Abstract:Laser powder bed fusion of metals is increasingly used for fabricating complex parts requiring good mechanical properties. Simultaneously, researchers in the field are intensifying the efforts to reduce defects, such as internal porosities, which hinder a wider industrial adoption of this technology, urging process monitoring to a pivotal role in defect identification and mitigation. Therefore, understanding the correlation between in-process monitoring signals and post-process actual defects is fundamental to taking informed decisions and potential corrective actions during the process.This work focuses on developing models to predict spatter-related defects from specific process signatures detected through off-axis long-exposure imaging. Layer-wise images were properly aligned with corresponding cross-sections from tomographic reconstructions to investigate the relationship between spatter-related signatures and actual defects measured by X-ray computed tomography. This relationship was used as a knowledge basis to develop an analytical image-processing approach and a machine learning-based methodology, which were then compared in terms of their correlation performances. The advantages and limitations of both methods are discussed in the paper.Both approaches led to promising results in the prediction of lack-of-fusion defects caused by spatters, with the machine learning approach showing a prediction accuracy in the order of 90% for defects with equivalent diameter above 90 μm, while the analytical model needed equivalent diameters larger than 130 μm to reach a prediction accuracy in the order of 80%. Furthermore, the machine learning method led to strong results regarding early defect detection, with most of the investigated defects properly predicted by analysing two consecutive layers after the signature detection.

A novel spatter detection algorithm based on typical cellular neural network operations for laser beam welding processes

A Spatio-Temporal Ensemble Deep Learning Architecture for Real-Time Defect Detection during Laser Welding on Low Power Embedded Computing Boards

Supervised deep learning for real-time quality monitoring of laser welding with X-ray radiographic guidance

Neural network based image segmentation for spatter extraction during laser-based powder bed fusion processing

Embedded processing methods for online visual analysis of laser welding

Spatter detection and tracking in high-speed video observations of laser powder bed fusion

Applying edge artificial intelligence to closed-loop real time control and monitoring of laser based battery pack welding

Spatter formation in laser beam welding using laser beam oscillation

Laser Welding Penetration Monitoring Based on Time-Frequency Characterization of Acoustic Emission and CNN-LSTM Hybrid Network

Research on a Feature Point Detection Algorithm for Weld Images Based on Deep Learning

Neural-network-assisted in situ processing monitoring by speckle pattern observation

Deep Learning Based Monitoring of Spatter Behavior by the Acoustic Signal in Selective Laser Melting

Real-time monitoring of high-power disk laser welding statuses based on deep learning framework

Online monitoring system for welding states of bottom-locking joints in high-speed trains via multi-information fusion and 3DCNN

Machine‐Learning‐Based Monitoring of Laser Powder Bed Fusion

Prediction of spatter-related defects in metal laser powder bed fusion by analytical and machine learning modelling applied to off-axis long-exposure monitoring

Spatter feature-based in-situ identification of disturbances in process parameters during laser powder bed fusion

Beam Offset Detection in Laser Stake Welding of Tee Joints Using Machine Learning and Spectrometer Measurements

A Novel Visual Image Sensor for CO2short Circuiting Arc Welding and Its Application in Weld Reinforcement Detection

Deep learning-based monitoring system for predicting top and bottom bead widths during the laser welding of aluminum alloy

Detection of solidification crack formation in laser beam welding videos of sheet metal using neural networks