Venkatavaradan Sunderarajan
Advisor: Prof. Suman Das
will defend a doctoral thesis titled,
Multi-Scale Materials Characterization and Analysis of In-Situ Process Monitoring Data towards enabling Multivariate Statistical Process Control
in Laser Powder Bed Fusion Metal Additive Manufacturing
On
Thursday, November 30 at 3:15 p.m.
at MRDC Room 4211
and
Virtually via Zoom
https://gatech.zoom.us/j/97282558604?pwd=cS91QUJ5ZWpZZE5SMUh5djc2N3BXZz09
Meeting ID: 972 8255 8604
Passcode: 113736
Committee
Prof. Suman Das, Advisor, ME/MSE
Prof. Hamid Garmestani, MSE
Prof. Sandra Magnus, AE/MSE/INTA
Prof. Jianjun Shi, ISyE/ME
Prof. Preet Singh, MSE
Abstract
Widespread adoption and industrial scaling of Laser Powder Bed Fusion (L-PBF) Metal Additive Manufacturing (AM) is currently challenged by a variety of issues including dimensional and form errors, undesired (and oftentimes stochastic) porosity, delamination of parts, extreme variations in part properties, undesirable failure rates, and significant costs to optimize processes for gaining acceptable part quality, albeit with limited statistical confidence. Currently accepted industry standards, such as ASTM F3001, do not address the inherent variations present in the parts produced by L-PBF. Such property variations, present even in an extensively researched alloy such as Ti6Al4V, greatly hinder the acceptance of L-PBF to produce mission critical parts without substantially expensive qualification procedures. To date, a significant volume of research has focused on the capabilities of AM, but there is a lack of research focused on its repeatability and reproducibility. The objective of this dissertation is to evaluate variations in post-process multi-scale part characterization and to utilize in-situ process monitoring in an industrial L-PBF Metal AM platform to enable a framework for Qualification, Validation and Verification (QVV) by:
i. Studying variations in static tensile properties of parts manufactured across multiple builds under the same regimen.
ii. Studying variations in critical dimensional features of multiple parts manufactured in a single build under the same regimen.
iii. Utilizing heterogeneous data input from multi-fidelity in-situ monitoring sensors to develop process control charts using multivariate statistics.
The tensile properties thus investigated highlight the variations encountered even while using a combination of machine, process and material that has been fully operationally qualified to produce mission critical parts. Property dependence on controlled process parameters as well as challenges to repeatability and reproducibility in L-PBF Metal AM are presented, and pathways to address the same are proposed. Finer layer thicknesses, coupled with locations either closer to the shielding gas source or at the center of the build plate, even while incorporating powder re-use between successive builds offer the least coefficients of variations in tensile properties.
The dimensional measurement of critical geometric features proved that the practically achievable limit of resolution and repeatability for producing thin-walled and thin-gap specimens using this platform is around 500 µm. An effective geometric constraint can be provided by a base to provide tighter control in dimensional variations, although it may not be a design permissible for practical applications. For features with nominal dimensions less than 500 µm, the scatter in the measurements is equal in value or is a low multiple of the largest diameters of the feedstock powder. Therefore, this may pose a challenge that cannot be overcome easily, especially without innovative post-processing methods. The repeatability of fine feature dimensions is favored by larger powder layer thicknesses, whereas the opposite is generally true for obtaining a smoother surface finish. The build location at the center of the build plate exhibits the tightest control over variations and results in the most predictable outcomes.
Four considerations are critical for the successful implementation of a robust in-situ monitoring methodology that can also benefit from scaling up and successful adoption in the production of mission-critical and/or safety-critical AM parts. These include targeting the reliable detection of stochastic flaws representative of real-world applications as opposed to seeded defects; developing methods to collect, analyze and act on data collected over the duration and volume of an entire build; developing process control methods that rely on fundamental statistics with robust and transparent algorithms as opposed to “black-box” ML based algorithms requiring tremendous time and financial investment to actually realize the promise they offer; and interpreting results from less complex statistical algorithms will enable a higher confidence in the detected outcomes as well as a higher likelihood of the eventual method gaining traction for adoption in the manufacturing of service critical parts.
Simultaneous usage of multi-fidelity in-situ monitoring sensors enabled an enhanced capture of process signatures to understand deviations and develop corrective actions. 2D-Wavelet transformation-based approach to identify recoater streaking defects from optical images of the powder bed and a Sobel gradient operator-based approach to detect hot and cold spots from photodiode signals are useful examples demonstrating the capability of direct image processing techniques to identify defects encountered during a build. Such methods enable extraction of layer-wise statistical measures for subsequently monitoring the L-PBF Metal AM process via multivariate statistical process control charts.
The outcomes of this dissertation provide valuable insights into the inherent complexities of the L-PBF Metal AM process and underscore the challenges in achieving repeatability and reproducibility. Future work can extend the proposed QVV framework using robust in-situ monitoring methodologies to enhance process reliability enabling the widespread industrial adoption of L-PBF Metal AM for to produce mission-critical parts.