Intelligent Manufacturing: A Fully Annotated Synthetic Dataset for Fiber Orientation Estimation

Area of Science:

• Materials Science
• Computer Vision
• Data-Driven Manufacturing

Background:

• Accurate fiber orientation estimation is crucial for predicting composite material properties.
• Existing methods for characterizing microstructure can be time-consuming and lack precise ground-truth data.

Purpose of the Study:

• To introduce a fully annotated synthetic dataset for machine learning-based fiber orientation estimation.
• To provide a benchmark for evaluating deep learning models in composite materials analysis.

Main Methods:

• Generation of 55,092 synthetic microscopy-like images using a physics-informed simulation pipeline.
• Calibration of image parameters (aspect ratio, orientation, areal fraction) using X-ray microscopy data.
• Analytical computation of the second-order orientation tensor for noise-free ground-truth annotations.

Main Results:

• A comprehensive dataset with high-resolution synthetic images and precise ground-truth orientation tensors.
• The dataset enables reproducible evaluation of machine learning models for orientation prediction.
• Open-source code and documentation facilitate dataset reuse and extension.

Conclusions:

• The synthetic dataset supports the development of automated workflows for composite characterization.
• This resource accelerates research in data-driven manufacturing and materials science.
• Facilitates advancements in intelligent manufacturing through improved microstructure analysis.