Lifetime Prediction of UV LEDs Based on Bayesian MCMC and Other Models
Objective The ultraviolet light-emitting diode(UV LED)is a semiconductor device capable of emitting ultraviolet light,with the development history dating back to the 1960s.During the 1960s and 1970s,UV light sources primarily relied on traditional sources such as fluorescent lamps and xenon lamps,which were often bulky and featured high power consumption and short lifetimes,thus limiting their applications.As more people become aware of the dangers of mercury pollution,the United Nations Environment Programme adopted the Minamata Convention on Mercury in 2013,which aims to control and reduce mercury emissions globally.In 2016,this convention was approved at the 20th meeting of the Standing Committee of the Twelfth National People's Congress.Nowadays,UV LEDs emerge as a new generation of semiconductor lighting devices to replace mercury lamps and related mercury-containing products.With continuous performance improvements,UV LEDs are gradually expanding their application range and have been widely employed in fields such as drone aerial cameras,UV disinfection devices,light-curing equipment,and optical sensors.As LED technology progresses,ensuring the long-term reliable operation of UV LEDs necessitates the adoption of appropriate reliability evaluation methods.A critical aspect of these methods is lifetime prediction.Due to the unique mechanism of radiant flux decay in UV LEDs,which involves rapid initial decay followed by a gradual slowdown,there is currently no fixed standard for reliability evaluation methods.Methods Typically,the reliability evaluation methods for white LEDs are applied,but the mainstream lifetime prediction methods for white LEDs and the determination standards for certain parameters do not entirely suit UV LEDs.Therefore,we propose a lifetime prediction method for UV LEDs based on the Bayesian MCMC and other functional models,comparing it with the TM-21 exponential model to validate the method and accuracy of the proposed models.Our study involves conducting aging tests on UV LEDs,analyzing the parameter changes before and after aging,and assessing the influence of different parameter fitting methods on the accuracy of the decay data of UV LED radiant flux over time.Meanwhile,appropriate UV LED test samples are selected for aging tests in working conditions.The parameter changes before and after the tests are analyzed.Exponential models are utilized to fit the radiant flux data over time for different data segments.Additionally,we conduct lifetime prediction and analysis of the test data by adopting the exponential model.Hypothesis testing indicates that the lifetime follows a normal distribution,from which the mean and standard deviations of the normal distribution are calculated to obtain the mean time to failure(MTTF).By combining other functional models such as power function models and logarithmic function models,the Bayesian MCMC model is applied for the first time to UV LED lifetime prediction,with model formulas proposed for comparison.We introduce the principles of the Bayesian MCMC model and combine the model with UV LED lifetime data.The accuracy of different models is compared via the model prediction results to verify the applicability of various methods.The methods and accuracy of selecting UV LED lifetime prediction models are further clarified.Results and Discussions The test results in working conditions show that the actual test and the predictions from the exponential model,Bayesian MCMC,power function model,and logarithmic function model all yield the sample lifetime L70,leading to an average lifetime.Compared with the actual test value of 6565 h,the L70 predicted by the exponential model is 5517 h,with a prediction error of 16.0%.The L70 predicted by the Bayesian MCMC model is 7041 h(Fig.10),with a prediction error of 7.2%.The power function model has a predicted value of 6849 h(Fig.11),with a prediction error of 4.3%.The value predicted by the logarithmic function model is 6219 h(Fig.12),with a prediction error of 5.3%.It can be concluded that compared to the exponential fitting model,the Bayesian MCMC,power function model,and logarithmic function model are more accurate,with the power function model being the most precise.Conclusions During data fitting,it is found that the Bayesian MCMC model's fitting value is relatively stable due to the influence of the initial fitting values.However,due to the randomness of the sampling process,there are slight differences in the sampled data each time,but these differences fall within a reasonable error range.The power function model can fit data segments similarly to the exponential model but must exclude the initial rapidly decaying data segments.The logarithmic function model is significantly affected by the data segments adopted for fitting,requiring careful selection of data segments.In the prediction of related samples,each of the above models has its merits.For accurate predictions that are not influenced by data segments,the Bayesian MCMC model can predict the lifetime L70 of related devices.
optical devicesultraviolet light-emitting diodelifetime predictionBayesian Markov Chain Monte Carlopower function model
NETL
NSTL
万方数据
