Imbalanced quality monitoring of selective laser melting using acoustic and photodiode signals

Abstract

Selective laser melting (SLM) has become a promising additive manufacturing technique to fabricate metal parts with complex structures. However, the poor part consistency and process repeatability still hinders the wider adoption of the SLMed parts, necessitating the need of in-situ process sensing for quality monitoring. Unfortunately, the acquired sensing datasets in real-world experiments may be imbalanced, which can largely affect quality monitoring. In this study, a deep learning (DL)-based imbalanced data generation approach is proposed to realize in-situ quality monitoring with imbalanced datasets in the SLM process. Two imbalanced datasets of acoustic and photodiode signals are first collected and created by the developed in-situ sensing system. Then, a Generative adversarial network-based data generation model is developed to generate the minority samples. Finally, a DL-based quality classification model is established to classify the augmented balanced datasets. Results show that the generated high-quality samples can significantly improve the model performance. Besides, it takes only 0.19 ms for the proposed DL-based quality classification model to classify a sample with an average classification accuracy above 93 %, providing a great potential for imbalanced quality monitoring using the acoustic and photodiode signals in the SLM process.