A Fusion Fault Diagnosis Method Based on Digital Twin

Abstract

The present invention discloses a fusion fault diagnosis method based on digital twins, comprising the following steps: (1) establishing a digital twin model of equipment; (2) updating the digital twin model of equipment; (3) running the digital twin model of equipment to simulate the fault operation state of equipment; (4) establishing a random forest fault diagnosis model to output equipment fault classification and fault prediction results; (5) preparing a verification data set, recording and evaluating the performance and indicators of the random forest fault diagnosis model; (6) evaluating the model performance indicators; (7) using a fusion fault diagnosis algorithm, using the optimized fault diagnosis model to correct the output of the digital twin model, and evaluating the model performance indicators; (8) judging whether a drone will fail based on the output results; the present invention can achieve the purpose of real-time detection of the equipment operation state and fault diagnosis of the equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of equipment fault diagnosis, and specifically relates to a fusion fault diagnosis method based on digital twins. Background Art

[0002] Fault diagnosis monitors the condition of components in industrial systems by analyzing information and data from sensors, enabling online monitoring and operational maintenance management of equipment. Currently, fault diagnosis technology has found significant applications in high-end robotics, aerospace, industrial manufacturing, and various other fields. Fault diagnosis techniques are broadly categorized into mathematical model-based, empirical knowledge-based, and data-driven approaches. As equipment structures become increasingly complex and diverse, single fault diagnosis methods face limitations in terms of model consistency, algorithm applicability, and accuracy, failing to meet the high-precision and high-reliability requirements for fault diagnosis. Therefore, relying solely on a single digital twin model is insufficient to accurately describe the structure and mechanisms of equipment. Deeper algorithm integration and optimization are necessary to better understand equipment mechanisms and data characteristics, and to improve the accuracy of fault diagnosis.

[0003] Currently, the cross-model diesel engine combustion chamber fault diagnosis method and system application number: CN202211471504.2, does not establish a three-dimensional model of the equipment, cannot intuitively and quickly repair the equipment, and the accuracy of the diagnostic results is limited by the existing data set, and the diagnostic accuracy for mixed faults is low.

[0004] Existing algorithms have problems such as low modeling accuracy, poor model consistency, low data utilization, and insufficient applicability of fault diagnosis algorithms. Summary of the Invention

[0005] In order to overcome the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a fusion fault diagnosis method based on digital twins, which has the characteristics of high synchronization and high consistency between the model and the equipment, high accuracy of the diagnosis results, high effective utilization rate of data, and strong algorithm adaptability. It can achieve the purpose of real-time detection of the equipment operating status and diagnose the equipment faults.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A fusion fault diagnosis method based on digital twins includes the following steps:

[0008] (1) Establish a digital twin model of the equipment, which takes the equipment linearization model as input, guides the operation of the equipment three-dimensional model, and outputs the relevant parameters of the actual operation fault state of the equipment;

[0009] (2) Update the digital twin model of the device in step (1) based on the state parameters of the actual device;

[0010] (3) Run the digital twin model of the equipment to simulate the equipment fault operation state to obtain sufficient fault simulation data sets and prepare the historical fault data sets generated by actual operation;

[0011] (4) Establish a random forest fault diagnosis model, output equipment fault classification and fault prediction results, use simulation data sets to train the random forest fault diagnosis model, and use historical fault data sets to update the random forest fault diagnosis model;

[0012] (5) Prepare validation datasets, record and evaluate the performance and metrics of the random forest fault diagnosis model;

[0013] (6) Use the particle swarm algorithm to optimize the random forest fault diagnosis model, use the validation data set to train the random forest fault diagnosis model optimized by the particle swarm algorithm, and evaluate the model performance indicators;

[0014] (7) Using a fusion fault diagnosis algorithm, the optimized fault diagnosis model is used to correct the output of the digital twin model and evaluate the model performance indicators;

[0015] (8) Determine whether the UAV will fail based on the output results of the fusion fault diagnosis algorithm in step (7). If a failure occurs, output the possible failure and output warning information to notify the staff to perform equipment maintenance. Otherwise, continue to update the data in the twin.

[0016] In the step (1), the concept of digital twin is used to establish a digital twin model of the device. The Unity3D software is used to accurately describe the appearance and internal structure of the device, and a three-dimensional simulation model of the device is established. The linearization model equation and the fault model equation of the device are derived through the device dynamics and kinematics equations. The operation status of the three-dimensional simulation model is guided by the mathematical model to establish a digital twin model of the device. In combination with the operation status data transmitted by the real-time communication of the physical device, the relevant parameters of the three-dimensional model of the device are updated in real time, the operation status of the device is grasped in real time, and the linearization model expression of the device is established as follows:

[0017]

[0018] y(t)=Cx(t)

[0019] x(t) is the device state parameter, u(t) is the system input parameter, y(t) is the system output parameter, A is the system state parameter matrix, B is the system input matrix, and C is the system output matrix. The fault model equation of the device is as follows:

[0020]

[0021] y(t)=Cx(t)

[0022] Where x(t) is the equipment fault state parameter, u(t) is the fault system input parameter, y(t) is the fault system output parameter, A is the fault system state parameter matrix, B is the fault system input matrix, C is the fault system output matrix, and f a ∈R m Fault vector representing equipment failure, n out (t) is the fault function of the device, and its specific form varies depending on the fault type of different devices.

[0023] In step (3), the fault injection into the device includes the following steps:

[0024] (1) Establish equipment failure model equations. According to the internal mechanism structure and failure type analysis of the equipment, the equipment failure model equations simulate different types of equipment failures. The failure functions of different equipment vary according to different equipment.

[0025] (2) Simulate the output of the fault model equation under different fault types, and the output value is the relevant fault state value of the equipment parameters;

[0026] (3) The fault state value is used to guide the simulation operation of the three-dimensional simulation model to achieve the purpose of fault injection, thereby obtaining a large number of simulated fault data sets.

[0027] In the step (4), the simulated fault data set generated in the step (3) is used to train the random forest fault diagnosis model. The random forest fault diagnosis model can output the fault type of the device and the expected possible faults based on the relevant fault state values in the data set.

[0028] The performance evaluation indicators in step (5) are root mean square error (RMSE), mean absolute error (MAE), mean error deviation (MBE), R 2 The score is calculated as follows:

[0029]

[0030]

[0031]

[0032]

[0033] Where n is the sample size, y i are sample parameters, and all error parameters are the errors between the predicted values and the original values.

[0034] RMSE, MAE, and MBE are important indicators for measuring the accuracy of a regression model. They reflect the difference between the actual value and the expected value. In multiple modeling processes, each simulation result will produce a set of error values. The smaller the error value, the higher the accuracy of the regression model.

[0035] Coefficient of determination R 2 The value is an indicator to measure the degree of fit of the regression model. The value is in the interval [0,1]. 2 The closer it is to 1, the stronger the explanatory power of the variable on the output, that is, the better the data fit; R 2 The closer it is to 0, the worse the model fit is.

[0036] The process of optimizing the fault diagnosis model using the particle swarm algorithm in step (6) includes the following steps:

[0037] (1) The depth D, number of trees T, and feature subset size N of the decision tree in the random forest algorithm are used as particles in the particle swarm optimization algorithm. The update equations for the position and speed of the particle swarm optimization algorithm are as follows:

[0038]

[0039]

[0040] (2) Select RMSE root mean square error as the fitness function. The calculation formula of RMSE is as follows:

[0041]

[0042] (3) According to the fitness function, calculate the fitness value of the decision tree depth D, the number of trees T, and the feature subset size N;

[0043] (4) Compare the current fitness value of each particle with its individual extreme value. If the current fitness value is better than the individual extreme value, the individual extreme value will be updated;

[0044] (5) Compare the current fitness value of each particle with its group extreme value. If the current fitness value is better than the group extreme value, the group extreme value will be updated;

[0045] (6) If the requirements for equipment fault classification and prediction of possible future equipment faults can be met, and the fault diagnosis performance requirements can be met, the optimized model is analyzed; if the performance requirements cannot be met, the model is adjusted and improved until the performance requirements are met.

[0046] The step (7) uses a fusion fault diagnosis algorithm, including the following steps:

[0047] (1) By establishing a virtual simulation model of the equipment digital twin, its expression is as follows:

[0048] MDT=(PE,VE,Ss,DD,CN)

[0049] MDT stands for device digital twin, PE represents physical entity, VE represents virtual entity, Ss represents service, DD represents twin data, and CN represents the connection between components. The device digital twin communicates with the device entity to update parameters. Running the model can obtain sufficient simulation data, and the model simulation output data is used as system state input 1;

[0050] (2) By establishing a linear model of the device, the internal state of the device system is calculated and used as the input of the virtual sensor signal 2;

[0051] (3) Using the fault diagnosis model optimized by the particle swarm algorithm as the correction model;

[0052] (4) The correction value obtained by the optimized model is used to correct the system state input 1 and the virtual sensor signal input 2, and a more accurate diagnosis result is obtained.

[0053] Beneficial effects of the present invention:

[0054] The present invention achieves accurate, virtual-real synchronous digital representation of physical entities by establishing a digital twin model of the equipment, thereby improving the timeliness of the real-time status detection and fault diagnosis results of the digital twin model. The fault diagnosis model is trained using a simulation data set generated by running the model simulation, and the pre-trained model obtained by updating the real UAV fault data set is used. The random forest model optimized by the particle swarm algorithm is then used as a correction model to correct the output results driven by the digital twin model simulation according to the observed values. The digital twin model is combined with data-driven technology to ensure the operation speed while improving the calculation accuracy of the random forest. At the same time, the dependence of the digital twin model on sample data is reduced, and the generalization ability of the digital twin model is enhanced.

[0055] This paper uses a fusion fault diagnosis method to explore effective ways to integrate the digital twin model with the data-driven algorithm. It then uses correction values obtained from the fault diagnosis model optimized by a particle swarm algorithm to calibrate the digital twin model's simulation output, improving the accuracy and applicability of the fault diagnosis algorithm. While overcoming the shortcomings of traditional methods, it also enhances the accuracy of complex equipment fault diagnosis and the feasibility of full lifecycle management and control of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of a fusion fault diagnosis method based on digital twins in the present invention.

[0057] Figure 2 This is a schematic diagram of the fault diagnosis process based on particle swarm optimization.

[0058] Figure 3 This is a flowchart of the training process of the fusion fault diagnosis model.

[0059] Figure 4 This is the algorithm block diagram of the fusion fault diagnosis method based on digital twins. DETAILED DESCRIPTION

[0060] The present invention will be described in further detail below with reference to the accompanying drawings.

[0061] like Figure 1-Figure 4 In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings. The present invention proposes a fusion fault diagnosis method based on digital twins, which includes the following steps:

[0062] S1: Establish a digital twin model of the equipment. Digital twins establish virtual twins and use computer communication technology, multi-source information fusion perception technology, and 3D modeling technology to map physical entities and describe the real-time status of the equipment, thereby enabling equipment status monitoring and fault diagnosis. By promoting the interactive fusion and optimization of virtual and real space data information, intelligent operation and maintenance management of the equipment is achieved.

[0063] S2: Update the digital twin model of the equipment. The real-time requirements of digital twin technology are not only reflected in the real-time communication of data, but also in the synchronization of the real-time status of the physical device and the device twin model, forming a high degree of mapping between the physical object and the digital virtual body, thereby achieving the purpose of real-time status monitoring of the equipment.

[0064] S3: Run the model simulation to inject faults to obtain sufficient fault simulation data sets and prepare the historical fault data sets generated by actual operation;

[0065] S4: Use the simulation data set to train the fault diagnosis model and use the historical fault data set to update the fault diagnosis model;

[0066] S5: Prepare validation datasets, record and evaluate the performance and metrics of the pre-trained fault diagnosis model;

[0067] S6: Use the validation data set to train the fault diagnosis model optimized by the particle swarm algorithm and evaluate the model performance indicators;

[0068] S7: Adopting the fusion fault diagnosis algorithm, using the optimized fault diagnosis model to correct the output of the digital twin model and evaluate the model performance indicators;

[0069] S8: Determine whether the drone will fail based on the output results of the fusion fault diagnosis algorithm. If a failure occurs, output the possible failure and output warning information to notify the staff to perform equipment maintenance. Otherwise, continue to update the data in the twin.

[0070] Furthermore, the process of optimizing the fault diagnosis model using the particle swarm optimization algorithm includes the following steps:

[0071] S61: Using the depth D, number T of decision trees and size N of feature subset as particles of particle swarm optimization algorithm;

[0072] S62: Select RMSE root mean square error as the fitness function;

[0073] S63: Calculate the fitness value of the particle according to the fitness function;

[0074] S64: Compare the current fitness value of each particle with its individual extreme value. If the current fitness value is better than the individual extreme value, update the individual extreme value.

[0075] S65: Compare the current fitness value of each particle with its group extreme value. If the current fitness value is better than the group extreme value, the group extreme value will be updated.

[0076] S66: If the fault diagnosis performance requirements are met, the optimal particle parameters are output and the diagnosis model is trained; if the performance requirements are not met, the model is adjusted and improved until the performance requirements are met;

[0077] Furthermore, a fusion fault diagnosis algorithm is used, which includes the following steps:

[0078] S71: By establishing a virtual simulation model of the device digital twin and communicating with the device entity to update parameters, sufficient simulation data can be obtained by running the model, and the model simulation output data is used as system state input 1;

[0079] S72: Calculate the internal state of the device system by establishing a device linearization model, and use it as input 2 of the virtual sensor signal;

[0080] S73: Using the fault diagnosis model optimized by the particle swarm algorithm as a correction model;

[0081] S74: The system state input 1 and the virtual sensor signal input 2 are corrected by the correction value obtained through the optimized model, and a more accurate diagnosis result is obtained.

[0082] The present invention is further described below in conjunction with relevant background technologies and implementation steps:

[0083] In step S1 of the present invention, Unity 3D software is used to accurately describe the device's appearance and internal structure, creating a three-dimensional model of the device. The device's dynamics and kinematics equations are used to derive a linearized model and a fault model. The mathematical model then guides the three-dimensional model to create a digital twin.

[0084] In step S2, real-time operational data is used to support the continuous iterative update of the digital twin model to achieve a complete mapping of the physical entity.

[0085] In step S3, a fault simulation module is established, and sufficient fault simulation data sets can be obtained by running the module.

[0086] In step S4, while using the simulation data set to train the random forest fault diagnosis module, the historical fault data set generated by actual flight is introduced to update the random forest fault diagnosis model, and the performance evaluation indicators are recorded, including the root mean square error RMSE, mean absolute error MAE, mean error deviation MBE, and R2 value, which will be used as the standard for subsequent performance evaluation.

[0087] At present, equipment fault diagnosis methods based on digital twin technology are the main direction of research and application in the field of fault diagnosis. The reasonable and effective introduction of digital twin technology can optimize the structural mapping of physical entities in virtual space, thereby achieving accurate fault diagnosis and effectively promoting the rapid development of research in the field of fault diagnosis.

[0088] The fusion fault diagnosis method based on digital twins will cause error accumulation in model simulation, resulting in a significant difference between the simulation results and the actual output. The random forest model optimized by the particle swarm algorithm is introduced to provide feedback and correction on the output results of the digital twin model.

[0089] In order to solve the above problems, a fusion fault diagnosis algorithm is proposed.

[0090] (1) Optimize the fault diagnosis algorithm

[0091] In step S6, to improve the performance of the fault diagnosis model, a particle swarm optimization algorithm is introduced. The depth, number of trees, and feature subset size of the decision tree of the random forest fault diagnosis model are used as particles in the particle swarm optimization algorithm. The root mean square error (RMSE) is selected as the fitness function. The smaller the root mean square error, the more accurate the diagnosis result obtained by the model. The particle swarm optimization algorithm is used to iteratively optimize the above parameters and select appropriate parameters to optimize the random forest fault diagnosis model.

[0092] Complex equipment often has harsh working environments, complex operating conditions, numerous system parameters, nonlinear relationships among various parameters, weak correlations, and difficulty in constructing mathematical models. Therefore, using any one method cannot fully meet the needs of accurate fault diagnosis. It is necessary to integrate and optimize the algorithm at a deeper level in order to better understand the equipment mechanism and data characteristics and improve the accuracy of fault diagnosis.

[0093] (2) Combining digital twin technology with fault diagnosis algorithms

[0094] In step S7, Figure 4 As shown in Figure 1, by establishing a digital twin virtual simulation model and communicating with the entity to update parameters, sufficient simulation data can be obtained by running the model, and the model simulation output data is used as the system state input 1. By establishing a device linearization model, the internal state of the device system is calculated and used as the input of the virtual sensor signal 2.

[0095] Using the correction values obtained from the optimized random forest fault diagnosis model using a fusion algorithm, we corrected Inputs 1 and 2, achieving a relatively accurate diagnosis. Based on the output of the fusion fault diagnosis algorithm, we determine whether the drone will malfunction. If a malfunction occurs, we output the potential fault information and a warning message to notify personnel to perform equipment maintenance. Otherwise, we continue to update the data in the twin.

[0096] In summary, the present invention mainly solves two technical problems: the first is the poor consistency between the digital twin model and the physical device and the poor timeliness of fault diagnosis; the second is the low accuracy of equipment fault diagnosis and poor algorithm applicability;

[0097] Table 1: Evaluation indicators of the digital twin-based fusion fault diagnosis method on the validation set

[0098]

[0099] In order to verify the performance of the fusion fault diagnosis method based on digital twins, RMSE, MAE, MBE, R 2 Compared to other methods such as SVM, RF, and PSO-RF, the evaluation metric shows a large error between the predicted and actual values. However, the fusion fault diagnosis method based on digital twins achieves a smaller error between the predicted and actual values, effectively improving the accuracy of the fault prediction model. This fusion fault diagnosis method based on digital twins has a lower prediction error and can address the issues of lack of fusion optimization between models and data, as well as the low accuracy of data-driven algorithms, resulting in more accurate and reliable prediction results.

[0100] The above disclosure is only an example of the present invention and certainly cannot be used to limit the scope of the present invention. Ordinary technicians in this field can understand the process of implementing the above example and make equivalent changes in accordance with the claims of the present invention.

Claims

1. A fusion fault diagnosis method based on digital twins, characterized in that: The following steps are included: (1) Establish a digital twin model of the equipment, which takes the equipment linearization model as input, guides the operation of the equipment three-dimensional model, and outputs the relevant parameters of the actual operation fault state of the equipment; (2) Update the digital twin model of the device in step (1) based on the state parameters of the actual device; (3) Run the digital twin model of the equipment to simulate the equipment fault operation state to obtain sufficient fault simulation data sets and prepare historical fault data sets from actual operation; (4) Establish a random forest fault diagnosis model, output equipment fault classification and fault prediction results, use simulation data sets to train the random forest fault diagnosis model, and use historical fault data sets to update the random forest fault diagnosis model; (5) Prepare validation datasets, record and evaluate the performance and metrics of the random forest fault diagnosis model; (6) Use the particle swarm algorithm to optimize the random forest fault diagnosis model, use the validation data set to train the random forest fault diagnosis model optimized by the particle swarm algorithm, and evaluate the model performance indicators; (7) Using a fusion fault diagnosis algorithm, the optimized fault diagnosis model is used to correct the output of the digital twin model and evaluate the model performance indicators; (8) Determine whether the UAV will fail based on the output of the fusion fault diagnosis algorithm in step (7). If a failure occurs, output the possible failure and output warning information to notify the staff to perform equipment maintenance. Otherwise, continue to update the data in the twin. The step (7) uses a fusion fault diagnosis algorithm, including the following steps: (1) By establishing a virtual simulation model of the equipment digital twin, its expression is as follows: MDT=(PE,VE,Ss,DD,CN) MDT is the device digital twin, PE represents the physical entity, VE represents the virtual entity, Ss represents the service, DD represents the twin data, and CN represents the connection between the components. The device digital twin communicates with the device entity to update parameters. The model is run to obtain sufficient simulation data, and the model simulation output data is used as the system state input 1; (2) By establishing a linear model of the device, the internal state of the device system is calculated and used as the input of the virtual sensor signal 2; (3) Using the fault diagnosis model optimized by the particle swarm algorithm as the correction model; (4) The correction value obtained by the optimized model is used to correct the system state input 1 and the virtual sensor signal input 2, and a more accurate diagnosis result is obtained.

2. A fusion fault diagnosis method based on digital twins according to claim 1, characterized in that: In the step (1), the concept of digital twin is used to establish a digital twin model of the equipment. The appearance and internal structure of the equipment are accurately described through Unity 3D software, and a three-dimensional simulation model of the equipment is established. The linearization model equation and the fault model equation of the equipment are derived through the equipment dynamics and kinematic equations. The operation status of the three-dimensional simulation model is guided by the mathematical model to establish a digital twin model of the equipment. In combination with the operation status data transmitted by real-time communication of the physical equipment, the relevant parameters of the three-dimensional model of the equipment are updated in real time, and the operation status of the equipment is grasped in real time.

3. A fusion fault diagnosis method based on digital twins according to claim 2, characterized in that: The linear model expression of the equipment is as follows: y(t)=Cx(t) x(t) is the device state parameter, u(t) is the system input parameter, y(t) is the system output parameter, A is the system state parameter matrix, B is the system input matrix, and C is the system output matrix. The fault model equation of the device is as follows: y(t)=Cx(t) Where x(t) is the equipment fault state parameter, u(t) is the fault system input parameter, y(t) is the fault system output parameter, A is the fault system state parameter matrix, B is the fault system input matrix, C is the fault system output matrix, and f a ∈R m Fault vector representing equipment failure, n out (t) is the fault function of the device, and its specific form varies depending on the fault type of different devices.

4. A fusion fault diagnosis method based on digital twins according to claim 1, characterized in that: In step (3), the fault injection into the device includes the following steps: (1) Establish equipment failure model equations. According to the internal mechanism structure and failure type analysis of the equipment, the equipment failure model equations simulate different types of equipment failures. The failure functions of different equipment vary according to different equipment. (2) Simulate the output of the fault model equation under different fault types, and the output value is the relevant fault state value of the equipment parameters; (3) The fault state value is used to guide the simulation operation of the three-dimensional simulation model to achieve the purpose of fault injection, thereby obtaining a large number of simulated fault data sets.

5. A fusion fault diagnosis method based on digital twins according to claim 4, characterized in that: In the step (4), the simulated fault data set generated in the step (3) is used to train the random forest fault diagnosis model. The random forest fault diagnosis model can output the fault type of the device and the expected possible faults based on the relevant fault state values in the data set.

6. A fusion fault diagnosis method based on digital twins according to claim 1, characterized in that: The performance evaluation indicators in step (5) are root mean square error (RMSE), mean absolute error (MAE), mean error deviation (MBE), R 2 The score is calculated as follows: Where n is the sample size, y i are sample parameters, and all error parameters are the errors between the predicted values and the original values.

7. A fusion fault diagnosis method based on digital twins according to claim 6, characterized in that: In multiple modeling processes, each simulation result will produce a set of error values. The smaller the error value, the higher the accuracy of the regression model.

8. The digital twin-based fusion fault diagnosis method according to claim 6 is characterized in that: Coefficient of determination R 2 The value is an indicator to measure the degree of fit of the regression model. The value is in the interval [0,1]. 2 The closer it is to 1, the stronger the explanatory power of the variable on the output, that is, the better the data fit; R 2 The closer it is to 0, the worse the model fit is.

9. The digital twin-based fusion fault diagnosis method according to claim 1 is characterized in that: The process of optimizing the fault diagnosis model using the particle swarm algorithm in step (6) includes the following steps: (1) The depth D, number of trees T, and feature subset size N of the decision tree in the random forest algorithm are used as particles in the particle swarm optimization algorithm. The update equations for the position and speed of the particle swarm optimization algorithm are as follows: (2) Select RMSE root mean square error as the fitness function. The calculation formula of RMSE is as follows: (3) According to the fitness function, calculate the fitness value of the decision tree depth D, the number of trees T, and the feature subset size N; (4) Compare the current fitness value of each particle with its individual extreme value. If the current fitness value is better than the individual extreme value, the individual extreme value will be updated; (5) Compare the current fitness value of each particle with its group extreme value. If the current fitness value is better than the group extreme value, the group extreme value will be updated; (6) If the requirements for equipment fault classification and prediction of possible future equipment faults can be met, and the fault diagnosis performance requirements can be met, the optimized model is analyzed; if the performance requirements cannot be met, the model is adjusted and improved until the performance requirements are met.

Images