Session: 17-01-01: Research Posters

Paper Number: 149723

149723 - Multi-Class Defect Identification and Localization in Cfrp Composite Material Using Infrared Images and Mask Rcnn 

Composite materials are gaining popularity in diverse engineering applications due to their outstanding strength-to-weight ratio. These materials can be tailored to achieve desired mechanical properties since they allow for customized composition and structure, making them suitable for various applications. Despite having such structural benefits, composite materials are prone to subsurface defects. Detecting the defects lying in the subsurface laminate is challenging. Diagnosing existing defects within the composite laminate structure with non-destructive testing (NDT) methods has recently become indispensable for structural health monitoring. Hence, Infrared (IR) thermographic analysis comes into the picture. IR image captures subsurface heat variation, proving effective in spotting delamination, cracks, and voids in composite structures up to a certain depth. Yet, IR thermography faces obstacles due to the limited availability of IR image databases and the diverse nature of subsurface defects. Additionally, IR images are low resolution and are prone to fish-eye effects. Due to these limitations, developing a comprehensive defect detection approach using IR thermography alone is impossible. However, mask region-based convolutional neural network (Mask-RCNN) is a promising algorithm that can be used to detect defects in the IR images. By segmenting each item at the pixel level, Mask R-CNN creates a high-fidelity mask for every heat variation by identifying the borders of heat variation or potential subsurface defects within IR images. Using this feature of Mask R-CNN that combines object detection and instance segmentation, even abstract shapes of these potential subsurface defects can be identified. Thus, the implementation of Mask-RCNN along with IR thermography is proposed in this research. The research question developed for this study is: Can Mask R-CNN identify and localize irregularities automatically from thermal images of CFRP specimens collected experimentally? Two specific aims were developed to address this research question: 1) Create a robust and highly accurate framework using Mask R-CNN for IR images, and 2) Automatically localize and identify the irregularities from IR images of the CFRP dataset. In this study, an IR dataset of composite materials with Teflon inserts of various sizes was used. First, 120 out of 150 images were annotated using annotation software to create the training dataset, and the rest of the images were unannotated for testing the algorithm. The accuracy was calculated for all 30 test images. The model's performance was analyzed with the increasing number of training epochs. Results indicated significant improvements in accuracy and loss over the increasing number of epochs, with accuracy increasing from 77.26% at 30 epochs to 97.91% at 100 epochs and loss decreasing from 0.3537 to 0.2753. This approach offers a practical, non-destructive method for condition monitoring, crucial for creating a maintenance schedule to monitor the integrity of various structural components over time. Additionally, this framework can be implemented in different applications where detecting subsurface defects is a concern. 

Presenting Author: Aditi Barua University of Arkansas at Little Rock

Presenting Author Biography: Aditi Barua is a master's student in the Department of Mechanical Engineering at the University of Arkansas at Little Rock (UALR). She was awarded the Donaghey College of Science, Technology, Engineering and Mathematics (DCSTEM) Graduate Assistantship in Fall 2023. Aditi is now the graduate research assistant (GRA) at Dr. Dabetwar's Material Intelligence and Prognostics (MIP) Lab. Her research focuses on the automated detection of subsurface defects in composite materials, utilizing IR thermography and artificial intelligence. Through her research, she aims to develop highly accurate, noninvasive, automated condition-monitoring techniques and significantly contribute to the Structural Health Monitoring sector.

Aditi has demonstrated strong leadership qualities in various organizations. She has won two national-level championships in creative writing and has been awarded merit-based scholarships four times by the Government of Bangladesh. In her leisure time, Aditi enjoys singing, painting, and swimming.

Authors:

Aditi Barua University of Arkansas at Little Rock
Shweta Dabetwar University of Arkansas at Little Rock