Defect identification in composite materials via thermography and deep learning techniques

Abstract

Composite materials are widely used in aircraft, vehicle, and various industries due to their excellent mechanical properties. A thermography-based nondestructive test is often employed for diagnosis of defects in composite laminates, while the test results are largely affected by the environmental conditions, and show significant dependence to the inspector, instruments being used, and the test objectives. To overcome the limitation, the present study proposes a framework for identifying defects in composite materials by integrating a thermography test with a deep learning technique. A dataset of thermographic images of composite materials with defects were collected from literatures and were used for training the system to identify defects from given thermographic images. The versatile application of the proposed technique was validated by testing it on composite specimens produced by resin transfer molding and thermoplastic injection molding, using a combination of carbon/organo fabrics and thermoset/thermoplastic resins. The performance of the proposed system was evaluated by assessing its ability to identify defects from the specimens with artificial defects, and is discussed in light of average precision for identification of defects.