README.md

metadata

license: mit
configs:
  - config_name: default
    data_files:
      - split: original
        path: data/original-*
      - split: augmented
        path: data/augmented-*
dataset_info:
  features:
    - name: image
      dtype: image
    - name: L
      dtype: float32
    - name: Fx
      dtype: float32
    - name: Fy
      dtype: float32
    - name: xP
      dtype: float32
    - name: Fy2
      dtype: float32
    - name: xQ
      dtype: float32
    - name: Ax
      dtype: float32
    - name: Ay
      dtype: float32
    - name: By
      dtype: float32
    - name: overloaded
      dtype:
        class_label:
          names:
            '0': safe
            '1': overloaded
  splits:
    - name: original
      num_bytes: 173067
      num_examples: 30
    - name: augmented
      num_bytes: 22170328
      num_examples: 3000
  download_size: 22306940
  dataset_size: 22343395

Beam Image HW1

Dataset Summary

beam_image_hw1 contains free-body diagram style images of simply supported beams with applied point loads.
Each image is paired with features (beam span, load magnitude, positions) and a binary target: safe vs overloaded.

• Original split: 30 images (224×224 PNG) generated from parameterized cases
• Augmented split: 300+ images created via label-preserving transformations (rotation, brightness shift, noise, mirroring)

---

Purpose

This dataset is designed for introductory computer vision tasks in an engineering context.
Models can experiment with classifying safe vs overloaded beam cases based on diagram inputs.

---

Composition & Collection

Features:

• image (PNG, 224×224): beam free-body diagram
• L (float32): beam span length (m)
• Fx, Fy, Fy2 (float32): applied loads (N)
• xP, xQ (float32): load positions (m)
• Ax, Ay, By (float32): support reactions (N)
• overloaded (class): safe (0), overloaded (1)

Original images: generated programmatically via a Python renderer
Augmented images: expanded with simple transformations

---

Preprocessing & Augmentation

• Original diagrams resized to 224×224 pixels
• Augmentation methods:
  • Random flips and rotations (±10°)
  • Brightness/contrast shifts
  • Gaussian noise injection
• Labels preserved because augmentations do not affect statics

---

Splits

• original: 30 rows
• augmented: 300+ rows

---

Intended Use / Limits

• Educational only
• Not suitable for real engineering decision-making

---

Ethical Notes

• Fully synthetic diagrams, no sensitive content
• No copyrighted imagery used

---

License

MIT License

---

AI Tool Disclosure

ChatGPT (OpenAI, GPT-5) was used for:

• Code support
• Documentation writing

All diagrams, dataset generation, and labeling logic were created and validated by the author.