[Paper Review] Influence of Spattering on In-process Layer Surface Roughness during Laser Powder Bed Fusion

Haolin Zhang, Chaitanya Krishna Prasad Vallabh+1 more|arXiv (Cornell University)|Mar 1, 2023

Additive Manufacturing Materials and Processes4 citations

TL;DR

This study introduces an in-situ monitoring framework combining fringe projection profilometry and off-axis imaging to quantify the correlation between melt pool spattering and in-process layer surface roughness in laser powder bed fusion (LPBF). Using machine learning to extract spatter signatures—such as spatter count, ejection angle, and melt pool center—alongside regression modeling, the authors demonstrate that spatter-informed predictions reduce surface roughness error by over 50% compared to conventional process parameter-only models.

ABSTRACT

Laser powder bed fusion (LPBF) holds promise to efficiently produce metal parts. However, LPBF incurs stochastic melt pool (MP) spattering, which would roughen workpiece in-process surface, thus weakening inter-layer bonding and causing issues like porosity, powder contamination, and recoater intervention. Understanding the consequential effect of MP spattering on layer surface remains difficult due to the lack of process monitoring capability for concurrently tracking MP spatters and characterizing layer surfaces. In this work, using our lab-designed LPBF-specific fringe projection profilometry and an off-axis camera, we quantify the correlation between MP spattering and in-process layer surface roughness for the first time to reveal the influence of MP spatters on process anomaly and part defects. A method of extracting and registering MP spattering metrics is developed by machine learning of the in-situ imaging data. Each image is analyzed to obtain the MP center location and the spatter count and ejection angle. These MP spatter signatures are registered for each monitored MP across each layer. Regression modeling is used to correlate registered MP spatter signature and its processing parameters with layer surface topography measured by the in-situ FPP. We find that the attained MP spatter feature profile can help predict the layer surface roughness more accurately (> 50% less error), in contrast to the conventional approaches that would only use nominal process setting. This is because the spatter information can reflect key process changes including the deviations in actual laser scan parameters and their effects. The results corroborate the importance of spatter monitoring and the distinct influence of spattering on layer surface roughness. Our work paves a foundation to thoroughly elucidate and effectively control the role of MP spattering in defect formation during LPBF.

Motivation & Objective

• To address the lack of concurrent monitoring for melt pool spattering and layer surface topography during LPBF.
• To investigate how spattering events influence in-process surface roughness and contribute to defects such as porosity and poor inter-layer bonding.
• To develop a real-time, data-driven method for extracting and registering spatter features from in-situ images.
• To improve the accuracy of surface roughness prediction by integrating spatter metrics with process parameters, rather than relying solely on nominal settings.

Proposed method

• Employed a custom-designed LPBF-compatible fringe projection profilometry (FPP) system to measure in-process layer surface topography with high resolution.
• Used an off-axis high-speed camera to capture in-situ melt pool dynamics and spattering events during laser scanning.
• Applied machine learning to analyze each image frame, extracting key spatter signatures: melt pool center location, spatter count, and ejection angle.
• Developed a registration method to track individual melt pools and their spattering characteristics across layers and scan paths.
• Built a regression model correlating the registered spatter features and processing parameters (e.g., laser power, scan speed) with measured surface roughness from FPP.
• Validated the predictive performance of the spatter-informed model against conventional models based only on nominal process settings.

Experimental results

Research questions

• RQ1 How does melt pool spattering correlate with in-process layer surface roughness during LPBF?
• RQ2 To what extent can spatter signatures extracted from in-situ imaging improve the accuracy of surface roughness prediction compared to nominal process parameters?
• RQ3 What specific spatter features (e.g., count, ejection angle, location) are most predictive of surface roughness variations?
• RQ4 Can real-time monitoring of spattering enable early detection of process anomalies that lead to defects?
• RQ5 How do deviations in actual laser scan parameters manifest in spattering behavior and surface quality?

Key findings

• The integration of in-situ spatter signatures with processing parameters reduced surface roughness prediction error by more than 50% compared to models using only nominal process settings.
• Spatter count and ejection angle were identified as key indicators of process instability and surface quality degradation.
• The melt pool center location and spatter distribution patterns were consistently correlated with localized roughness variations on the layer surface.
• The machine learning-based spatter feature extraction method enabled reliable, frame-by-frame tracking of spattering events across multiple layers.
• The study revealed that spattering is not a random byproduct but a measurable process indicator that reflects actual deviations in laser scan behavior.
• The results demonstrate that real-time spatter monitoring can serve as a robust proxy for assessing surface integrity and predicting defect formation in LPBF.