先進的機器學習驅動的高通量方法，用於合理設計多共振熱激活延遲螢光材料;Advanced Machine Learning-Driven High-Throughput Approach for the Rational Design of Multi-Resonance Thermally Activated Delayed Fluorescence Materials

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/94694

Title:	先進的機器學習驅動的高通量方法，用於合理設計多共振熱激活延遲螢光材料;Advanced Machine Learning-Driven High-Throughput Approach for the Rational Design of Multi-Resonance Thermally Activated Delayed Fluorescence Materials
Authors:	黃義生;Huang, YI-Sheng
Contributors:	化學學系
Keywords:	機器學習;熱活化延遲螢光;有機發光二極體;Machine Learning;Thermally Activated Delayed Fluorescence;Organic Light Emitting Diodes
Date:	2024-07-29
Issue Date:	2024-10-09 15:24:25 (UTC+8)
Publisher:	國立中央大學
Abstract:	熱活化延遲熒光 (TADF) 技術通過提高效率、色純度和成本效益，特別在智慧型手機和液晶顯示器 (LCD) 中，增強了有機發光二極體 (OLEDs)。含有硼的多共振 (MR) TADF 材料具有色純度高和窄的發射光譜。有效的反系統間跨越 (RISC) 過程對實現 100% 的內部量子效率至關重要。然而，設計潛在的 MR-TADF 材料仍具挑戰，因其靈活的分子設計以及準確預測 RISC 速率的困難性。在此，我們提出了一種先進的機器學習 (ML) 驅動的高通量虛擬篩選 (HTVS) 方法，用於合理設計 MR-TADF 材料。通過利用 156 個同行評審的 MR-TADF 材料數據庫，我們開發了基於 XGBoost、ANN 和 CNN 學習器的 ML 預測模型，使用從密度泛函理論 (DFT) 計算得出的分子描述符集合 (MDS)。我們的模型在kRISC的預測精度上達到了很高的水平，其中考慮了激發態特性的 ANN/MDS-B* 模型表現最為優越。我們使用這些模型篩選了基於 11 個 MR-TADF 核心設計的 2,288 個新 MR-TADF 候選物，識別出具有高 kRISC 值的有希望分子，並推導出進一步的化學設計規則。本研究強調了基於 ML 的 HTVS 在加速發現高性能 MR-TADF 材料方面的有效性，闡明了設計實驗性 OLEDs 所需的重要供體 MR-TADF 核心架構。;Thermally activated delayed fluorescence (TADF) technology enhances OLEDs with efficiency, color purity, and cost-effectiveness, especially in smartphones and LCDs. Multi-Resonance (MR) TADF materials with boron offer high color purity and narrow emission spectra. Efficient reverse intersystem crossing (RISC) processes are crucial for achieving 100% internal quantum efficiency. However, designing potential MR-TADF materials remains challenging due to their flexible molecular design and the difficulty in accurately predicting RISC rates. Here, we present an advanced machine learning (ML) driven high-throughput virtual screening (HTVS) approach for the rational design of MR-TADF materials. By utilizing a database of 156 peer-reviewed MR-TADF materials, we developed ML predictive models based on XGBoost, ANN, and CNN learners, employing molecular descriptor sets (MDS) derived from density functional theory (DFT) calculations. Our models achieved high predictive accuracy for kRISC, with the ANN/MDS-B* model, which considers the excited-state properties, demonstrating superior performance. We utilized these models to screen 2,288 newly designed MR-TADF candidates based on 11 MR-TADF cores, identifying promising molecules with high kRISC, values and deriving further chemical design rules. The study underscores the effectiveness of ML-based HTVS in expediting the discovery of high-performance MR-TADF materials, elucidating important donor MR-TADF core architectures for designing advanced organic light-emitting diodes (OLEDs) for experimental discovery.
Appears in Collections:	[Graduate Institute of Chemistry] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	6	View/Open

社群 sharing

Loading...