Ultrasound in situ characterization of hybrid additively manufactured Ti6Al4V^a)

Hybrid processing in metal additive manufacturing (AM) enables localized tailoring of material heterogeneity including the grain structure and substructure, anisotropy, and residual stress.^1,2 Hybrid processes incorporate converging processes and/or energy sources applied either cyclically or at the same time during manufacturing to produce distinct local microstructure and mechanical property patterns and manipulate global performance.³ The most common hybrid AM processes combine additive and subtractive steps through cyclic machining of side and/or top surfaces. Besides being an accessible secondary process, machining maintains the surface quality of complex internal features, preserves geometrical accuracy, and provides an improved surface for subsequent additive steps.³ Machining introduces mechanical energy to the work piece, promoting plastic deformation just below the milled surface, which results in work hardening and microstructure deformation.^4,5 These effects may be used to improve part performance; for example, Feldhausen et al.⁶ showed that intermittent milling improved density and toughness in DED 316 L stainless steel. Similarly, powder directed energy deposition (DED) systems offer several advantages for hybrid AM processing including higher deposition rates than powder bed systems, the ability to print on different substrates, and the possibility of incorporating multiple materials in the same build.² In fact, most commercial hybrid systems consist of laser based DED and machining.^7,8

Ti6Al4V is amongst the most studied alloy systems in metal AM given its broad applicability from healthcare to aerospace solutions.^9,10 Ti6Al4V is an α + β alloy, in which the α phase exhibits a hexagonal closed pack (HCP) crystal lattice and the β phase has a body centered cubic (BCC) lattice. At temperatures higher than approximately half the melting point, also known as the beta transus temperature (BTT), the crystal structure of Ti6Al4V shows a phase transformation from α to β.^11–13 Additionally, high cooling rates, such as those present in the DED process, result in the formation of a martensitic α′ phase, while the continuous thermal cycling also promotes the formation of a metastable α + β structure.¹⁴ Thus, the expected phase transitions due to cooling after laser processing are β → α′ and α + β, while the microstructure is dominated by the α′ phase.¹¹ The microstructure of DED Ti6Al4V is also characterized by a drastic epitaxial growth of the prior β phase, accompanied by strong texture.⁹ These microstructural effects are the result of the temperature gradient and growth rate that depend on the heat input (i.e., laser power), cooling rate, and scan strategy, and result in strong anisotropy in the mechanical performance of Ti6Al4V samples.^9,15 Hybrid processing offers avenues to control these microstructural effects.

Despite their potential advantages, the complexity of metal and hybrid AM processes continues to pose challenges related to the repeatability, integrity, and quality control of components. Due to the high value, customization, and manufacturing time associated with metal AM, in-process quality control is preferable.¹⁶ Thus, in situ real-time solutions for characterization of AM samples are desirable. “In situ” refers to measurements concurrent with the AM process, while real time indicates that the measurement completion time and AM build time scales are equivalent.¹⁷ Reliable in situ real-time characterization, is a prerequisite for the implementation of effective feedback loops to correct or stop the process if a quality requirement or desired property is not met. Multiple options have been proposed to inspect and qualify metal AM components in situ, using a variety of measurements including optical, thermal, x-ray, and acoustic/ultrasonic waves. Several comprehensive reviews on the topic are available. Tapia and Elwany¹⁸ highlighted different examples of commonly used sensors for monitoring metal AM processes. Their review emphasized thermal measurements because the part microstructure, density, and mechanical properties derive from the temperature field. Nonetheless, they noted the limitations of temperature measurements for real time control due to the need for computationally intensive simulations to make predictions. Building upon this work, Everton et al.¹⁹ described in situ metrology in detail for DED and laser powder bed fusion (LPBF) processes. Their review focused on examples of thermal and optical in situ applications, while the potential for electromagnetic and ultrasonic methods was also discussed. In particular, the authors highlighted the need for in situ metrology beyond the component surface. Hossain and Taheri²⁰ detailed the use of acoustic emission techniques and signal processing for in situ monitoring of AM processes as an efficient alternative for in situ monitoring of AM processes. Vora and Sanyal²¹ provided a broad overview of ex situ and in situ metrology for various AM technologies and materials. They discussed relevant destructive and nondestructive measurements from the feedstock stage to the finished product including composition, morphology, geometry, defect content, and micro/meso/macro structures. Finally, Honarvar and Varvani-Farhani²² provided a summary of ultrasonic testing applications in polymer and metal AM both in situ and ex situ. There is disagreement regarding the feasibility and use of in situ ultrasound monitoring in AM.^20–22 However, the discussion has been primarily focused on defect detection and porosity quantification, while the need and potential for material characterization are scarcely considered in this context. Ultrasonic measurements offer several advantages. The interaction of ultrasonic waves with material grain boundaries provides a means to assess the microstructure of the manufactured piece in addition to its elastic properties and defect content.²³ Also, bulk ultrasonic waves typically offer deeper inspection depths within the material.²² The accessibility, portability, and low energy requirements of ultrasonic equipment make these methods easily transferable to industrial and remote applications.

While several ultrasonics-based measurements have been proposed,^24–27 only a few examples of in situ ultrasound measurements collected in a metal AM environment have been realized.^17,28–34 Rieder et al.²⁸ used contact transducers adhered to the bottom of a build plate to monitor the wave speed and spectral amplitude obtained from LPBF samples in situ. They also demonstrated sensitivity to porosity imparted by changes in the laser power during the build.²⁹ Koester et al.³⁰ attached contact acoustic emission transducers to a custom-made fixture for a DED system. The build plate was mounted on the fixture and the transducers collected signals at varying build conditions including the inactive state of the system, powder spray only, optimum laser power, and less than ideal laser power and powder feed. Using a data classification approach, Taheri et al.³¹ later demonstrated that each build condition could be identified based on its acoustic signature. Plotnikov et al.³² also incorporated acoustic emission sensors on an intermediate plate located under the build plate of an LPBF system. The signals were correlated with laser power and corresponding porosity, and the combined interpretation of acoustic and infrared data were proposed.³² Nadimpalli et al.³³ used a broad band contact transducer attached to the bottom of the baseplate to monitor the quality of parts made with ultrasonic additive manufacturing (UAM) and showed in situ ultrasonic velocity and attenuation measurements. They also proposed an interfacial spring model for layered media and inversion algorithm as well as demonstrated the use of the model and ultrasonic measurements to determine bond quality.³³ In a more recent work, Nadimpalli et al.³⁴ modified their measurement system to avoid any impacts on the quality of the UAM piece due to raising the baseplate. In this case, they used an immersion transducer placed in a chamber filled with high temperature lubricant oil within an aluminum block placed under the baseplate. They showed that they were able to distinguish interfacial defects in situ, and proposed a methodology for online repair of base/build defects as well as inter-track defects.³⁴ Kube et al.¹⁷ also utilized immersion transducers beneath the build plate, in this case in a pitch-catch configuration and immersed in ethylene glycol, to measure the shear wave scattering from melt pools in real time and obtain correlations between the scattering amplitudes and the corresponding laser duration and melt pool depth.

To enable in process quality control in metal AM and hybrid AM, the purpose of this work is to demonstrate the capabilities of in situ real-time ultrasound measurements in a powder DED system for characterization of Ti6Al4V samples during a hybrid AM process. Despite their relevance and advantages, in situ real-time ultrasonic material characterization in powder DED systems and in hybrid processes has not been practically demonstrated.³⁵ The ultrasonic characterization system described here was successfully used to monitor Ti6Al4V samples continuously in a DED system that incorporated hybrid processing. The potential to quantify roughness and solid state phase changes during the manufacturing downtimes introduced by the hybrid process is demonstrated. Also, the acquired signals were used to observe the evolution of wave speed and attenuation in the material throughout the build.

Two cube samples (12.7 mm side length) were manufactured using a DED system equipped with a three axis machining center (Optomec LENS HY20-CA). As shown schematically in Fig. 1(a), this system ejects powder through four nozzles which surround a laser source used to melt the material. The structure of the system was adapted for research purposes, such that additional instrumentation and wiring were incorporated without disrupting the inert atmosphere (argon) required for the AM process.

The samples were made with a hybrid process combining DED with milling, and using Ti6Al4V powder (Advanced Powders and Coatings, GE Additive) with a particle size of 92.9 ± 15.6 µm as measured through scanning electron microscopy images and image processing.³⁶ The main DED parameters were laser power, spot size, scanning speed, and layer thickness of 500 W, 700 µm, 17 mm/s, and 457 µm, respectively (Table I). These parameters resulted in a energy density³⁷ of 42 J/mm². For each layer, the sample outline was deposited first followed by a bidirectional raster pattern which was rotated 90° with respect to the previous layer. The entire surface was milled without coolant every four build layers with a spindle speed of 800 RPM, cutting speed of 1.02 m/min, and axial depth of cut of 450 µm (Table I). The milling depth of cut was defined such that the inherent roughness imparted by the DED process was removed. Thus, every hybrid layer involved four build layer steps and one milling step for a total nominal height of 1.2 mm, and each sample required a total of 11 hybrid layers Fig. 1(a). The process steps as defined for purposes of this work are listed in Table II. Note that these steps include the manufacturing downtimes after printing and after milling that are necessary to switch manufacturing heads. Both samples had identical manufacturing and processing history, and they were nominally equal. Hereon, the samples are referred to as sample 1 and sample 2 based on the order in which they were made.

After completion of the build, the density of both samples was measured using both the Archimedes method and an x-ray CT instrument (Nikon XT H225 ST).³⁸ The percent density results from each method are summarized in Table III, and in all cases the density of both samples was above 98%. Additionally, the spatial distribution of porosity in both samples obtained with x-ray CT is depicted in Fig. 2. Although both samples underwent identical processing and preparation, sample 1 had evenly distributed small pores throughout its geometry [Fig. 2(b)] whereas sample 2 had regions of concentrated larger pores [Fig. 2(d)].

The samples were sectioned using wire electron discharge machining (EDM), mounted in Bakelite (Buehler) and hand polished in a rotating polishing wheel. The grinding steps were done using 400, 600, and 800 grit SiC paper. Final polishing was completed using 9 µm diamond suspension on a TEXPAN polishing pad and Titanium Attack Polish on a BLACKCHEM2 polishing pad (Pace Technologies). Then micrographs were collected with a laser confocal microscope (Keyence VK-X200K). The general microstructure is depicted in Fig. 1(c) with respect to the sample geometry. Representative micrographs of samples 1 and 2 are shown in Figs. 2(a) and 2(c), respectively, for qualitative assessment. The visible grain boundaries in these micrographs correspond to prior β grains, while the subgrain structure is dominated by acicular α′ martensite. In comparison with as-printed DED cases,³⁹ the applied hybrid process resulted in the disruption of grain growth, and most prior β grain boundaries were observed to stop and start at the hybrid layer interfaces. Finally, no significant differences were observed between the microstructures of sample 1 [Fig. 2(a)] and sample 2 [Fig. 2(c)], although this observation is limited to the specific location studied.

Inspired by previous work,^{17,28–31,33} ultrasonic transducers were arranged under the build plate to collect information during the hybrid AM process. This study utilized unfocused contact transducers (Olympus, Waltham, MA) in a pulse-echo mode, with the purpose of tracking the back wall reflection from the sample. A different transducer was used for each measurement, both with a nominal center frequency of 7.5 MHz and element diameter of 12.7 mm. A DPR 300 pulser/receiver (JSR Ultrasonics, Pittsford, NY) and a 100 MHz analog to digital A/D AL8xGT card (Acquisition Logic, Chantilly, VA) were used to pulse, receive, and digitize the signals. A personal computer operating utwin software (Mistras, Schoolcraft, MI) was used to control the data acquisition. Signals were acquired at a rate of 40 Hz with a repetition average of 20.

To improve the measurement flexibility and process sustainability, a fixture was designed to hold the transducers in place without having to permanently attach them to the build plate. In this manner, the transducers and the build plate could be reused, and the transducers could be placed in different locations with respect to the build plate to align with the build locations of different samples. The fixture consisted of three layers as shown in Fig. 1(b), including a bottom metal layer with a polymer transducer holder, a delay line, and the build plate. The bottom metal layer was used to secure the ensemble to the AM system vice. The polymer transducer holder was designed to be inserted in the bottom metal layer. The holder also had cavities to accommodate four contact transducers, including their wiring, and was manufactured using a high-accuracy multi-jet 3D printer (Object 500 Connex3, Stratasys). Each insert design was specific to the transducers and build locations. For each print, the transducer was placed in a predetermined insert cavity. Then a small amount of high temperature viscous couplant (Versasonic Multipurpose High Temperature Gel) was applied to the transducer face. A 19 mm thick fused silica glass plate was placed on top of the transducers to serve as a delay line and protect the transducers from heat damage. Last the build plate was placed on top of the delay line and secured to the bottom metal layer with bolts. All dimensions were controlled such that the transducers were placed directly underneath the planned sample build locations. Videos of the ultrasonic system collecting information during laser processing and milling steps are available in the supplementary material.⁴⁰ 

The ultrasonic signals were characterized by several reflections associated with the different layers of the fixture for this experimental setup. Figure 3(a) shows a typical signal collected during the experiment after rectification, for which the time axis is equivalent to the sample build direction. The initial pulse is seen at t = 0 µs. Then, at t = 7.2, 14.4, and 21.6 µs, the first and consecutive reflections of the delay line back wall are identified. The reflection of interest in this case is located at t = 18 µs. At the start of the experiment this pulse corresponded with the reflection from the surface of the build plate. As material was added, this reflection denoted the top of the sample. Stray reflections at t = 9.7 and 11.9 µs can be observed as well, and are attributed to reflections due to shear waves in the system from the delay line. The time window of interest for this experiment was defined between the second and third delay line reflections (14.4 < t < 21.6 µs). Thus, the delay line geometry places a constraint on the height that can be continuously built and interrogated. This constraint was not an issue for the samples built for these experiments. However, modifications would be needed for other build geometries.

Several steps were required to analyze the signals. First, although the measurement system was mostly static, slight vibrations during the manufacturing process resulted in minor reflection movement. To remove these effects, the signal at t = 14.4 µs (i.e., the second delay line reflection) was used as the basis for cross correlation of all signals. Then, the signals were aligned according to the shifted values found from this cross correlation. This signal was selected because it was both a signal constant for all times and close in time to the reflection of interest. After all signals were corrected, the average of the signals was calculated and subtracted from the entire ensemble to minimize static reflections and noise. To facilitate the analysis, the information before and after the window of interest was removed to reduce the size of the data file. In this manner, the most prominent reflection in the signals could be associated with the top of the sample. The result of this process is shown in Fig. 3(b).

The reduced signals containing the sample back wall reflection (i.e., the reflection from the top of the sample) were then analyzed in different manners. The Hilbert and Fourier transforms of the signals were calculated to show the evolution of the signals in the time and frequency domains as a function of build time, both as material was added and with respect to different processing steps. The Hilbert transform was calculated for the entire window of interest and its magnitude is shown in Fig. 3(b). To calculate the Fourier transform for each signal, a pulse width defined as 3λ (i.e., three wavelengths) and centered at the maximum of the back wall reflection was used, the rest of the signal was padded with zeroes. This width was defined because the electronic noise disturbed the signal spectrum during most steps except post-printing and post-milling. Then, the spectral centroid location and amplitude were calculated for all signals. An example spectrum and corresponding spectral centroid are shown in Fig. 3(c). The amplitude and time shift of the signals were obtained to study their change as a function of manufacturing time and processing steps. Amplitude based analyses considered the maximum amplitudes in the time and frequency domain as well as the spectral centroid amplitude. The time shift was obtained using cross correlation [Fig. 3(d)]. Finally, the frequency domain maximum amplitude and the time domain shifts were used to compute the wave speed and relative attenuation as a function of added layers with respect to the build plate.

The data collected for the two samples during the build are shown in the time and frequency domains in Fig. 4 as a function of experiment (i.e., manufacturing) time. Figures 4(a) and 4(b) correspond with samples 1 and 2, respectively, while Fig. 4(c) contains the detail of a single process cycle collected from sample 2. The higher amplitude regions correspond with the back wall signal and the dark regions belong to the background. In this case, the data were analyzed after their collection due to data acquisition system constraints. However, the results shown in this section could be easily obtained during the manufacturing process without substantial computational requirements. In this manner, the ability to monitor the AM process in situ and in real time is demonstrated.

The top row in Fig. 4 shows the time domain data in microseconds. Note that the vertical time axis denotes the travel time inside the sample. If the wave speed is known, the vertical axis could be expressed in units of distance. Thus, as the experiment time progressed the signal from the back wall arrived later in accordance with the added distance traveled due to the added material. The time domain detail in the top row of Fig. 4(c) shows that the different process steps involved in each cycle can be discerned by the varying amplitude and shifts in arrival time. First, the back wall signal arrival time increases as layers are added. During the post-printing downtime, there is a shift to earlier arrival times and a slight increase in amplitude. At the milling step, there is a sharp discontinuity leading to an earlier arrival time and drastic increase in amplitude. Finally, during the post-milling downtime, slight increases in amplitude and arrival time drifts are observed.

The bottom row in Fig. 4 shows the data in the frequency domain. In this case the vertical axis corresponds with the frequency in MHz. Although the nominal transducer frequency was 7.5 MHz, the observed center frequency of both transducers was approximately 5 MHz. The transducer used to collect the data from sample 2 exhibited another high amplitude region at about 7.5 MHz. In general, the center frequency did not appear to shift drastically as the experiment time progressed. However, the amplitude of the frequency spectra varied throughout the build. The detail in the bottom row of Fig. 4(c) displays how the frequency spectra changed with processing steps within one cycle. As layers were added, the start of each layer was characterized by a low amplitude followed by a slight amplitude increase as each layer was built. During the post-printing downtime, the amplitude initially increased then remained steady. There was a sharp decrease in amplitude at the milling step followed by a sharp increase in amplitude. Finally, during the post-milling downtime the amplitude remained high and steadily increased. The characteristics of the time and frequency domain data with each processing step are further described in Sec. III B, with a focus on the post-print and post-mill steps. These time and frequency domain features may be used to establish thresholds for normal and abnormal build conditions for process monitoring based on desired material properties.

As shown in Fig. 4, different cycles and processing steps have distinct characteristics that were analyzed in both the time and frequency domains. Specifically, the sample back wall time shift, frequency content, and frequency-domain amplitude were studied.

The sample back wall time shift was defined in a cumulative sense as the shift of each signal with respect to the build plate signal. Figure 5(a) shows the progression of the cumulative time shift with experiment time for one sample. Note that the first and last cycles are not displayed due to coarseness in the data causing discontinuities. The cumulative time shift of the delay line is also included in Fig. 5(a) for comparison purposes. While the time shift measured from the sample visibly varied with manufacturing steps and cycles, the time shift measured from the delay line remained between zero and 0.01 µs (the limit defined by the sampling rate) throughout the experiment. Analogous to the observations in Fig. 4(a), the time shift increased with experimental time because the added material increases the travel time within the sample. Each manufacturing step and processing cycle are clearly distinguishable including layers built, post-printing downtimes, milling, and post-milling downtimes. A drastic decrease in the time shift is observed in each post-printing down time step, as circled in Fig. 5(a). An example of this observation is shown in Fig. 5(b), where the data clearly display a steady decrease in post-printed time shift. During this post-printed downtime, the sample did not experience any processing and the only changes occurring in the sample were due to thermal effects, namely, cooling.

The simplest model to describe cooling of an object is given by an exponential temperature decay. This behavior is displayed by the post-print time shift [Fig. 5(b)]. The post-print time shift was fit with a decaying exponential given by

where $tshift$ is the post-print time shift of each cycle normalized by the maximum time shift observed in the cycle, $T exp *$ is the experimental time (i.e., manufacturing time) defined as $T exp *=T exp −τ$ such that $T exp *$ starts at zero, and $A$ and $β$ are the exponential model fit parameters. As shown in Fig. 5(b), the exponential decay model provides an excellent fit to the data. Moreover, the exponential fits of the post-print time shift for all cycles are plotted in Fig. 5(c). While the values for $A$ decreased with increasing cycle number, i.e., added material, the value for $α$ in both cases remained fairly constant with $β1$ = 2.83 ± 0.34 for sample 1, and $β2$ = 2.87 ± 0.26 for sample 2.

To understand the nature of the observed post-print time shift, the maximum difference in time shift $Δtmax$ was quantified for all cycles and for both samples. In all cases $Δtmax$ decreased with increasing experimental time (downtime). Sample reduction due to thermal cooling could result in a time shift drop because, with decreasing travel distance, ultrasonic waves would arrive earlier. The maximum linear thermal expansion for the samples was calculated as

where $ΔLmax$ is the maximum change in length due to thermal expansion, $Lmax$ is the longest sample dimension, $ΔTmax$ is the maximum change in temperature defined as the difference between the melting point of Ti6Al4V and room temperature (20 °C), and $α*$ is the linear coefficient of thermal expansion for Ti6Al4V. In this case $ΔLmax$ = 0.15 mm, which corresponds to a maximum normalized time shift of 0.01 µs/µs or 1%, assuming the most drastic conditions. In contrast, the measured $Δtmax$ was between 5% and 14%. Therefore, after correcting for shifts in the system (i.e., delay line) and thermal expansion effects, the observed post-print time shift drop was still not fully explained.

Cooling after laser processing is also expected to result in solid state phase changes.¹⁴ While the high cooling rates associated with DED of small Ti6Al4V structures promote rapid phase transformations,⁴¹ larger build volumes undergoing thermal cycling are more likely to experience variable cooling rates⁹ throughout the geometry, affecting the rate, volume, and type of phase transformation occurring in the sample. Due to differences in crystal lattice each phase exhibits different elastic properties. The β phase has the lowest elastic modulus while the α phase has the highest. The α′ phase behaves similarly to the α phase in terms of its elastic properties, although displaying a slightly lower modulus.¹¹ Changes in elastic properties affect the ultrasonic wave speed, in this case promoting an increase in wave speed which in turn will result in a time shift drop such as those observed during the post-printing downtimes.

To verify this hypothesis, the observed post-print time shifts were corrected for thermal expansion effects from the sample and build plate. Then, the corresponding increase in wave speed was calculated using the sample geometry information. This increase in wave speed with respect to manufacturing cycle is shown for both samples in Fig. 5(d). The wave speed increase is between 24 and 65 m/s, decreasing with increasing sample height. This effect may be explained by the fraction of the sample volume undergoing the most drastic phase change (β → α′). Earlier in the build this fraction is large, and it monotonically decreases as more material is added. Because less material is undergoing a phase transformation with respect to the bulk, the corresponding increase on the bulk wave speed is expected to drop.

Similar results have been obtained in other studies of phase transformation in conventionally manufactured titanium alloys. Kumar et al.¹³ observed an increase in ultrasonic wave speed of ∼48 m/s between Ti6Al4V microstructure dominated by metastable β phase and microstructure including α′ phase, which they obtained through heat treatment. Ogi et al.⁴² reported the temperature dependent single crystal elastic constants for pure titanium. They showed a monotonically decreasing modulus of the α phase with increasing temperature and a drop in modulus of the β phase. Additionally, Li et al.⁴³ reported a temperature dependent density of Ti6Al4V from liquid to solid phases. Using these results, the wave speed increase due to the solid state transformation from the BCC phase (β) to the HCP phase (α) is estimated to be 100.4 m/s. The wave speed increase observed after laser processing and displayed in Fig. 5(d) is in agreement with the expected trends and magnitudes reported in these studies. Therefore, solid state phase changes offer an explanation for the post-print time shift observed. However, simultaneous ultrasound and temperature measurements of the process are necessary to corroborate this hypothesis more clearly.

The entire data set for both samples was divided with respect to processing steps and cycles. Each sample underwent 11 cycles with each composed of the steps outlined in Table II. The frequency spectra for each step within each cycle were averaged. The maximum amplitude and corresponding center frequency of the averaged spectra for each step and cycle are plotted in the top row of Fig. 6 for both samples. Similarly, the spectral centroid amplitude and frequency are shown in the bottom row of Fig. 6. In Fig. 6 different colors correspond to different processing steps, while cycles are distinguished by different markers. Both the maximum amplitude and spectral centroid result in some separation between processing steps in the frequency space, which indicates that each step has a distinct spectral signature. This separation is more evident for the post-mill and layer 1 steps, while the data for the remaining steps show considerable overlap. Nonetheless the data for each step appear to follow the same trend in all cases. The amplitude and frequency decrease with added layers, they increase slightly during the post-print downtime, and sharply surge upward during the post-mill downtime. Additionally, the spectral centroid appears to capture the distinction between steps, especially for sample 2. The ability to observe these distinctions could be because the spectral centroid takes into account the spectrum shape in addition to its amplitude and frequency content.

For these samples, the changes in amplitude with processing steps were attributed to scattering due to the microstructure and due to the surface roughness. Surface roughness is expected in parts made with AM processes due to the layer-by-layer approach to manufacturing as well as variations in manufacturing parameters and conditions. The top surface roughness of the samples was measured with a surface profilometer (Dektak-XT Stylus Surface Profiling System), resulting in an RMS roughness of 100 µm. Li et al.⁴⁴ derived expressions to account for the contribution of losses due to surface roughness in material attenuation, based on corrections applied to the reflection and transmission coefficients. The only surface relevant to this analysis is the top surface of the sample, and its corrected reflection coefficient may be written as

where $Ri$ is the corrected reflection coefficient, $Ri0$ is the ideal reflection coefficient, $h$ is the RMS roughness, and $kL$ is the longitudinal wave number in the material.

To quantify the surface roughness effect, the ratio of post-print to post-mill average spectral amplitudes was considered for all cycles and both samples. This ratio is defined as

and includes the effect of material removal which reduces surface roughness. Assuming the roughness of the milled surface can be neglected, its reflection coefficient would be equal to the ideal reflection coefficient, that is $Ri0$ = 1. Following the method proposed by Li et al.⁴⁴ and assuming the roughness measured from the top surfaces is representative of the layer roughness during the manufacturing process, the corrected reflection coefficient is $Ri$ ≈ 0.77. In other words, the surface roughness causes a drop in amplitude of about 23%. In Fig. 7(a), the theoretical corrected reflection coefficient $Ri$ and the ratio $ri$ for both samples are shown. $ri$ fluctuates randomly with respect to cycle for both samples, indicating that the sample roughness also fluctuates. Additionally, $ri$ for sample 1 (0.44 ± 0.06) is consistently lower than that of sample 2 (0.54 ± 0.05). That is, the amplitude drop is between 46% and 56%. This amplitude drop is greater than the anticipated drop due to roughness (23%), however, before milling the distance traveled by the ultrasonic waves in the material is also longer. Thus, the added losses may be attributed to loss due to scattering in the material.

The losses due to added material were quantified by calculating the ratio of layer to post-mill average spectral amplitude for all layers (⁠$X$ = 1, 2, 3, 4) as

Then $riX$ was averaged for all cycles as

where $n$ is the number of cycles.

Figure 7(b) shows the amplitude ratio $rX$ for all layers and both samples. As expected, this ratio decreases as the layer number increases due to losses caused by the added material. Also, sample 1 is consistently below sample 2, suggesting greater losses due to microstructure scattering.

The observed time domain shifts and amplitudes in the frequency domain were used to determine the ultrasonic material characteristics for both samples. Due to the uncertainty in sample height during the printing process, only data from the post-mill process step were considered. After milling, the height of the sample is known with accuracy and the amplitude of the signal is not influenced by surface roughness effects. The cumulative time shift with respect to the build plate signal was used in combination with the known sample height to calculate the corresponding ultrasonic wave speed for each hybrid layer. Similarly, the average post-mill spectral amplitude was compared with the average spectral amplitude of the build plate signal and the corresponding relative attenuation was obtained as

where $αi$ is the relative attenuation for hybrid layer $i$ and $di$ is the corresponding sample height. Note that $αi$ quantifies the loss due to microstructure scattering in the added material with respect to the original signal obtained with the measurement system. Additionally, the multilayered structure of the measurement system complicates losses due to diffraction and coupling.⁴⁵ For these reasons, $αi$ is a relative measurement, and is denoted as such.

The evolution of wave speed and relative attenuation for both samples is shown in Fig. 8. In both cases the wave speed steadily increases until reaching a plateau at 6.1 mm/µs for sample 1 and 6.0 mm/µs for sample 2. Similarly, the relative attenuation early in the sample displays drastic and/or not physically meaningful behavior for both samples, but stabilizes after the fourth hybrid layer at 4.8 mm. Sample 1 plateaus at 0.022 Np/mm and sample 2 at 0.019 Np/mm. As shown in Sec. II B, sample 1 has a more uniform spatial distribution of small pores while sample 2 includes regions of agglomerated larger pores which is in agreement with their respective bulk sample wave speeds. Additionally, while optical microscopy did not show significant qualitative differences in the microstructure of both samples, the results are limited to the location sampled for microscopy and the difference in attenuation requires further exploration. The unusual behavior observed before the 4th hybrid layer in both the wave speed and relative attenuation values may be a geometric effect due to the interaction between the ultrasonic wave front and the short sample geometry. This behavior stabilizes after about four wavelengths.

In this work, a detachable fixture was designed to introduce ultrasonic sensors in a hybrid powder DED manufacturing platform. With this fixture, ultrasonic measurements were collected during the hybrid manufacturing of two Ti6Al4V samples. The results demonstrate the ability to quantify material properties such as wave speed and attenuation throughout the sample in a hybrid DED system. Different processing steps were observed to have distinct spectral characteristics, which depended on both the surface roughness and the build height. Manufacturing downtimes due to the hybrid processing enabled observations of a time of flight shift that, after correction for thermal expansion, exhibited the expected BCC → HCP solid state phase transformation behavior.

The importance of quality control for microstructure and mechanical properties is emphasized in the context of hybrid AM and manufacturing of functional and patterned material domains. For these advanced applications, defect detection approaches on their own are insufficient for quality control. The approach introduced here showcases the ability of ultrasonic measurements to interrogate build volumes and quantify microstructure and elastic property changes as a means of monitoring the AM process. Given desired material properties that can be translated into ultrasonic quantities, the measurements shown in this work may be used to define thresholds of acceptable and unacceptable process signatures. Thus, the capabilities and advantages of in situ real time ultrasound measurements in a hybrid DED process are demonstrated.

Limitations of this approach include the geometric constraints imposed by the fixture and the measurements as well as the fact that the entire information from a given layer is given in terms of the average waveform. Therefore, any kind of plane spatial mapping is not feasible. Future work may involve measurements from a statistically representative set of samples. Ultrasound in situ measurements may also be related to varying manufacturing process parameters and multimaterial builds. Data clustering algorithms may be used to improve the sensitivity to process steps as well as the ability to establish thresholds for normal and abnormal conditions. Finally, alternative measurement setups as well as comparison with other process signatures such as acoustic emissions, thermal data, and process power consumption may be used to improve upon characterization flexibility and accuracy.

This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. 1610400. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. Manufacturing and characterization analysis were performed at the NanoEngineering Research Core Facility (NERCF), which is partially funded by the Nebraska Research Initiative. The author's also wish to thank Ziyad Smoqi for his support in x-ray CT measurements.