Multi-feature fusion imaging via machine learning for laser ultrasonic based defect detection in selective laser melting part

Abstract

Laser ultrasonic (LU) testing is a promising technique that meets the constraints of additive manufacturing (AM) online monitoring, which is vital for the quality control of AM products and the practical application of AM. However, the scattering noise caused by surface roughness of AM parts and the ambient noise can generally result in defect imaging with low resolution. This is not conducive to defect detection. In terms of this problem, a multi-feature fusion imaging methodology is proposed instead of the traditional method containing a single feature, i.e., maximum amplitude within a set window of all A-scans. The most discriminative features in each A-scan are automatically extracted by principal component analysis (PCA). These features are subsequently fused by a trained artificial neural network (ANN) to obtain an evaluated value that determines whether a A-scan is the normal case or defective case (denoted by 0 and 1, respectively). Finally, a C-scan image with high signal to noise ratio (SNR) can be produced by plotting the binary dataset of 0 and 1. LU inspection on a selective laser melting (SLM) component with the pre-induced micro hole defects on the rough surface is performed to validate the effectiveness of the methodology. The defects are distributed in 2 rows and 6 columns where the diameter of one row is 100 μm and the other is 50 μm. It is demonstrated that the proposed methodology can detect and size all micro defects by the obtained high SNR C-scan.