A Printing-Inspired Digital Twin for the Self-Driving, High-Throughput, Closed-Loop Optimization of Roll-to-Roll Printed Photovoltaics
Description
Datasets related to the manuscript 'A Printing-Inspired Digital Twin for the Self-Driving, High-Throughput, Closed-Loop Optimization of Roll-to-Roll Printed Photovoltaics' by Ng et al. The hypothesis of this research is that integrating high-throughput roll-to-roll (R2R) fabrication with machine learning (ML) models can significantly enhance the scalability and performance of printed organic photovoltaic (PV) devices. By employing a digital twin approach, fabrication parameters can be optimized in real-time, improving power conversion efficiency (PCE) and overall device performance through closed-loop feedback. The dataset includes raw data from the initial fabrication and characterization of 11,800 devices, along with refined data for ML modeling. The PV devices were fabricated using an R2R process, with layers deposited onto polyethylene terephthalate (PET) strips, and characterized under 1 sun conditions (AM 1.5G). The raw data contains 649,055 data points, including fabrication parameters, characterization results, and environmental conditions. The dataset comprises two main parts: Dataset 1: The raw data collected from the initial high-throughput fabrication and characterization of 11,800 organic photovoltaic devices. Dataset 2: The refined and cleaned dataset used for machine learning modeling, which includes training, validation, and testing subsets. Data Collection Methodology Fabrication: The photovoltaic devices were fabricated using a roll-to-roll process, with functional layers deposited onto a strip of polyethylene terephthalate (PET) with a patterned transparent conducting electrode (TCE). Multiple sensors embedded in the MicroFactory collected data from 36 manufacturing features, including deposition parameters, temperatures, and environmental conditions. Characterization: Each device was characterized in-line under standard 1 sun conditions (AM 1.5G), measuring key performance metrics such as short-circuit current density (Jsc), open-circuit voltage (Voc), fill factor (FF), and power conversion efficiency (PCE). Raw Data (Dataset 1): Contains 649,055 data points, including fabrication parameters, characterization results, and environmental conditions for each of the 11,800 devices. Cleaned Data (Dataset 2): Includes selected variables from Dataset 1 that were used for machine learning modeling, divided into training, validation, and testing sets.