Session: 16-01-01: Poster Session: NSF-Funded Research (Grad & Undergrad)

Paper Number: 99498

99498 - In-Situ Monitoring of Printed Layer Surface Topography During Laser Powder Bed Fusion via Fringe Projection Profilometry 

In laser powder bed fusion (LPBF) additive manufacturing (AM) the surface topography of a being-printed part critically determines the part properties including internal porosity, surface defects and mechanical strength. Conventionally, exhausting and expensive ex-situ characterization and testing of the final printed parts would be required to identify these defects and evaluate other related performance metrics. Thus, in-situ, layer-wise monitoring of surface topography is desired for online qualifying the print process and the print part with efficiency and reliability. To accomplish this, we first developed a cost-effective and non-destructive Fringe Projection Profilometry (FPP) system specifically for LPBF AM. FPP is typically used for measuring finer features and reconstructing 3D topography of objects. However, to use the FPP method for measuring the dynamic topography of powder bed and printed layers during a LPBF process, unique challenges exist due to the ambient conditions in the build chamber and the variations of material properties (reflectivity etc.). We enhance the discernibility and feature recognition capability of the FPP system in the LPBF scenario by employing an equipment based High Dynamic Range (HDR) method integrated with machine learning (ML) frameworks. The projector based HDR method is applied to ease the shadow and intensity saturation problems by projecting varying intensities of sinusoidal fringes. Often employed FPP methods are limited by the measurement resolution, to improve this we develop a machine learning framework to improve the surface topography resolution that is limited by camera resolution to match ex-situ optical profilometry resolution (~5 µm), by training and testing various Super resolution (SR) neural networks (NN), using the in-situ FPP measurement data and ex-situ profilometry data. The designed SR NN, unlike the traditional feature extraction based convolutional NN (CNN), is the generative model which is capable of synthesizing high resolution (~5 µm) details that is not latently available from FPP methodology. The developed NNs achieve the accurate performance with Peak signal to noise ratio (PSNR) score of 32.24 and mean squared error (MSE) of 6.55 micron².

Furthermore, since powders melting and solidification directly contribute to the formation of printed layer surface, we exploit this significant cause-and-effect relationship to develop another method of estimating in-process layer surface topography from in-situ melt pool (MP) signatures. Specifically, we correlate the MP signature maps (MPSMs) obtained from our coaxial high-speed single-camera based two-wavelength imaging pyrometry (STWIP) system with the FPP system measured surface topography by a long short-term memory (LSTM) neural network. The signatures from the STWIP system include MP temperature, MP intensity and MP area. A machine learning aided image analysis algorithm is employed to retrieve the spatial distribution of MPs within the corresponding part’s coordinates system. Then, the MP signature maps (MPSMs) are reconstructed by mapping the STWIP measured MP signatures to the registered MP coordinates. The LSTM neural network is trained by the developed in-situ FPP measurement data and employed for estimating in-process layer surface topography directly from the registered MPSMs with 92.88 % accuracy. This part of work demonstrates a promising approach of efficiently monitoring both MP and surface properties altogether by a single MP monitoring system provided a corresponding FPP-trained neural network.

Both of the presented methods are expected to measure more capably - the surface topography of the printed layers during LPBF-AM, thus advancing the existing state-of-the-art methods towards the desired online inspection of LPBF print defects and process anomalies via in-situ FPP and MP signatures.

Presenting Author: Haolin Zhang University of Pittsburgh

Presenting Author Biography: I am a graduate student researcher at University of Pittsburgh studying Additive manufacturing process and in-situ monitoring implementation and control for LPBF process. Furthermore, I am proficient in machine learning model development and data driven model construction.

Authors:

Haolin Zhang University of Pittsburgh
Chaitanya Krishna Vallabh University of Pittsburgh
Xiayun Zhao University of Pittsburgh