Self-awareness enhanced drone digital twin modular simulation method

Abstract

This invention proposes a modular simulation method for UAV digital twins aimed at enhancing self-awareness. The method includes: constructing a modular simulation library based on a reduced-order model; for each geometric module, employing a static condensed reduction primitive method to calculate multiple simulation models under different parameter states, thus building the modular simulation library; classifying and identifying the state of the physical machine based on dynamic sensors; and using a trained optimal decision tree to estimate the most matching simulation model in the module simulation library using observation data from the physical assets, which serves as the state of that module in the digital twin at that moment. The method provided by this invention, based on modularization of states within the digital twin, significantly improves simulation speed, thereby greatly enhancing the self-awareness of the UAV.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) twin simulation, and in particular to a modular simulation method for UAV digital twins with enhanced self-awareness. Background Technology

[0002] As a high-tech industry, unmanned aerial vehicles (UAVs) possess distinct characteristics such as high technological level, capital intensity, high product added value, and strong industrial radiation and driving capacity, thus holding enormous development potential. With the significant increase in the complexity and sophistication of UAVs, the requirements and difficulties for their testing, training, and maintenance have also increased, leading to increased testing costs and extended testing time. Furthermore, due to various limitations of physical testing, some testing projects may be impossible to conduct. Digital twins offer a possible solution to these problems. As UAVs become increasingly autonomous, their self-awareness—the ability to sense changes in their internal state and make corresponding adjustments—is also growing. Figure 1 Therefore, it is becoming increasingly important.

[0003] Self-aware drones can constantly perceive their own structural data, estimate their structural state, predict combat capabilities, and dynamically plan missions, significantly enhancing their survivability and effectiveness in harsh environments. This self-awareness is also crucial for other digital twin applications. For example, in cities, this self-awareness can aid in daily traffic management and emergency crowd evacuation. In short, the significance of digital twins lies not merely in the simple replication of the real world, but more importantly in simulating and predicting it, enabling proactive decision-making and responses to potential emergencies. Real-time simulation and prediction of drone attributes such as stress are prerequisites for achieving this self-awareness. However, digital twin systems are typically complex mega-systems, characterized by multiple elements, multiple domains, uncertainty, and diverse model granularities. This complexity poses significant challenges to the efficiency and accuracy of simulations. Furthermore, due to the complex analytical tasks involving internal and external environments, the decision-making process is also highly complex, leading to uncontrollable and uncertain behavior that poses a significant challenge to reliability. Summary of the Invention

[0004] The main objective of this invention is to overcome the aforementioned deficiencies in the prior art and propose a modular simulation method for UAV digital twins aimed at enhancing self-awareness. Based on the modularization of the state of the digital twin, the simulation speed is greatly improved, thereby significantly enhancing the self-awareness of the UAV.

[0005] The present invention adopts the following technical solution:

[0006] Modular simulation methods for self-awareness-enhanced drone digital twins include:

[0007] The module simulation library is constructed based on the reduced-order model. For each geometric module, the static condensed reduction primitive method is proposed to calculate multiple simulation models under different parameter states and construct the module simulation library.

[0008] Based on the classification and identification of the physical machine's state using dynamic sensors, and using the trained optimal decision tree, the most matching simulation model in the module simulation library is estimated through the observation data on the physical asset, which serves as the state of that module in the digital twin at that moment.

[0009] Specifically, the construction of the modular simulation library based on the reduced-order model involves using the static condensed reduction primitive method for each geometric module to calculate multiple simulation models under different parameter states. The module simulation library also includes:

[0010] For each 3D module asset A i Associate it with a partial differential equation and specify the corresponding external boundary conditions:

[0011]

[0012] in, These are the parameters for the assembled complete unmanned system. For each module, the parameter i is the module number in the SCRBE model; a(u, V; μ) and f(V; μ) are bilinear and linear, respectively, and X(μ) is the function space; finite element approximation is performed on the equation to find a set of solutions u h (μ)=X h (μ) makes:

[0013]

[0014] X h (μ)∈X(μ) is the system-level finite element approximation, corresponding to the overall mesh of the unmanned system; let N FE =dim(X h (μ) The degrees of freedom approximated by the finite element method: Then, the above formula is discretized into a linear system:

[0015] K(μ)U(μ)=F(μ).

[0016] Specifically, based on the classification and identification of the physical machine's state using dynamic sensors, and utilizing a trained optimal decision tree, the most matching simulation model in the module simulation library is estimated using observation data from the physical assets. This model serves as the state of the module in the digital twin at that moment. This also includes:

[0017] For each 3D module asset A i Each simulation model S associated with this module in the module simulation library ij Corresponding to 3D module asset Ai A state;

[0018] Train the optimal decision tree T: X→M, where M represents the set of simulation models associated with the digital twin components; through observation data on physical assets. Estimate the best-fitting simulation model, which serves as the state d of the module in the digital twin at time t. i t :

[0019]

[0020] Specifically, training the optimal decision tree includes:

[0021] According to each simulation model M t ∈M, calculate the predicted value x t And by comparing it with To measure the difference of M t The degree of approximation to the actual physical state; sampling the forward mapping F: M→X to obtain a series of (x) values ​​used for training. t M t The noise impact of the observed sensor data is represented by a Gaussian model V. The positive mapping function in the presence of noise is:

[0022] Another embodiment of the present invention provides a modular simulation system for a drone digital twin aimed at enhancing self-awareness, comprising:

[0023] Building Unit: Based on the reduced-order model, a modular simulation library is constructed. For each geometric module, the static condensed reduction primitive method is to be used to calculate multiple simulation models under different parameter states, and to construct the modular simulation library.

[0024] Simulation Unit: Based on the dynamic sensors, the state of the physical machine is classified and identified. Using the trained optimal decision tree, the simulation model that best matches the simulation model in the module simulation library is estimated through the observation data on the physical asset, which serves as the state of the module at that moment in the digital twin.

[0025] Another embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described modular simulation method for a self-awareness-enhanced unmanned aerial vehicle digital twin.

[0026] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described modular simulation method for a self-awareness-enhanced unmanned aerial vehicle digital twin.

[0027] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0028] This invention provides a modular simulation method for UAV digital twins aimed at enhancing self-awareness. The method includes: constructing a modular simulation library based on a reduced-order model; for each geometric module, employing a static condensed reduction primitive method to calculate multiple simulation models under different parameter states, thus constructing the modular simulation library; classifying and identifying the state of the physical machine based on dynamic sensors; and using a trained optimal decision tree to estimate the most matching simulation model in the module simulation library using observation data from the physical assets, which serves as the state of that module in the digital twin at that moment. The method provided by this invention, based on modularization of states within the digital twin, significantly improves simulation speed, thereby greatly enhancing the self-awareness of the UAV. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the self-awareness of a drone in the prior art provided by the present invention;

[0030] Figure 2 These are module simulation model diagrams under different parameters provided in the embodiments of the present invention, wherein Figure (a) is a module simulation model diagram and Figure (b) is another module simulation model diagram;

[0031] Figure 3 Figure (a) is a schematic diagram of the simulation module dynamic adaptation of an embodiment of the present invention, wherein Figure (b) is a schematic diagram of selecting the most matching physical simulation model; and Figure (c) is a schematic diagram of selecting a model based on sensor data.

[0032] Figure 4 This invention provides an architecture diagram of a cross-user behavior recognition transfer learning system.

[0033] Figure 5 A schematic diagram of an electronic device provided in an embodiment of the present invention;

[0034] Figure 6 This is a schematic diagram of an embodiment of a computer-readable storage medium provided in this invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0036] Digital twins. Despite their relatively recent emergence, digital twins have garnered significant attention from both industry and academia, finding widespread application in various fields such as equipment manufacturing and smart cities. In equipment manufacturing, digital twins are used to optimize all stages of the product manufacturing process and lifecycle. In China, institutions such as the China Electronics Technology Group Corporation, the Academy of Military Sciences, the North China Institute of Computing Technology, the China Aero Engine Research Institute, the China Aero Engine Power Research Institute, the Beijing Simulation Center, and the Beijing Institute of Electronic Engineering, among others, have conducted extensive research on this topic.

[0037] Thanks to extensive research in academia and sustained focused investment from industry, digital twin technology is developing rapidly at a remarkable pace. However, existing research still has some shortcomings or areas that have not yet been addressed: its modeling process still requires significant human resources; due to the complexity of the system, the efficiency of simulation is far from meeting the real-time requirements of self-awareness enhancement; and the reliability of the decision-making model, which is the core of self-awareness, and the simulation process it relies on, still lacks effective analytical and evaluation methods.

[0038] Modular modeling. Developing user-friendly and powerful 3D modeling software has always been a key issue and research hotspot in computer-aided design and computer graphics. Most commercial software, such as ProE and 3DS MAX, adopts a modeling method based on low-level geometric operations. Although powerful, it is very complex, requiring users to undergo a lengthy training process to create 3D models freely, and the modeling process is also quite cumbersome. After self-taught researchers first proposed a new framework for modular 3D modeling, this method of extracting parts from existing 3D shapes and combining them to construct new shapes has attracted considerable attention from researchers due to its ability to balance ease of use and functionality. Much research has been conducted from different perspectives, such as the natural integration of parts, module recommendation, and model set evolution.

[0039] While this modular 3D modeling approach based on reuse significantly lowers the barrier to entry for 3D modeling and facilitates designers' exploration of various possible combinations and design solutions, its application is relatively limited in CAD fields such as equipment manufacturing, where digital twins are more prevalent, primarily due to its focus on applications in film and games.

[0040] Based on this technological background, this invention proposes a modular simulation method for UAV digital twins aimed at enhancing self-awareness, specifically including:

[0041] A modular simulation library is constructed based on a reduced-order model. For each geometric module, the Static-Condensation Reduced-Basis-Element (SCRBE) method is used to calculate multiple simulation models under different parameter states, thus constructing the modular simulation library. The core idea of ​​the SCRBE method is to first use a substructure method to construct a corresponding simulation system for each component, such as... Figure 2 (a) Based on this, a reduced-basic finite element method with guaranteed accuracy is applied to each component. This fully leverages the advantages of the reduced-basic method in terms of accuracy, speed, and parameters, while also possessing the scalability and flexibility inherent in a component-based approach. For example... Figure 2 As shown in (b), in the SCRBE model, each component is connected by a set of parameters. Define the component, where i is its part number. These parameters can be geometric parameters affecting the spatial shape of the component, or non-geometric parameters such as material properties, depending on the specific application. A component with a specified set of parameters and value ranges is called a prototype component, while a component with specified parameter values ​​is called an instantiation of the prototype component. Taking the linear elastic case as an example, examine the structural response of the module under a given load, and then gradually extend to other scenarios. First, for each 3D module asset A... i Associate it with a partial differential equation, and specify the corresponding external boundary conditions when necessary:

[0042]

[0043] in, These are the parameters for the assembled complete unmanned system. The parameters for each module are given. a(u,v;μ) and f(v;μ) represent the bilinear and linear functions, respectively, and X(μ) is the function space. A finite element approximation is performed on the equation to find a set of solutions u. h (μ)=X h (μ) makes:

[0044]

[0045] X h (μ)∈X(μ) is the system-level finite element approximation, corresponding to the global mesh of the unmanned system. Let N FE =dim(x h (μ) The degrees of freedom approximated by the finite element method: Then the above formula can be discretized into a linear system:

[0046] K(μ)U(μ)=F(μ).

[0047] The simulation module dynamically adapts based on the optimal classification tree. After the simulation library is built, it will quickly select the most suitable physical simulation model for each component on the virtual twin based on the observation data acquired by the dynamic sensors on the physical machine, such as... Figure 3 (a) This means that the model parameters are closest to the corresponding sensor state values. Specifically, assume that during the operation of the physical machine, it can access p observation data sources, which provide knowledge about the basic state of the physical machine. These observations can be real-valued (e.g., sensor readings, inspection data) or categorical (e.g., a fault detection system reports a nominal, warning, or fault, represented by integers 0, 1, and 2, respectively). These observations are combined into a feature vector, denoted as... Where t represents the time index and X represents the feature space, i.e., the space of all possible observations. This data can be used to estimate which physics-based model best matches the physical asset, as it best explains the observed data. For each 3D module asset A... i Each simulation model S associated with this module in the simulation library ij Corresponding to A i A state. The project will train an optimal decision tree T: X → M ( Figure 1-3 (b) M represents the set of simulation models associated with the digital twin components. This is achieved through observation data on physical assets. Estimate the best-fitting simulation model, which serves as the state d of the module in the digital twin at time t. i t :

[0048]

[0049] like Figure 3 As shown in (b), based on the data from sensor 22, it can be preliminarily determined whether the corresponding component is in a high-damage state (<429) or a low-damage state (>=429). Based on this, it is divided into two branches. The high-damage state is further divided into 40% or 80% damage, with a corresponding sensor threshold of 495. Similarly, the low-damage state is determined sequentially based on its threshold, with corresponding models of 0%, 20%, or 40% damage. Subsequently, the corresponding damage models can be used as initial values ​​to quickly simulate wing stress and determine its next trend, thus supporting immediate decision-making.

[0050] Training a decision tree T requires corresponding training data. This is based on each simulation model M. t ∈M, the predicted value x can be calculated. t And by comparing it with To measure the difference of M tThe degree of closeness to the actual physical state. This allows us to sample the forward mapping F: M → X to obtain a series of (x) values ​​for training. t M t Considering that the actual observed sensor data generally contains a certain degree of noise, the influence of this noise can be represented by a Gaussian model V. Thus, the positive mapping function under noisy conditions becomes:

[0051] like Figure 4 Another aspect of this invention provides a modular simulation system for a drone digital twin aimed at enhancing self-awareness, comprising:

[0052] Building Unit: Based on the reduced-order model, a modular simulation library is constructed. For each geometric module, the static condensed reduction primitive method is to be used to calculate multiple simulation models under different parameter states, and to construct the modular simulation library.

[0053] Simulation Unit: Based on the dynamic sensors, the state of the physical machine is classified and identified. Using the trained optimal decision tree, the simulation model that best matches the simulation model in the module simulation library is estimated through the observation data on the physical asset, which serves as the state of the module at that moment in the digital twin.

[0054] Figure 5 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 520 and executable on the processor 520. When the processor 520 executes the computer program 511, it implements a cross-user behavior recognition transfer learning method provided by the embodiment of the present invention.

[0055] Since the electronic device described in this embodiment is the device used to implement the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the methods described in the embodiments of the present invention. Therefore, how the electronic device implements the methods in the embodiments of the present invention will not be described in detail here. Any device used by those skilled in the art to implement the methods in the embodiments of the present invention is within the scope of protection of the present invention.

[0056] Please see Figure 6 , Figure 6 This is a schematic diagram of an embodiment of a computer-readable storage medium provided in this invention.

[0057] like Figure 6 As shown, this embodiment provides a computer-readable storage medium 600, on which a computer program 611 is stored. When the computer program 611 is executed by a processor, it implements the cross-user behavior recognition transfer learning method provided in this embodiment of the invention.

[0058] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0059] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This invention provides a modular simulation method for UAV digital twins aimed at enhancing self-awareness. The method includes: constructing a modular simulation library based on a reduced-order model; for each geometric module, employing a static condensed reduction primitive method to calculate multiple simulation models under different parameter states, thus constructing the modular simulation library; classifying and identifying the state of the physical machine based on dynamic sensors; and using a trained optimal decision tree to estimate the most matching simulation model in the module simulation library using observation data from the physical assets, which serves as the state of that module in the digital twin at that moment. The method provided by this invention, based on modularization of states within the digital twin, significantly improves simulation speed, thereby greatly enhancing the self-awareness of the UAV.

[0061] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A self-awareness enhanced drone digital twin modular simulation method, characterized in that, The method comprises the following steps: The module simulation library is constructed based on a reduced-order model, and for each geometric module, a static condensed reduction primitive method is used to calculate multiple simulation models under different parameter states to construct the module simulation library; The state of the physical machine is classified and identified according to a dynamic sensor, and the optimal decision tree is trained to estimate the most matched simulation model in the simulation library through observation data on the physical asset as the state of the module in the digital twin at the current time; The module simulation library is constructed based on a reduced-order model, and for each geometric module, a static condensed reduction primitive method is used to calculate multiple simulation models under different parameter states to construct the module simulation library, and the method further comprises the following steps: For each three-dimensional module asset an associated partial differential equation is associated, specifying the corresponding external boundary conditions: , where, are parameters of the complete unmanned system after splicing, are parameters of each module, is the module number in the SCRBE model; and are bilinear and linear, respectively, is the function space; the finite element approximation is performed on the equation to find a set of solutions such that: , For the system level finite element approximation, the corresponding to the overall grid of the unmanned system; Let The degrees of freedom of this finite element approximation: Then the above formula is discretized into a linear system: ； The state of the physical machine is classified and identified according to a dynamic sensor, and the optimal decision tree is trained to estimate the most matched simulation model in the simulation library through observation data on the physical asset as the state of the module in the digital twin at the current time, and the method further comprises the following steps: for each three-dimensional module asset each simulation model in the module simulation library associated with the module corresponds to a state of the three-dimensional module asset ; Training optimal decision trees , represents a set of simulation models associated with the digital twin component; through observation data on the physical asset estimate the most matching simulation model as the state of this module in the digital twin at time t : ； The optimal decision tree is trained, and the training specifically comprises the following steps: According to each simulation model , the predicted value is calculated and the closeness to the actual physical state is measured by comparing the difference between the predicted value and the actual value ; the forward mapping is sampled to obtain a series of training ; the noise effect of the observed sensor data is represented by a Gaussian model , and the forward mapping function under the noise condition is: .

2. The self-awareness enhanced UAV digital twin modular simulation system implemented by the self-awareness enhanced UAV digital twin modular simulation method of claim 1, wherein, The method comprises the following steps: The construction unit: the module simulation library is constructed based on a reduced-order model, and for each geometric module, a static condensed reduction primitive method is used to calculate multiple simulation models under different parameter states to construct the module simulation library; The simulation unit: the state of the physical machine is classified and identified according to a dynamic sensor, and the optimal decision tree is trained to estimate the most matched simulation model in the simulation library through observation data on the physical asset as the state of the module in the digital twin at the current time.

3. An electronic device, comprising: The method comprises the following steps: The memory, the processor and the computer program stored in the memory and executable on the processor, wherein the processor implements the method of claim 1 when executing the computer program.

4. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium, and the computer program is executed by the processor to implement the method of claim 1.

Images