solar fault classifier

this project implements a convolutional neural network (cnn) for classifying solar panel faults from images. the model detects four types of faults:

• bypass diode activated
• junction box
• multicell hotspot
• single cell hotspot

project structure

.
├── app.ipynb              # jupyter notebook for model training
├── test.ipynb             # jupyter notebook for model evaluation
└── images_classed         # image dataset

installation

1. clone the repository:

   git clone https://git.sr.ht/~obk/solar-fault-classifier

2. install dependencies:

   pip install -r requirements.txt

note: a gpu is recommended for faster training.

usage

1. model training (app.ipynb)

1. run all cells in app.ipynb:
   • trains cnn model with data augmentation
   • monitors training progress through accuracy/loss metrics
   • saves final model as solar_fault_classifier.h5

2. model evaluation (test.ipynb)

1. execute all cells in test.ipynb:
   • evaluates model performance (using same images due to data limitations)
   • displays accuracy metrics and visualizations
   • shows prediction comparisons with actual labels

3. prediction examples

key components

• app.ipynb: model training notebook

  • cnn architecture with convolutional layers and dropout
  • uses imagedatagenerator for augmentation/validation
  • saves trained model weights

• test.ipynb: model evaluation notebook

  • loads pretrained model
  • generates performance metrics
  • creates visualizations of predictions

• dataset structure (images_classed/):

  images_classed/
  ├── bypass_diode_activated
  ├── junction_box
  ├── multicell_hotspot
  └── single_cell_hotspot