A Machine Learning Based Model for Monitoring of Composites’ Drilling-Induced Defects During Assembly Production Using Terahertz Imaging Data

Amini, Amin orcid iconORCID: 0000-0001-7081-2440 and Gan, Tat-Hean (2022) A Machine Learning Based Model for Monitoring of Composites’ Drilling-Induced Defects During Assembly Production Using Terahertz Imaging Data. In: 2022 IEEE Workshop on Complexity in Engineering (COMPENG), 18-20 July 2022, Florence, Italy.

[thumbnail of AAM] PDF (AAM) - Accepted Version
Restricted to Repository staff only

393kB

Official URL: http://dx.doi.org/10.1109/COMPENG50184.2022.990543...

Abstract

The composite materials are becoming more popular due to their advantages over traditional materials, including being lightweight, high stiffness-to-density and high strength-to-weight ratios. As a result, composite materials have been widely used in manufacturing sector for various industries including aerospace, automotive, marine and energy. Nonetheless, as machining of composites is unavoidable for assembly purposes, defects can be induced at various stages of manufacturing process. Drilling of fiber-reinforced composites is a complex task due to their anisotropic, inhomogeneous, and highly abrasive characteristics. Defects form drilling process including delamination and fiber pull-out can significantly affect the strength and performance of composites. There have been a wide variety of non-destructive testing (NDT) methods playing a major role in testing of composite materials. However, the current NDT solutions for in-service inspection are largely complex, which leads to higher inspection costs. The proposed solution uses artificial intelligence (AI) based algorithm utilizing Terahertz imaging data to detect drilling-induced defects in composite materials during manufacturing and assembly. A machine learning (ML) model has been developed to process the data obtained from Terahertz scanning to automatically detect and report the defects in composite drillings. In order to achieve such a system, a ML model based on Faster R-CNN neural network for drill holes’ defects detection has been developed. This automated solution will have the ability to reduce the manual inspection time of the operator and the costs of inspection process of drilling holes. The developed system proved to have a statistically significant efficiency in both performance and speed as well as reducing the sub-quality products.


Repository Staff Only: item control page