Monitoring and Control of Metal Additive Manufacturing Processes Using Ultrasound

Abstract

Laser powder bed fusion (LPBF) and Directed energy deposition (DED) are two classes of additive manufacturing (AM) that are widely used to fabricate metal components. In these processes, metal powder or wire and a heat source are applied to form a moving melt pool, followed by solidified tracks to form the component in a layer-by-layer fashion. The solidified melt pool track gradually adds material to build a component from scratch or can be used to add material to a damaged component to refurbish it. DED via powder-feeding (DED-PF) uses metal powder as a feedstock and usually uses a laser as an energy source. A DED-PF machine using a laser energy source is referred to as laser DED-PF (LDED-PF). DED-PF has further advantages compared to other DED methods, such as being able to gradually change the volume fraction of different powders to produce functionally graded materials, which can be used to make high-performance aerospace components that are resistant to high temperatures for example. The first part of this thesis focuses on techniques for in-situ detection of near-surface defects as they are being produced, which may be used for either LPBF or DED. In particular, techniques for LPBF using laser ultrasound (LU); a non-contact method that generates and measures sound waves reflected from defects such as voids inside a metal sample. This is advantageous for in-situ AM metal inspection, since placing and removing a sensor in contact with the high-temperature topmost layer or track could substantially increase the fabrication time. A reconstruction method in the frequency domain was implemented (phase shift migration) that also accounts for the offset between the generation and detection lasers spots for detecting defects that are close to the surface. The method was tested for both AlSi10Mg and Ti6Al4V samples with three different types of artificial defects to determine which artificial defects are most appropriate to test LU setups. The surface roughness induced during these AM processes was mimicked by printing the samples without up-skin. The effect of the sound generation laser on the final surface roughness, which may determine the spreadability of the powder in subsequent layers, was also investigated. Novel computationally efficient methods to filter the expected unwanted signals in the initial ultrasound data in the frequency domain as well as the fast computation of a stationary phase approximation within the reconstruction were also developed. Defects with a size range from about 200 to 300 µm printed in Ti6Al4V can be recognized from the background noise of the reconstruction. The computation time of the stationary phase approximation with a more accurate initial estimate of its stationary point was reduced by 33%. The second part of the thesis focuses on finer control of the powder stream profile for the DED-PF processes. This is done using sound radiation forces produced by high-intensity ultrasound. These forces can be applied to the powder stream to change the area and/or cross-section of the powder stream to indirectly affect the melt pool and increase powder catchment efficiency. This could be used to change the melt pool size, shape, and/or internal convective flows, to increase printing speeds and/or reduce defects. An initial proof-of-concept study was carried out with Ti6Al4V and SS 316L powders to demonstrate control of the narrowing of the powder stream from a gravity hopper with a DED-PF nozzle analog. The experimental results were compared with an in-house software library, that can quickly calculate the sound force fields from an ultrasound array using automatic differentiation, and can perform particle tracking. Also, a novel Eulerian-Eulerian approach that can model the powder as a dilute phase was developed, which can more accurately model the effect of a force field, by using an initial condition for the particle speeds. An implementation was also carried out in a modified LDED-PF machine to increase the cross-sectional area, and related catchment efficiency, of Ti6Al4V tracks. A novel model of the powder stream subject to a force field was also developed. Using particle track statistics, the powder stream from a DED-PF nozzle can be modeled using a Gaussian beam model, which is conventionally used to model laser beams. Within this model, it is shown that one can model the powder stream particle concentration distribution (PCD) while being subject to a focusing force field with an equivalent optical transfer matrix. For the initial proof of concept study, using Ti6Al4V and SS 316L particles at a particle speed of 0.6 m/s, the downstream powder stream width was reduced by at least 30%. For the experiment implemented in an LDED-PF machine using Ti6Al4V particles at a particle speed of 3.6 m/s, the powder stream width at the substrate was reduced by 38%. This produced an increase in track height of 72%, and cross-sectional area of 111%, for an original average height of 0.4 mm and area of 0.93 mm squared, when using a laser power of 900 W. The Gaussian beam model was able to produce more accurate cross-section profiles of the powder stream when subject to a force field, while also reducing the simulation time from 34 minutes (using a Lagrangian particle simulation implemented in C++) to 0.2 seconds (using the Gaussian beam model implemented in MATLAB). With these two techniques, in-situ and ex-situ monitoring and control may be carried out for metal AM processes. Both these research areas could be integrated to produce a novel DED-PF system that has superior sensing capabilities to find defects in situ as well as finer control of the melt pool, allowing for better quality control. One could use such a system, together with the appropriate control algorithms to tune the process in situ to reduce defects or even correct them. This new system has the potential to produce parts at a lower cost due to less material waste.