A gearbox online fault diagnosis method based on digital twinning
By establishing a digital twin model of the gearbox, the virtual-real linkage of the gearbox equipment is realized, which solves the problems of insufficient accuracy and real-time performance in the existing technology, and realizes efficient and low-cost online fault diagnosis and monitoring, thus extending the service life of the equipment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KUNMING UNIV OF SCI & TECH
- Filing Date
- 2023-02-08
- Publication Date
- 2026-06-23
AI Technical Summary
Existing gearbox fault diagnosis methods are insufficient in terms of accuracy and real-time performance, making it difficult to meet the requirements of high efficiency, low cost, and stable operation in industrial environments, especially in the case of the difficulty in identifying and diagnosing multiple faults under harsh working conditions.
By establishing a digital twin model of the gearbox that highly maps to its actual operating state, the virtual-real linkage of the gearbox equipment is realized. The model parameters are optimized using cosine similarity and trust region algorithms, and fault feature extraction and real-time monitoring are performed using convolutional neural networks to achieve online fault diagnosis.
It improves the accuracy and real-time performance of gearbox fault diagnosis, enables efficient and low-cost online monitoring and fault identification, and extends the service life of equipment.
Smart Images
Figure CN116046384B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent and digital mechanical fault diagnosis technology, specifically relating to an online fault diagnosis method for gearboxes based on digital twins. Background Technology
[0002] A gearbox is an integrated transmission mechanism with advantages such as compact structure, large transmission ratio, and high load capacity. It is frequently used as the transmission hub in many important mechanical devices, such as helicopters, ships, armored vehicles, and wind turbines. The health of the gearbox is crucial to the normal operation of the entire machine, affecting economic efficiency and personnel safety. Gearboxes are generally used in high-load and harsh environments, resulting in a relatively high failure rate. For example, in a typical wind turbine, 20% of downtime is due to gearbox failure, which can have serious consequences. Utilizing effective fault diagnosis methods to quickly and accurately identify gearbox faults can save significant time and maintenance costs, improving work efficiency. Therefore, real-time and effective monitoring and fault diagnosis of the gearbox's operating status are of great importance.
[0003] Currently, methods for gearbox fault diagnosis are mainly divided into two categories. The first category is model-based fault diagnosis. Fan et al. established a dynamic model of torsional vibration of a helicopter planetary transmission gear train with a cracked planetary carrier. They considered nonlinear factors such as time-varying meshing stiffness, gear backlash, and viscous damping in simulating the influence of the planetary carrier crack on the planetary pile angle offset and the dynamic parameters of the gear train. Cheng et al. established dynamic models of torsional vibration of a helicopter planetary transmission gear train with normal operation and sun gear tooth failure. Based on the characteristic parameters proposed by the model, they quantitatively detected the degree of sun gear tooth failure. These model-based fault diagnosis methods can improve the understanding of various physical responses under gearbox transmission gear train fault modes and enrich fault diagnosis methods and data sources. However, due to the many assumptions and simplifications in the models, they cannot cover the influence of various uncertain factors, and there is still a certain gap between the simulation data and the actual situation, so the accuracy of fault diagnosis needs to be improved. The second category is signal processing-based fault diagnosis. Meng Lingxia et al. proposed a method based on Gabor rearrangement of logarithmic time-frequency ridge manifolds, which can suppress cross terms and effectively extract early fault features of gearboxes under varying operating conditions. Wang Zhile et al. drew on the advantages of signal processing techniques such as envelope analysis, windowed synchronous averaging, and order ratio calculation tracking, and proposed the envelope angular domain windowed synchronous averaging technique to effectively demodulate the weak fault characteristic signal of the sun gear and successfully extract the fault characteristics of the sun gear.
[0004] The aforementioned fault diagnosis methods based on signal processing, however, are limited because both the information detection methods and fault analysis methods are designed for specific signals and specific faults. They cannot systematically address and resolve problems, and struggle to identify multiple faults occurring simultaneously. Furthermore, using raw vibration signals as input makes the accuracy of identification susceptible to harsh working conditions and environmental uncertainties. They also rely heavily on expert experience, lack universality, and are insufficient to meet the needs of real-world industrial applications.
[0005] Therefore, to address the above problems, this invention provides a gearbox online fault diagnosis method based on digital twins. By establishing a gearbox digital twin model that highly maps to the actual operating state, the "virtual-real linkage" of the gearbox equipment is realized, enabling real-time online monitoring of the gearbox's operating status and achieving online fault diagnosis of the gearbox, thereby maximizing the maintenance of efficient, low-cost, and stable system operation. Summary of the Invention
[0006] To address the aforementioned technical issues, this invention proposes a digital twin-based online fault diagnosis method for gearboxes. By establishing a digital twin model of the gearbox that highly maps to its actual operating state, the method achieves "virtual-real linkage" of the gearbox equipment, enabling real-time online monitoring of the gearbox's operating status and realizing online fault diagnosis. This maximizes the efficiency, cost-effectiveness, and stability of the system operation.
[0007] To achieve the above-mentioned technical objectives, the present invention is implemented through the following technical solution: a method for online fault diagnosis of gearboxes based on digital twins, comprising the following steps:
[0008] S1: Collect different data during the operation of the gearbox and build a digital twin model of the gearbox;
[0009] S2: Dynamically update the digital twin model of the gearbox, introduce a trust region algorithm to iteratively optimize the model parameters, and use cosine similarity as a similarity metric between the simulated signal and the measured vibration signal of the digital twin model. When the cosine similarity is greater than the preset threshold, the iteration stops and the undetermined parameter values are obtained. The digital twin model is corrected according to the relevant parameters to obtain a high-fidelity model of the gearbox's virtual-real linkage.
[0010] S3: By predefining fault types, a fault module is embedded in the digital twin model of the gearbox to obtain the simulation signal of the gearbox fault state. Fault diagnosis algorithm is used to obtain fault feature information and upload it to the cloud database.
[0011] S4: Collect vibration signals from the physical equipment, match and compare them with simulation data in the twin database, and finally obtain the operating status of the physical equipment to realize online monitoring and fault diagnosis of the physical equipment;
[0012] S5: If a gearbox is detected to be in a faulty state, the equipment maintenance rules are invoked to perform edge calculations on the fault information and adjust the relevant operating parameters of the physical equipment to achieve stable and efficient operation of the gearbox and extend the service life of the equipment.
[0013] Preferably, the data collected in S1 includes static data and dynamic data; the static data includes acquired and definite information such as historical operating data of the gearbox, equipment characteristics, geometric morphology data, and initial material properties; the dynamic data includes transient time-varying information related to the operating state of the gearbox.
[0014] Preferably, the step of constructing the digital twin model of the gearbox in S1 is as follows:
[0015] S1.1: The initial numerical model is constructed by integrating gearbox mechanism information and static data. This process can be described as follows:
[0016] MO = {I 历史数据 I 机理 I 形态 P 装备} (1)
[0017] In the formula, P 装备 Indicates the characteristics of the equipment, I 历史数据 I 机理 and I 形态 These represent historical data, mechanism, and morphological information, respectively. MO represents the initial numerical model constructed by the combined driving force of historical data, mechanism, and morphological information.
[0018] The definition of is as follows:
[0019] P 装备 ={P 多学科 P 多物理 P 多尺度 , ..., P 多参量 P 多源数据 P 概率性} (2)
[0020] In the formula, P 装备 The characteristic parameters representing the gearbox include multidisciplinary, multiphysical, multiscale, multiparameter, multi-source data, and probabilistic characteristics.
[0021] S1.2: Using the measured dynamic data and combining it with the current mechanism information of the gearbox, the geometric shape in the initial numerical model of the gearbox is updated in real time to construct a digital twin model of the gearbox. This process can be described as follows:
[0022]
[0023] In the formula, Indicates the characteristics currently possessed by the equipment, I在线线数 MO represents the current on-site monitoring data, MO represents the model constructed in equation (1), and MOcurr represents the gearbox digital twin model.
[0024] Preferably, the specific steps for iterative optimization of the gearbox digital twin model in S2 are as follows:
[0025] S2.1: Generate simulated vibration signals based on the digital twin model of the gearbox;
[0026] S2.2: Collect measured vibration signals of physical equipment under the same working conditions and at the same time;
[0027] S2.3: Cosine similarity is used as a metric for matching analysis to determine whether the digital twin model of the gearbox and the physical entity have a high degree of consistency. The cosine similarity expression is as follows:
[0028]
[0029] In the formula, Vp and Vs are the vibration responses of length N obtained from physical equipment testing and digital twin model simulation, respectively;
[0030] S2.4: If the similarity between the two is less than the preset threshold, the trust region algorithm is introduced to optimize the key feature parameters of the gearbox digital twin model, iteratively solve the optimal parameters, and correct the twin model until the cosine similarity judgment requirement is met.
[0031] Preferably, in S2.4, the specific steps of the trust region algorithm for optimization are as follows:
[0032] S2.4.1: Initialize all parameters;
[0033] S2.4.2: Set the conditions for stopping iteration, including the stopping criteria for function values, gradient norm, and maximum number of iterations;
[0034] S2.4.3: Solving the subproblem, the trust region subproblem at the Kth iteration can be expressed as:
[0035]
[0036] g k =▽f P(X) (6)
[0037] In the above formula, B k Let g be the Hessian matrix of P(X). k It is the gradient operator of P(X), where P(X) is the objective function;
[0038] S2.4.4: Constructing r k The function is used to evaluate X k +Dk Can it be used as the next iteration point? If r k If the value is close to 1, then q k Within the trust region, it infinitely approximates the objective function, and its calculation formula is:
[0039]
[0040] S2.4.5: Update the trust region radius and construct the trust region radius set:
[0041]
[0042]
[0043] S2.4.6: Output the optimal parameters after the iteration conditions are met.
[0044] Preferably, in step S3, fault feature information is obtained through a convolutional neural network (CNN).
[0045] The beneficial effects of this invention are:
[0046] This invention constructs a gearbox twin model based on digital twin technology, realizing the virtual-real mapping of gearbox equipment. Then, by dynamically correcting the twin model, the high fidelity of the gearbox equipment twin model is ensured. Finally, by matching and comparing real-time twin data with measured signals, online fault diagnosis of the gearbox equipment is achieved. This invention solves the problems of low modeling accuracy, insufficient diagnostic mechanism basis, and lack of real-time collaboration. The constructed gearbox digital twin model has an accuracy of up to 99%, effectively improving the simulation accuracy of the model. This facilitates real-time monitoring of the gearbox equipment's operating status and online fault diagnosis, and has significant implications for engineering applications. Attached Figure Description
[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0048] Figure 1 This is a flowchart of the online fault diagnosis method for gearboxes based on digital twins according to the present invention;
[0049] Figure 2 This is a graph showing the change in wear and tear of the updated model and the initial model, as well as the physical data, over time.
[0050] Figure 3 This is a diagram of the gearbox experimental data acquisition device of the present invention. Detailed Implementation
[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] Example 1
[0053] This embodiment uses a planetary gearbox with a missing sun gear tooth as an example. The equipment parameters are as follows: transmission ratio 6.18, sun gear 1 with 17 teeth, planet gears 3 with 34 teeth, ring gear 88 teeth, motor speed 2000 r / min, load 200 N / m, and sampling frequency 24 kHz. Using the theoretical formula for gearbox meshing frequency, the meshing frequency of the gearbox in this case is calculated as: fm = 474.92 Hz. The order of the planetary gearbox meshing frequency after normalizing the planet carrier frequency is 88.
[0054] S1: Collect different data during the operation of the gearbox and build a digital twin model of the gearbox.
[0055] Data from the planetary gearbox experimental setup was collected using a data acquisition system. The gearbox experimental setup is shown in the attached figure. Figure 3 As shown, the system includes a drive motor, coupling, transparent protective cover, gearbox, magnetic powder brake, and data acquisition device. Complete data information includes both static and dynamic data. Static data includes historical gearbox operating data (stress, strain, speed, fatigue life, etc.), equipment characteristics (component assembly relationships, coupling relationships, etc.), geometric data (tooth width, number of teeth, pressure angle, tooth height coefficient, etc.), and initial material properties (density, shear modulus, elastic modulus, etc.)—all already acquired and definite information. Dynamic data includes transient, time-varying information related to the gearbox's operating state, such as speed, displacement, stress, and strain. This data needs to be monitored and acquired in real time. Preprocessing operations such as data storage, cleaning, transformation, and dimensionality reduction are used to remove irrelevant information with minimal impact on gearbox modeling, ensuring the timeliness and reliability of the dynamic data.
[0056] A digital twin model of a planetary gearbox is constructed using static and dynamic data collected by S1, enabling dual-driven modeling of the gearbox's mechanistic information and data.
[0057] The initial numerical model is constructed by integrating gearbox mechanism information and historical data. This process can be described as follows:
[0058] MO = {I 历史数据 I 机理 I 形态 P装备} (1)
[0059] In the formula, P 装备 Indicates the characteristics of the equipment, I 历史数据 I 机理 and I 形态 These represent historical data, mechanism, and morphological information, respectively. MO represents the initial numerical model constructed by the combined driving force of historical data, mechanism, and morphological information.
[0060] Among them, P 装备 The definition is as follows:
[0061] P 装备 ={P 多学科 P 多物理 P 多尺度 , ..., P 多参量 P 多源数据 P 概率性} (2)
[0062] In the formula, P 装备 These represent the characteristic parameters of the gearbox.
[0063] Using on-site monitoring data and combining it with the current mechanism information of the gearbox, the geometry in the initial numerical model of the gearbox is updated in real time to construct a digital twin model of the gearbox. This process can be described as follows:
[0064]
[0065] In the formula, Indicates the characteristics currently possessed by the equipment, I 在线线数 MO represents the current on-site monitoring data, MO represents the model constructed in equation (1), and MOcurr represents the gearbox digital twin model.
[0066] S2: Using cosine similarity as a metric for matching analysis, the simulation data generated by the digital twin model constructed in S1 is compared with the actual operating data of the physical entity under the same variables. The trust region algorithm is introduced to iteratively optimize the feature parameters of the digital twin model. When cosθ>0.74, the iterative solution is stopped and the optimized parameter value is obtained, thus completing the correction of the high-fidelity model of the gearbox digital twin model.
[0067] The formula for calculating cosine similarity is:
[0068]
[0069] In the formula, Vp and Vs are the vibration responses of length N obtained from physical equipment testing and online digital twin model simulation, respectively.
[0070] like Figure 2As shown, after updating the twin model using the method proposed above, the wear amount of the gear tooth surface is closer to the actual measured wear amount compared with the initial unupdated version, indicating that the updated gearbox twin model has high accuracy.
[0071] S3: By predefining fault types, the fault model is embedded in the digital twin model of the gearbox to obtain the simulation signal of the gearbox under different fault states. The fault simulation signal is normalized and divided into several sub-signals of the same length in the time domain. The kurtosis index of each sub-signal is calculated. The fault feature information is obtained by using a convolutional neural network (CNN) and uploaded to the cloud twin database.
[0072] The formula for calculating kurtosis is as follows:
[0073]
[0074] S4: Real-time acquisition of data from physical entities, processing of measured fault signals in the same way as fault simulation signals, matching and comparing with twin fault data, and finally determining the fault type and severity of the physical equipment, thereby realizing real-time online monitoring and fault diagnosis of physical equipment.
[0075] S5: Based on the diagnostic results obtained in S4, call the equipment maintenance rules to perform edge calculations on the fault information, adjust the relevant operating parameters of the planetary gearbox, achieve stable and efficient operation of the gearbox, and extend the service life of the equipment.
[0076] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not describe all details exhaustively, nor do they limit the invention to the specific implementations described.
Claims
1. A method for online fault diagnosis of gearboxes based on digital twins, characterized in that, Includes the following steps: S1: Collect different data during the operation of the gearbox and build a digital twin model of the gearbox; S2: Dynamically update the digital twin model of the gearbox, introduce a trust region algorithm to iteratively optimize the model parameters, and use cosine similarity as a similarity metric between the simulated signal and the measured vibration signal of the digital twin model. When the cosine similarity is greater than the preset threshold, the iteration stops and the undetermined parameter values are obtained. The digital twin model is corrected according to the relevant parameters to obtain a high-fidelity model of the gearbox's virtual-real linkage. S3: By predefining fault types, a fault module is embedded in the digital twin model of the gearbox to obtain the simulation signal of the gearbox fault state. Fault diagnosis algorithm is used to obtain fault feature information and upload it to the cloud database. S4: Collect vibration signals from the physical equipment, match and compare them with simulation data in the twin database, and finally obtain the operating status of the physical equipment to realize online monitoring and fault diagnosis of the physical equipment; S5: If a gearbox is detected to be in a faulty state, the equipment maintenance rules are invoked to perform edge calculations on the fault information and adjust the relevant operating parameters of the physical equipment to achieve stable and efficient operation of the gearbox and extend the service life of the equipment. The specific steps for iterative optimization of the gearbox digital twin model in S2 are as follows: S2.1: Generate simulated vibration signals based on the digital twin model of the gearbox; S2.2: Collect measured vibration signals of physical equipment under the same working conditions and at the same time; S2.3: Use cosine similarity as a metric for matching analysis to determine whether there is a high degree of consistency between the digital twin model of the gearbox and the physical entity; the cosine similarity expression is as follows: （4） In the formula, and The vibration response of length N is obtained from physical equipment testing and digital twin model simulation, respectively. S2.4: If the similarity between the two is less than the preset threshold, the trust region algorithm is introduced to optimize the key feature parameters of the gearbox digital twin model, iteratively solve the optimal parameters, and correct the twin model until the cosine similarity judgment requirement is met. In S2.4, the specific steps of the trust region algorithm for optimization are as follows: S2.4.1: Initialize all parameters; S2.4.2: Set the conditions for stopping iteration, including the stopping criteria for function values, gradient norm, and maximum number of iterations; S2.4.3: Solving the subproblem, the trust region subproblem at the Kth iteration is expressed as: （5） （6） In the above formula, yes The Hessian matrix, yes gradient operator, It is the objective function; S2.4.4: Construction The function is used for evaluation Can it be used as the next iteration point? If If the value is close to 1, then Within the trust region, it infinitely approximates the objective function, and its calculation formula is: （7） S2.4.5: Update the trust region radius and construct the trust region radius set: （8） （9） S2.4.6: Output the optimal parameters after the iteration conditions are met.
2. The online fault diagnosis method for gearboxes based on digital twins according to claim 1, characterized in that, The data collected in S1 includes static data and dynamic data; static data includes historical operating data of the gearbox, equipment characteristics, geometric data, and known information on the initial properties of materials; dynamic data includes transient time-varying information related to the operating state of the gearbox.
3. The online fault diagnosis method for gearboxes based on digital twins according to claim 2, characterized in that, The steps for constructing the digital twin model of the gearbox in S1 are as follows: S1.1: Integrating gearbox mechanism information and static data to complete the initial numerical model construction, this process is described as follows: （1） In the formula, Indicates the characteristics of the equipment. , and These represent historical data, mechanism, and morphological information, respectively. This represents the initial numerical model constructed by the combined influence of historical data, mechanistic information, and morphological information. The definition of is as follows: （2） In the formula, The characteristic parameters representing the gearbox include multidisciplinary, multiphysical, multiscale, multiparameter, multi-source data, and probabilistic characteristic parameters; S1.2: Using the measured dynamic data and the current mechanism information of the gearbox, the geometric shape in the initial numerical model of the gearbox is updated in real time to construct a digital twin model of the gearbox. This process is described as follows: （3） In the formula, This indicates the characteristics that the equipment currently possesses. This indicates the current on-site monitoring data. The model constructed in expression (1) This represents the digital twin model of the gearbox.
4. The online fault diagnosis method for gearboxes based on digital twins according to claim 1, characterized in that, In S3, fault feature information is obtained through a convolutional neural network (CNN).