Virtual-real fusion life and reliability evaluation method based on Bayesian theory

Abstract

The invention relates to a Bayesian theory-based virtual-real fusion life and reliability evaluation method, which introduces a Bayesian theory to carry out life and reliability evaluation research aiming at the problems of limited sample size, incomplete test data, more uncertainty factors and the like in life evaluation of electronic products. According to the technical scheme, the method comprises the following steps: firstly, dividing prior distribution and sample data according to obtained simulation data and test data; secondly, combining prior distribution with sample information according to a Bayesian theory, and deriving posterior probability distribution of parameters; and finally, comparing and verifying the precision of the fusion evaluation method according to actual test data. The patent mainly realizes effective fusion of electronic product simulation priori knowledge and newly acquired test data, and improves the evaluation accuracy and robustness under the conditions of small samples and high uncertainty.

Description

Technical Field

[0001] This invention relates to a virtual-real fusion lifetime and reliability assessment method based on Bayesian theory, which combines digital simulation models with accelerated test data to achieve accurate characterization of the performance degradation process and effective calibration of the assessment model. Background Technology

[0002] In the field of equipment health management and life prediction, reliability assessment is a crucial step in ensuring the safe and stable operation of systems. Traditional reliability assessment methods primarily rely on experimental data, namely, obtaining product failure data and performing statistical modeling through accelerated life testing, environmental stress screening, and other methods. However, with the increasing complexity of modern equipment, relying solely on measured data is insufficient to comprehensively depict its failure patterns. Simultaneously, the development of physical modeling and numerical simulation technologies has provided new tools for reliability assessment. Integrating simulation and experimental data to form a multi-source information fusion reliability assessment method has become a hot research and engineering application area.

[0003] Experimental data is highly realistic, directly reflecting actual product operation or environmental simulation, and can capture failure mechanisms in complex systems that are difficult to model. However, experiments are often costly, time-consuming, and have poor repeatability, and it is difficult to obtain sufficient failure samples in the early design stage. Simulation data, on the other hand, builds physical models based on material properties, structural design, and environmental conditions. Through finite element analysis (FEA), thermal analysis, performance model building, and degradation injection, it predicts the system's response and potential failure behavior under different loads or conditions, offering advantages such as high efficiency, repeatability, and scalability.

[0004] Combining the two, simulation data provides broad coverage and theoretical support, while experimental data is used for model calibration and error correction, enabling a closed loop from "prediction" to "verification" and improving the accuracy and applicability of the evaluation results. Reliability assessment based on the fusion of simulation and experimental data typically involves the following steps: First, failure mechanism analysis is conducted: identifying the main failure modes of critical components or systems and clarifying the controlling factors affecting lifespan; based on this, physical modeling and simulation calculations are performed: a physical model is established based on the product structure, material properties, and loading conditions, and multi-field coupled simulations such as thermo-mechanical-environment are conducted to obtain intermediate quantities such as stress-strain, temperature distribution, and damage evolution; then, experimental design and data acquisition are carried out: accelerated life tests or operating condition simulation tests are conducted to obtain degradation data or failure time distribution under real-world conditions; based on simulation and testing, data fusion and model correction are performed: the simulation results are corrected for biases to better match actual experimental observations, and Bayesian methods, small sample statistics, transfer learning, and other techniques are used to achieve joint analysis of simulation and experimental data; finally, lifespan prediction and reliability assessment are conducted: lifespan prediction is performed on the fused model, the model reliability and confidence interval are quantified, and decision support is provided for the reliability assessment of real products. Summary of the Invention

[0005] To address the challenges of limited sample size, incomplete experimental data, and numerous uncertainties in the lifespan assessment of electronic products, Bayesian theory is introduced to conduct lifespan and reliability assessment research. Bayesian theory can effectively integrate prior knowledge with newly acquired experimental or simulation data, dynamically updating assessment results and improving the accuracy and robustness of assessments under small sample sizes and high uncertainty conditions, making it suitable for the assessment needs of high-reliability electronic products.

[0006] To achieve the above objectives, the present invention employs the following technical solution: First, prior distribution and sample data are divided based on the acquired simulation data and experimental data; second, the prior distribution and sample information are combined according to Bayesian theory to derive the posterior probability distribution of the parameters; finally, the accuracy of the fusion evaluation method is verified by comparing actual experimental data.

[0007] The patent primarily aims to effectively integrate prior knowledge of electronic product simulation with newly acquired experimental data, thereby improving the accuracy and robustness of evaluations under conditions of small sample size and high uncertainty.

[0008] The specific steps are as follows: Step 1: Obtain electronic product simulation data and experimental data, and classify the prior distribution and sample data.

[0009] Step 2: Calculate the characteristic lifetime and scale parameters of the prior distribution and sample data.

[0010] Step 3: Perform Monte Carlo random sampling to obtain posterior data and its distribution.

[0011] Step 4: Verify the accuracy of lifetime and reliability assessments using reliable real-world test data. Attached Figure Description

[0012] Figure 1 The lifetime distribution of fiber optic gyroscopes based on Bayesian theory, where virtual and real data are combined. Detailed Implementation

[0013] A Bayesian sampling method is applied to conduct a virtual-real integrated assessment of the lifespan and reliability of electronic products. The failure lifespan obtained from simulation is used as the prior distribution, and the failure lifespan obtained from accelerated testing is used as the proposed distribution. The simulation data of the virtual model is used to correct the prediction results of the test data and obtain the posterior distribution, thereby improving the accuracy of lifespan and reliability assessment.

[0014] The core idea of ​​Bayesian statistical theory is to treat unknown parameters as random variables with probability distributions. This methodology comprises three key steps: constructing an initial probability distribution (prior distribution) of the parameters based on prior knowledge; obtaining sample data through observation; and deriving the posterior probability distribution of the parameters by combining the prior distribution with the sample information according to Bayes' theorem. This process achieves a probabilistic update from prior knowledge to posterior knowledge.

[0015] The essence of Bayesian statistical inference lies in using the posterior probability distribution as the basis of analysis: In the formula, For the prior distribution of parameters, Let be the likelihood function. This represents the marginal likelihood (normalization constant).

[0016] Because the normalization constant of the posterior distribution is calculated using integration in a high-dimensional parameter space. This is usually not feasible and requires relying on random sampling methods to bypass analytical computation. This paper uses the Monte Carlo method for random sampling, which generates samples that follow a target distribution. We use a sample, i.e., an independent sample from the posterior distribution, for calculation, thus transforming the integration problem into the calculation of the sample mean: According to the law of large numbers, when At that time, the sample mean can be considered as Converging to the true expectation .

[0017] And when it is not possible to directly from When performing sampling, a known distribution, i.e., the proposed distribution, can be introduced. We approximate the target distribution using weighted samples. The target expectation can then be rewritten as: Define importance weights The estimated value can then be written as: The sample failure lifetime of the digital simulation model of electronic products is used as prior data, as shown in Table 1; the failure lifetime obtained by accelerated testing is used as sample data, as shown in Table 2.

[0018] Table 1. Failure lifetime of fiber optic gyroscope simulation model samples

[0019] Table 2 Failure lifetime of fiber optic gyroscope accelerated test samples

[0020] The prior data had an evaluation lifetime and standard deviation of 12.3 years and 2.6 years, respectively; the sample data used a two-parameter model of the Weibull distribution, with the probability density function of the Weibull distribution being: In the formula, For test time, shape parameters Characterizing the shape and scale parameter of the distribution curve Characterizing the lifetime, e is the base of the natural logarithm. A Weibull distribution is fitted to the sample data, with a shape parameter of 6.8 and a scale parameter of 28.8.

[0021] The Monte Carlo sampling number was set to 1000 times, and the sampling results were used to determine whether to accept the results. Finally, the virtual-real fusion lifetime distribution of the fiber optic gyroscope based on Bayesian theory was obtained. Figure 1 ).

[0022] Based on this, the distribution model parameters and reliability of the virtual-real fusion data are calculated. The Weibull distribution is still used as the distribution model, and the calculated shape parameter m is 5.2, and the scale parameter is... It is 13.3.

[0023] The failure lifetimes of two failed samples from the accelerated test were selected to verify the virtual-real fusion lifetime and reliability accuracy. The failure lifetimes of the two failed samples were 11.3 years and 11.6 years, respectively. The characteristic lifetime of 11.4 years was taken as the evaluation index, and the fusion accuracy was 83.3%.

[0024] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for evaluating the lifetime and reliability of virtual-real fusion based on Bayesian theory, characterized in that, include: Step 1: Obtain electronic product simulation data and experimental data, and classify the prior distribution and sample data; Step 2: Calculate the characteristic lifetime and scale parameters of the prior distribution and sample data; Step 3: Perform Monte Carlo random sampling to obtain posterior data and its distribution; Step four: Verify the accuracy of lifetime and reliability assessments using reliable actual test data.

2. The method according to claim 1, characterized in that: By integrating data from different sources, such as simulation data and experimental test data, life and reliability assessments can be conducted.

3. The fusion lifetime and reliability assessment according to claim 2, characterized in that: By applying the Bayesian sampling method, the failure lifetime obtained from simulation is used as the prior distribution, and the failure lifetime obtained from accelerated testing is used as the proposed distribution. The simulation data of the virtual model is used to correct the prediction results of the test data and obtain the posterior distribution.

Images