bearing and the auxiliary bearing in real time at a preset receiving frequency. The determination module is used to determine the characteristic data of the test bearing and the monitoring bearing at each receiving time based on the vibration data of the test bearing and the monitoring bearing at each receiving time. The module is used to construct a bearing causal digital twin model. The bearing causal digital twin model includes an interconnected causal parameter prediction model and a causal relationship model. The causal parameter prediction model is used to predict the causal parameters at each reception time based on the feature data of the test bearing and the companion bearing at that reception time. The causal relationship model is used to predict the feature data of the test bearing and the companion bearing at that reception time based on the causal parameters at that reception time. The causal parameters include the causal effect of the companion bearing on the test bearing and the causal effect of the test bearing on the companion bearing. The acquisition module is used to acquire the characteristic data of the test bearing and the auxiliary bearing at N consecutive reception times. The N consecutive reception times include the current reception time and the N-1 reception times before the current reception time. The training module is used to train the bearing causal digital twin model using feature data of the test bearing and the auxiliary bearing at N consecutive receiving times, so as to obtain a trained bearing causal digital twin model.
7. The bearing causal digital twin modeling device for a vertical rotor test bench according to claim 6, characterized in that, The vibration data of the test bearing and the sample bearing received at each receiving moment includes the vibration acceleration of multiple sampling points of the test bearing and the vibration acceleration of multiple sampling points of the sample bearing obtained by sampling the vibration acceleration of the test bearing and the sample bearing at a preset sampling frequency during the time between the receiving moment and the previous receiving moment.
8. The bearing causal digital twin modeling device for a vertical rotor test bench according to claim 7, characterized in that, Based on the vibration data of the test bearing and the monitoring bearing at each receiving moment, the characteristic data of the test bearing and the monitoring bearing at that receiving moment are determined, specifically including: The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as the characteristic data of the test bearing at that receiving moment. The maximum value, mean value, waveform factor, margin factor, kurtosis, and root mean square value of each vibration acceleration of the test bearing at each receiving moment are obtained as characteristic data of the test bearing at that receiving moment.
9. The bearing causal digital twin modeling device for a vertical rotor test bench according to claim 7, characterized in that, The bearing causal digital twin model is trained using feature data from the test bearing and the accompanying bearing at N consecutive receiving times, resulting in a well-trained bearing causal digital twin model. Specifically, this includes: For each of the N consecutive receiving times, the feature data of the test bearing and the accompanying bearing at that receiving time are used as the first sample of that receiving time; the feature data of the test bearing and the accompanying bearing at that receiving time and the k receiving times before that receiving time are used as the second sample of that receiving time. The first sample at each receiving time is input into the causal parameter prediction model so that the causal parameter prediction model outputs the predicted value of the causal parameter at that receiving time. The predicted values of the causal parameters at each receiving time and the second sample at that receiving time are input into the causal relationship model so that the causal relationship model outputs the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving time. The loss for each receiving moment is constructed based on the first sample at each receiving moment and the predicted values of the feature data of the test bearing and the auxiliary bearing at that receiving moment; With the goal of minimizing the loss at each receiving time, the parameters of the causal parameter prediction model and the causal relationship model are updated to obtain a trained bearing causal digital twin model.
10. The bearing causal digital twin modeling device for a vertical rotor test bench according to claim 7, characterized in that, Also includes: The prediction module is used to input the feature data of the test bearing and the companion bearing at the current receiving time into the trained bearing causal digital twin model to obtain the causal parameters at the current receiving time.