Exploring the Potential of Thermographic Data and Machine Learning for Defect Detection in Automated Fiber Placement (AFP)

Abstract

(Part1) The increasing demand for composites in the industry has led to the development of new automated manufacturing techniques such as Automated Fiber Placement (AFP). However, owing to some factors, small defects can occur during the AFP process which may compromise the structural integrity of the product. Traditionally these defects are detected manually by inaccurate and time-consuming inspection methods. In order to resolve this problem, it is necessary to implement effective automated defect detection systems. As an attempt to contribute to the solution of the problem, this paper reviews the current state of automated defect detection in AFP, and explores the potential of machine learning for improving defect detection. The review provides background information on composite materials, traditional and automated composite manufacturing methods, and defects associated with AFP. It also discusses traditional inspection techniques used in AFP and existing techniques for automated defect detection, in particular machine learning-based techniques. In addition, it also covers different data acquisition approaches for training ML algorithms. The paper aims to contribute towards advancing automated defect detection in AFP by utilizing machine learning techniques coupled with thermographic datasets in order to offer fast, accurate and cost-effective defect detection methods.(Part2) This comprehensive paper, presented in two parts, addresses the critical challenge of automated defect detection in the Automated Fiber Placement (AFP) process for composite manufacturing. The first part provides a detailed review of the current state of automated defect detection, emphasizing the potential of machine learning to enhance this process. Covering essential background information on composite materials, traditional and automated manufacturing methods, and AFP-associated defects, the paper explores traditional inspection techniques and machine learning-based approaches. It also delves into different data acquisition strategies for training machine learning algorithms, aiming to contribute to advancing automated defect detection in AFP.

Building upon the foundation established in Part 1, the second part extends the exploration by focusing on the application and evaluation of machine learning algorithms using thermographic datasets. Objectives include developing efficient methods for collecting and preparing thermographic data and rigorously assessing various machine learning models in defect detection scenarios. The research aims to create a robust thermographic dataset capable of identifying common defects, such as gaps in AFP. Tested models demonstrate significantly improved defect detection accuracy, highlighting the efficacy of thermographic data. Through experimental analysis and comparison with comparable systems, this study showcases the potential of thermographic data to enhance AFP defect detection, providing valuable insights into best practices for dataset creation and model evaluation. The findings pave the way for more advanced AFP inspection techniques and broader applications in composite material manufacturing.