System fault probability prediction method based on Bayesian network and SPICE

Abstract

The invention provides a system fault probability prediction method based on a Bayesian network and SPICE, and relates to the technical field of aerospace science, and the prediction method comprises the steps: obtaining a Bayesian network probability model and function module performance state evidence data, corresponding to a preset operation environment, of each function module in a target system, inputting each piece of evidence data into a probability model, outputting probability distribution data of each functional module in a fault or in different performance degradation states in a preset operation environment, and converting the probability distribution data into corresponding SPICE circuit simulation parameter disturbance values; each SPICE circuit simulation parameter disturbance value is applied to a corresponding circuit node in the corresponding SPICE circuit simulation model for simulation, and a fault probability prediction result is obtained. According to the method, the Bayesian network is combined with SPICE circuit simulation, effective combination from probabilistic reasoning to physical behavior simulation is achieved, and therefore the overall fault probability of a complex system in an uncertain operation environment can be predicted.

Description

Technical Field

[0001] This invention relates to the field of aerospace science and technology, and more specifically, to a system failure probability prediction method based on Bayesian networks and SPICE. Background Technology

[0002] With the rapid development of commercial spaceflight, especially the rise of low-cost commercial satellites, ensuring the long-term reliable operation of satellite systems in the complex and harsh space environment while strictly controlling development costs and timelines has become one of the core challenges restricting the industry's development. Commercial satellites often use advanced commercial off-the-shelf components to reduce costs, size, and power consumption. However, these components often lack complete space radiation hardening and verification, resulting in significant uncertainties regarding performance degradation and failure risks in space-based radiation environments.

[0003] Currently, the mainstream methods for on-orbit reliability assessment and prediction of satellite electronic systems still have serious shortcomings: Existing simulation methods struggle to quantify system-level failure probabilities. Physics-based circuit simulation, while capable of high-fidelity simulation of a system's electrical behavior under deterministic conditions, is inherently a deterministic tool. It cannot directly address the uncertainties and probabilistic degradation of device performance caused by factors such as radiation and aging. Engineers typically only perform "worst-case" analyses, often leading to overly conservative designs, unnecessary cost and weight increases, and still failing to provide a quantitative probability of system failure.

[0004] Probabilistic reliability models (such as fault trees) can handle uncertainty and describe the logical relationships of component failures. However, these methods typically remain at the level of system functional logic or static reliability block diagrams. They lack the ability to model the dynamic physical processes within the system (such as circuit feedback, load matching, and signal integrity), making it difficult to accurately assess how a component's non-fatal performance degradation (such as a decrease in gain or an increase in leakage current) propagates through circuit interactions and ultimately leads to system-level functional failure. In other words, traditional probabilistic models cannot answer the crucial question of "how probabilistic deviations in component performance lead to probabilistic overperformance of system performance."

[0005] Therefore, in cost-intensive fields such as commercial satellites, there is an urgent need for a reliability method that can, during the design phase, use computational simulation as the primary means to both quantify the uncertainty degradation of devices and accurately simulate the chain effects of this degradation at the system physical level, thereby ultimately predicting the overall failure probability of the system. Summary of the Invention

[0006] The problem solved by this invention is one or more of the aforementioned related technical problems.

[0007] To address the above problems, this invention provides a system fault probability prediction method based on Bayesian networks and SPICE, comprising: Obtain a pre-constructed Bayesian network probability model; the nodes of the Bayesian network probability model correspond to the functional modules of the target system, and the Bayesian network probability model is used to predict the probabilistic dependency between the performance status of each functional module and the environmental factors affecting the performance of the functional module; Obtain performance status evidence data of each functional module corresponding to the preset operating environment, and input the performance status evidence data of each functional module into the Bayesian network probability model, and output the probability distribution data of each functional module failing or being in different performance degradation states under the preset operating environment. The probability distribution data are converted into corresponding SPICE circuit simulation parameter perturbation values. Based on the circuit connection relationship of the target system, a corresponding SPICE circuit simulation model is constructed, and the disturbance values ​​of each SPICE circuit simulation parameter are applied to the corresponding circuit nodes in the SPICE circuit simulation model for simulation, so as to obtain the fault probability prediction result of the target system under the preset operating environment.

[0008] Optionally, the process of constructing the Bayesian network probability model includes: The target system is decomposed into modules, and the functional modular circuit connection relationships are obtained; the functional modular circuit connection relationships are used to characterize the electrical connection relationships between the decomposed functional modules. Based on the functional modular circuit connection relationship, the nodes and topology of the Bayesian network are determined, wherein the nodes correspond to each functional module, and the directed edges between the nodes correspond to the signal or functional dependencies between each functional module. Obtain the test database of each functional module and the preset operating environment data of the target system; Based on the experimental database and the preset operating environment data, a conditional probability table corresponding to each of the functional modules is generated. The conditional probability table is integrated into the topology of the Bayesian network to form the Bayesian network probability model.

[0009] Optionally, the functional module performance status evidence data includes the expected performance degradation state; the functional module performance status evidence data is obtained in the following ways: Based on the test database, the expected performance degradation state of each functional module in the target operating environment is determined according to the target operating environment data.

[0010] Optionally, the process of acquiring the test database includes: A data retrieval framework is constructed based on the device status data of each functional module and the preset test condition data; Based on the data retrieval framework, performance response data of each functional module under various test conditions are obtained from a preset data source, and the performance response data is integrated to form the corresponding test database.

[0011] Optionally, the test condition data includes one or more of the following: irradiation dose, dose rate, bias voltage, temperature, and hydrogen concentration.

[0012] Optionally, the preset operating environment data includes space environment particle data, which includes particle energy spectrum and total particle fluence at different confidence levels.

[0013] Optionally, converting the probability distribution data of the fault or performance degradation states of each functional module into corresponding SPICE circuit simulation parameter perturbation values ​​includes: Based on the probability distribution data, determine the change in performance parameters corresponding to the performance degradation state of each functional module; Based on the preset performance parameter-SPICE parameter mapping relationship, the change in the performance parameter is mapped to the corresponding SPICE circuit simulation parameter perturbation value.

[0014] Optionally, the performance parameter-SPICE parameter mapping relationship includes at least one of the following: the mapping relationship between the module output electrical performance parameters and the SPICE controlled source parameters, and the mapping relationship between the module internal gain and the SPICE device model parameters.

[0015] The beneficial effects of the system failure probability prediction method based on Bayesian networks and SPICE of the present invention are: By transforming the module-level probabilistic fault prediction results output by Bayesian networks into parameter perturbation values ​​identifiable by SPICE circuit simulation and injecting them into the system-level circuit model for simulation, an effective combination of probabilistic reasoning and physical behavior simulation is achieved. This enables quantitative prediction of the overall failure probability of complex systems under uncertain operating environments. This method overcomes the limitations of traditional reliability assessments that rely solely on the accumulation of component failure rates or simulation of fixed failure modes. It can realistically reflect the circuit coupling effects and system-level failure behaviors between the probabilistic performance degradation of multiple modules, significantly improving the accuracy and engineering applicability of predictions. At the same time, this method can replace a large number of high-cost ground tests with simulations during the design phase, supporting rapid iterative reliability assessments of design schemes, component selection, and orbital environments. This provides an efficient and low-cost reliability design and verification method for cost-sensitive fields such as commercial spacecraft.

[0016] To address the above problems, this invention provides a system failure probability prediction device based on Bayesian networks and SPICE, comprising: An acquisition unit is used to acquire a pre-constructed Bayesian network probability model; the nodes of the Bayesian network probability model correspond to the functional modules of the target system, and the Bayesian network probability model is used to predict the probabilistic dependency between the performance status of each functional module and the environmental factors affecting its performance. The acquisition unit is also used to acquire the performance status evidence data of each functional module corresponding to the preset operating environment, and input the performance status evidence data of each functional module into the Bayesian network probability model, and output the probability distribution data of each functional module failing or being in different performance degradation states in the preset operating environment. The processing unit is used to convert the probability distribution data into corresponding SPICE circuit simulation parameter perturbation values. The processing unit is also used to construct a corresponding SPICE circuit simulation model based on the circuit connection relationship of the target system, and to apply the disturbance values ​​of each SPICE circuit simulation parameter to the corresponding circuit node in the SPICE circuit simulation model for simulation, so as to obtain the fault probability prediction result of the target system under the preset operating environment.

[0017] The system failure probability prediction device based on Bayesian networks and SPICE described in this invention has the same advantages over the prior art as the system failure probability prediction method based on Bayesian networks and SPICE, and will not be repeated here.

[0018] To address the aforementioned problems, this invention provides a system fault probability prediction device based on Bayesian networks and SPICE, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the system fault probability prediction method based on Bayesian networks and SPICE.

[0019] The system failure probability prediction device based on Bayesian networks and SPICE described in this invention has the same advantages over the prior art as the system failure probability prediction method based on Bayesian networks and SPICE, and will not be repeated here. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a system failure probability prediction method based on Bayesian networks and SPICE according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the construction process of a Bayesian network probability model according to an embodiment of the present invention. Figure 3 This is a flowchart of a system failure probability prediction method using Bayesian networks and SPICE according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the transformation and injection of circuit parameters in a SPICE simulation according to an embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0022] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0023] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are only used to distinguish different devices, functional modules, or units, and are not intended to limit the order of functions performed by these devices, functional modules, or units or their interdependencies.

[0024] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0025] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0026] Relying on expensive and time-consuming physical testing: To ensure reliability, traditional space missions typically conduct extensive ground-based radiation tests, accelerated life tests, and environmental simulation tests on critical components and even the entire spacecraft. While this method yields intuitive and reliable results, it is extremely costly, time-consuming, and cannot cover all possible on-orbit operating conditions, fundamentally conflicting with the needs of commercial spaceflight for economic efficiency and rapid iteration.

[0027] To address the problems existing in the aforementioned related technologies, embodiments of the present invention provide a system failure probability prediction method based on Bayesian networks and SPICE.

[0028] like Figure 1 As shown in the figure, an embodiment of the present invention provides a system fault probability prediction method based on Bayesian networks and SPICE, comprising: Step S100: Obtain a pre-constructed Bayesian network probability model; the nodes of the Bayesian network probability model correspond to the functional modules of the target system, and the Bayesian network probability model is used to predict the probabilistic dependency between the performance status of each functional module and the environmental factors affecting the performance of the functional module.

[0029] Specifically, step S100 is the starting point of the prediction process of this invention, and its core is the introduction of a Bayesian network probabilistic model specifically customized for the target system. This model is not a general-purpose tool, but a dedicated inference engine pre-built based on the specific architecture of the target system. Its construction idea is to abstract the physical structure of the target system into a probabilistic graphical model.

[0030] Specifically, the target system (such as the photoelectric detection unit of a satellite) is first conceptually decomposed into its core functional modules (e.g., laser source, photoelectric converter, signal amplifier, etc.). In a pre-constructed Bayesian network model, each such functional module is mapped as a node in the network. The connections between these nodes (directed edges) reflect the actual signal flow, functional dependencies, or physical coupling relationships between the modules, thus transforming the physical topology of the system into the topology of a probabilistic network.

[0031] The core function of this model is to encapsulate and quantify the "uncertainty" knowledge regarding the impact of environmental factors on the performance of each module. For example, through learning or predefinition, it has mastered relationships such as "what is the probability that the laser source output power will decrease by 10% when the space radiation dose reaches a certain level." Therefore, when specific environmental conditions (evidence) are obtained, the model can perform rigorous probabilistic reasoning based on its internal conditional probability relationships, outputting the probability that each module is in various performance degradation states, such as "the probability that the amplifier gain will decrease by more than 20% is 15%." This process completes the crucial transformation from deterministic environmental input to a module-level probabilistic state description.

[0032] The direct benefit of acquiring and using the pre-built Bayesian network probability model described in step S100 is that it provides a standardized and computable probabilistic front end for subsequent system-level analysis. It clearly correlates complex and uncertain environmental effects with the states of each module within the system, and outputs the potential risk of each independent module in a rigorous mathematical form (probability distribution). This transforms the continuous and fuzzy performance degradation problem in reality into a discrete state probability problem that can be handled by subsequent deterministic simulation tools. This avoids the problem of unmanageable uncertainty caused by directly inputting environmental physical parameters into the physical simulator, achieving the first information leap from "environment" to "module state probability," and laying an indispensable, quantitative input foundation for subsequent accurate simulations of system-level physical behavior that consider uncertainty.

[0033] Step S200: Obtain performance status evidence data of each functional module corresponding to the preset operating environment, input the performance status evidence data of each functional module into the Bayesian network probability model, and output the probability distribution data of each functional module failing or being in different performance degradation states under the preset operating environment.

[0034] Specifically, this involves acquiring performance status evidence data for functional modules that precisely corresponds to a pre-defined operating environment (e.g., the radiation spectrum, total fluence, and temperature cycles encountered by a three-year low Earth orbit mission). This evidence data is not the raw environmental parameters, but rather a judgment or preliminary mapping result based on domain knowledge of "the most likely direct physical state exhibited by the module under this specific environment." For example, for this orbital environment, the evidence might be stated as "the laser diode is expected to experience a total ionizing dose effect equivalent to X krad(Si) (equivalent to X × 1000 rad of absorbed radiation dose deposited in silicon material, where X is a variable representing a specific value (e.g., 100, 500, etc.))" or "a certain type of bipolar transistor base-emitter junction is expected to be exposed to high dose rate irradiation under Y-bias conditions." This preliminarily interpreted evidence provides clear input conditions for each functional module.

[0035] Subsequently, this specific evidence data is input into the Bayesian network probability model corresponding to the system structure obtained in step S100. Upon receiving the evidence, the model performs global probability propagation and update calculations based on the dependencies encoded in its network topology and the pre-stored conditional probability tables within each node. Its output is no longer a single judgment, but rather a complete probability distribution of each functional module regarding its performance state. For example, the output might indicate: "Under the current orbital environment, the probability of the photoelectric converter efficiency decreasing by more than 15% is 8%, the probability of decreasing by 5%-15% is 25%, and the probability of maintaining normal operation is 67%." This process essentially completes the deterministic reasoning from "environment-evidence" to "module-probability."

[0036] This process achieves a precise transformation from specific scenarios to quantifiable risks and the separation of uncertainties. By converting a deterministic operating environment into a module-level probability distribution output, this method successfully decomposes complex, coupled system reliability problems into a series of clear, quantifiable descriptions of module state uncertainties. This provides crucial input for subsequent steps: no longer the vague "may be affected," but the explicit "at what probability and at what level of degradation." This decomposition and quantification enables traditional circuit simulation tools (SPICE), which deal with deterministic relationships, to systematically handle probabilistic failure problems for the first time. This lays an indispensable and mathematically rigorous data foundation for ultimately conducting system-level reliability simulations that realistically reflect the coupling effects of random failures, thus bridging the traditional gap between probabilistic analysis and physical simulation.

[0037] Step S300: Convert the probability distribution data into corresponding SPICE circuit simulation parameter perturbation values.

[0038] Specifically, probability distribution data is a mathematical expression describing "how likely each functional module is to be at what performance level" (e.g., "the probability of an amplifier gain decreasing by 10% is 30%"). However, standard SPICE circuit simulators cannot directly understand and process this probabilistic language; they can only accept deterministic circuit parameters (such as resistance values, current source strength, and transistor model parameters) for simulation calculations.

[0039] Therefore, the abstract probability distribution is parsed and solidified into a series of specific, executable instruction parameters for the SPICE simulator, i.e., SPICE circuit simulation parameter perturbation values. This process is not a simple mapping, but rather based on a deep understanding of circuit principles. For example, for the probabilistic conclusion that "the output power of the laser source has a 40% probability of decreasing by 15%", the conversion logic will determine its most representative scenario (such as taking the expected value or the most likely state) as "the output current decreases by 15%", and then translate this change in physical quantity into a deterministic operation instruction to connect a controlled current source in parallel with the original laser in the SPICE model and set its output value to the corresponding attenuation amount. In this way, the probabilistic performance degradation expectation of each module is transformed into one or more component parameter modification or additional component insertion instructions with explicit values ​​that can be injected into the SPICE netlist.

[0040] The above process constructs a "bridge" connecting uncertainty analysis and deterministic physical simulation, overcoming the fundamental technical obstacle that probabilistic fault prediction results cannot directly drive high-fidelity circuit system simulation. By parsing and solidifying probability distribution data into specific parameter perturbation values ​​identifiable by SPICE, it enables module-level risks derived from probabilistic reasoning to be "instantiated" as observable, deterministic performance degradation events in circuit simulation. This transformation makes it possible to subsequently use mature SPICE tools for system-level simulation, thereby enabling accurate assessment of the propagation and superposition effects of these probabilistic faults under real circuit interactions.

[0041] Step S400: Based on the circuit connection relationship of the target system, construct a corresponding SPICE circuit simulation model, and apply the disturbance values ​​of each SPICE circuit simulation parameter to the corresponding circuit node in the SPICE circuit simulation model for simulation, so as to obtain the fault probability prediction result of the target system under the preset operating environment.

[0042] Specifically, based on the actual circuit connection relationship of the target system (such as the schematic diagram), a high-fidelity SPICE circuit simulation model is first constructed in the SPICE simulation environment. This model accurately reflects the electrical interconnection and interaction between the functional modules.

[0043] Subsequently, the SPICE circuit simulation parameter perturbation values ​​generated in step S300, representing the potential performance degradation of each module, are precisely applied to the corresponding circuit nodes or device parameters in this simulation model. For example, the perturbation value representing "decreased light source output current" is applied to the control terminal of the equivalent current source of the corresponding laser diode in the model; the perturbation value representing "transistor gain degradation" is reflected in adjusting the current amplification factor (β) parameter in its SPICE model. After these injections are completed, the SPICE simulator is driven to run.

[0044] The simulation is not performed only once, but rather through multiple sampling simulations or Monte Carlo analyses based on various possible scenarios implied by the probability distribution. By observing and statistically analyzing whether key system-level output indicators (such as final output voltage, signal bit error rate, and logic function correctness) exceed acceptable ranges under various combinations of disturbances, the frequency of failure of the entire target system under the preset operating environment is statistically determined, i.e., the quantified failure probability prediction result. For example, if the system output indicators fail to meet the standards 85 times out of 1000 simulations with random disturbances, the predicted failure probability is 8.5%.

[0045] The previous process enabled large-scale, automated testing and quantitative statistical analysis of the system-level cascading effects caused by probabilistic component failures on a precise physical model. By injecting module-level probabilistic perturbations into the system-level SPICE model for simulation, this method can realistically reveal the fault propagation and amplification phenomena caused by inherent physical mechanisms such as circuit coupling, load matching, and feedback loops. This yields a system-level fault probability that is far more realistic than the simple summation of the independent failure rates of individual modules. This allows designers to accurately locate system weaknesses and assess the effectiveness of design redundancy before manufacturing and launch, ultimately making optimal balance decisions between cost and reliability. This significantly reduces the risks and costs associated with traditionally relying on expensive physical system testing and "over-design."

[0046] In this embodiment, the system failure probability prediction method based on Bayesian networks and SPICE effectively combines uncertain probabilistic reasoning with deterministic physical simulation, solving the industry problem of difficulty in quantitatively assessing the overall reliability of complex systems. Traditional methods are either limited to module-level probabilistic analysis (unable to assess system behavior) or can only perform circuit simulations with fixed failure modes (unable to handle probabilistic degradation). This embodiment creatively forms a complete technical chain from "probability" to "physical behavior" to "system probability" by first using Bayesian networks to output module-level probabilistic failure predictions, then converting them into parameter perturbations that can be handled by SPICE simulation and performing system-level simulations. This enables quantitative prediction of the overall failure probability of a system under uncertain environments.

[0047] Furthermore, this method overcomes the one-sidedness of simply summing up module failure rates or making worst-case assumptions for system evaluation. By injecting the probability prediction results into the SPICE model for simulation, it can realistically simulate the system-level failure behavior caused by physical mechanisms such as circuit coupling, load changes, and feedback regulation when multiple modules experience different degrees of performance degradation simultaneously or successively. This allows for the discovery of potential fault propagation paths and system weaknesses, making the prediction results closer to actual on-orbit operation.

[0048] Meanwhile, it provides an efficient and low-cost reliability design iteration tool for cost-sensitive fields such as commercial spacecraft. The core advantage of this method lies in its ability to replace a large number of expensive and time-consuming ground-based radiation tests and system-level tests with simulation calculations during the design phase. Engineers can predict the system reliability performance under different design options and orbital schemes by adjusting the input (environmental data) of the Bayesian network or quickly replacing modules in the SPICE model. This allows for rapid optimization trade-offs between cost, performance, and reliability, significantly shortening the development cycle and reducing overall costs, making it particularly suitable for the development of commercial satellites where strict cost control is essential.

[0049] like Figure 2 As shown, optionally, the process of constructing the Bayesian network probability model includes: Step S101: Decompose the target system into modules and obtain the functional modular circuit connection relationship; the functional modular circuit connection relationship is used to characterize the electrical connection relationship between the decomposed functional modules. Step S102: Based on the functional modular circuit connection relationship, determine the nodes and topology of the Bayesian network, wherein the nodes correspond to each functional module, and the directed edges between the nodes correspond to the signal or functional dependencies between each functional module. Step S103: Obtain the test database of each functional module and the preset operating environment data of the target system; Step S104: Based on the experimental database and the preset operating environment data, generate a conditional probability table corresponding to each of the functional modules; Step S105: Integrate the conditional probability table into the topology of the Bayesian network to form the Bayesian network probability model.

[0050] In some embodiments, the target system (such as the photoelectric detection unit of a satellite) is first functionally decomposed into a series of independent or semi-independent functional modules (e.g., SLD light source, photodiode, transimpedance amplifier, filter circuit, etc.), and the functional modular circuit connections between these modules are clarified (i.e., how signals flow from one module to another). This step transforms the physical system into a modular network.

[0051] Based on the module connection relationships obtained in the previous step, they are mapped to a Bayesian network topology. Each functional module becomes a network node. The signal flow or functional dependency between modules (e.g., the output of a light source is the input of a photodiode, and the output of a photodiode is the input of an amplifier) ​​is abstracted as directed edges between nodes. This directed edge represents a probabilistic dependency or influence relationship (a parent node influences a child node), thus transforming the physical connection graph of the system into a probabilistic dependency graph.

[0052] Building the model requires two types of core data: Test Database: A dedicated "history" database is established for each functional module. This database is built by collecting historical test data, consulting publicly available literature, and conducting device simulations (such as TCAD simulations). The core of the database records the module's performance response (such as parameter degradation curves) under various preset test conditions (such as different irradiation doses and temperatures).

[0053] Preset operating environment data: The specific environmental data that the target system will encounter when running on the predetermined orbit is calculated by analysis software (such as space environment analysis tools), such as the energy spectrum and total flux of space radiation particles.

[0054] Conditional Probability Table (CPT) Generation: For each functional module node, quantitative analysis is performed by combining its experimental database and preset operating environment data. For example, based on the total orbital radiation fluence, the degradation range of module performance parameters (such as gain and dark current) under that fluence is determined by searching or interpolation in the experimental database. Then, this physical degradation range is quantized and discretized into different "states" (such as "normal," "mild degradation," and "severe degradation"), and their probability is evaluated, thereby generating the conditional probability table for that node. CPT defines the probability of the module being in various performance states under the "environmental conditions" (parent node state) it experiences.

[0055] The conditional probability table generated by each module node is then filled into the corresponding node in the previously established Bayesian network topology. At this point, a complete and computable Bayesian network probabilistic model containing system structure information and uncertainty quantification knowledge is constructed. This model encapsulates all the probabilistic rules regarding "how the environment affects modules" and "how modules are related."

[0056] In some specific embodiments, taking the construction of a simplified Bayesian network model of a photodetector system as an example, the construction process of the corresponding Bayesian network probability model includes: Decomposition: The system is decomposed into three modules: SLD light source, PIN photodiode, and transimpedance amplifier. The connection relationship is: SLD light source -> PIN photodiode -> transimpedance amplifier.

[0057] Mapping topology: Establish three nodes, with directed edges from the SLD light source to the PIN photodiode, and then to the transimpedance amplifier.

[0058] Data preparation: Experimental data on the changes in optical power, responsivity, and gain of the three devices under different gamma irradiation doses were collected. The total ionizing dose of the system orbit over three years was calculated to be 50 krad (Si).

[0059] CPT generation: For the root node SLD light source, based on the experimental data of 50 krad(Si), it is possible to find that the probability of output power decrease by more than 10% is 0.7 ("degraded" state), and the probability of the opposite is 0.3 ("normal" state).

[0060] For PIN photodiodes, their state is affected by the state of the SLD light source and their own irradiation. Their CPT needs to be defined as: the probability of a decrease in responsivity when the SLD light source is "normal" and irradiated with 50 krad (Si); and the probability of a decrease in responsivity when the SLD light source is "degraded" and irradiated with the same amount of light.

[0061] For transimpedance amplifiers, the CPT is similar, and its state depends on the state of the preceding PIN photodiode and the effect of its own irradiation.

[0062] Integration: Assign the above CPTs to the three nodes respectively, and the model is now complete.

[0063] The core benefit of this model construction process lies in its systematic and structured integration of the physical architecture of complex systems, multi-dimensional environmental stress data, and component-level experimental / simulation performance responses into a unified, probabilistically reasonable digital knowledge model. By decomposing and mapping the system to a network topology and combining environmental and experimental data to generate a quantified conditional probability table, this process creates a "digital twin" that not only reflects the system's composition but also characterizes the uncertainty transmission relationships between the environment and modules. This provides a standardized inference engine for subsequent steps, enabling automated and quantitative assessment of implicit, context-dependent failure risks within the system, fundamentally changing the limitations of traditional reliability analysis that relies on isolated component data and empirical judgment.

[0064] Optionally, the functional module performance status evidence data includes the expected performance degradation state; the functional module performance status evidence data is obtained in the following ways: Based on the test database, the expected performance degradation state of each functional module in the target operating environment is determined according to the target operating environment data.

[0065] Specifically, the core of the functional module performance status evidence data is the expected performance degradation state. This is not subjective speculation, but a deterministic judgment derived from objective data correlation calculations. The acquisition method follows a clear logical chain: The process relies on two prepared datasets: 1) an experimental database for each functional module, which stores in a structured form the performance response of the module under various controllable experimental conditions (such as different total irradiation dose, dose rate, and temperature) (such as the change curves or data points of parameters such as output current, gain, and dark current); 2) target operating environment data, which is the environmental parameters that the target system will actually experience on its specific mission trajectory (such as particle energy spectrum and total flux) calculated by professional analysis tools.

[0066] Environment-to-performance mapping calculation: The core operation for obtaining evidence is to use preset operating environment data as "query conditions" and perform matching, interpolation, or model calculations in the experimental database of the corresponding functional module. Its goal is to answer a specific question: "In this specific real-world environment, what performance state is this module most likely to exhibit?" The relationship between operating environment data and experimental conditions is as follows: the former represents the "actual problem" to be evaluated, while the latter represents the "pre-set test questions" in the database. Both are data from the same physical dimension, but their purposes and formats differ.

[0067] Operating environment data: This is a quantitative description of the environment that the target system will face in its actual, specific mission. For example, "The satellite will receive a total ionizing dose of 80 krad(Si) over its three-year mission period." Characteristics: Specific, unique, and mission-specific. It is a (or a set of) definite values ​​representing the actual challenges faced by the system.

[0068] Test conditions are a series of discrete environmental stress levels artificially set or simulated when building a test database to test or simulate individual functional modules. For example, to build a database for a transistor, the performance of the transistor was tested in a laboratory or simulation at total doses of 0, 10, 50, and 100 krad (Si). Its characteristics are: discrete, standardized, and module-specific. It is a pre-defined sequence of values ​​that serves as the index key for the database.

[0069] For example, the target operating environment data (80 krad) can be used as a query value to match or interpolate between discrete "test conditions" (0, 50, 100 krad) to find or deduce in the database where the corresponding performance data point is when the module is subjected to a dose of 80 krad.

[0070] To put it simply: the experimental conditions are like latitude and longitude coordinates marked on a map, and the operating environment data is like your real-time GPS location. You need to use your GPS location (the actual environment) to find the nearest known coordinate point on the map (the experimental conditions) to locate yourself.

[0071] The expected performance degradation state is the interpreted and formatted output after querying performance response data based on target operating environment data. It is a concise qualitative or quantitative description of the module's performance most likely to exhibit in future tasks. For example, the conclusion might be: Under an 80 krad task environment, the transistor current gain is expected to decrease to approximately 75% of its original value (or specifically 72). Performance response data is the raw, objective record stored in the test database. It describes the specific numerical values ​​of the module's performance parameters under a particular test condition. For example, the record might read: "Under a 50 krad test condition, the transistor current gain was measured to be 85" (the original value might be 100).

[0072] Through the above mapping calculations, abstract environmental parameters (such as "total orbital flux of 100 krad(Si)") are transformed into quantified physical state descriptions (i.e., expected performance degradation states) for specific modules, such as "the output power of the laser diode is expected to decrease by 12%" or "the current gain of a certain type of transistor is expected to decrease to 75% of its original value." This specific state description serves as the evidence data input into the Bayesian network model.

[0073] The process of acquiring functional module performance status evidence data establishes an objective and repeatable standardized procedure, accurately transforming macroscopic, continuous environmental parameters into microscopic, discrete module performance status assertions, providing a solid and unambiguous input foundation for probabilistic inference models. Through quantitative mapping based on experimental databases, it completely avoids the subjectivity and inconsistency inherent in traditional reliability analysis that relies on qualitative estimations based on expert experience, ensuring the scientific rigor and accuracy of the "environment" to "evidence" conversion process. This significantly improves the reliability and credibility of subsequent Bayesian network inference results, making it a crucial link in the entire prediction method chain for achieving a reliable connection from the "physical world" to the "probabilistic model."

[0074] Optionally, the process of acquiring the test database includes: A data retrieval framework is constructed based on the device status data of each functional module and the preset test condition data; Based on the data retrieval framework, performance response data of each functional module under various test conditions are obtained from a preset data source, and the performance response data is integrated to form the corresponding test database.

[0075] Optionally, the test condition data includes one or more of the following: irradiation dose, dose rate, bias voltage, temperature, and hydrogen concentration.

[0076] Specifically, firstly, the two dimensions of the data index for each functional module are defined: Device status data: These are the inherent attributes of the module itself, serving as the "identity identifier" of the data. This includes process technology (e.g., 0.18μm CMOS), structure type (e.g., NPN bipolar transistor), package type (e.g., ceramic flat package), production batch, and manufacturer. This dimension ensures that the acquired data is highly relevant to the target module.

[0077] Preset test condition data: These are the external stresses that the module may experience during the evaluation, serving as the "environmental labels" for the data. These include, but are not limited to, irradiation dose, dose rate, bias voltage, operating temperature, and hydrogen concentration. This dimension constructs a test scenario matrix that simulates the real-world operating environment.

[0078] Based on the above "identity-environment" retrieval framework, the corresponding performance response data is systematically retrieved or generated from the following preset data sources: Historical test data: Retrieve measured performance data (such as current gain versus dose curves) from the internal or cooperative institutions’ previous radiation test and reliability test databases to match the device state with the test conditions.

[0079] Public literature data: Extract performance degradation data of the same or similar devices under publicly reported test conditions from academic papers, technical reports, and device manuals.

[0080] Defect data based on device simulation: Using physical simulation tools such as TCAD, a numerical model of the device containing defect models is established, and its electrical performance response is simulated under preset test conditions (especially extreme or combined conditions where historical data is missing).

[0081] Performance response data (such as data tables, graphs, and simulation result files) obtained from various sources and in different formats will be cleaned, normalized (e.g., standardized units and benchmarks), and correlated and calibrated. Then, this data will be structured, reorganized, and stored according to a "device state-test condition" framework, ultimately forming the test database for each module. This database is essentially a multidimensional lookup table or response surface model, capable of explicitly answering the question, "How will the key performance parameters change when a device in a certain state is subjected to a certain set of test conditions?"

[0082] By integrating inherent device properties, multi-dimensional environmental stress, and diverse heterogeneous performance data, a high-fidelity digital performance degradation knowledge base covering a wide range of operating conditions was generated for each functional module. This process systematically solves the inherent problems of high cost, long cycle time, and sparse data points associated with single data sources (such as relying solely on expensive ground tests). By integrating historical data, literature knowledge, and simulation predictions, the "state-response" knowledge graph of each module is greatly enriched, providing an indispensable and high-quality data foundation for the subsequent accurate generation of conditional probability tables and evidence mapping. This is the premise and guarantee for the realization of the entire probabilistic prediction method.

[0083] Optionally, the preset operating environment data includes space environment particle data, which includes particle energy spectrum and total particle fluence at different confidence levels.

[0084] Specifically, the core of the preset operating environment data in the specific application scenario of this application is space environment particle data. This data is not a single value, but a set of physical quantities generated through professional analysis that can accurately characterize the radiation environment that the target system will encounter at a specific location and time in space. It mainly includes two key dimensions: Particle energy spectrum: refers to the energy distribution of various high-energy particles (such as protons, electrons, and heavy ions) existing in space. Particles of different energies cause damage to electronic devices in different ways and to varying degrees (for example, high-energy protons are prone to ionization damage, while heavy ions are prone to single-event effects). Energy spectrum data describes the particle flux in various energy ranges along the mission orbit, which is the basis for assessing the type and depth of damage.

[0085] Total particle fluence at different confidence levels: This refers to the total number of particles that are expected to be received per unit area during the mission period. Due to natural fluctuations and uncertainties in the space environment, this data is usually given in probabilistic form. For example, "At a 50% confidence level, the total fluence is 1 × 10⁻⁶". 11 p / cm 2 At a 90% confidence level, the total betting volume is 5 × 10. 11 p / cm 2 Data at different confidence levels allows reliability analysis to cover various risk scenarios, from the most likely scenario to the most severe scenario, providing a more comprehensive basis for design decisions.

[0086] This data is typically calculated using specialized space environment analysis software (such as SPENVIS, OMERE, etc.) or models, combined with the target spacecraft's predetermined orbital parameters (such as altitude, inclination, mission lifetime) and mission duration. For example, for a commercial satellite planned to operate in Low Earth Orbit (LEO) for three years, by inputting the orbital parameters, the software can calculate the proton energy spectrum of galactic cosmic ray and solar particle events encountered in its orbit, as well as the equivalent total ionizing dose (TID) accumulated over three years at different confidence levels (such as 50% and 95%). The calculated "particle energy spectrum" and "total fluence" constitute the quantified, pre-defined operational environment data used for subsequent analysis.

[0087] By clearly defining and quantifying the pre-defined operating environment data into space environment particle data that includes particle energy spectra and total fluence at different confidence levels, the core benefit lies in providing a high-fidelity, quantifiable environmental input benchmark covering the range of uncertainties for the entire reliability prediction process. This transforms abstract space radiation threats into precise engineering parameters that can be processed by probabilistic and physical models. On the one hand, this ensures that the entire chain from probabilistic inference (Bayesian networks) to physical simulation (SPICE) is built on a real and rigorous foundation of space physics, greatly improving the accuracy and reliability of prediction results. On the other hand, by providing data at different confidence levels, it supports comprehensive reliability margin analysis from "typical conditions" to "extreme conditions," enabling designers to make more refined and scientific trade-offs between cost, performance, and risk. This provides crucial data support for achieving the core objective of commercial space missions: ensuring reliability while controlling costs.

[0088] Optionally, converting the probability distribution data of the fault or performance degradation states of each functional module into corresponding SPICE circuit simulation parameter perturbation values ​​includes: Based on the probability distribution data, determine the change in performance parameters corresponding to the performance degradation state of each functional module; Based on the preset performance parameter-SPICE parameter mapping relationship, the change in the performance parameter is mapped to the corresponding SPICE circuit simulation parameter perturbation value.

[0089] Optionally, the performance parameter-SPICE parameter mapping relationship includes at least one of the following: the mapping relationship between the module output electrical performance parameters and the SPICE controlled source parameters, and the mapping relationship between the module internal gain and the SPICE device model parameters.

[0090] Specifically, the internal gain of a module refers to its ability to amplify an input signal; it is a specific performance parameter. Examples include the voltage gain of an amplifier, the current gain (β or hFE) of a transistor, and the electro-optical conversion efficiency of a laser. It is a parameter with a definite physical meaning and unit. The change in performance parameter refers to the change in any physical quantity that characterizes the degree of performance degradation of a module, derived from probability distribution data. A change in the internal gain of a module (e.g., "current gain decreased by 20%)" is a specific manifestation of this change in performance parameter. The concept of performance parameter change encompasses all parameters that may degrade, including but not limited to: Gain-related parameters: decreased internal gain (e.g., amplifier gain, transistor β value); Output characteristic-related parameters: decreased output current / voltage / power (e.g., light source optical power); Static parameter-related parameters: increased leakage current, increased on-resistance, threshold voltage drift, etc.; Dynamic parameter-related parameters: slower response speed, narrower bandwidth, etc.

[0091] The relationship between SPICE controlled source parameters, SPICE device model parameters, and SPICE circuit simulation parameter perturbation values: The essence of the relationship is that the first two are two specific "tools" or "carriers" for realizing perturbation, while the "perturbation value" is the "specific setting value" required when using these tools.

[0092] SPICE circuit simulation parameter perturbation values: This is the target data that this invention ultimately aims to generate. It is a set of instructions or values ​​that guide the SPICE simulator on how to modify its simulation model. It is an abstract concept representing "modifications that need to be applied," but its specific content must be achieved by modifying certain inherent SPICE parameters.

[0093] SPICE controlled source parameters and SPICE device model parameters: These are two specific technical paths or operational objects for achieving the above-mentioned "perturbation".

[0094] SPICE controlled source parameters: These refer to the attribute parameters of controlled source components (such as voltage-controlled voltage sources E, current-controlled current sources F, etc.) in the SPICE simulator. The most crucial parameter is its gain coefficient or transfer function value. By changing this value, the changes in the output characteristics of a module can be simulated equivalently.

[0095] SPICE device model parameters: These refer to the internal physical parameters of semiconductor device models (such as transistor and diode models) in the SPICE simulator. Examples include the "BF" (ideal forward gain) of a bipolar transistor and the "KP" (transconductance parameter) of a MOSFET. Directly modifying these parameters can precisely change the intrinsic characteristics of the device.

[0096] The probabilistic conclusions describing uncertainty, output by the Bayesian network, are transformed into deterministic instructions that the SPICE simulator can recognize and execute. This involves two steps: Step 1: Extract the most representative physical change from the probability distribution: The input is the probability distribution data of faults or performance degradation states of each functional module (e.g., "the probability of an amplifier gain drop exceeding 20% ​​is 30%, and the probability of a drop of 10%-20% is 50%"). First, from this distribution, the most representative specific performance degradation scenario and its corresponding physical quantity change are analyzed and determined. This is typically achieved by taking the expected value (mean), mode (most likely state), or selecting a conservative value based on engineering judgment. For example, from the above distribution, it might be determined that "the most likely performance degradation state is a 15% gain drop." Thus, the abstract probability distribution is solidified into a specific performance parameter change (e.g., "a 15% gain drop" or "a 0.5mA drop in output current").

[0097] Step 2: Map physical quantities to SPICE executable instructions according to predefined rules: After obtaining the specific performance parameter changes, they need to be "translated" into a "language" that the SPICE simulator can understand. This relies on a pre-defined performance parameter-SPICE parameter mapping library. This mapping library defines how the performance changes of different types of functional modules should be implemented in the SPICE netlist. It mainly includes two core mapping strategies: The mapping relationship between module output electrical performance parameters and SPICE controlled source parameters applies to changes in the module's port output characteristics. For example, "the laser source output optical power decreases" is mapped to "adjusting the output current value of the controlled current source in its SPICE equivalent model".

[0098] The mapping relationship between the module's internal gain and SPICE device model parameters: applicable to changes in the module's internal amplification or transmission characteristics. For example, mapping "transistor current gain attenuation" to "directly modifying the current amplification factor (such as the BF parameter) in its SPICE transistor model", or mapping it to "adding a controlled voltage source to the base to achieve an equivalent gain change effect".

[0099] By applying the corresponding mapping relationship, the changes in performance parameters are ultimately converted into one or more specific SPICE circuit simulation parameter perturbation values ​​with clear numerical values, that is, instructions to modify specific parameters in the SPICE netlist or model.

[0100] The core benefit of this conversion process lies in its construction of a sophisticated "translator" and "adapter," successfully resolving the inherent interface mismatch between probabilistic fault prediction results and deterministic physical simulation tools. By parsing probability distributions into specific physical quantities and converting them into SPICE-recognizable parameter perturbation commands using predefined mapping relationships, this method enables module-level risks derived from uncertainty reasoning to be accurately and automatically "instantiated" into observable and analyzable deterministic circuit events in high-fidelity system simulations. This is not merely a formal data conversion, but a crucial methodological bridge, making it possible to subsequently use SPICE for system-level simulations to assess fault coupling effects. This achieves a leapfrog, refined assessment from "the probability that a component might fail" to "the probability that the system will fail as a result."

[0101] In some embodiments, such as Figure 3 The flowchart shown is for the system failure probability prediction method using Bayesian networks and SPICE. The entire process can be divided into three main stages, and the specific steps are as follows: Phase 1: Data Preparation and Experimental Database Construction Input data: Device status (device status data): including process, structure, package type, batch, manufacturer, etc. This is the inherent attribute data of the functional module.

[0102] Test conditions (test condition data): including dose, dose rate, bias, temperature, hydrogen concentration, etc.

[0103] Data processing: The above data is input into the "device model" and multi-source data such as "defect database", "literature results" and "TCAD simulation" are used to model, retrieve and simulate the performance response of the device.

[0104] Core output: By integrating data from all the above sources, a structured experimental database is formed. This database systematically stores the performance response patterns of various devices (modules) under various experimental conditions.

[0105] Phase Two: Environmental Analysis and Bayesian Network Probabilistic Inference Environmental input: The "on-orbit environment" and "mission time" are combined to obtain the "cumulative dose" (i.e., total fluence) through analysis and calculation. This "cumulative dose" and the "on-orbit environment" data behind it are the defined preset operating environment data (space environment particle data, including particle energy spectrum and total fluence).

[0106] Reasoning Input and Process: The device (module) types obtained from the first stage decomposition are input together with the calculated cumulative dose into a Bayesian network (BN).

[0107] The conditional probability table (CPT) inside the Bayesian network associates environmental data with knowledge from the experimental database to perform uncertain reasoning.

[0108] Core output: The Bayesian network outputs the probability distribution of the performance state of each module.

[0109] This stage comprehensively implements steps S100 and S200. S100 acquires the pre-constructed Bayesian network probability model, while S200 transforms the environmental data into "functional module performance status evidence data" (e.g., determined cumulative dose) input to the model, ultimately "outputting the probability distribution data of each functional module...". The process of generating the conditional probability table corresponds to step S104.

[0110] Phase 3: System Modeling, Fault Injection, and SPICE Simulation Verification System modeling: Based on information such as the "product manual" or "similar samples", "circuit selection" and "circuit diagram drawing (circuit decomposition)" are performed. This directly corresponds to step S101, which is to decompose the target system into modules and obtain the functional modular circuit connection relationships.

[0111] Fault injection: The probabilistic conclusions output by the Bayesian network are transformed into parameter changes affecting specific circuit nodes through node injection. This relates to step S300 (converting probability distribution data into SPICE circuit simulation parameter perturbation values) and the defined performance parameter-SPICE parameter mapping relationship.

[0112] System simulation and result analysis: The circuit model after the injection of faults (parameter disturbances) is sent into the SPICE environment for circuit simulation.

[0113] The simulation results are analyzed, and the system performance under different fault injection scenarios is statistically analyzed to obtain a quantitative system-level reliability assessment result.

[0114] In other embodiments, such as Figure 4 The diagram shown illustrates the process of converting circuit parameters into SPICE simulation parameters, using a photodetector system as a specific example. Figure 4 The blue dashed line represents a simplified circuit of the photodetector system. This diagram clearly illustrates how the probabilistic module performance degradation predicted by a Bayesian network can be transformed into executable, deterministic circuit parameter modifications in SPICE simulations. The specific explanation is as follows: Injection processing for SLD (Superluminescent Diode) light sources: Performance degradation scenario: The output characteristics of the SLD light source (such as output optical power or drive current) will decrease as the total radiant flux increases.

[0115] Equivalent circuit mapping: Figure 4 The method shown does not directly modify the complex physical model of the SLD device itself, but instead employs a clever external equivalent. It equates the decrease in the output current of the SLD light source to "adding an additional controlled current source" to the original circuit node, and the output value of this current source is set to the attenuation amount that varies with the irradiance.

[0116] This method directly corresponds to the first mapping relationship, namely the mapping relationship between the module output electrical performance parameters and the SPICE controlled source parameters. It maps the change of the performance parameter "source output current" to the change of the parameter (output current value) of a "controlled current source" in SPICE, thereby generating a specific "SPICE circuit simulation parameter disturbance value" (i.e., the current setting value of the new current source).

[0117] Injection processing for bipolar transistors: Performance degradation scenario: When a bipolar transistor is exposed to radiation, its key performance parameter - current gain (usually expressed as β or hFE) will decrease.

[0118] Equivalent Circuit Mapping: The figure illustrates another typical equivalent approach. The attenuation of transistor current gain can be simulated by "adding an extra voltage to the base." This extra voltage source alters the base bias, thus producing the same circuit behavior effect as the gain reduction (such as a decrease in collector current) in SPICE simulations.

[0119] It can be viewed both as a realization of the mapping relationship between the module's internal gain and the SPICE device model parameters (equivalent to gain change by altering the internal operating point through external voltage) and as a simulation of internal performance using a "SPICE controlled source parameter" (a voltage source). This demonstrates the practicality and creativity of the mapping relationship.

[0120] Figure 4 More than just a simple example, it vividly reveals the core technical idea of ​​this invention: to physicalize abstract probabilities and circuitize device behavior. It demonstrates that through a creative circuit equivalence method, any module-level performance degradation predicted by a probabilistic model can be transformed into a deterministic perturbation that a SPICE simulator can understand and process.

[0121] This invention provides a system failure probability prediction device based on Bayesian networks and SPICE, comprising: An acquisition unit is used to acquire a pre-constructed Bayesian network probability model; the nodes of the Bayesian network probability model correspond to the functional modules of the target system, and the Bayesian network probability model is used to predict the probabilistic dependency between the performance status of each functional module and the environmental factors affecting its performance. The acquisition unit is also used to acquire the performance status evidence data of each functional module corresponding to the preset operating environment, and input the performance status evidence data of each functional module into the Bayesian network probability model, and output the probability distribution data of each functional module failing or being in different performance degradation states in the preset operating environment. The processing unit is used to convert the probability distribution data into corresponding SPICE circuit simulation parameter perturbation values. The processing unit is also used to construct a corresponding SPICE circuit simulation model based on the circuit connection relationship of the target system, and to apply the disturbance values ​​of each SPICE circuit simulation parameter to the corresponding circuit node in the SPICE circuit simulation model for simulation, so as to obtain the fault probability prediction result of the target system under the preset operating environment.

[0122] This invention provides a system fault probability prediction device based on Bayesian networks and SPICE, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to implement the system fault probability prediction method based on Bayesian networks and SPICE as described above when the computer program is executed.

[0123] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the system fault probability prediction method based on Bayesian networks and SPICE as described above.

[0124] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A system failure probability prediction method based on Bayesian networks and SPICE, characterized in that, include: Obtain a pre-constructed Bayesian network probability model; the nodes of the Bayesian network probability model correspond to the functional modules of the target system, and the Bayesian network probability model is used to predict the probabilistic dependency between the performance status of each functional module and the environmental factors affecting the performance of the functional module; Obtain performance status evidence data of each functional module corresponding to the preset operating environment, and input the performance status evidence data of each functional module into the Bayesian network probability model, and output the probability distribution data of each functional module failing or being in different performance degradation states under the preset operating environment. The probability distribution data are converted into corresponding SPICE circuit simulation parameter perturbation values. Based on the circuit connection relationship of the target system, a corresponding SPICE circuit simulation model is constructed, and the disturbance values ​​of each SPICE circuit simulation parameter are applied to the corresponding circuit nodes in the SPICE circuit simulation model for simulation, so as to obtain the fault probability prediction result of the target system under the preset operating environment.

2. The system failure probability prediction method based on Bayesian networks and SPICE according to claim 1, characterized in that, The process of constructing the Bayesian network probability model includes: The target system is decomposed into modules, and the functional modular circuit connection relationships are obtained; the functional modular circuit connection relationships are used to characterize the electrical connection relationships between the decomposed functional modules. Based on the functional modular circuit connection relationship, the nodes and topology of the Bayesian network are determined, wherein the nodes correspond to each functional module, and the directed edges between the nodes correspond to the signal or functional dependencies between each functional module. Obtain the test database of each functional module and the preset operating environment data of the target system; Based on the experimental database and the preset operating environment data, a conditional probability table corresponding to each of the functional modules is generated. The conditional probability table is integrated into the topology of the Bayesian network to form the Bayesian network probability model.

3. The system fault probability prediction method based on Bayesian networks and SPICE according to claim 2, characterized in that, The functional module performance status evidence data includes the expected performance degradation state; the methods for obtaining the functional module performance status evidence data include: Based on the test database, the expected performance degradation state of each functional module in the target operating environment is determined according to the target operating environment data.

4. The system fault probability prediction method based on Bayesian networks and SPICE according to claim 2, characterized in that, The process of acquiring the experimental database includes: A data retrieval framework is constructed based on the device status data of each functional module and the preset test condition data; Based on the data retrieval framework, performance response data of each functional module under various test conditions are obtained from a preset data source, and the performance response data is integrated to form the corresponding test database.

5. The system fault probability prediction method based on Bayesian networks and SPICE according to claim 4, characterized in that, The test condition data include one or more of the following: irradiation dose, dose rate, bias voltage, temperature, and hydrogen concentration.

6. The system failure probability prediction method based on Bayesian networks and SPICE according to claim 2, characterized in that, The preset operating environment data includes space environment particle data, which includes particle energy spectrum and total particle fluence at different confidence levels.

7. The system failure probability prediction method based on Bayesian networks and SPICE according to claim 1, characterized in that, The step of converting the probability distribution data of the fault or performance degradation states of each functional module into corresponding SPICE circuit simulation parameter perturbation values ​​includes: Based on the probability distribution data, determine the change in performance parameters corresponding to the performance degradation state of each functional module; Based on the preset performance parameter-SPICE parameter mapping relationship, the change in the performance parameter is mapped to the corresponding SPICE circuit simulation parameter perturbation value.

8. The system failure probability prediction method based on Bayesian networks and SPICE according to claim 7, characterized in that, The performance parameter-SPICE parameter mapping relationship includes at least one of the following: the mapping relationship between the module output electrical performance parameters and the SPICE controlled source parameters, and the mapping relationship between the module internal gain and the SPICE device model parameters.

9. A system failure probability prediction device based on Bayesian networks and SPICE, characterized in that, include: An acquisition unit is used to acquire a pre-constructed Bayesian network probability model; the nodes of the Bayesian network probability model correspond to the functional modules of the target system, and the Bayesian network probability model is used to predict the probabilistic dependency between the performance status of each functional module and the environmental factors affecting its performance. The acquisition unit is also used to acquire the performance status evidence data of each functional module corresponding to the preset operating environment, and input the performance status evidence data of each functional module into the Bayesian network probability model, and output the probability distribution data of each functional module failing or being in different performance degradation states in the preset operating environment. The processing unit is used to convert the probability distribution data into corresponding SPICE circuit simulation parameter perturbation values. The processing unit is also used to construct a corresponding SPICE circuit simulation model based on the circuit connection relationship of the target system, and to apply the disturbance values ​​of each SPICE circuit simulation parameter to the corresponding circuit node in the SPICE circuit simulation model for simulation, so as to obtain the fault probability prediction result of the target system under the preset operating environment.

10. A system failure probability prediction device based on Bayesian networks and SPICE, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the system failure probability prediction method based on Bayesian networks and SPICE as described in any one of claims 1 to 8.

Images