Digital twinning detection platform and method based on meta-universe technology

Abstract

The invention discloses a digital twinborn detection platform and method based on the meta-universe technology, and belongs to the technical field of detection platforms. A multi-source sensor acquisition and processing unit; the virtual-real mapping module is used for mapping the collected detection data to a virtual space so as to obtain the twinning points of the elementary cosmic space; defect evolution modeling is conducted, specifically, potential defects are evolved into dominant recognizable defects from recessive micro-defects based on a defect evolution modeling equation set of multi-physics field coupling; a defect identification and prediction module; a multi-dimensional interaction and simulation display module; and the virtual-real interaction protocol interface is used for feeding back the prediction result to the production line controller. According to the intelligent detection platform, defect prediction is realized by fusing defect evolution of element universe, digital twinning and multi-physics field coupling and variational time sequence prediction. The method can be applied to high-reliability industrial fields such as wafer manufacturing, packaging detection, micro-nano photoelectricity, automotive electronics and aircraft devices.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of detection platforms, and in particular to a digital twin detection platform and method based on metaverse technology. BACKGROUND

[0002] With the gradual entry of chip manufacturing technology into the 5nm node, the wafer density and structural complexity grow exponentially, and any slight process deviation and defect evolution can lead to the failure of the entire device. Traditional image recognition and single-point process analysis methods have not met the requirements. The current technical route faces the following bottlenecks: first, the unpredictability of defect evolution mechanism, leading to process window optimization relying on a large number of repeated experiments; second, the weak correlation between chip detection results and previous process parameters, making it difficult to form a closed-loop feedback; third, in the mass production process, complex multi-physical factors act on the defect generation process, and existing modeling tools are difficult to realize integrated coupling simulation across scales and physical quantities.

[0003] In recent years, with the development of metaverse technology and digital twin system, a new idea has been provided for high-precision industrial process simulation and real-time visualization. Metaverse is not only an immersive three-dimensional virtual space, but also a dynamic scene evolution platform driven by multi-source heterogeneous data. Its underlying capabilities include physical modeling, data synchronization, behavior prediction, virtual-real interaction, and other dimensions. Especially in the field of industrial detection, by constructing a highly simulated and dynamic feedback twin space, the defect formation process, time sequence changes and multi-dimensional parameter coupling behavior of the detection object can be completely restored, and further guide the optimization of equipment operation in the physical space.

[0004] At present, although metaverse technology and digital twin have been tentatively deployed by some enterprises, it is still a great challenge to build a complete "virtual-real mapping + evolution prediction" in a chip-level microscopic detection system.

[0005] Therefore, the present application designs a digital twin detection platform and method based on metaverse technology to solve the above problems. SUMMARY

[0006] In view of the above shortcomings of the prior art, the present application provides a digital twin detection platform and method based on metaverse technology.

[0007] To achieve the above purpose, the present application realizes the following technical solutions:

[0008] The digital twin detection platform based on metaverse technology comprises:

[0009] a digital twin body model;

[0010] Multi-source sensor acquisition and processing unit;

[0011] The virtual-reality mapping module maps the collected detection data into the virtual space to obtain the twin points of the metaverse space;

[0012] Defect evolution modeling, based on a set of defect evolution modeling equations coupled with multiple physical fields, can transform potential defects from hidden micro-defects to explicit and recognizable defects;

[0013] Defect identification and prediction module: using historical defect database Establish a variational time series prediction network for defect prediction;

[0014] A multi-dimensional interaction and simulation display module is used to highlight identified defects in the digital twin model through three-dimensional coordinate mapping;

[0015] The virtual-reality interaction protocol interface is used to feed back the prediction results to the production line controller.

[0016] Furthermore, in the virtual-reality mapping module, the physical space state set is defined for:

[0017]

[0018] in, Detection image at time t; P t : process parameter tensor; σ t : stress tensor; v t : Indicates the operating speed of the device; H indicates the height of the detection image; W indicates the width of the detection image; T t represents the temperature field at time t;

[0019] The physical space state set Mapping function f through twin points map Mapping to virtual space:

[0020]

[0021] Among them, z t : The twin points of the metaverse space obtained after mapping.

[0022] Furthermore, the defect evolution modeling equations adopt the following coupled differential equations:

[0023]

[0024] Where: x represents the spatial coordinate, which is used to describe the location of defects or physical fields in the material; σ(x, t) is the stress field; T(x, t) is the thermal field; μ(x, t) is the microdefect density function; D(·) is the diffusion coefficient function; Q(x, t) is the heat source term; Φσ (x, t): material elastoplasticity change term; material elasticity modulus tensor; u: displacement field; k is thermal conductivity; σ is stress tensor; T is temperature field; J e is current density; μ is micro-defect density function; R is defect reaction term.

[0025] Further, it also includes a finite element solver for numerically solving the coupled differential equation set and projecting the solution result back into the digital twin model.

[0026] Further, the variational time series prediction network:

[0027]

[0028] where q φ (z 1：T | μ 1：T ) is the approximate posterior distribution; p θ (μ 1：T | z 1：T ) is the decoder; μ 1：T is the observed defect time series data; z 1：T : latent variable sequence; φ: encoder parameter; θ: decoder parameter; T1: time step.

[0029] Further, in the defect identification and prediction module, the loss function is defined as:

[0030]

[0031] where p(z 1：T ) is the prior assumption of hidden variables; is the log-likelihood expectation of the generation model; KL(q||p) is the KL divergence.

[0032] The operation method of the digital twin detection platform based on the metaverse technology uses the digital twin detection platform based on the metaverse technology, which includes the following steps:

[0033] Step 1, constructing a digital twin model;

[0034] Step 2, accessing multi-source sensors to collect detection signals;

[0035] Step 3, the virtual-real mapping module maps the collected detection data to the virtual space to obtain a metaverse space twin point;

[0036] Step 4, based on the defect evolution modeling equation set coupled with multiple physical fields, the latent defect is evolved from an implicit micro-defect to an explicit identifiable defect; the defect evolution modeling equation set adopts the following coupled differential equation set:

[0037]

[0038] wherein: x represents a spatial coordinate, used to describe the location of defects or physical fields in the material, sigma(x, t) is a stress field; T(x, t) is a thermal field; mu(x, t) represents a micro-defect density function; D(·): a diffusion coefficient function; Q(x, t): a heat source term; Phi σ (x, t): a material elastoplasticity change term; material elasticity modulus tensor; u: displacement field; k is thermal conductivity (Thermal conductivity), unit W / (m·K); sigma is stress tensor (Stress tensor), unit Pa; T is temperature field (Temperature field), unit K; J e is current density (Current density), unit A / m 2 ; mu is micro-defect density (Micro-defect density), unit m -3 (defect quantity per unit volume); R is defect reaction term (Defectreaction term), which describes the generation / annihilation rate of defects (such as R(mu) = k1mu-k2mu 2 ).

[0039] The finite element method is used by the finite element solver to numerically solve the above coupling differential equation set, and the solution result is projected back to the digital twin model;

[0040] Step 5, the defect identification and prediction module uses a variational time series prediction network for defect identification and prediction;

[0041] The variational time series prediction network is:

[0042]

[0043] wherein, q φ (z 1：T | mu 1：T ) is an approximate posterior distribution; p θ (mu 1：T | z 1：T ) is a decoder; mu 1：T is observed defect time series data; z 1：T : latent variable sequence; phi: encoder parameter; theta: decoder parameter; T1: time step;

[0044] Step 6, the multi-dimensional interaction and simulation display module highlights the identified defects in the digital twin model through three-dimensional coordinate mapping;

[0045] Step 7, the prediction result is fed back to the production line controller through the virtual-real interaction protocol interface.

[0046] Further, step 1 is specifically: high-precision three-dimensional modeling is performed on the detected workpiece to generate a corresponding digital twin model.

[0047] Further, step 2 is specifically: the data of various sensors deployed in the physical production line are uniformly collected through edge computing nodes, and data alignment, time synchronization and structure standardization are performed, and the detection data is output to the virtual-real mapping module.

[0048] Further, in step 3, the physical space state set is defined as

[0049]

[0050] Among them, the detection image at time t; P t : process parameter tensor; sigma t : stress tensor; v t : represents the running speed of the device; H represents the height of the detection image; W represents the width of the detection image; T t represents the temperature field at time t;

[0051] The physical space state set is mapped to the virtual space through the twin point mapping function f map

[0052]

[0053] Among them, z t : the metaverse space twin point obtained after mapping.

[0054] Compared with the prior art, the present application has the beneficial effects that: the present application establishes a chip detection virtual mapping mechanism based on metaverse space, which first maps the chip detection process to an immersive metaverse space and defines a twin point mapping function. This method not only contains geometric dimension information, but also contains material physical field attributes and historical state evolution trend, breaking through the limitation of traditional detection system only "image feature-classification label" plane mapping, so that each chip has three-dimensional, dynamic and time sequence consistent mapping features.

[0055] The present application establishes a multi-physical field coupled defect evolution modeling equation set. Traditional detection is only based on image appearance features for judgment, which is difficult to explain the root cause of defect generation. The present application is based on the stress-heat-material microstructure three-field linkage evolution mechanism, and proposes a coupled differential equation set. Defects are regarded as diffusion behavior of microstructure, and are dynamically coupled with process parameters. The possible formation area and time window of defects can be predicted by physical mechanism as the driving force, which greatly improves the model explainability.

[0056] ​​The application establishes a defect time evolution prediction network based on variational inference, overcomes irregularity of time series and ambiguity of defect features, generates future defect patterns using historical state latent variables, considers accuracy and generalization ability in the risk prediction process, and greatly enhances decision robustness.

[0057] The application uses a lightweight virtual-real interactive protocol interface (X-VDE), supports structured state uploading, parameter writing back and control logic interruption, achieves a good balance in communication efficiency and system compatibility, and is easy to integrate into existing industrial informatization systems (such as MES and SCADA).

[0058] The application can be applied to high-reliability industrial fields such as wafer manufacturing, packaging detection, micro-nano optoelectronics, automotive electronics and aviation devices. DETAILED DESCRIPTION

[0059] To make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0060] Embodiment one: a digital twin detection platform based on meta-universe technology, comprising:

[0061] A digital twin model is generated by using CAD drawings, structured light scanning, laser measurement and other technologies to perform high-precision three-dimensional modeling on the detected workpieces (such as engines, glass bottles, PCB boards, etc.), and the corresponding digital twin model is generated; this model realizes three-dimensional entity construction in the meta-universe virtual space, and can dynamically bind various physical properties (such as temperature, stress distribution, material defect development function, etc.);

[0062] A multi-source sensor acquisition and processing unit is used to uniformly acquire various sensor data deployed in the physical production line through edge computing nodes, and perform data alignment, time sequence synchronization and structure standardization processing, and output detection data;

[0063] For example, in a chip manufacturing production line, multi-source data acquisition nodes are arranged, including: a defect image acquisition unit (such as an AOI system); a process parameter acquisition module (temperature, pressure, current density); a material property sampling sensor (stress probe, particle analyzer); a real-time control state recorder (installed work station, speed feedback, etc.); the above information is uploaded to the edge computing server through the industrial network, and is simultaneously sent to the digital twin detection platform of the application for processing.

[0064] A virtual-real mapping module maps the collected detection data to the virtual space;

[0065] A set of real space states is defined As:

[0066]

[0067] Wherein, The detection image at time t; P t : Process parameter tensor; sigma t : Stress tensor; v t : Denotes the running speed of the device; H denotes the height (Height) of the detection image, in pixels (pixels); W denotes the width (Width) of the detection image, in pixels (pixels); T t Denotes the temperature field at time t.

[0068] The set of real space states Through the twin point mapping function f map Is mapped to the virtual space:

[0069]

[0070] Wherein, z t : The metaverse space twin point obtained after mapping, used to drive the geometric, material, and state attribute changes of the space entity object (i.e. the virtual chip model).

[0071] The core role of the twin point: data fusion and dimension reduction: the twin point z t Is the compression and feature extraction (such as through the neural network f map ) of the original detection data , which unifies multi-modal data (images, stresses, temperatures, etc.) into a high-dimensional vector, facilitating subsequent modeling. Time sequence evolution dependence: defect evolution is a time sequence process, which needs to rely on the historical twin point sequence z 1：t-1 To initialize the initial state (such as the distribution of hidden micro-defects).

[0072] The present application establishes a chip detection virtual mapping mechanism based on the metaverse space, which first maps the chip detection process to an immersive metaverse space and defines a twin point mapping function. This method not only contains geometric dimension information, but also contains material physical field attributes and historical state evolution trends, breaking through the limitations of traditional detection systems that only map "image features-classification labels" in a plane, so that each chip has three-dimensional, dynamic, and time sequence consistent mapping features.

[0073] Defect evolution modeling, based on the defect evolution modeling equation set of multi-physical field coupling, simulates the whole process of potential defects evolving from "hidden micro-defects" to "visible and identifiable defects", which specifically adopts the following coupled differential equation set:

[0074]

[0075] where: x represents the spatial coordinate, used to describe the location of defects or physical fields in the material, sigma (x, t) is the stress field; T(x, t) is the thermal field; mu (x, t) represents the micro-defect density function; D(·): diffusion coefficient function; Q(x, t): heat source term; Phi σ (x, t): material elastoplasticity change term; material elastic modulus tensor; u: displacement field; k is the thermal conductivity (Thermal conductivity), unit W / (m·K); sigma is the stress tensor (Stress tensor), unit Pa; T is the temperature field (Temperature field), unit K; J e is the current density (Current density), unit A / m 2 ; mu is the micro-defect density (Micro-defect density), unit m -3 (defect quantity per unit volume); R is the defect reaction term (Defectreaction term), which describes the generation / annihilation rate of defects (such as R(mu) = k1mu-k2mu 2 ).

[0076] The input of the coupled differential equation system is the twin point z t output from the virtual-real mapping module: the stress field sigma (x, t) is analytically extracted from z t The distribution of the stress tensor sigma t is extracted from z t The temperature field data T t is extracted from z 1：t-1 The micro-defect density mu (x, t) is initialized or iteratively updated by the history twin point sequence z The output of the coupled differential equation system (such as mu (x, t)) is input to the variational prediction network as an observation variable to further predict the future defect state.

[0077] Further, the finite element method (FEM) is used by the finite element solver to numerically solve the above coupled differential equation system, and the solution result is projected back into the digital twin model.

[0078] The present application no longer simply relies on image classification models or black box deep networks, but by combining stress fields, temperature fields, electromigration and microstructure parameters, the above coupled differential equation system is derived and numerically solved, the model explicitly proposes the evolution of the defect density field and the dynamic coupling of the physical state, theoretically realizes the possibility of "physically driven prediction of defect generation process", and projects it to the virtual space through the finite element solver for visual rendering.

[0079] The application establishes a multi-physical field coupling defect evolution modeling equation set, and traditional detection is only based on image apparent features for judgment, which is difficult to explain the root cause of the defect. Based on the stress-heat-material microstructure three-field linkage evolution mechanism, a coupled differential equation set is proposed, the defect is regarded as the diffusion behavior of the microstructure, and is dynamically coupled with the process parameters, which can predict the possible formation area and time window of the defect driven by the physical mechanism, and greatly improves the model interpretability.

[0080] Defect identification and prediction module: utilize historical defect database A defect time evolution prediction network based on variational inference is established to realize future defect risk prediction, and the following variational time series prediction network is specifically adopted:

[0081]

[0082] Wherein, q φ (z 1：T |μ 1：T ) is the approximate posterior distribution defined by the neural network parameter phi; p θ (μ 1：T |z 1：T ) is the decoder defined by the parameter theta; mu 1：T is the observed defect time series data (such as defect density, stress field, etc.), the role: input data, drive model learning and prediction; z 1：T : latent variable sequence, encode the implicit features of defect evolution, the role: connect the encoder and the decoder, realize the time series feature extraction and generation; phi: encoder parameter (neural network weight), the role: learn the mapping from observed data to latent space; theta: decoder parameter (neural network weight), the role: learn the mapping from latent space to defect prediction; T1: time step, the role: define the time series window length of modeling.

[0083] In order to overcome the irregularity of time series and the fuzziness of defect features, the application establishes a defect time evolution prediction network based on variational inference, generates future defect patterns using historical state latent variables, considers accuracy and generalization ability in risk prediction process, and greatly enhances decision robustness.

[0084] Further, the loss function is defined as:

[0085]

[0086] Wherein, p(z 1：T ) is the prior assumption of latent variable, which restricts the latent variable space; To generate the log-likelihood expectation of the model, the maximum is approximated by Monte Carlo sampling; KL(q||p) is the KL divergence, which measures the difference between the variational distribution q and the prior distribution p, and ensures the rationality of the hidden variable.

[0087] During the training process, the historical defect image data and the parameter sequence will be used to back-propagate and optimize the parameters of the prediction network φ, θ.

[0088] The multi-dimensional interaction and simulation display module is used to highlight the identified defects in the digital twin model through three-dimensional coordinate mapping, and the platform can show the whole process from the initial micro-crack to the expansion failure of the defects. Users can freely drag and rotate to observe, and can also query the historical data at any position.

[0089] The virtual-real interaction protocol interface (X-VDE) is used to feed back the prediction results to the production line controller, and the production line controller automatically generates control instructions such as alarm prompts, production line suspension, maintenance scheduling, etc. according to the prediction results.

[0090] The virtual-real interaction protocol is defined as follows:

[0091] [Message Header]-32Bytes

[0092] [Chip_ID]-8Bytes

[0093] [Defect_Risk_Map]-1024Bytes

[0094] [Recommendation_OpCode]-2Bytes

[0095] [Checksum]-2Bytes

[0096] In order to realize real-time interaction between detection data and simulation space, the application uses a lightweight virtual-real interaction protocol interface (X-VDE) to support structured state uploading, parameter writing back and control logic interruption, and achieves a good balance in communication efficiency and system compatibility, which is easy to integrate into existing industrial information systems (such as MES, SCADA).

[0097] The application can be applied in high-reliability industrial fields such as wafer manufacturing, packaging detection, micro-nano optoelectronics, automotive electronics and aviation devices.

[0098] Embodiment two: a running method of the digital twin detection platform based on the meta-universe technology, comprising the following steps:

[0099] Step 1, constructing a digital twin model: using CAD drawings, structured light scanning, laser measurement, etc. to conduct high-precision three-dimensional modeling on the detected workpiece (such as engine, glass bottle, PCB board, etc.), and generate the corresponding digital twin model; this model realizes three-dimensional entity construction in the meta-universe virtual space, and can dynamically bind various physical properties (such as temperature, stress distribution, material defect development function, etc.);

[0100] Step 2, access multi-source sensor to collect detection signals: collect various sensor data deployed in the physical production line through edge computing nodes, and perform data alignment, time synchronization and structure standardization, and output detection data to the virtual-real mapping module;

[0101] For example, in the chip manufacturing production line, multi-source data collection nodes are arranged, including: defect image collection unit (such as AOI system); process parameter collection module (temperature, pressure, current density); material property sampling sensor (stress probe, particle analyzer); real-time control state recorder (installed work station, speed feedback, etc.); the above information is uploaded to the edge computing server through the industrial network, and is simultaneously sent to the digital twin detection platform of the present application for processing.

[0102] Step 3, the virtual-real mapping module maps the collected detection data to the virtual space;

[0103] Define the real space state set As follows:

[0104]

[0105] Wherein, The detection image at time t; P t : Process parameter tensor; σ t : Stress tensor; v t : represents the running speed of the device; H represents the height (Height) of the detection image, in pixels (pixels); W represents the width (Width) of the detection image, in pixels (pixels); T t represents the temperature field at time t.

[0106] The real space state set is mapped to the virtual space by the twin point mapping function f map :

[0107]

[0108] Wherein, z t : the meta-universe space twin point obtained after mapping, used to drive the geometric, material and state attribute changes of the space entity object (i.e. virtual chip model).

[0109] Step 4, the defect evolution modeling equation set based on multi-physical field coupling simulates the whole process of potential defects evolving from "implicit micro-defects" to "explicit identifiable defects", and the following coupled differential equation set is used:

[0110]

[0111] Wherein: x represents the spatial coordinate, used to describe the position of the defect or physical field in the material, σ(x, t) is the stress field; T(x, t) is the thermal field; μ(x, t) represents the micro-defect density function; D(·): diffusion coefficient function; Q(x, t): heat source term; Φ σ (x, t): material elastic-plastic change term; Material elastic modulus tensor; u: displacement field; k is the thermal conductivity (Thermal conductivity), unit W / (m·K); σ is the stress tensor (Stress tensor), unit Pa; T is the temperature field (Temperature field), unit K; J e Is the current density (Current density), unit A / m 2 ; μ is the micro-defect density (Micro-defect density), unit m -3 (Defect quantity in unit volume); R is the defect reaction term (Defectreaction term), which describes the generation / annihilation rate of defects (such as R(μ)=k1μ-k2μ 2 ).

[0112] Further, the finite element method (FEM) is used by the finite element solver to numerically solve the above coupled differential equation set, and the solution result is projected back to the digital twin model.

[0113] Step 5, the defect identification and prediction module uses a variational time series prediction network for defect identification and prediction.

[0114] The variational time series prediction network is:

[0115]

[0116] Wherein, q φ (z 1：T |μ 1：T ) is the approximate posterior distribution, which is defined by the neural network parameter φ; p θ (μ 1：T |z 1：T ) is the decoder, which is defined by the parameter θ; μ 1：T Is the observed defect time series data (such as defect density, stress field, etc.), the function is: input data, drive model learning and prediction; z 1：T: latent variable sequence, encoding the hidden features of defect evolution, function: connecting the encoder and the decoder, realizing the temporal feature extraction and generation; φ: encoder parameters (neural network weights), function: learning the mapping from the observed data to the latent space; θ: decoder parameters (neural network weights), function: learning the mapping from the latent space to the defect prediction; T1: time step, function: defining the length of the time window for modeling.

[0117] Step 6: The multi-dimensional interaction and simulation display module highlights the identified defects in the digital twin model through three-dimensional coordinate mapping. The platform can show the entire process from the initial micro-crack to the expansion failure of the defect. Users can freely drag and rotate to observe, and can also query historical data at any position.

[0118] Step 7: The prediction results are fed back to the production line controller through the virtual-real interaction protocol interface (X-VDE). The production line controller automatically generates control instructions such as alarm prompts, production line suspension, maintenance scheduling, etc. according to the prediction results, realizing closed-loop intervention.

[0119] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. The digital twin detection platform based on Metaverse technology is characterized by: include: Digital twin models; Multi-source sensor acquisition and processing unit; The virtual-reality mapping module maps the collected detection data into the virtual space to obtain the twin points of the metaverse space; Defect evolution modeling, based on a set of defect evolution modeling equations coupled with multiple physical fields, can transform potential defects from hidden micro-defects to explicit and recognizable defects; Defect identification and prediction module: using historical defect database Establish a variational time series prediction network for defect prediction; A multi-dimensional interaction and simulation display module is used to highlight identified defects in the digital twin model through three-dimensional coordinate mapping; The virtual-reality interaction protocol interface is used to feed back the prediction results to the production line controller.

2. The digital twin detection platform based on metaverse technology according to claim 1 is characterized in that: In the virtual-reality mapping module, define the physical space state set for: in, Detection image at time t; P t : process parameter tensor; σ t : stress tensor; v t : Indicates the operating speed of the device; H indicates the height of the detection image; W indicates the width of the detection image; T t represents the temperature field at time t; The physical space state set Mapping function f through twin points map Mapping to virtual space: Among them, z t : The twin points of the metaverse space obtained after mapping.

3. The digital twin detection platform based on metaverse technology according to claim 1 is characterized in that: The defect evolution modeling equations use the following coupled differential equations: Where: x represents the spatial coordinate, which is used to describe the location of defects or physical fields in the material; σ(x, t) is the stress field; T(x, t) is the thermal field; μ(x, t) is the microdefect density function; D(·) is the diffusion coefficient function; Q(x, t) is the heat source term; Φ σ (x, t): elastic-plastic change of material; Material elastic modulus tensor; u: displacement field; k is thermal conductivity; σ is stress tensor; T is temperature field; J e is the current density; μ is the micro-defect density function; R is the defect reaction term.

4. The digital twin detection platform based on metaverse technology according to claim 3 is characterized in that: A finite element solver is also included for numerically solving the system of coupled differential equations and projecting the results of the solution back into the digital twin model.

5. The digital twin detection platform based on metaverse technology according to claim 1 is characterized in that: Variational Time Series Prediction Network: Among them, q φ (z 1:T |μ 1:T ) is the approximate posterior distribution; p θ (μ 1:T |z 1:T ) is a decoder; μ 1:T is the observed defect time series data; 1:T : sequence of latent variables; φ: encoder parameters; θ: decoder parameters; T1: time step.

6. The digital twin detection platform based on metaverse technology according to claim 5 is characterized in that: In the defect recognition and prediction module, the loss function is defined as: Among them, p(z 1:T ) is the prior hypothesis of latent variables; is the log-likelihood expectation of the generative model; KL(q‖p) is the KL divergence.

7. A method for operating a digital twin detection platform based on Metaverse technology, utilizing the digital twin detection platform based on Metaverse technology according to any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1: Build a digital twin model; Step 2: Connect multiple source sensors to collect detection signals; Step 3: The virtual-reality mapping module maps the collected detection data to the virtual space to obtain the metaverse space twin point; Step 4: Based on the defect evolution modeling equations coupled with multiple physical fields, the potential defects are evolved from hidden micro-defects to explicit identifiable defects. The defect evolution modeling equations use the following coupled differential equations: Where: x represents the spatial coordinate, which is used to describe the location of defects or physical fields in the material; σ(x, t) is the stress field; T(x, t) is the thermal field; μ(x, t) is the microdefect density function; D(·) is the diffusion coefficient function; Q(x, t) is the heat source term; Φ σ (x, t): elastic-plastic change of material; Material elastic modulus tensor; u: displacement field; k is thermal conductivity (W / (m·K); σ is stress tensor (Pa); T is temperature field (K); J e Current density, unit A / m 2 ; μ is the micro-defect density function (Micro-defect density), unit m -3 (number of defects per unit volume); R is the defect reaction term, which describes the rate of defect creation / annihilation (e.g. R(μ) = k1μ - k2μ 2 ). The coupled differential equations are numerically solved using the finite element method using a finite element solver, and the solution results are projected back into the digital twin model; Step 5: The defect recognition and prediction module uses a variational time series prediction network to perform defect recognition and prediction; The variational time series prediction network is: Among them, q φ (z 1:T |μ 1:T ) is the approximate posterior distribution; p θ (μ 1:T |z 1:T ) is a decoder; μ 1:T is the observed defect time series data; 1:T : sequence of latent variables; φ: encoder parameters; θ: decoder parameters; T1: time step; Step 6: The multi-dimensional interaction and simulation display module highlights the identified defects in the digital twin model through three-dimensional coordinate mapping; Step 7: The prediction results are fed back to the production line controller through the virtual-reality interaction protocol interface.

8. The method for operating a digital twin detection platform based on metaverse technology according to claim 7, characterized in that: Step 1 is specifically: perform high-precision three-dimensional modeling of the workpiece to be inspected and generate a corresponding digital twin model.

9. The method for operating a digital twin detection platform based on metaverse technology according to claim 7, characterized in that: Step 2 is specifically as follows: various sensor data deployed in the physical production line are uniformly collected through edge computing nodes, and data alignment, timing synchronization and structure standardization are performed, and the detection data is output to the virtual-reality mapping module.

10. The method for operating a digital twin detection platform based on metaverse technology according to claim 7, characterized in that: In step 3, define the physical space state set for: in, Detection image at time t; P t : process parameter tensor; σ t : stress tensor; v t : Indicates the operating speed of the device; H indicates the height of the detection image; W indicates the width of the detection image; T t represents the temperature field at time t; The physical space state set Mapping function f through twin points map Mapping to virtual space: Among them, z t : The twin points of the metaverse space obtained after mapping.

Images