MACHINE LEARNING-GUIDED SYNTHESIS OF ROOM-TEMPERATURE PHOSPHORESCENCE CARBON DOTS WITH OPTICAL PROPERTIES PREDICTION

• Main Author: Addini M M, Diva
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70310
• Institution: Institut Teknologi Bandung

Carbon dots (CDs) have received tremendous interest due to their properties and numerous promising applications. CDs have a quasi-spherical morphology with a diameter of less than 20 nm and are composed of a sp2 hybrid carbon core conjugated with organic functional groups decorating their surface. CDs have good properties, including adjustable wavelength, non-toxic, significant stoke shift, and long luminescence lifetime. This CDs property are crucial in chemical sensors, bioimaging, anti-counterfeiting, optoelectronic devices and so on. To optimally utilize CDs, a strategy is needed to regulate their optical properties so that they have photoluminescence, including fluorescence and phosphorescence. One that influences the optical properties of CDs is the parameter of material synthesis. A breakthrough is needed in formulating the synthesis process to find a suitable synthesis formulation to produce CDs with optimal optical properties so that the experiments carried out can be effective and efficient. One way that can be done is to conduct experimental and computational studies using machine learning to study and predict the optical properties of CDs. Synthesis of materials with a minimum number of trials is crucial for accelerating the development of CDs because it requires considerable time, effort, and cost. In analyzing the relationship between synthesis parameters and the optical properties of CDs, reliable computational methods are needed. Machine learning (ML) is a computational method that has begun to be implemented in various fields, including materials science, to study CDs' experimental data. Therefore, in this research, a study of the effect of synthesis parameters on the optical properties (fluorescence and phosphorescence) of CDs was carried out guided by machine learning. Machine learning can study extensive data with high complexity to produce precise predictions. In addition, the use of machine learning to guide material synthesis in predicting CDs' phosphorescence has never been done before. This study reports the synthesis results of CDs, which produce room-temperature phosphorescence (RTP) using microwave radiation with ethylenediamine and phosphoric acid as precursors. Synthesis parameters such as precursor concentration and irradiation duration are varied and produce CDs with various optical characteristics, which will later be used as the initial ML data set. The results of Transmission Electron Microscopy (TEM) and FourierTransform Infrared (FTIR) characterization confirm the formation of CDs, which are shown from the dispersed points in the TEM image and the presence of peak in the FTIR spectrum, indicating the vibration of the sp3 C-H and C=C bonds indicating the formation of a ? conjugation system as the core structure of CDs. The synthesis results of CDs showed blue fluorescence and green phosphorescence when excited at a wavelength of 365 nm. CDs' phosphorescent can last up to 7 seconds and be seen with the naked eye. The results of the Photoluminescence (PL) spectrophotometer characterization showed that the emission peaks from CDs were at wavelengths of 420 nm and 515 nm for fluorescence and phosphorescence, respectively. From the results of various synthesis parameters, the highest phosphorescent lifetime was obtained, up to 1.566 seconds, indicating ultralong room temperature phosphorescence (URTP) on CDs. Experimental data were then studied using three ML algorithms, including extreme gradient boost (XGB), random forest (RF) and decision tree (DT). The evaluation results of the ML model show that XGB has the best coefficient of determination (R2) and mean absolute error (MAE) among the other models. The results of the XGB model hyperparameter settings show R2 to 0.904 dan 0.915 for the test and train data sets, respectively. The XGB model is used to predict the average lifetime of new data containing CDs synthesis parameters which have never been done before. The most optimal prediction results were then validated by re-experimenting, and the prediction results obtained the highest average lifetime of up to 1.597 s with a measurement error of up to 7.33%.