Semi-supervised multi-label feature selection algorithm for online monitoring of laser metal deposition manufacturing quality

Abstract

Laser metal deposition (LMD) manufacturing, as a technique of laser metal additive manufacturing, has the advantages of short production periods, good economy and unrestricted forming shapes in the direct manufacturing of metal parts. However, the poor accuracy of single-object monitoring and the lack of labelled data make quality monitoring a huge challenge. Accordingly, this study extracts multi-features from the multi-object, including melt pool dimensions, spatters and temperature field, and proposes a semi-supervised multi-label feature selection (SSMLFS) algorithm. For SSMLFS algorithm, a multiple regression quality model is firstly proposed to produce different quality levels for unlabelled data based on microstructure. Then, to achieve feature dimension reduction and filter out highly relevant features, a quality correlation evaluation function is developed to calculate the contribution and ranking of the various features. Moreover, a local search algorithm based on quality is designed to improve the search speed of feature subsets and speed up the convergence of the SSMLFS algorithm. Experimental validation is conducted on BeAM Magic 800 machine and uses several commonly monitoring algorithms based on machine learning. Related experimental results prove that SSMLFS can significantly improve the accuracy and efficiency of quality monitoring compared to single-feature and full-feature monitoring. The proposed SSMLFS algorithm provides a semi-supervised feature selection framework to handle the low accuracy of single-feature monitoring and the complexity of multi-feature monitoring in the quality monitoring of LMD manufacturing.