Nanocrystal-based optical diffusers for white LED lighting - inverse design achieved by machine learning - Master of Science thesis

Aug 4, 2026

White lighting-emitting diodes (WLEDs) demand efficient control of spatial light distribution, typically achieved using lenses and mirrors. This thesis proposes optical diffusers using nanocrystal nanocomposite films to tune scattering properties by varying nanocrystal concentration, enabling tailored illumination performance. Machine learning is used for inverse design of symmetric (2D) diffuser patterns and extension to non-symmetric (3D) designs. Demonstrations include several predefined WLED intensity patterns to validate the proposed neural network architectures and design workflow.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages83

Reading Time3 h 29 min

AccessFree to download

Chapter 1. Introduction

1.1 Introduction to white light-emitting diodes (WLED)

1.2 Improving the performance of WLEDs by employing cellulose-nanocrystal (CNC)-based optical diffusers

Nanocrystal-based optical diffusers for white LED lighting: inverse design achieved by machine learning technologies by Gangyi Li A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Microsystems and Nanodevices Department of Electrical and Computer Engineering University of Alberta © Gangyi Li, 2023 Abstract Illumination receives a great deal of attention as white lighting-emitting diodes (WLEDs) become energy-efficient light sources in households and commercial buildings, on streets and highways, and at stadiums and construction sites. In general, lenses and mirrors are used to control the spatial distribution ofWLED light. In this thesis, we propose to use optical diffuser, the key optical device in scattering optics, to achieve the desired figures of merit for WLEDs. Optical diffusers are typically used to create soft light (similar brightness from any angle of view), however, here we can alter the concentration of nanocrystals in the nanocomposite film (optical diffuser) to control its optical property. Machine learning is employed to achieve the inverse design of the optical diffuser pattern on WLEDs, and this design task is beyond human capacities which are generally carried out using the brute force approach (solving a problem through exhaustion) . In this thesis work, we have constructed the neural network architectures for machine learning and used them to achieve the inverse design of a symmetric pattern of optical diffusers (twodimensional design) on WLED modules. Furthermore, we focus on achieving the inverse design of non-symmetric patterns of optical diffusers (three-dimensional design), and several predefined patterns ofWLED light intensity are demonstrated for showing the success of our efforts. Preface This thesis is submitted for the degree of the Master of Science at University of Alberta. This Master thesis contains results of the research undertaken in the Department of Electrical and Computer Engineering, University of Alberta, from September 2021 to December 2022, under the supervision of Professor Xihua Wang and Professor Hao Liang. Chapter 2 of this thesis is based on a journal publication-Gangyi L, Yuan L, Qiwei X, Hao L, and Xihua W, "Deep learning enabled inverse design of nanocrystal-based optical diffusers for efficient white LED lighting," Appl. Opt. 61, 8783-8791 (2022) . I was responsible for building optical module and carrying out simulation works with the instructions from Qiwei X. Yuan L led the work on the design and training of neural network models, and I also participated invalidation of neural network models. Wang X and Hao L were the corresponding authors. Chapter 3 of this thesis is based on a paper titled “Scattering optics for LED lighting achieved by nanocrystals and machine learning”, and the manuscript is in preparation. I and Yuan L are cofirst authors of the paper. My work focused on constructing optical module and carrying out simulation. Yuan L was responsible for designing the entire deep learning approach and training the models. Wang X and Hao L are the corresponding authors. Acknowledgement As the farewell for my life at University of Alberta, foremost, I would like to express my deepest respect and thanks to my supervisors, Professor Xihua Wang and Professor Hao Liang, who give me the chance to pursue my Master degree at University of Alberta in Canada. I still could remember scene I met with them at first time, they gave me a lot of advice to help me get used to the new environment and the chances to participate in their research groups. In the past two years, When I encounter difficulties in my research work or taking courses, they will always provide me with patient and excellent guidance and instructions, even sacrificing their rest time. The two professors always give me advice and training on my weakest part communication and presentation skills, which have been really helpful for

FAQs

What problem does this thesis address for white LED lighting?

It addresses how to achieve desired spatial illumination figures of merit for WLEDs beyond conventional lens and mirror approaches.

How does the proposed optical diffuser work?

By altering the concentration of nanocrystals in a nanocomposite film used as an optical diffuser to control its optical scattering properties.

How is machine learning used in the inverse design process?

Neural network architectures learn to generate diffuser patterns for WLEDs, supporting inverse design for symmetric (2D) and non-symmetric (3D) diffuser configurations.

Aug 4, 2026

Upload by River Wang