Automatic void content assessment of composite laminates using a machine-learning approach

Voids have a substantial impact on the mechanical properties of composite laminates and can lead to premature failure parts. Optical microscopy is commonly employed imaging technique assess void content parts, as it reliable less expensive than alternative options. Usually, image thresholding techniques are used parse acquired images automatically; however, these very sensitive acquisition conditions type material used. Additionally, algorithms be calibrated before each analysis, in order provide accurate results. This work proposes machine-learning approach, based convolutional neural network architecture, with objective providing robust tool capable automatically parsing optical images, without need parameter tuning. Results from training testing datasets composed extracted three distinct types confirm that proposed approach parses more accurately traditional algorithm, previous calibration step. shows promising, despite sometimes lower expected precision individual statistics.