Intelligent system failure analysis method based on fault identification

Abstract

The invention discloses an intelligent system failure analysis method based on fault recognition, and relates to the technical field of intelligent system reliability analysis, and the analysis method comprises the specific steps: employing Modelica to construct a multi-physical domain digital twinborn model, combing historical data, and building a structured fault mode library; developing a fault injection algorithm to acquire data to form a standardized data set; preprocessing data to construct a causal link and a probability flow network; carding task scenes to construct a reliability evaluation model and training the reliability evaluation model; inputting actual parameters to trace faults, generating an optimization plan, and outputting an analysis report and a plan after verification; according to the method, the problem of insufficient fault correlation analysis caused by modeling scale splitting in a traditional method is solved by constructing a multi-physical domain digital twinborn model covering microcosmic, mesoscopic and macroscopic and establishing a structured fault mode library of a standardized feature description system containing three types of faults of hardware, software and environment interaction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent system reliability analysis, in particular to an intelligent system failure analysis method based on fault identification. BACKGROUND

[0002] With the deep application of intelligent systems in aerospace, industrial automation, and autonomous driving high safety critical fields, their reliability is directly related to mission success and personnel safety. Such systems are usually composed of electronic components, functional modules, and environmental interaction interfaces, involving electromagnetic, thermodynamic, mechanical, and other multi-physical domain coupling, and need to run continuously in dynamic task scenarios. The fault mode presents a cross-scale propagation characteristic: microscopic transistor parameter degradation may cause mesoscopic circuit delay, which in turn leads to macroscopic software process crash or system-level task failure; environmental interaction faults may directly trigger multi-module cascading failure. Traditional analysis methods lack cross-level modeling capability and cannot accurately capture the dynamic coupling effect of faults, leading to a large deviation between failure prediction and actual task, so innovative technology breakthroughs are urgently needed.

[0003] The existing technology has many limitations: first, the modeling scale is fragmented, hardware simulation and software analysis are usually performed independently, ignoring the cross-level interaction of electronic components, functional modules, and whole machine environment, making it difficult to associate microscopic parameter degradation with macroscopic system behavior when tracing faults; second, the analysis method is static, fault trees or Markov models rely on preset logical relationships and cannot dynamically simulate the nonlinear coupling of hardware wear, software anomalies, and environmental parameter changes; third, the evaluation lacks scene adaptability, general models are difficult to cover complex task environments, leading to a disconnect between risk level classification and actual task requirements, and low efficiency of plan optimization. In addition, physical fault injection is costly and has low reproducibility, which cannot cover complex fault scenarios, limiting the comprehensiveness of reliability verification. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide an intelligent system failure analysis method based on fault identification. A multi-physical domain digital twin model is constructed using the Modelica language, historical fault data is sorted to establish a structured fault mode library, a multi-scale fault injection algorithm is developed to collect data, a causal link and probability flow network is constructed, a scenario-based reliability evaluation model is trained based on the multi-scale network, actual scenario parameters are input to trace the fault source, an optimized plan is automatically generated, and an analysis report and executable plan are output after simulation verification, effectively improving the accuracy and efficiency of intelligent system failure analysis.

[0005] To solve the above technical problems, the present application provides the following technical solution: an intelligent system failure analysis method based on fault identification, the specific steps of which are as follows: S100, system modeling and fault library construction: a multi-physical domain digital twin model covering micro, meso, macro and environmental interface is constructed using the Modelica language; historical fault data is sorted out, and a structured fault mode library containing hardware, software and environmental interaction faults is established; S200, fault injection and data collection: based on the digital twin model and the fault mode library, a multi-scale control fault injection algorithm is developed, and faults are injected in the simulation environment through a three-level control logic; hardware, software and task execution data are collected at a frequency of 1 kHz to form a standardized data set; S300, causal link and probability flow network construction: pre-process the collected data, filter and denoise to extract hardware and software failure characteristics; build a causal link network covering micro, meso and macro three layers; and define the fault probability as a flow to build a source node-intermediate node-sink node topology network; S400, reliability evaluation model training: sort out the task scenarios of intelligent systems, analyze the environmental, task and resource constraint characteristics, and build a scenario feature parameter library; based on the multi-scale causal link and probability flow network, a scenario-based reliability evaluation model containing fault probability prediction, task impact assessment and risk level division is constructed, and the model is trained by an adaptive active learning strategy until the prediction error is below the set threshold; S500, failure analysis and plan output: input the actual task scenario parameters into the reliability evaluation model, trace the fault source, calculate the fault mode occurrence probability and diffusion path, grade the risk and identify the key nodes and high-risk modes, automatically generate multiple sets of optimized plans, and output the system failure analysis report and executable plans after simulation verification by the digital twin model.

[0006] Further, in the S100, system modeling and fault library construction, the multi-physical domain digital twin model is refined into a three-level structure: the micro level includes transistor, resistor and capacitor electronic element models, among which the transistor model accurately depicts the carrier mobility, channel length modulation effect and temperature sensitivity parameters; the meso level includes sensor, actuator and control board function module models, the sensor model integrates noise characteristics, response delay and time drift parameters, and the actuator model includes transmission efficiency, hysteresis characteristics and load characteristic parameters; the macro level includes whole machine structure models and environmental interaction interface models, the whole machine structure model covers the stiffness of inter-module mechanical connection and vibration transmission characteristics, and the environmental interface model includes the action boundary conditions of temperature field, radiation field and electromagnetic interference field; the models at each level realize data interaction through parameterized interfaces, the micro level outputs component performance degradation data to the meso level, the meso level outputs module cooperative working state data to the macro level, and the macro level feeds back environmental influence parameters to each level.

[0007] Further, in the system modeling and fault library construction, the structured fault mode library establishes a complete feature description system for each type of fault: the hardware fault records physical damage features, parameter deviation range and failure precursor; the software fault contains code abnormal position, data abnormal feature and running environment parameter; the environmental interaction fault clearly shows the mutation amplitude of environmental parameter, the action time and the response state of system protection mechanism; each fault mode is associated with the corresponding fault tree analysis diagram, which contains the logical relationship of basic events, intermediate events and top events, and all fault information is defined according to the unified data dictionary standard.

[0008] Further, in the fault injection and data collection, the fault injection algorithm contains three-level control logic: the micro-level control logic is for transistor-level faults, which configures threshold voltage drift, leakage current increment and fault trigger time, and realizes fault injection by modifying the electrical parameters of corresponding elements in the Modelica model; the meso-level control logic is for functional unit faults, which contains sensor data drift, actuator response delay and communication module packet loss rate, and realizes it by inserting interference signals at module interaction interfaces; the macro-level control logic is for system-level faults, which sets environmental parameter mutation and multi-module collaborative failure timing, and realizes it by adjusting the environmental boundary conditions and module collaboration rules of the digital twin model.

[0009] Further, in the causal link and probability flow network construction, the specific content of the causal link network covering micro, meso and macro three layers is: the micro layer focuses on the conduction relationship between transistor-level hardware wear and circuit logic error, and calculates the influence coefficient of electron mobility attenuation on circuit delay through the fault physical model; the meso layer analyzes the correlation law between circuit module fault and software process abnormality, and establishes the mapping relationship between hardware error injection amount and software crash probability; the macro layer quantifies the conduction path of environmental parameter change and system-level fault, and clearly shows the contribution of temperature gradient change to multi-module collaborative error, the three-layer network realizes dynamic coupling through cross-scale interface parameters, and the causal strength of each layer is quantified by the multi-scale causal strength quantification algorithm, whose calculation formula is: wherein, represents the micro, meso or macro scale the lower fault from node the causal strength of the node to node the feature variable the mutual information of and the information entropy the scale the physical or logical distance between nodes The scale attenuation coefficient is determined by statistical analysis of historical fault data based on the fault propagation characteristics of different levels of the intelligent system. It is set to 0.8 for microscale, 0.5 for mesoscale, and 0.3 for macroscale.

[0010] Furthermore, in S300, the causal link and probabilistic flow network construction defines the failure probability as a flow. The specific steps for building the source node-intermediate node-sink node topology network are as follows: Classifying network node types: the source node is the initial failure occurrence unit, the intermediate node is the failure transmission carrier, and the sink node is the final failure state; Constructing topology connections: the source node and intermediate nodes are connected through hardware interfaces or software links, intermediate nodes are connected according to module interaction protocols, and intermediate nodes and sink nodes are mapped according to failure consequence association rules; Simulating the dynamic transmission process using probabilistic flow transmission equations; Quantifying failure characteristics based on a nonlinear propagation model; Optimizing parameters through network iterative training, and outputting a probability distribution containing the association probabilities between each failure mode. Matrix and failure propagation range curves at different times, This represents the total number of failure modes.

[0011] Furthermore, in S300, the calculation formula for the probability flow transfer equation in the construction of the causal link and probability flow network is as follows: ,in, For a moment The failure probability flow of module m in task phase k. Let be the failure probability flow of module m at time t during task phase k. The self-holding coefficient is determined by the proportion of persistent faults in the module's historical fault data, and its value ranges from [value missing]. , To transfer gain, Let m be the set of upstream associations of module m. To transmit the efficiency matrix, It is a time-delay correction function. It is a module Modules in the upstream associated set.

[0012] Furthermore, in S300, the calculation formula for the nonlinear propagation model in the construction of causal links and probabilistic flow networks is as follows: ,in, for The extent of failure propagation at any given moment. The initial fault influence domain is determined based on the physical or logical range of the initial fault occurrence unit. It is time. At the diffusion threshold moment, the fault influence range first exceeds [a certain threshold] in the fault injection simulation. The statistical determination was made at twice the time point. The diffusion rate coefficient is determined by statistical analysis of the time it takes for a fault to expand from its initial influence area to its maximum extent, based on historical data. The scale contribution factor is determined based on the weighted contribution of different scales to failure diffusion. For scale Maximum causal strength It is the total number of scales; For specific scales.

[0013] Furthermore, in the S400 reliability assessment model training, the scenario-based reliability assessment model includes: an input layer, a core layer, and an output layer; the input layer receives environmental characteristic parameters, task characteristic parameters, and resource constraint parameters; the core layer connects a fault probability prediction module, a task impact assessment module, and a risk level classification module, and outputs a fault probability time series curve, task indicator impact value, system risk level, and a list of high-risk modules; the output layer generates four types of result data.

[0014] Furthermore, in S400, the specific steps for training the reliability assessment model using an adaptive active learning strategy are as follows: (1) Randomly select 10% of the samples from the scene feature parameter library as the initial training set. The samples cover the full range of values ​​of environment, task and resource constraint features. Input the samples into the scene-based reliability assessment model for initial training to obtain the initial model parameters. (2) Calculate the prediction error of the initial model on the validation set, and identify the sample point with the highest information gain by using the prediction variance index; (3) The identified high information gain samples are added to the training set, the model is retrained and the parameters are updated. At the same time, the generalization ability of the model is evaluated and the prediction error in each scenario is recorded. (4) Repeat steps (2)-(3), adding 5% of the training set sample size in each iteration, until the average prediction error of the model in typical scenarios is less than 5% in three consecutive iterations; (5) Save the final training model parameters and generate a model accuracy report. The report includes the prediction error distribution and maximum error value for each scene type.

[0015] Compared with existing technologies, this intelligent system failure analysis method based on fault identification has the following advantages: I. The present application overcomes the problem of insufficient fault correlation analysis caused by modeling scale fragmentation in traditional methods by constructing a multi-physical domain digital twin model covering micro, meso and macro scales, and establishing a structured fault mode library containing standardized feature description systems of three types of faults, i.e. hardware, software and environmental interaction; combined with a fault injection algorithm with micro, meso and macro three-level control logic, it can simulate complex composite fault scenarios from transistor level parameter drift, module function abnormality to environmental mutation induced system level failure in a simulation environment with high fidelity, providing high precision data basis and model support for failure mechanism research and fault source tracing.

[0016] II. The dynamic causal link and probability flow network constructed by the cross-scale causal strength quantification algorithm and the probability flow transfer equation can capture the nonlinear coupling effect between hardware performance degradation, software process abnormality and environmental parameter disturbance in real time, dynamically predict the flow path of fault probability and the range of failure propagation, and is superior to the static analysis method relying on preset logic; the scenario-based reliability evaluation model constructed based on this network and combined with the task scene feature library can automatically evaluate risks, identify key nodes and high-risk modes, and quickly generate multiple sets of optimized plans verified by digital twin model simulation for specific task environments through adaptive active learning strategy training, thereby greatly shortening the system failure analysis response time and enhancing the task reliability and fault tolerance capability of intelligent systems in complex dynamic environments.

[0017] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, in some degree of respect, based on the study of the following, or can be taught from the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creating any creative labor.

[0019] Figure 1 The total flow chart of intelligent system failure analysis method based on fault identification; Figure 2 The hierarchical structure of multi-physical domain digital twin model and the correlation diagram of fault mode library; Figure 3 The data interaction and output relationship diagram of each link of intelligent system failure analysis. DETAILED DESCRIPTION

[0020] To further illustrate the technical means and effects taken by the present application to achieve the predetermined object of the invention, the specific embodiments, structures, features and effects according to the present application are described in detail as follows in combination with the drawings and preferred embodiments.

[0021] Embodiment one: Implementation of intelligent system failure analysis method based on fault identification in space probe.

[0022] A multi-physics domain digital twin model covering the micro, meso, macro and environmental interfaces of the space probe is constructed. The micro layer includes models of transistors, resistors, capacitors and other electronic components inside the probe, which accurately reflect the carrier mobility of transistors and the basic electrical properties of resistors and capacitors. The meso layer covers models of functional modules such as on-board sensors, propulsion system actuators and control board cards. The sensor model integrates the noise and response delay characteristics in the space environment, and the actuator model reflects the transmission efficiency and load characteristics of the propeller. The macro layer includes a probe structure model and an interface model interacting with the space environment. The structure model involves the mechanical connection stiffness and vibration transmission between modules, and the environmental interface model includes the temperature field, radiation field and electromagnetic interference field boundary conditions in space. The models at different levels realize data interaction through parameterized interfaces. The micro layer transmits component performance degradation data to the meso layer, the meso layer feeds back the module cooperative working state to the macro layer, and the macro layer feeds back the influence parameters of the space environment to each level.

[0023] At the same time, the historical fault data of the space probe is sorted out, and a structured fault mode library is established. The hardware fault records the physical damage characteristics of the solar cell panel, the parameter deviation range of the attitude control system and the failure precursor. The software fault includes the abnormal position of the on-board computer code, the data processing abnormal characteristics and the running environment parameters. The environmental interaction fault specifies the amplitude of the sudden temperature change in space, the duration of cosmic radiation and the response state of the probe protection mechanism. Each fault mode is associated with a corresponding fault tree analysis diagram, which includes the logical relationship of basic events, intermediate events and top events. All fault information is standardized and defined according to a unified data dictionary.

[0024] Based on the constructed digital twin model and the fault mode library, a multi-scale control fault injection algorithm is developed to inject faults in the simulation environment through a three-level control logic. The micro-level control logic targets transistor-level faults and configures relevant fault parameters to achieve fault injection by modifying the electrical parameters of the corresponding elements in the digital twin model, such as simulating the abnormal performance of a transistor under radiation influence. The meso-level control logic targets functional unit faults, such as causing data drift in a satellite-borne sensor, response delay in a propulsion system actuator, and packet loss in a communication module, by inserting interference signals at the module interaction interface. The macro-level control logic targets system-level faults by setting sudden changes in space environment parameters (such as sudden temperature rise and fall) and multi-module coordinated failure timing (such as abnormal coordination between attitude control systems and propulsion systems), and achieves this by adjusting the environmental boundary conditions and module coordination rules of the digital twin model.

[0025] The data of the detector hardware (such as power supply data of solar panels, thrust data of thrusters), software (such as running data of satellite-borne computers, data processing progress data), and task execution (such as orbit adjustment accuracy, scientific exploration task completion) are collected at a frequency of 1 kHz to form a standardized data set.

[0026] The collected data set is preprocessed after filtering and denoising, and the failure characteristics of hardware and software are extracted, such as unstable power supply characteristics of solar panels and data processing error characteristics of satellite-borne software. A three-layer causal link network is constructed, focusing on the conduction relationship between transistor-level hardware wear and circuit logic errors at the micro-level, analyzing the influence of electronic component performance changes on circuit operation; analyzing the correlation between circuit module faults and software process abnormalities at the meso-level, such as the relationship between sensor faults and satellite-borne software data processing abnormalities; quantifying the transmission path of space environment parameter changes and system-level faults at the macro-level, such as the influence degree of temperature changes on the coordination error of the probe attitude control system and the propulsion system. The three-layer network is dynamically coupled through cross-scale interface parameters, and the causal strength of each layer is quantified by a multi-scale causal strength quantification algorithm. The calculation formula of the multi-scale causal strength quantification algorithm is: wherein, represents the micro-, meso-, or macro-scale the lower fault node the causal strength from the lower fault node to the upper node the feature variable the mutual information of the information entropy the physical or logical distance between nodes at the scale ​​​The scale attenuation coefficient is determined by statistical analysis of historical fault data based on the fault propagation characteristics of different levels of the intelligent system. It is set to 0.8 for microscale, 0.5 for mesoscale, and 0.3 for macroscale.

[0027] Failure probability is defined as a flow. A source node-intermediate node-sink node topology network is constructed, and network node types are defined: the source node is the initial failure unit (such as a faulty transistor), intermediate nodes are the failure propagation carriers (such as related circuit modules or software processes), and the sink node is the final failure state (such as probe attitude loss). Topological connections are constructed: source nodes and intermediate nodes are connected through hardware interfaces or software links; intermediate nodes are connected according to module interaction protocols; and intermediate nodes and sink nodes are mapped according to failure consequence association rules. The dynamic propagation process of failure probability is simulated using a probability flow propagation equation. The calculation formula for the probability flow propagation equation is as follows: ,in, For a moment The failure probability flow of module m in task phase k. Let be the failure probability flow of module m at time t during task phase k. The self-holding coefficient is determined by the proportion of persistent faults in the module's historical fault data, and its value ranges from [value missing]. , To transfer gain, Let m be the set of upstream associations of module m. To transmit the efficiency matrix, It is a time-delay correction function. It is a module Modules in the upstream associated set; failure characteristics are quantified based on a nonlinear propagation model, the calculation formula of which is: ,in, for The extent of failure propagation at any given moment. The initial fault influence domain is determined based on the physical or logical range of the initial fault occurrence unit. It is time. At the diffusion threshold moment, the fault influence range first exceeds [a certain threshold] in the fault injection simulation. The statistical determination was made at twice the time point. The diffusion rate coefficient is determined by statistical analysis of the time it takes for a fault to expand from its initial influence area to its maximum extent, based on historical data. The scale contribution factor is determined based on the weighted contribution of different scales to failure diffusion. For scale Maximum causal strength It is the total number of scales; For specific dimensions; through network iteration training optimization parameters, output contains the matrix of each failure mode correlation probability and the failure diffusion range curve at different time.

[0028] The task scenarios of the space probe are combed, such as near-earth orbit detection, deep space exploration, rendezvous and docking, the environmental characteristics in different scenarios (such as the difference in radiation intensity between near-earth orbit and deep space), the task characteristics (such as the accuracy requirement of the detection task), and the resource constraint characteristics (such as fuel reserves and power supply) are analyzed, and a scenario characteristic parameter library is constructed.

[0029] Based on the multi-scale causal link and the probability flow network, a scenario-based reliability evaluation model containing failure probability prediction, task impact evaluation and risk level division is constructed, an adaptive active learning strategy is used to train the model, 10% of the samples in the scenario characteristic parameter library are randomly selected as the initial training set, the initial model parameters are obtained by inputting the model for initial training, the prediction error of the initial model on the validation set is calculated, the sample point with the highest information gain is identified through the prediction variance index, the sample with high information gain is supplemented to the training set, the model is retrained and the parameters are updated, and the generalization ability of the model is evaluated, the above process is repeated, and the sample amount of the training set is supplemented by 5% each time, until the average prediction error of the model in the typical scenario is less than the set threshold value in the continuous 3 iterations, and the model training is completed.

[0030] The actual task scenario parameters of the space probe (such as the orbit parameters of a deep space exploration task, the target star environment parameters and the task execution steps) are input into the reliability evaluation model, the model traces the fault source, such as determining that the communication interruption of a certain probe is caused by a communication module transistor affected by radiation, calculates the occurrence probability and diffusion path of the failure mode, such as the failure may spread to the data transmission system and the attitude control system, and evaluates the risk level, and identifies the key nodes (such as the communication relay module) and the high-risk mode (such as large-scale failure of the circuit under long-term strong radiation).

[0031] Multiple sets of optimized plans are automatically generated, such as adjusting the communication frequency to avoid the strong radiation period and starting the standby communication module, the digital twin model is used to simulate and verify these plans, the running state of the probe under different plans is simulated, and finally the system failure analysis report and executable plans are output, which provides support for the reliable operation of the space probe.

[0032] In summary, in the reliability analysis of the space probe, a multi-physical domain digital twin model and a structured fault library are constructed, faults are injected and data are collected by using a three-level control logic, a three-layer causal link and a probability flow network are built, a scenario-based reliability evaluation model is trained by combining an adaptive active learning strategy, and finally the fault source can be traced, the probability and the diffusion path can be calculated, the failure analysis report and the executable plan can be generated after simulation verification, and the reliability analysis and response ability of the space probe in the complex space environment are comprehensively improved.

[0033] Example Two: Failure analysis of spacecraft intelligent control system.

[0034] A multi-physical domain digital twin model of the spacecraft intelligent control system is constructed using the Modelica language, which is refined into a three-level structure. The microscopic layer includes transistor, resistor, capacitor, and other electronic component models in the spacecraft control circuit. The mesoscopic layer covers functional module models such as on-board sensors, thruster actuators, and central control boards. The sensor model integrates noise, response delay, and other characteristics, and the actuator model includes transmission efficiency and load characteristics. The macroscopic layer includes spacecraft whole machine structure models and space environment interaction interface models. The environment interface model covers the boundary conditions of temperature field, radiation field, and electromagnetic interference field. Each level interacts through parameterized interfaces. The microscopic layer outputs component performance degradation data to the mesoscopic layer, the mesoscopic layer feeds back module cooperation status to the macroscopic layer, and the macroscopic layer transmits space environment influence parameters to each level, such as Figure 3

[0035] At the same time, historical spacecraft failure data is sorted out to establish a structured failure mode library. Hardware failure records physical damage of circuit boards and parameter deviation of components. Software failure includes abnormal code position of control program, data transmission abnormality characteristics, and running environment parameters. Environmental interaction failure specifies the action time of space radiation intensity mutation, temperature sudden change, and system protection mechanism response state. Each failure mode is associated with a corresponding fault tree analysis diagram, and all information is standardized according to a unified data dictionary.

[0036] Based on the above digital twin model and failure mode library, a multi-scale control fault injection algorithm is developed. Faults are injected through three-level control logic. The microscopic layer injects faults into transistors by modifying the electrical parameters of transistors in the Modelica model. The mesoscopic layer injects data drift faults into on-board sensors and response delay faults into thruster actuators by inserting interference signals into the module interaction interface. The macroscopic layer simulates system-level faults such as space temperature sudden drop and multi-module cooperative failure by adjusting the environmental boundary conditions and module cooperation rules of the digital twin model.

[0037] Spacecraft hardware operation parameters, software program logs, and task execution state data are collected at a frequency of 1 kHz to form a standardized data set.

[0038] ​The collected data is pre-processed, and the hardware and software failure features are extracted after filtering and denoising. A three-layer causal link network is constructed: the micro layer analyzes the conduction relationship between transistor wear and circuit logic error; the middle layer studies the correlation between sensor failure and control software process anomaly; the macro layer quantifies the conduction path of spatial environment parameter changes and system-level failure. The three-layer network is dynamically coupled through cross-scale interface parameters, and the causal strength of each layer is calculated using a multi-scale causal strength quantification algorithm. The calculation formula of the multi-scale causal strength quantification algorithm is: .

[0039] The failure probability is defined as a flow, and a source node-intermediate node-sink node topology network is built. The source node is the initial failure of electronic components or modules, the intermediate node is the carrier of failure transmission (such as communication link, software process), and the sink node is the final failure state (such as attitude control failure). The topological connection relationship between nodes is constructed, and the dynamic transmission process is simulated using a probability flow transmission equation. The calculation formula of the probability flow transmission equation is: The failure features are quantified by a nonlinear propagation model, and the calculation formula of the nonlinear propagation model is: After network iteration training, the failure mode correlation probability matrix and failure propagation range curve are output.

[0040] The mission scenarios of spacecraft (such as orbit maintenance, attitude adjustment, and payload deployment) are sorted out, and the characteristics of spatial environment (such as radiation intensity and temperature), mission objectives (such as orbit accuracy and attitude stability), and resource constraints (such as energy and computing resources) are analyzed under each scenario. A scenario feature parameter library is constructed.

[0041] Based on the multi-scale causal link and probability flow network, a scenario-based reliability evaluation model is constructed. This model includes failure probability prediction, task impact assessment, and risk level division modules. An adaptive active learning strategy is used to train the model: an initial training set is selected from the scenario feature parameter library for initial training, high information gain samples are identified by prediction variance and supplemented to the training set, the training is repeated and the generalization ability is evaluated by 5-fold cross-validation, until the average prediction error of continuous 3 iterations is lower than the set threshold, the model parameters are saved and the precision report is generated.

[0042] The actual scenario parameters of the spacecraft "orbit maintenance" task (such as current orbit position, spatial radiation intensity, and energy remaining) are input into the reliability evaluation model, the failure source is traced, the occurrence probability and diffusion path of each type of failure mode are calculated, the risk level is divided, and the key nodes (such as the thruster control module) and high-risk modes (such as sensor data drift) are identified.

[0043] The model automatically generates multiple sets of optimization plans (such as switching to backup sensors and adjusting thruster operation mode), which are verified by digital twin simulation, and outputs system failure analysis report and executable plans.

[0044] To sum up, the embodiment is directed to a spacecraft intelligent control system, through multi-physical domain digital twin model construction, the microscopic, macroscopic and environmental interface are comprehensively covered, combined with the structured fault mode library to lay the foundation for fault analysis, the standardized data is obtained through fault injection and data acquisition, the fault conduction law is determined by means of causal link and probability flow network, the prediction accuracy is improved through the scene reliability evaluation model training, and finally the failure analysis report and executable plan are output, the whole process relies on multi-scale causal strength quantitative algorithm, probability flow transfer equation and other tools, and the failure tracing and risk control in the complex scene of the spacecraft are effectively realized.

[0045] The above is only a preferred embodiment of the present application, not any form of limitation on the present application, although the present application has been disclosed as above with a preferred embodiment, however, it is not intended to limit the present application, any person skilled in the art, without departing from the scope of the technical scheme of the present application, can make some changes or modifications to the above disclosed technical content to make equivalent embodiments with equivalent changes, but as long as it does not deviate from the technical scheme content of the present application, any brief introduction, modification, equivalent change and modification of the above embodiments according to the technical essence of the present application, all still belong to the scope of the technical scheme of the present application.

Claims

1. A failure analysis method of an intelligent system based on fault identification, characterized by, The specific steps of the analysis method are: S100, system modeling and fault library construction: a multi-physical domain digital twin model covering micro, meso, macro and environment interface is constructed using Modelica language; historical fault data is sorted out to establish a structured fault mode library containing hardware, software and environment interaction faults; S200, fault injection and data collection: based on the digital twin model and fault mode library, a multi-scale control fault injection algorithm is developed, and faults are injected in the simulation environment through a three-level control logic; hardware, software and task execution data are collected at a frequency of 1 kHz to form a standardized data set; S300, causal link and probability flow network construction: after preprocessing and denoising the collected data, the hardware and software failure characteristics are extracted; a causal link network covering micro, meso and macro three layers is constructed; And define the fault probability as a flow, build a source node-intermediate node-sink node topology network; S400, reliability evaluation model training: sort out the task scenarios of intelligent systems, analyze the environmental, task and resource constraint characteristics, and construct a scene characteristic parameter library; based on the multi-scale causal link and probability flow network, a scenario-based reliability evaluation model containing fault probability prediction, task impact assessment and risk level division is constructed, which is trained by adaptive active learning strategy until the prediction error is below the set threshold; S500, failure analysis and plan output: input the actual task scenario parameters into the reliability evaluation model, trace the fault source, calculate the fault mode occurrence probability and diffusion path, grade the risk and identify the key nodes and high-risk modes, automatically generate multiple sets of optimization plans, and output the system failure analysis report and executable plans after simulation verification by the digital twin model.

2. The intelligent system failure analysis method based on fault identification according to claim 1, characterized in that, In the S100, system modeling and fault library construction, the multi-physical domain digital twin model is refined into a three-level structure: the micro level includes transistor, resistor, and capacitor electronic element models, among which the transistor model accurately depicts the carrier mobility, channel length modulation effect, and temperature sensitivity parameters; the meso level includes sensor, actuator, and control board function module models, the sensor model integrates noise characteristics, response delay, and time drift parameters, and the actuator model includes transmission efficiency, hysteresis characteristics, and load characteristic parameters; the macro level includes whole machine structure model and environment interaction interface model, the whole machine structure model covers module mechanical connection stiffness, vibration transmission characteristics, and the environment interface model includes temperature field, radiation field, and electromagnetic interference field boundary conditions; Each level model realizes data interaction through parameterized interface, the micro level outputs element performance degradation data to the meso level, the meso level outputs module cooperative working state data to the macro level, and the macro level feeds back environmental influence parameters to each level.

3. The method of claim 1, wherein, The S100, in system modeling and fault library construction, the structured fault mode library establishes a complete feature description system for each type of fault: the hardware fault record physical damage characteristics, parameter deviation range and failure precursor; the software fault contains code exception position, data anomaly characteristics and running environment parameters; the environmental interaction fault clearly shows the mutation amplitude of the environmental parameter, the action time and the response state of the system protection mechanism; each fault mode is associated with a corresponding fault tree analysis diagram, which contains the logical relationship of basic events, intermediate events and top events, and all fault information is standardized and defined according to a unified data dictionary.

4. The method of claim 1, wherein, The S200, in fault injection and data collection, the fault injection algorithm contains three-level control logic: the microcosmic layer control logic is for transistor-level faults, which configures threshold voltage drift, leakage current increment and fault trigger time, and realizes fault injection by modifying the electrical parameters of the corresponding elements in the Modelica model; the mesoscopic layer control logic is for functional unit faults, which contains sensor data drift, actuator response delay and communication module packet loss rate, and is realized by inserting interference signals at the module interaction interface; the macroscopic layer control logic is for system-level faults, which sets the environmental parameter mutation and multi-module cooperative failure timing, and is realized by adjusting the environmental boundary conditions and module cooperation rules of the digital twin model.

5. The method of claim 1, wherein the method further comprises: The S300, in the construction of the causal link and the probability flow network, the specific content of the three-layer causal link network of micro, meso and macro is: the micro layer focuses on the conduction relationship between transistor level hardware wear and circuit logic error, and calculates the influence coefficient of electron mobility attenuation on circuit delay through the fault physical model; the meso layer analyzes the correlation law of circuit module fault and software process anomaly, and establishes the mapping relationship between hardware error injection quantity and software crash probability; the macro layer quantifies the conduction path of environmental parameter change and system level fault, and determines the contribution degree of temperature gradient change to multi-module collaborative error, the three-layer network realizes dynamic coupling through cross-scale interface parameters, and the causal strength of each layer is quantified by multi-scale causal strength quantification algorithm, the calculation formula is: Wherein, represents the micro, meso or macro scale the causal strength of the lower fault from node to node is the mutual information of characteristic variables and is the information entropy, is the distance between nodes in scale is the scale attenuation coefficient, based on the fault propagation characteristics of different levels of intelligent system, it is determined by statistical analysis of historical fault data, the micro is 0.8, the meso is 0.5, and the macro is 0.3.​​​ 6. The method of claim 1, wherein, S300, in the construction of causal link and probability flow network, the fault probability is defined as flow, and the specific steps of building the source node-intermediate node-sink node topology network are: dividing network node types: the source node is the initial fault occurrence unit, the intermediate node is the fault transmission carrier, and the sink node is the final failure state; building topology connection relationship: the source node and the intermediate node are connected through hardware interface or software link, the intermediate nodes are connected according to module interaction protocol, and the intermediate node and the sink node are mapped according to failure consequence correlation rules; the dynamic transmission process is simulated by using probability flow transmission equation; the failure characteristics are quantified based on nonlinear propagation model; the parameters are optimized through network iteration training, and the matrix and failure diffusion range curve at different times, is the total number of fault modes.

7. The method of claim 6, wherein the method further comprises: The S300, in the causal link and the probability flow network construction, the calculation formula of the probability flow transfer equation is: Wherein, is the failure probability flow of module m at time t, is the failure probability flow of module m at task stage k, is the failure probability flow of module m at time t, is the self-maintaining coefficient, which is determined by the proportion of the failure persistence in the historical failure data of the module, and the value range is , is the transfer gain, is the upstream associated set of module m, is the transfer efficiency matrix, is the time delay correction function, is the module in the upstream associated set of module m.

8. The intelligent system failure analysis method based on fault identification according to claim 6, characterized in that, The S300, in the causal link and probability flow network construction, the calculation formula of the nonlinear propagation model is: Wherein, is the failure diffusion range at time t, is the initial failure influence domain, which is determined according to the physical or logical range of the initial failure occurrence unit, is time, is the diffusion threshold time, which is determined by the time point at which the failure influence range first exceeds 2 times of the initial failure influence domain in the fault injection simulation, is the diffusion rate coefficient, which is determined by the time statistics of the fault spreading from the initial influence domain to the maximum range in the historical data, is the scale contribution factor, which is determined based on the contribution weight of different scales to failure diffusion, is the maximum causal strength of the scale is the total number of scales; is a specific scale.​​​ 9. The failure analysis method of claim 1, wherein, The S400, in reliability evaluation model training, the scenario-based reliability evaluation model includes: input layer, core layer and output layer; the input layer: receives environmental feature parameters, task feature parameters and resource constraint parameters; the core layer: connects fault probability prediction module, task influence evaluation module and risk level division module, outputs fault probability time curve, task index influence value, system risk level and high-risk module list; the output layer: generates four types of result data.

10. The intelligent system failure analysis method based on fault identification according to claim 1, characterized in that, The S400, in reliability evaluation model training, the specific steps of training the model using the adaptive active learning strategy are: (1) randomly select 10% of the samples from the scene feature parameter library as the initial training set, which covers the full value range of environmental, task and resource constraint features, input the samples into the scenario-based reliability evaluation model for initial training, and obtain the initial model parameters; (2) calculate the prediction error of the initial model on the validation set, identify the sample point with the highest information gain through the prediction variance index; (3) supplement the identified high-information-gain sample to the training set, retrain the model and update the parameters, evaluate the model's generalization ability, and record the prediction error in each scene; (4) repeat steps (2)-(3), supplement 5% of the training set sample size each time, until the average prediction error of the model in the typical scene is less than 5% in the last 3 iterations; (5) save the final trained model parameters, generate a model accuracy report, which contains the prediction error distribution and maximum error value of each scene type.

Images