基于深度学习的LED荧光粉层结构设计方法  

Deep Learning Based Design Method for LED Phosphor Layer Structure

作　　者：蔡义昕 贾婧媛 王榕梓 范鸿吉 齐成祥 苏莹 曹暾 CAI Yixin;JIA Jingyuan;WANG Rongzi;FAN Hongji;QI Chengxiang;SU Ying;CAO Tun(School of Optoelectronic Engineering and Instrumentation Science,Dalian University of Technology,Dalian Liaoning 116024,CHN)

机构地区：[1]大连理工大学光电工程与仪器科学学院,辽宁大连116024

出　　处：《光电子技术》2024年第4期289-297,305,共10页Optoelectronic Technology

基　　金：国家重点研发计划(2019YFA0709100)。

摘　　要：针对传统的荧光粉LED(Light Emitting Diodes)模型荧光粉层结构分布特征设计效率低且生产成本高的问题,提出一种基于深度学习的荧光粉层结构分布特征设计方法。通过建立荧光粉LED的仿真模型,探究荧光粉层分布特征与出光效果的映射关系并建立数据集。之后基于数据集和深度神经网络建立正向网络和反向网络。正向网络可以用于建立荧光颗粒分布特征与出光效果的映射关系;反向网络基于上述的映射关系,可以根据给定的出光效果反向输出对应的分布特征,用于指导荧光粉层的制备工艺。根据训练结果,网络模型的预测结果真实值和预测值的拟合度较好,正向网络的平均相关系数R2为0.996 4,反向网络的平均相关系数R2为0.948 3,表明所训练的深度学习模型具有良好的泛化能力。Aiming at the low efficiency and high production cost of the traditional phosphor LED(Light Emitting Diodes)model phosphor layer structure and distribution feature design,a deep learn-ing-based phosphor layer structure distribution feature design method was proposed.By establishing a simulation model of phosphor LED,the mapping relationship between the phosphor layer distribution features and the light emission performance was explored and a dataset was established.After that,forward network and inverse network were established based on the dataset and deep neural network.The forward network could be used to establish the mapping relationship between the distribution char-acteristics of phosphor particles and the optical output properties;the inverse network,based on the mapping relationship above,could output the corresponding distribution characteristics according to the given optical properties,which could be used to guide the preparation process of the phosphor lay-er.According to the training results,the network model had a good fit between the real and predicted values of the prediction results,with an average correlation coefficient R2 of 0.9964 for the forward network and 0.9483 for the inverse network,which indicated that the trained deep learning model had a good generalization ability.

关 键 词：深度学习 神经网络 发光二极管 荧光粉 逆向设计 

分 类 号：TN702[电子电信—电路与系统] TP302.1[自动化与计算机技术—计算机系统结构]