Session: 01-05-02: Applications of Artificial Intelligence/Machine Learning for Aerospace Structures

Paper Number: 137043

137043 - Unsupervised and Supervised Machine Learning Algorithms for Low Velocity Impact Damage Quantification in Cfrp Composites 

Low velocity impact damage in composites is one of the critical considerations in the design of damage tolerant composite structures. As a result, damage detection and evaluation are areas of active research that continue to attract a lot of attention in the research and applied engineering communities. Particularly precarious situations arise when no apparent damage or barely visible damage at the impacted surface of the composite structure is accompanied by a significant reduction in structure’s strength, stiffness, and durability.

Micro Computed Tomography (micro-CT) emerged as one of the most advanced nondestructive evaluation methods that enables one to obtain high resolution 3D image data. Development of commercial micro-CT systems made the micro-CT technology a widely accessible tool for damage detection, visualization and assessment in composite structures. Image analysis in materials typically consists of three main steps: (i) image processing, (ii) segmentation, and (iii) microstructure and damage evaluation. One of the most critical steps in image analysis is image segmentation, which is essentially a discretization of the grayscale set of images. The objective of the discretization is to accurately discern regions/features of interest. Segmentation is one of the most important steps in the image processing. In the case of damage analysis in the materials and structures, the role of segmentation is to isolate the damaged region. Therefore, an accurate assessment of the damage from CT data depends to a great extent on the image segmentation.  At the same time, image analysis of the micro-CT data in composite materials is not standardized, and accurate interpretation of various damage features (e.g. cracks, voids, delamination) and boundary of the damage zone depends to a great extent on the qualifications of an analyst as gray-value distributions vary significantly from specimen to specimen.

In the present work, we develop a novel machine learning (ML) approach to automatic image segmentation of low velocity impact damage in carbon fiber reinforced polymer (CFRP) composite. The approach is based on synergy of unsupervised and supervised ML algorithms, which combines strong mathematical rigor of the unsupervised ML methods with flexibility and accessibility of the supervised ML algorithms. The unsupervised ML algorithm relies on the use non-parametric statistical methods in conjunction with the so-called intensity-based segmentation, which enable to determine the thresholds of image histograms and isolate damage. Rigorous unsupervised ML algorithm serves as initial benchmarks for more flexible, but less rigorous supervised ML algorithms. The supervised ML methods include U-Net Deep Learning Architecture and Fully Convolutional DenseNet (FC-DenseNet).

The developed ML approach has been applied to the analysis of low velocity impact damage in carbon fiber reinforced polymer (CFRP) composites. The grayscale images from the CT scans of the impacted CFRP specimens has been analyzed to identify and isolate impact damage.  

Presenting Author: Olesya Zhupanska University of Arizona

Presenting Author Biography: Dr. Zhupanska is Professor in the Department of Aerospace and Mechanical Engineering a the University of Arizona