SURFACE DAMAGE IDENTIFICATION IN 3D PRINTED METAL PARTS USING CONVOLUTIONAL NEURAL NETWORK

Abstract

Surface Response to Excitation (SuRE) is an active Structural Health Monitoring (SHM) method used in this study for the detection and quantification of the artificial damages created by the milling operation on additively manufactured metal plates. In this method, one piezoelectric element is bonded to one end of the test specimen to excite it with surface waves and the dynamic response to excitation is recorded by another piezoelectric at the other end of the part. The excitation signal is a sweep sine wave with a duration of 1 ms and a frequency range of 50-120 kHz. Using a Markforged metal 3D printer, five stainless steel plates of identical size (195×54×2.5 mm) were created. The data was recorded when all the parts were in a healthy condition and when they were face milled at 3 different lengths. The collected sensory data in the time domain were converted to time-frequency representation images using continuous wavelet transform (CWT). Different data augmentation methods have been implemented for expanding the size of the dataset. The image dataset was used as input to train a Two-Dimensional Convolutional Neural Network (2D-CNN) for the detection of damage and also for quantifying the damage length. The CNN could detect the damaged parts with 97.4% accuracy and classify the damage length with an overall accuracy of 98.7%.