Research On High-power LED Intelligent Coating Technology

White LED is known as the fourth-generation lighting sources due to its low energy consumption,durability,environmental protection,quick response and other characteristics.It has a very broad prospect in the lighting fields such as modern agriculture,optical wireless communication and daily life.Countries attach great importance to the development of white LED.At the same time,due to the rapid development of the white LED industry,people’s requirements for the performance of white LED products have become higher and higher.Among them,determining the coating thickness of phosphor powder has a great impact on the performance of white LED products,and slight changes in the coating thickness of phosphor powder may lead to the decrease of the performance of white LED products.In this paper,by collecting the data of the ratio of phosphor and glue,phosphor powder coating,lens plastic sealing modeling and spectral separation,and combining the knowledge of colorimetry with the relevant data analysis of white LED,a color coordinate prediction model for white LED is proposed.Connect the color coordinate of the blue LED chip and the color coordinate of the thickest phosphor powder coated on the blue LED chip to form a straight line,then the color coordinate of the phosphor powder coated on the blue LED chip must be on the straight line.So,we just focus on the x-coordinate,and then we plug the predicted x-coordinate into the equation of the line,and we get the corresponding y-coordinate.Through formula calculation,the conversion of CIE1931-XYZ to HSV color space is obtained,and the corresponding hue,saturation and lightness are obtained.Then the luminous flux can be converted into brightness by formula calculation.In this paper,the color coordinate-X prediction model and the flux prediction model of white LED light efficiency are constructed in two ways: multiple linear regression(MLR)prediction and deep confidence network(DBN)prediction model.Multiple linear regression prediction is mainly realized by LINEST function in EXCEL.In the Deep Belief Network,layer-by-layer greedy training algorithm is used for forward training,and then gradient descent algorithm is used for reverse fine-tuning.The number of network layers and the number of hidden layer nodes are determined by the experimental method,then the network structure of the optimal DBN prediction model for the color coordinate-X prediction model and the optical flux prediction model for the white light efficiency are determined as 3-18-10-1 and 3-20-26-3 respectively.By calculating and comparing the root mean square error(RMSE)and mean absolute error(MAE)of the prediction results of the four models,it can be obtained that no matter which prediction is made,it is obtained that the prediction result of DBN model is better than that of MLR model.DBN was used to establish the color coordinate-X prediction model and the flux prediction model to predict the luminous effect of white LED.By comparing with the expected luminous effect,the corresponding optimization method of phosphor powder coating thickness is put forward,thus an intelligent diagnosis and feedback system is formed for packaging process quality detection such as white LED phosphor powder formula,phosphor powder coating thickness scheme and spectral separation,so as to realize intelligent quality control in packaging production process and greatly reduce the number of experiments.