Deep learning for augmented process monitoring of scalable perovskite thin-film fabrication

This repository contains the code for the paper Deep learning for augmented process monitoring of scalable perovskite thin-film fabrication. The project explores the use of deep learning for material classification, device performance prediction, and process control in the fabrication of blade-coated perovskite solar cells.

adapted from original publication

Dataset Overview

The dataset, In Situ Photoluminescence Dataset for Exploring Material and Processing Variabilities in Blade-Coated Perovskite Photovoltaics, includes:

• Time-resolved in situ photoluminescence (PL) and diffuse reflection data.
• Material-level properties: precursor composition (molarity and molar ratio).
• Device-level performance metrics: power conversion efficiency (PCE), open-circuit voltage (VOC), short-circuit current density (JSC), and fill factor (FF).

The dataset can be downloaded from Zenodo.

Getting Started

Prerequisites

• Required packages can be installed using: pip install -r requirements.txt

Setup

1. Clone the repository: git clone https://github.com/PerovskitePV/DL-Fabrication-Monitoring cd DL-Fabrication-Monitoring

2. Download the dataset: Save data.h5 from Zenodo in the data/ folder.

3. Download the forecasting models: Save the random forest models from Zenodo in the models/forecasting_models/ folder.

Citation

If you use this repository or the dataset in your work, please cite our paper:

@article{https://doi.org/10.1039/D4EE03445G,
  title = {Deep learning for augmented process monitoring of scalable perovskite thin-film fabrication},
  ISSN = {1754-5706},
  url = {http://dx.doi.org/10.1039/D4EE03445G},
  DOI = {10.1039/d4ee03445g},
  journal = {Energy & Environmental Science},
  publisher = {Royal Society of Chemistry (RSC)},
  author = {Laufer,  Felix and G\"{o}tz,  Markus and Paetzold,  Ulrich Wilhelm},
  year = {2025}
}

Reproducing the Results

The repository includes the following scripts to reproduce the results from the paper Deep learning for augmented process monitoring of scalable perovskite thin-film fabrication.

Structure

├── README.md                                                           
├── requirements.txt                                        - txt-file to install the environment
├── 00a_generate_Material_train_test_folds.ipynb            - Script to generate the same Material train-test-splits and the same folds for cross-validation
├── 00b_generate_PCE_train_test_folds.ipynb                 - Script to generate the same PCE train-test-splits and the same folds for cross-validation
├── 01_material_composition_monitoring_molar_ratio.ipynb    - Script to reproduce results shown in Fig 2 (and S6b)
├── 02_material_composition_monitoring_molarity.ipynb       - Script to reproduce results shown in Fig 2 (and S6a)
├── 03_device_performance_prediction.ipynb                  - Script to reproduce results shown in Fig 3 (and S11)
├── 04_signal_forecasting.ipynb                             - Script to reproduce results shown in Fig S13 and S14
├── 05_forecasting_and_performance_prediction.ipynb         - Script to reproduce results shown in Fig 5 & 6 (and S15)
├── data
|   └── ...                                                 - data.h5 has to be downloaded from zenodo
├── models
|   ├── regression_model     
|   |   └── model_weights.pth                               - weights for the regression model
|   └── forecasting_models   
|       ├── ...                                             - random forest model have to be downloaded from zenodo
|       └── ...                                             
└── images
    └── concept.png                                         - project's concept overview