FlowRatio Analyzer – CNN-Based Classification of Extrusion Flow Ratio in FDM 3D Printing

A system for classifying flow-ratio calibration blocks in FDM 3D printing using convolutional neural networks.

About

FlowRatio Analyzer is a machine-learning project developed for an engineering thesis.
The system analyzes images of 3D-printed calibration blocks and classifies them into three categories:

• Under-extrude
• Optimal extrusion
• Over-extrude

The project includes a complete pipeline: dataset preparation, model training, evaluation, and comparison of architectures.

Dataset

All images in this dataset were captured by me during my own FDM calibration runs.

FDM Extrusion Calibration Dataset

Features

• Image Preprocessing: cropping, resizing, normalization, augmentation
• Two Neural Network Architectures:
  • Custom CNN
  • MobileNetV2 (transfer learning)
• Cross-validation: Stratified K-Fold for reliable model evaluation

Results

• Custom CNN

  • Average Accuracy: 79.92%
  • Best Accuracy: 85.21% (Fold 2)

• MobileNetV2

  • Average Accuracy: 98.08%
  • Best Accuracy: 99.37% (Fold 5)

Technology Stack

• Language: Python
• Deep Learning: TensorFlow / Keras
• Data Handling: NumPy, PIL, scikit-learn
• Visualization: Matplotlib
• Environment: CUDA-accelerated GPU training (NVIDIA)

Author

Mateusz Andrzejewski
Engineering Thesis Project (2025)