Photoelectric coupler dynamic reliability evaluation method based on Bayesian data fusion
By integrating multi-source data through Bayesian data fusion, the problem of a single data source in the reliability evaluation of optocouplers is solved, and high-precision dynamic reliability evaluation and lifetime prediction are achieved.
Patent Information
- Application Number
- CN202511663288.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-13
- Publication Date
- 2026-02-10
AI Technical Summary
Existing methods for evaluating the reliability of optocouplers rely on a single data source, neglecting quality assurance and field data, resulting in high prediction uncertainty and a lack of dynamic update mechanisms.
A Bayesian data fusion framework is adopted to integrate accelerated degradation test, quality assurance and field data, and dynamically optimize reliability evaluation through Bayesian inference and likelihood function updates.
It improves the accuracy of optocoupler lifetime prediction, reduces uncertainty, and enables dynamic correction of reliability indicators, making it suitable for quality assurance and lifetime prediction of optoelectronic devices.
Smart Images
Figure CN121503255A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of photoelectric coupler reliability, and particularly relates to a photoelectric coupler dynamic reliability evaluation method based on Bayesian data fusion. BACKGROUND
[0002] As a key device for electrical isolation and signal transmission in electronic systems, photoelectric couplers are widely used in power electronics, communication systems, industrial control, aerospace and other fields. With the development of electronic products towards high performance and high reliability, the service environment of photoelectric couplers is becoming increasingly complex, often facing multiple field coupling effects such as temperature cycling, damp heat stress, mechanical vibration and irradiation, resulting in various failure modes such as Au - Al bonding point cracking, current transmission ratio degradation, common mode noise interference, etc. These failure modes show an accelerating trend under complex working conditions, making life prediction accuracy a focus of the academic and industrial communities.
[0003] Existing photoelectric coupler reliability evaluation methods mainly rely on laboratory accelerated degradation tests, such as obtaining degradation data through high-temperature storage or power cycle tests, and extrapolating life distribution using Arrhenius model. Although these methods can obtain life information within a limited time, they have the following shortcomings: 1. Single data source, relying only on laboratory data, ignoring quality assurance data and actual use data in the secondary screening process; 2. Unable to effectively handle small samples and high proportion of no-failure data, resulting in large prediction uncertainty; 3. Lack of dynamic updating mechanism, making it difficult to reflect the reliability changes of the device under real working conditions.
[0004] Some existing research attempts to improve the above shortcomings, but none of them have achieved the fusion of multiple data sources. Therefore, there is an urgent need for a dynamic reliability evaluation method that can integrate laboratory, quality assurance and actual working condition data. SUMMARY
[0005] The purpose of the application is to solve the problems of low data utilization, high uncertainty and lack of dynamic updating in existing photoelectric coupler reliability evaluation methods, and to propose a photoelectric coupler dynamic reliability evaluation method based on Bayesian data fusion.
[0006] The technical solution of the application is: a photoelectric coupler dynamic reliability evaluation method based on Bayesian data fusion, comprising the following steps: S1. Use accelerated degradation test data to establish a photoelectric coupler accelerated degradation model, and then obtain the probability density distribution of failure rate as the prior distribution of Bayesian inference; S2. Introducing large sample data of secondary screening link, constructing likelihood function, carrying out first time Bayes updating to prior distribution, obtaining intermediate posterior distribution; S3. Combining real life of external field with no failure data, carrying out second time Bayes updating, obtaining final posterior distribution; S4. Based on final posterior distribution, calculating average failure time, reliability and confidence interval of photoelectric coupler, completing dynamic reliability evaluation.
[0007] The beneficial effects of the application are: 1. Through the Bayes data fusion framework, the accelerated test, secondary screening and field data are integrated, the limitation of single data source is broken through, and the evaluation precision is improved; 2. The high proportion of no failure data is processed by using the censored likelihood function, the small sample uncertainty is reduced, and the case shows that the life prediction error is reduced from more than 15% of the traditional method to within 8%; 3. The dynamic correction of reliability index is realized, the average failure time and confidence interval are optimized with data accumulation, and it is more in line with the real working condition; 4. The method has strong universality, and can provide an engineering practical tool for quality assurance, storage life extension and life prediction of photoelectric coupler and other photoelectric devices.
[0008] As preferred, the accelerated degradation test data in step S1 includes current transfer ratio degradation data of the photoelectric coupler under high temperature storage and power cycle stress conditions; the accelerated degradation model adopts Arrhenius acceleration model for modeling, and the expression formula is: ; Wherein, t is the degradation time, represents the degradation rate of characteristic parameters, A is a pre-factor or frequency factor, which is related to the properties of materials and reaction types; Ea is activation energy, unit is joule / mol ( J / mol ), which represents the energy required to initiate the reaction; k is Boltzmann constant, which is 8.617*10 -4 e VK -1 , T is absolute temperature, unit is Kelvin ( K ). By fitting the degradation curve, the accelerated degradation model parameters are obtained, the pseudo life is calculated, and the life under the actual working temperature is extrapolated by using the following formula: ; The prior distribution adopts gamma distribution to fit the failure rate parameter (the inverse of life, the life obeys exponential distribution), and the probability density function is: ; wherein, is a shape parameter, is a rate parameter, is a gamma function.
[0009] As preferred, the large sample data of the secondary screening link in step S2 comprises the failure number and the test duration under the high temperature aging test condition; and the formula of the likelihood function is: ; ; wherein, and are likelihood functions obtained by processing the secondary screening data and the field use data respectively, and are the number of failure data and the number of non-failure data respectively; and are the first failure time and the first non-failure time respectively, i and j are the failure probability density function and the failure probability function respectively. The formula for the first Bayesian updating of the prior distribution to obtain the intermediate posterior distribution is:
[0010] ; ; wherein, the function is a posterior density function, is a prior density function, is a likelihood function.
[0011] As preferred, the field data in step S3 comprises the failure history data and the non-failure data under the actual working condition; and the second Bayesian updating adopts the same likelihood function construction method as that in step S2 to update the intermediate posterior distribution again.
[0012] As preferred, in order to realize the real-time and rapid prediction of the optoelectronic coupler and facilitate the deployment, the prior distribution in the Bayesian updating in steps S2 and S3 adopts an approximate conjugate prior distribution; when the conjugate prior distribution is a common distribution (gamma distribution), the constant term can be quickly obtained, and thus the posterior distribution is obtained, and the calculation expression is: ; In the formula, is the total number of failures, is the total duration of the secondary screening or the actual use.
[0013] Preferably, the reliability calculation formula in step S4 is: ; in, The conditional reliability function is an exponential distribution, and the mean time to failure is calculated using the following formula: ; The confidence interval is calculated based on the quantiles of the posterior distribution, and is expressed as: ; in, and It is a posterior distribution of and Quantiles It is the significance level.
[0014] The beneficial effects of the above preferred solution are: 1. The above-mentioned preferred scheme integrates accelerated testing, secondary screening, and field data. The Bayesian fusion framework makes full use of multi-source information, reduces prediction bias, and significantly improves the accuracy of optocoupler reliability evaluation. 2. By using a truncated likelihood function to process data without failures and combining it with the conjugate prior distribution to quickly calculate the posterior distribution, statistical uncertainty is reduced. This makes the confidence interval more compact and enhances the reliability of the evaluation. 3. The Bayesian iterative update mechanism enables reliability metrics to be dynamically optimized with new data, reflecting changes in real operating conditions, providing a practical tool for product lifecycle monitoring, and supporting quality assurance and life extension decisions. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of the method for establishing an accelerated degradation model.
[0016] Figure 2 It is the current transfer ratio of the optocoupler ( CTR Degradation process curve.
[0017] Figure 3 It is the failure probability density distribution before and after Bayesian data fusion.
[0018] Figure 4 These are the lifetime reliability and probability density distribution curves before and after Bayesian data fusion. Detailed Implementation
[0019] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.
[0020] Example: Figure 1 This is a diagram of the internal circuit structure of the optocoupler in this embodiment. The accelerated degradation model establishment method described in step S1 is as follows: Figure 1 As shown, accelerated degradation test data were selected under three temperature conditions (125℃, 150℃, and 175℃), with 5 samples under each condition, and measurements were taken. CTR Degradation process, such as Figure 2 As shown, the degradation trajectory is fitted using a three-parameter exponential function: ; in, for CTR Degradation amount, As the initial value, a , b These are the model parameters. The degradation rate is fitted using the Arrhenius model: ; The pseudo-lifetime was calculated and extrapolated to the actual operating temperature of 45℃. The degradation rate, pseudo-lifetime, actual operating lifetime, and failure rate parameters of the sample are shown in Table 1.
[0021] Table 1. Degradation rate, pseudo-lifetime, and predicted lifetime of the samples ; The expression for the failure rate probability density (prior distribution) obtained by fitting is: ; The average failure rate of the optocoupler calculated using the formula in step S4 is 981.5 × 10⁻⁶. -6 / week, with a mean failure time of 1018.9 weeks and a 90% confidence interval of [711.4, 1603.9] (weeks).
[0022] The first Bayesian update in step S2 introduced secondary screening data of 368 optocouplers in 4 batches. The test conditions were in accordance with GJB548C-2021 method 5004.2 (high temperature aging), as shown in Table 2. The total duration was 4071.9 weeks, and the number of failures was 2.
[0023] Table 2. High-Temperature Aging Test Conditions for Secondary Screening ; After constructing the likelihood function, the intermediate posterior distribution is obtained through Bayesian update as follows: ; The mean failure time of the optocoupler updated using the calculation formula in step S4 is 1127.1 weeks, with a 90% confidence interval of [800.98, 1725.48] (weeks).
[0024] Step S3, the second Bayesian update, collected historical failure data of the product under actual operating conditions. Two optocouplers failed at operating times of 1232 weeks and 1374 weeks, respectively. The failure density distribution was updated again through secondary data fusion. ; The final reliability evaluation result was obtained using the calculation formula in step S4: the mean time to failure of the optocoupler is 1143.9 weeks, with a 90% confidence interval of [825.47, 1710.85] (weeks). The failure probability density curves and lifetime probability density and reliability curves before and after Bayesian data fusion in this case are shown below. Figure 3 and Figure 4 As shown.
Claims
1. A method for evaluating the dynamic reliability of optocouplers based on Bayesian data fusion, characterized in that, Includes the following steps: S1. Using accelerated degradation test data, an accelerated degradation model of the optocoupler is established, and the probability density distribution of the failure rate is obtained as the prior distribution for Bayesian inference. S2. Introduce large sample data from the secondary screening stage, construct the likelihood function, perform the first Bayesian update on the prior distribution, and obtain the intermediate posterior distribution; S3. Combining the actual lifespan and failure-free data from field use, a second Bayesian update is performed to obtain the final posterior distribution; S4. Based on the final posterior distribution, calculate the mean failure time, reliability, and confidence interval of the optocoupler to complete the dynamic reliability evaluation.
2. The method for evaluating the dynamic reliability of optocouplers based on Bayesian data fusion according to claim 1, characterized in that, The accelerated degradation test data mentioned in step S1 includes the current transfer ratio degradation data of the optocoupler under high-temperature storage and power cycling stress conditions; the accelerated degradation model is modeled using the Arrhenius accelerated degradation model, and its expression formula is: ； in, t For degradation time, Indicates the degradation rate of the characteristic parameter. A It can be a pre-factor or a frequency factor; Ea The activation energy is expressed in joules per mole (J / mol). J / mol ); k It is the Boltzmann constant, which is 8.617 × 10⁻⁶. -4 e VK -1 , T It is absolute temperature, and the unit is Kelvin (K). K By fitting the degradation curve, the parameters of the accelerated degradation model are obtained, the pseudo-lifetime is calculated, and the lifetime is extrapolated to the actual operating temperature using the following formula: ； The prior distribution uses a gamma distribution to fit the failure rate parameters, and its probability density function is: ； in, For shape parameters, For rate parameters, This is a gamma function.
3. The method for evaluating the dynamic reliability of optocouplers based on Bayesian data fusion according to claim 2, characterized in that, The large sample data in the secondary screening step S2 includes the number of failures and the test duration under high-temperature aging test conditions; the likelihood function construction formula is: ； ； in, and These are the likelihood functions obtained using secondary screening data and field data processing, respectively. and These represent the number of failed data and the number of non-failed data, respectively. and The first i The failure time and the first j A time without failure, and These are the failure probability density function and the failure probability function, respectively; The formula for calculating the intermediate posterior distribution after performing the first Bayesian update on the prior distribution is as follows: ； Among them, the function For the posterior density function, For the prior density function, Let be the likelihood function.
4. The method for evaluating the dynamic reliability of optocouplers based on Bayesian data fusion according to claim 3, characterized in that, The field data mentioned in step S3 includes failure history data and no failure data under actual working conditions; the second Bayesian update uses the same likelihood function construction method as step S2 to update the intermediate posterior distribution again.
5. The method for evaluating the dynamic reliability of optocouplers based on Bayesian data fusion according to claim 4, characterized in that, In steps S2 and S3, during the Bayesian update, to achieve real-time and rapid prediction of the optocoupler and facilitate deployment, an approximate conjugate prior distribution is proposed to be used. When the conjugate prior distribution is a common distribution, the constant term is quickly obtained, thereby deriving the posterior distribution. The calculation expression is as follows: ； In the formula: This represents the total number of failures. This refers to the total duration of secondary screening or actual usage.
6. The method for evaluating the dynamic reliability of optocouplers based on Bayesian data fusion according to claim 4, characterized in that, The reliability calculation formula mentioned in step S4 is: ； in, The conditional reliability function is an exponential distribution, and the mean time to failure is calculated using the following formula: ； The confidence interval is calculated based on the quantiles of the posterior distribution, and is expressed as: ； in, and It is a posterior distribution of and Quantiles It is the significance level.