| Machine learning |
Material screening and composition prediction |
Rapid identification of stable, high-efficiency materials reduces experimental workload |
Requires large, high-quality datasets; may over fit to known materials |
153
|
| Neural networks |
Modelling degradation kinetics |
Model nonlinear relationships between fabrication parameters and PCE; adaptable to new data |
Often lacks interpretability; sensitive to noise in training data |
154
|
| Explainable AI |
Manufacturing optimization, material recovery, and decision support |
Provides insights into model decisions; improves trust and transparency |
Still emerging in materials science; limited toolkits |
155
|
| Support vector machines |
Classification of degradation patterns and failure modes |
Effective for small datasets; robust to overfitting |
Less effective for large, noisy datasets; limited scalability |
156
|
| Convolutional neural networks |
Image-based defect detection in perovskite films |
High accuracy in visual inspection enables real-time quality control |
Requires labeled image datasets; computationally demanding |
148
|
