In-situ monitoring of selective laser melting based on heterogeneous integration of acoustic signals and images

Abstract

Monitoring during the selective laser melting (SLM) process is the only way to achieve 100% Non Destructive Testing (NDT) for parts produced. This work provided a heterogeneous integration monitoring with acoustic signals and images. Acoustic signal features and image features could indicate different states of the melting and then their signal features were extracted by deep belief networks (DBN) and convolutional neural networks (CNN) respectively. Heterogeneous features were used to recognize the sates of the melting by Support vector machines (SVM) method.Classifications were carried out with acoustic signal and image features respectively for comparisons. The optimal classification rate came from fused features from acoustic signals and images. The results indicated that the fused features improved the defect diagnosis capabilities for the SLM process. It is effective and promising by recognizing heterogeneous integrated signals for the in-situ monitoring of the SLM process.