Complex equipment reliability real-time mapping and evaluation method based on PINNs and MBSE models

Abstract

The invention discloses a real-time mapping and evaluation method for the reliability of complex equipment, which is based on physical information neural networks (PINNs) and a model system engineering (MBSE) model, and is characterized in that the method comprises the following steps of: (1) carrying out real-time mapping and evaluation on the reliability of the complex equipment, and (2) carrying out real-time mapping and evaluation on the reliability of the complex equipment based on the MBSE model, and (3) carrying out real-time mapping and evaluation on the reliability of the complex equipment based on the MBSE model, and (4) carrying out real-time mapping and evaluation on the reliability of the complex equipment. The method comprises the following steps: firstly, constructing a system architecture model of equipment by utilizing MBSE, and defining reliability constraint parameters; a partial differential equation describing a component failure mechanism is extracted, a neural network agent model integrating physical information constraints is constructed, and a loss function of the neural network agent model is formed by weighting a data driving item and a physical residual item. Mapping real-time working condition parameters into PINNs input by establishing a data interface of a system architecture model and a PINNs agent model, and outputting a physical performance degradation state; and finally, calculating real-time reliability based on the probability statistical model, and dynamically feeding back the real-time reliability to the system architecture model to update a demand verification state and generate a control instruction. According to the method, the physical partial differential equation is introduced as the regularization constraint of the neural network, so that the problem of poor generalization ability of a pure data driving model under a small sample working condition is solved, and online evaluation and closed-loop control in an equipment operation stage are realized.

Description

Technical Field

[0001] This invention belongs to the fields of systems engineering, digital design and reliability assessment technology, and specifically relates to a method for real-time mapping, dynamic simulation and assessment of the reliability of complex equipment that combines physical information neural networks (PINNs) and model-based systems engineering (MBSE) models. Background Technology

[0002] With the increasing integration of complex systems in aerospace, energy equipment, and other fields, verifying non-functional requirements such as the "six characteristics" of equipment during the design phase is becoming increasingly difficult. The current common practice is to use MBSE (Model-Based Structure Design) for functional architecture design and evaluate it through numerical simulations such as finite element analysis. However, existing technologies have the following technical problems: The architecture is disconnected from the physical world. MBSE models are mostly at the logical abstraction layer, making it difficult to reflect the physical failure mechanisms under the coupling of multiple physics fields at the underlying level in real time.

[0003] Insufficient computational timeliness. High-fidelity numerical simulation of complex equipment involves large computational loads, making it difficult to meet the real-time monitoring needs during equipment operation.

[0004] Data dependency limitations. Purely data-driven neural networks are prone to violating physical laws and having low confidence levels when small sample failure samples are scarce.

[0005] Therefore, there is an urgent need for a reliability assessment method that can integrate prior physical knowledge and be linked with the system architecture model in real time. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method, system, and storage medium for real-time reliability mapping and evaluation of complex equipment based on PINNs and MBSE models. This invention embeds PINNs as a proxy model into the MBSE architecture, leveraging the physical consistency prediction capability of PINNs under small sample sizes to map system-level operating parameters to physical layer performance degradation states in real time, and automatically feeds this data back to the system requirement model to guide equipment control.

[0007] The first aspect of this invention provides a method for real-time mapping and evaluation of the reliability of complex equipment based on PINNs and MBSE models, comprising the following steps: Step S1: System-level reliability requirement modeling. A system architecture model of the equipment (MBSE) is constructed based on the SysML system modeling language. In the parameter diagram of the system architecture model, reliability constraint parameters are defined through constraint blocks, and the specified performance parameters of the equipment are associated with the parameters of the constraint blocks as numerical attributes. Step S2: Characterization and extraction of failure physical mechanisms. For the target subsystem of the equipment, identify its corresponding failure physical processes, and extract the partial differential equations describing the failure physical processes, as well as the corresponding boundary conditions and initial conditions; Step S3: Construction and training of the Physical Information Neural Network (PINN) surrogate model. A multilayer perceptron neural network containing an input layer, hidden layers, and an output layer is constructed. Automatic differentiation techniques are used to construct the physical constraint module; the total loss function L is defined. tota Including data-driven loss term L data And physical information constraint loss term L phys By minimizing L tota Optimize network weights; Step S4: Real-time data mapping between the architecture model and the physical agent model. A mapping channel is established between the MBSE model output port and the PINNs model input layer through a data interface protocol. Feature preprocessing and dimension alignment are performed on real-time operating condition variables to generate normalized input vectors. Step S5: Dynamic Reliability Assessment and Requirement Status Feedback. Real-time collected equipment operation data is input into the trained PINNs proxy model to obtain predicted physical performance degradation values; a pre-set reliability probability model is used to calculate real-time reliability indices, and these indices are fed back into the requirement verification matrix of the MBSE system architecture model, automatically updating the verification status identifier of reliability requirements; when the verification status identifier indicates failure, an alarm signal is triggered or an equipment operation control command is generated.

[0008] Furthermore, in step S3, the physical information constraint loss term L phys The residual function is constructed based on the partial differential equation extracted in step S2; the physical constraint module uses the automatic differential calculation network to output the partial derivative with respect to the input coordinates and substitutes it into the partial differential equation to calculate the physical residual value.

[0009] Furthermore, in step S4, the feature preprocessing and dimensional alignment specifically include: using principal component analysis or an autoencoder to perform dimensionality reduction processing on the multidimensional working condition space output by the MBSE model, extracting feature operators related to the failure physical process; and using normalization processing to adapt the data distribution to the sensitive range of the PINNs activation function.

[0010] Further, in step S5, the method for converting the predicted value into a real-time reliability index is as follows: based on the predicted value and its uncertainty distribution output by PINNs in step S3, a probability density function (PDF) about the component performance is generated by combining Monte Carlo sampling, and the instantaneous reliability R(t) is obtained by calculating the probability integral of the performance index falling outside the failure threshold.

[0011] Furthermore, the MBSE system architecture model includes a requirement verification matrix RVTM, which receives the real-time reliability index fed back from step S5 through a logic judgment unit; the triggering of the alarm signal or the generation of equipment operation control instructions specifically includes: when the verification status identifier is set to "failed", triggering the emergency state machine of the system architecture model and outputting control instructions to reduce the load level or shut down for maintenance.

[0012] Furthermore, the partial differential equations in step S2 encompass one or more coupled forms of the fatigue damage equation in structural mechanics, the heat conduction equation in thermodynamics, and the corrosion rate equation in fluid mechanics.

[0013] Furthermore, in step S3, the training process of the PINNs proxy model introduces an adaptive weight coefficient λ and a dynamic balancing L. data With L phys The adaptive weight coefficient λ is updated based on the statistical characteristics of the physical residual gradient during training.

[0014] A second aspect of the present invention provides a real-time reliability assessment system for complex equipment, comprising: Memory, used to store computer programs; A processor for executing the computer program; When the processor executes the computer program, it implements the steps of the method described in the first aspect above.

[0015] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0016] The present invention has the following beneficial effects: 1. Improve the physical consistency of the model. By introducing physical equations as regularization terms, the problem of pure data-driven models violating physical laws in predictions under small sample sizes or extreme conditions is solved.

[0017] 2. Enables rapid online evaluation of PINNs. The inference speed of trained PINNs is much faster than traditional numerical simulation, supporting real-time reliability calculations during equipment operation.

[0018] 3. Closed-loop verification and control of reliability for complex equipment. A closed loop was achieved for complex equipment, from "requirement definition" to "physical condition monitoring" and then to "control command feedback," thereby improving the operational reliability of complex equipment. Attached Figure Description

[0020] Figure 1This is a flowchart illustrating the overall technical process of a real-time reliability mapping and evaluation method for complex equipment based on PINNs and MBSE models, provided in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the parameter graph mapping logic in the MBSE system architecture model of this invention.

[0022] Figure 3 This is a schematic diagram illustrating the network structure and physical loss function construction of the PINNs proxy model in an embodiment of the present invention.

[0023] Figure 4 This is a data flow and transformation logic diagram of the real-time data mapping interface in this embodiment of the invention.

[0024] Figure 5 This is a closed-loop diagram illustrating the feedback of reliability assessment results to the requirement verification matrix in an embodiment of the present invention.

[0025] Explanation of main icon numbers: 101—MBSE System Architecture Modeling Module; 102—Failure Physical Mechanism Extraction Module; 103—PINNs Proxy Model Construction Module; 104—Real-time Data Mapping Interface; 105—Dynamic Reliability Assessment Module; 201—Equipment structure block; 202—Reliability constraint block; 203—Numerical attribute; 204—Constraint parameter binding connection; 301—Input Layer; 302—Hidden Layer; 303—Output Layer; 304—Physical Operator; 305—Loss Function Summary Node; 401—Raw operating condition data stream; 402—Mapping engine; 403—Feature preprocessing module; 404—Normalized input vector; 501—Requirement verification matrix; 502—Logical judgment unit; 503—Verification status indicator. Detailed Implementation

[0027] like Figure 1 As shown, in this embodiment of the invention, the complex equipment is first defined in terms of function and structure through the MBSE system architecture modeling module 101.

[0028] System-level reliability requirements modeling In this embodiment, a system architecture model of a certain type of aero-engine turbine blade is first constructed using the SysML language. For example... Figure 2As shown, the "blade assembly" and its numerical attributes such as temperature, rotational speed, and stress are defined in the parameter diagram 203. A reliability constraint block 202 is introduced, defining the reliability constraint parameter R(t)>Rreq, such as "the creep life of the blade in a 1200K environment is not less than 3000 hours", and the system attributes are associated with the constraint parameter through binding lines 204.

[0029] Construction of physical information neural networks For creep failure of turbine blades, the dominant physical equations are determined. For example... Figure 3 As shown, a PINNs network is constructed, with input layer 301 receiving coordinates and time, and output layer 303 outputting displacement and stress. Automatic differentiation is performed using the physics operator module 304 to calculate partial derivatives and substitute them into Norton's creep law. The total loss function, L, calculated by node 305, is expressed as: L=ω data •L data +ω phys •L pde Where L pde This is the physical residual term, used to constrain the prediction results to strictly adhere to the physical equations.

[0030] Regarding the adaptive weights as described in claim 7, this embodiment employs the following update strategy: In the k-th iteration of training, calculate |∇ θ L data | and | ∇ θ L pde Let the average gradient of | be | Update weights using the moving average method This allows for a dynamic balance between the contributions of the two loss terms.

[0031] Real-time mapping and simulation like Figure 4 As shown, communication is established through the real-time data mapping interface 104. The raw operating condition data stream 401 (including "speed" and "gas temperature") enters the mapping engine 402, where it is normalized and dimensionality reduced by the feature preprocessing module 403 to generate a standardized input vector 404 adapted to the neural network, automatically triggering the underlying PINNs to perform forward computation. Reliability index conversion and feedback. The stress distribution and strain rate output by PINNs are input into the dynamic reliability assessment module 105. The instantaneous reliability R(t) is calculated using the Monte Carlo method. Figure 5 As shown, the result is sent to the logic judgment unit 502. If the index is lower than 0.99, the unit 502 automatically updates the verification status flag 503 in the requirement verification matrix 501 to "not passed" and generates an instruction to trigger the engine controller to reduce the speed to extend its life.

[0032] Example 2: Reliability Assessment of Power Plants and Main Oil and Gas Transmission Pipelines This embodiment demonstrates the application of this method in large-scale infrastructure.

[0033] Establish a pipeline system MBSE architecture model The structural model of the main piping system was constructed using SysML in MBSE modeling module 101. The topology, consisting of pump stations, valves, straight pipe sections, elbows, and tees, was defined using a Structured Block Diagram (BDD). The transmission of fluid pressure P and temperature T was defined in the Integrated Block Diagram (IBD). Figure 2 The logic is defined in the parameter diagram using constraint block 202: "The annual thinning at the elbow of the high-pressure steam pipeline should be less than 0.2mm".

[0034] Physical equation extraction of failure mechanism For the most common failure mechanisms of pipelines, namely "erosion-corrosion" and "thermal stress fatigue", partial differential equations are extracted: Pipe wall stress equations: The radial and circumferential stress distribution of the pressurized pipe is described using Lame's equations; Corrosion kinetics equation: Characterizing the nonlinear relationship between chemical corrosion rate and temperature based on Arrhenius law; Fluid erosion constraint: The Navier-Stokes equations are introduced as background constraints for fluid-structure interaction.

[0035] Build and train a PINNs proxy model like Figure 3 As shown, a fully connected neural network with 5 hidden layers 302 is established. The input layer 301 vector includes the pipe coordinates (x, y, z), the internal medium pressure P, the fluid temperature T, and the running time t. The physical operator module 304 transforms the corrosion kinetic equation into residual terms to constrain the network training.

[0036] Real-time data mapping and reliability assessment like Figure 4 As shown, the data stream 401 collected by the Supervisory Control And Data Acquisition (SCADA) system is dynamically injected into the trained PINNs model through the real-time data mapping interface 104. The PINNs model outputs the stress distribution field and wall thickness degradation field of the entire pipe section in real time.

[0037] Main pipeline demand verification closed loop and decision recommendations like Figure 5As shown, when the evaluation results show that the reliability R(t) is lower than the warning threshold, the logic judgment unit 502 updates the requirement verification matrix 501, marks the "pipeline structure integrity" status identifier 503 as "Fail", and automatically generates a control signal of "recommended pressure reduction operation" or "planned shutdown maintenance" and sends it to the power plant main control system.

[0038] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time reliability mapping and evaluation of complex equipment based on PINNs and MBSE models, characterized in that, Includes the following steps: Step S1: System-level reliability requirement modeling. Based on a system modeling language, construct an MBSE system architecture model for the equipment. In the parameter diagram of the system architecture model, define reliability constraint parameters through constraint blocks, and associate the specified performance parameters of the equipment as numerical attributes with the parameters of the constraint blocks. Step S2: Characterization and extraction of failure physical mechanisms. For the target subsystem of the equipment, identify its corresponding failure physical processes, and extract the partial differential equations describing the failure physical processes, as well as the corresponding boundary conditions and initial conditions; Step S3: Construction and training of the Physical Information Neural Network (PINN) surrogate model. A multilayer perceptron neural network containing an input layer, hidden layers, and an output layer is constructed. An automatic differentiation module is used to construct the physical constraint module; the total loss function L is defined. total Including data-driven loss term L data And physical information constraint loss term L phys By minimizing L total Optimize network weights; Step S4: Real-time data mapping between the architecture model and the physical agent model. A mapping channel is established between the MBSE model output port and the PINNs model input layer through a data interface protocol. Feature preprocessing and dimension alignment are performed on real-time operating condition variables to generate normalized input vectors. Step S5: Dynamic Reliability Assessment and Requirement Status Feedback. Real-time collected equipment operation data is input into the trained PINNs proxy model to obtain predicted physical performance degradation values; a pre-set reliability probability model is used to calculate real-time reliability indices, and these indices are fed back into the requirement verification matrix of the MBSE system architecture model, automatically updating the verification status identifier of reliability requirements; when the verification status identifier indicates failure, an alarm signal is triggered or an equipment operation control command is generated.

2. The method according to claim 1, characterized in that: In step S3, the physical information constraint loss term L physics The residual function is constructed based on the partial differential equation extracted in step S2; the physical constraint module uses the automatic differential calculation network to output the partial derivative with respect to the input coordinates and substitutes it into the partial differential equation to calculate the physical residual value.

3. The method according to claim 1, characterized in that: In step S4, the feature preprocessing and dimensional alignment specifically include: using principal component analysis or an autoencoder to reduce the dimensionality of the multidimensional working condition space output by the MBSE model, extracting feature operators related to the failure physical process; and using normalization to adapt the data distribution to the sensitive range of the PINNs activation function.

4. The method according to claim 1, characterized in that: In step S5, the method for converting the predicted value into a real-time reliability index is as follows: based on the predicted value and its uncertainty distribution output by PINNs in step S3, a probability density function PDF about the component performance is generated by combining Monte Carlo sampling, and the instantaneous reliability R(t) is obtained by calculating the probability integral of the performance index falling outside the failure threshold.

5. The method according to claim 1, characterized in that: The MBSE system architecture model includes a requirement verification matrix RVTM, which receives the real-time reliability index from step S5 through a logic judgment unit. The triggering of the alarm signal or the generation of equipment operation control instructions specifically includes: when the verification status identifier is set to "failed", triggering the emergency state machine of the system architecture model and outputting control instructions to reduce the load level or shut down for maintenance.

6. The method according to claim 1, characterized in that: The partial differential equations in step S2 encompass one or more coupled forms of fatigue damage equations in structural mechanics, heat conduction equations in thermodynamics, and corrosion rate equations in fluid mechanics.

7. The method according to claim 1, characterized in that: In step S3, the training process of the PINNs proxy model introduces an adaptive weight coefficient λ and a dynamic balancing L. data With L phys The adaptive weight coefficient λ is updated based on the statistical characteristics of the physical residual gradient during training.

8. A real-time reliability assessment system for complex equipment, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program; When the processor executes the computer program, it implements the method steps as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method as described in any one of claims 1-7.

Images