Imaging pyrometry of metal fragments initially generated during the supersonic impact of reactive metal projectiles

The detonation of a cased high-explosive charge causes the casing to fracture into fragments that are typically accelerated to speeds between 0.5 and 2 km/s, a velocity range commonly denoted the explosively launched regime.¹ The explosively accelerated fragments, or projectiles, are intended to inflict physical damage upon any nearby surface they intercept. If the casing is made of a reactive material instead of the typical inert steel, the high-velocity impact of these projectiles may also initiate chemical reactions between the fragments generated upon impact and the oxygen in the surrounding atmosphere. If the additional chemical energy released in the vicinity of the target is significant, the loading on the target will be enhanced.

Within the explosively launched regime, projectiles made of a bulk reactive metal, which forms a class of reactive materials, have a specific chemical energy when they fully oxidize with oxygen that significantly exceeds their specific kinetic energy, as shown in Fig. 1. As such, the ability of a reactive metal projectile to release its chemical energy following an impact is of particular interest for terminal ballistics applications that seek to maximize the energy delivered to a target. More fundamentally, there is interest in elucidating the physical processes associated with the impact-initiated energy release and determining the associated timescales of energy release.

The high-velocity impact of a reactive material projectile in an oxidizing atmosphere may be conceptually divided into two stages, denoted the early-impact and post-impact stages, which are characterized by different timescales of energy deposition.^2,3 The early-impact stage spans the time just before impact to the period of contact between the projectile and the target. It occurs over a timescale of microseconds during which some of the projectile kinetic energy is promptly converted to other forms of energy as the projectile deforms and fragments. During this stage, the fragments may be heated to a level sufficient to initiate chemical reactions between the hot fragments and the surrounding oxygen. The post-impact stage includes the events that occur after the early-impact stage, including secondary impacts of the fragments with nearby obstacles such as chamber walls. However, it is primarily characterized by energy release from the continuous combustion of the fragments over a timescale ranging from a few hundred microseconds to a few milliseconds. Simpson et al.⁴ defined more formally the timescale for the early-impact stage as the ratio of the projectile radius to the impact velocity. In the present investigation (i.e., a 3.1 mm diameter projectile impacting typically at 1.2 km/s), this leads to a timescale of $τ=1.3μ s$⁠. The post-impact stage may be considered to begin at times greater than about $5τ$ or $6.5μ s$ in this example.

Much of the work on impact-initiated energy release of reactive material projectiles has utilized the so-called Vented Chamber Calorimetry (VCC) technique to classify the performance of reactive materials in terms of energy release efficiency.^2,5,6 The VCC technique consists of launching a projectile into a test section through a small entrance hole typically covered with a thin sheet of metal. The projectile perforates the sheet, which may lead to initial fragmentation, before impacting a solid target at the end of the chamber. The dimensions of the chamber are selected such that the energy released causes pressurization of the test section measured using fast-response pressure transducers. The key metric in this diagnostic technique is the quasi-static pressure attained globally by the gas, which is then related to the total energy released by the projectile using conservation of energy. Due to the nature of the measurement, the VCC technique is useful for characterizing the overall energy release during the post-impact stage. However, it provides little or no insight into the rapidly evolving early-impact stage, which requires optical diagnostics with sub-microsecond time resolution to investigate.

The active use of optical diagnostics by the terminal ballistics community is relatively recent,^3,7–9 in comparison to the geology and planetary science, and aerospace communities, which have routinely examined the light emission from hypervelocity impacts in vacuum to draw conclusions regarding the nature of the impact event.^10–12 Even within the lower velocity range of the explosively-launched regime, the impact of a reactive material projectile in the presence of oxygen will produce light emission, but of a different nature than that from a hypervelocity impact in vacuum. This light emission may be used to determine if and when chemical reactions between the fragments and the oxygen in the air initiate, at which temperature, and if these chemical reactions are sustained for a given duration.

In Idrici et al.,³ a spatially integrated three-color pyrometer was used to record the light emission signature at three different wavelengths following the impact of small cylindrical projectiles comprised of either Al, Ti, or Zr with an inert aluminum oxide target in air and inert environments. The emission signatures were used to infer the three-color temperature history of the metal fragments. In air, they burned at a temperature that was near their corresponding adiabatic temperature. Combustion was sustained for a few hundred microseconds for the case of the impact of Al projectiles. In contrast, for the impact of Ti and Zr projectiles, the signals lasted for several milliseconds, generally because of secondary impacts that generated more reacting metal fragments, extending the overall combustion time. An increase in impact velocity led to light emission with higher intensity and longer duration for all metals, attributed to a visible increase in the number of reacting particles. A notable difference was observed between the impacts of Al projectiles and those of Ti and Zr projectiles. For Al, the appearance of a stationary cloud of particles near the target was necessary for sustained combustion to occur. A critical impact velocity of 700 m/s was identified for the selected projectile and target, below which a burning cloud of particles did not form. In contrast, fragments produced by the impact of Ti and Zr projectiles burned continuously as they traveled through the test section. Upon impact, the temperatures measured in both air and argon exceeded the equilibrium adiabatic temperature of the metals, which was attributed to compressed gas at the interface between the projectile and the target. Although the temperature dropped rapidly to values comparable to the adiabatic temperature, the relaxation time of the photomultipliers and signal pileup in air, due to the superposition of light emissions from the shocked air and hot particles, made it difficult to determine the temperature of the fragments during the first few microseconds of impact. This paper extends the previous work³ by focusing on the early-impact stage. The early-impact stage, being a precursor to the post-impact stage, has a direct influence on whether longer timescale chemical energy release will occur.

The early-impact stage of Al projectiles in the hypervelocity regime at low atmospheric pressure, where combustion is not of particular interest, has been investigated. Simpson et al.⁴ showed that the impact of an aluminum sphere with a hard stainless-steel target at 3 km/s, in a partial vacuum (⁠ $∼$1.33 kPa), produced the strong Al doublet lines at 396.1 nm during the first microsecond of contact. By $7.3μ s$ after initial contact, an AlO band formed, indicating prompt oxidation even with a trace amount of oxygen. Both Simpson et al.⁴ and Ma et al.¹³ hypothesized that jets of metal form from the shocked target and/or projectile material that is trapped at the interface between the two and dragged outward. These jets may be in the solid phase or partially molten, but as they travel away from the impact location at hypervelocity speeds, they are believed to undergo aerothermal ablation, which causes them to break up into molten droplets that eventually vaporize, which is consistent with the presence of the Al atomic lines in the spectra. The vapor can react promptly with the trace amount of oxygen available to produce aluminum sub-oxides. Although the experimental conditions in their studies differed from the present study in many aspects, their spectroscopic results are useful for comparison.

This paper investigates the early-impact stage of bulk cylindrical metallic projectiles, comprised of either Mg, Al, Ti, or Zr, impacting a solid aluminum oxide target at speeds between 1.1 and 1.3 km/s, in oxidizing and inert atmospheres. The goals of the present work include determining the spatial distribution of the temperature of the fragments during the early-impact stage and inferring whether chemical reactions initiate promptly in oxidizing atmospheres. This is accomplished by using a three-color imaging pyrometer with an exposure time of 200 ns to capture a sequence of snapshots of the spatial temperature distribution of the fragments during the first $5μ s$ after contact between the projectile and the target.

High purity (99.5%, Alfa Aesar) magnesium (⁠ $∼$0.045 g), aluminum (⁠ $∼$0.065 g), titanium (⁠ $∼$0.11 g), and zirconium (⁠ $∼$0.16 g) cylindrical projectiles (3.1 mm diameter $×$ 3.1 mm long) were used. Each projectile was mounted and lightly epoxied to a polycarbonate sabot (4.8-mm-thick and 7 mm OD), with a Bridgman seal machined at the rear face. A small 1.5-mm-diameter neodymium magnet was embedded on the curved side of the sabot. The impact velocities were limited to the range 1.1–1.3 km/s, which are characteristic of material velocities from explosively launched fragments. The impact target was a fully dense (⁠ $3.9 g/ cm 3$⁠) 50-mm $2$⁠, 8-mm-thick aluminum oxide plate replaced after each trial and backed by a rigid steel support. $Al 2 O 3$ was selected because it is inert such that any chemical reaction is from the projectile, rather than the target.

In the present work, a single-stage helium-driven light-gas gun was used to accelerate each projectile. A schematic of the experimental setup, which includes the light-gas gun, the test section, and optical diagnostics, is shown in Fig. 2.

The light-gas gun was constructed from commercial high-pressure stainless-steel tubing and Swagelok connectors.³ It included a high-pressure driver gas section, with a double diaphragm firing system to control the launch of the projectile, and a barrel of 7.1 mm ID (inner diameter) and 2.4 m length. The barrel included a double coil gauge, followed by an exhaust valve, and a sabot stripper close to the test section. The double coil gauge was used to determine the velocity of the sabot-mounted projectile before it passed through the sabot stripper. The velocity determined with the gauge was comparable to that estimated from the high-speed video recordings of the projectile. Prior to firing, the barrel was evacuated independently from the test section. The use of a low-pressure gasdynamic sabot stripper, consisting of $∼$200 kPa of $N 2$ gas trapped between two mylar diaphragms, was found to be the easiest way to strip the sabot from the projectile to avoid disrupting the early-impact stage between the projectile and the target.

The impact occurred within a test section, 165 mm ID and 950 mm long, with opposing rectangular polycarbonate windows (80 mm wide $×$ 455 mm long). The test section could be evacuated and filled with a different gas independently from the gun barrel.

The optical diagnostics consisted of an HSFC-Pro camera system converted into an imaging pyrometer which is described in detail in Sec. III, a Shimadzu HPV-X2 high-speed camera to visualize the impact event, including both the early- and post-impact stages, a three-color spatially-integrated pyrometer described in Goroshin et al.,¹⁴ and two high-resolution spectrometers. The Shimadzu camera recorded at a rate of 5 Mfps, with a 200 ns exposure time.

Spectra were captured in a few experiments using two high-resolution HR4PRO spectrometers (4000 series) from Ocean Insight. One spectrometer was designed with a wavelength range between 260 and 400 nm (UV) and the second with a wavelength range between 435 and 565 nm (VIS), both with an entrance slit of $50μ m$⁠. The spectral calibration was performed by using a tungsten-halogen filament lamp from Ocean Insight (HL-2000-LL Light Source) with a known color temperature of 2800 K. Light was collected from a small field of view, a few centimeters in diameter, located in front of the target. A 1 mm quartz optical fiber and a $600μ m$ optical fiber were used for the UV and VIS spectrometers, respectively. When using the UV spectrometer, the polycarbonate windows were replaced by quartz windows. The spectrometers have a built-in minimum integration time of 4 ms so the time resolution was too coarse to track species production over time.

The schematic shown in Fig. 2 includes the interconnections used to trigger the imaging pyrometer, high-speed camera, and spectrometers. The optical diagnostics were all triggered using the sudden flash of light produced when the projectile intercepted the target. The light emission from the impact flash was collected using a modified 35-mm SLR (single-lens reflex) camera with a 50-mm lens shown in Fig. 3. The viewfinder of the camera provided a convenient system to select the area of interest. The fiber taper bundle (15–1 mm diameter) mounted in the camera film plane ensured that light was collected uniformly from a large area of interest. The SLR camera was positioned near the target (⁠ $∼$1 m away) so the light was collected over a circular area of 8 cm in diameter centered on the middle of the target. The light emission was transferred via the taper bundle to a 1-mm-diameter optical fiber (dashed blue line in Fig. 2), which transmitted the light to a three-color spatially-integrated pyrometer with photomultipliers and narrow bandpass filters centered at 430, 570, and 650 nm with a FWHM (full width at half maximum) of 10 nm.^3,14 All three signals were captured by an oscilloscope. The signal from the 650 nm channel was used to trigger the oscilloscope which sent a TTL signal to the optical diagnostics to trigger them (see green arrows in Fig. 2).

The location of the SLR camera with the fiber taper bundle and the settings of the pyrometer were maintained constant for all trials to allow comparison of the signals. The gain of each photomultiplier was set to a maximum to ensure that signals were also captured in both argon and nitrogen, and to maximize the intensity of the signal from the impact flash so that time zero for all the optical diagnostics would be as synchronized as possible with the contact of the projectile with the target. There was an uncertainty of at most $+$50 ns in the time at which the images were captured based on when the oscilloscope was triggered by the impact flash.

The impact tests were conducted in oxidizing atmospheres, including air and binary mixtures of argon and oxygen with the oxygen at a concentration of 21% or 40%, inert atmospheres, including argon and nitrogen, to isolate the effects of chemical energy release, and in a partial vacuum (300–1300 Pa) to also isolate hydrodynamic effects.

It is worth noting that the four metals under study were selected because they have been extensively studied as additives to high explosives.¹⁴ Reaction of the explosively dispersed metal particles can lead to enhanced blast and impact loading effects. Furthermore, the four metals represent two distinct combustion mechanisms encountered in solid fuel combustion. According to the classification by Glassman,¹⁵ the combustion temperature of a metal is limited by the boiling point of its metal oxide. Metals with a boiling point below the dissociation limit of their metal oxide should undergo homogeneous combustion in the diffusion-limited regime, while metals with a boiling point above the dissociation limit of their metal oxide are expected to burnthrough heterogeneous combustion.

As shown in Table I, in the diffusion-limited regime, magnesium and aluminum burn with a gas-phase flame lifted from the metal surface, where nano- and sub-micrometer metal oxides form, indicating homogeneous combustion. According to this classification, titanium should also burn with the formation of gaseous species. However, they were not observed during spectroscopic studies. Consequently, titanium and zirconium are expected to undergo heterogeneous combustion.

In the present experiments, the imaging pyrometry system utilized an HSFC-Pro imaging system (PCO Imaging), comprised of four image-intensified monochrome cameras. The light from the impact event was collected through a common lens (AF-S Nikkor 300 mm) and divided among four different channels by a series of beam splitters. Positioned behind four of the beam splitters were narrow bandpass filters, followed by lenses, image intensifiers (photocathode type S20, micro channel plate, and phosphor screen), and CCD plates (SuperVGA, 1280H $×$ 1024V pixels). A schematic of the imaging pyrometer is shown in Fig. 4.

The image intensifiers within the HSFC-Pro camera permitted the use of low exposure times despite the significant reduction in the intensity of the light emission transmitted due to the narrow bandpass filters. The exposure time was set to 200 ns for trials in oxidizing atmospheres and 400 ns for trials in inert atmospheres. The main disadvantage of the image intensifiers was that each channel of the HSFC-Pro could only capture two snapshots per impact event, which limited the number of temperature maps per trial to four (for two-color pyrometry) or two (for three- or four-color pyrometry). This was nonetheless sufficient to investigate the first few microseconds of the impact event. In the present experiments, the HSFC-Pro was configured as a three-color imaging pyrometer, although all four channels were used. The images were acquired through the CamWare V1.21 software and recorded in 12-bit TIFF format with a maximum pixel count of 4095.

The narrow bandpass filters (Andover Corporation) selected had wavelengths centered about 430 nm (430FS10-50), 570 nm (570FS10-50), 630 nm (630FS10-50), and 670 nm (670FS10-50), with a FWHM of 10 nm, and a diameter of 50 mm. In a few trials, the 430 nm filter was replaced with a 480 nm (480FS10-50) narrow bandpass filter. The narrow bandpass filters were chosen based on the emission spectra of the metals under study, which were captured prior to these trials using a low resolution USB4000 spectrometer (Ocean Insight) with a wavelength range of 400–850 nm. All filters selected fell outside the atomic lines and molecular bands emitted by the metals of interest.

The metals under investigation were approximated as gray bodies, which introduced an error in the measured temperatures. To minimize this error, the minimum and maximum wavelengths were selected to be as far apart as possible. The importance of choosing narrow bandpass filters with widely spaced wavelengths was demonstrated in several studies, where the color temperature estimated using a two- or three-color pyrometer was compared to that estimated with a spectrometer utilizing a much wider wavelength range. In Badiola et al.,¹⁶ the peak combustion temperatures for Ti and Zr particles deduced using a two-color pyrometer (532 and 589 nm, FWHM not disclosed) were much lower than their adiabatic temperatures, and more than 400 K lower than the temperatures deduced using a spectrometer (350–700 nm). In another study conducted with iron particles, the global temperature deduced with continuous spectra in the range of 1000–1600 nm was about 150 K higher than the peak temperatures determined using two-color pyrometry (850 and 950 nm, FWHM of 10 nm).¹⁷ Selecting narrow bandpass filters with wavelengths that are far apart from each other also has the advantage of increasing the dynamic range of the imaging pyrometer.

To obtain a color temperature map from the impact images, several steps were followed to calibrate the imaging pyrometer. For brevity, the data processing procedure is described in detail in  Appendix A. Readers unfamiliar with imaging pyrometry are encouraged to review  Appendix A first.

To begin the analysis, sequences of images captured with the HSFC-Pro camera to visualize the early-impact stage are presented in Figs. 5–7. Figure 5 shows the first $2μ s$ of the contact between an Al projectile and an $A l 2 O 3$ target with an impact velocity of 1.2 km/s in air. A dashed blue contour was added to the first three Al images to delineate true light emission from light reflected by the surface of the projectile. Figure 6 shows the first $5μ s$ of the impact of a Zr projectile at 1.05 km/s in air. Figure 7 shows the first $1.5μ s$ of the impact of Zr projectiles in argon (a) and (b) and nitrogen (c) and (d). Note that the contrast was increased in all images to enhance certain features.

From the sequences of images captured in both oxidizing and inert atmospheres, several conclusions may be drawn about the early-impact stage prior to generating the temperature maps.

First, light emission began suddenly upon contact between the projectile and the target with the nature of the light emission depending on the atmosphere. Brighter light emission was observed at the periphery of the projectile at time zero in air [Figs. 5(a) and 6(a), 200 ns exposure], argon [Fig. 7(a), 400 ns exposure], and nitrogen [Fig. 7(c), 400 ns exposure], than in vacuum. In a vacuum, the imaging pyrometer could not be triggered directly upon impact due to insufficient strength of the 650 nm signal. Consequently, in gaseous atmospheres, some of the light emitted immediately upon contact between the projectile and the target was attributed to gas dynamic effects that were not present in vacuum. In gaseous environments, the projectile drives a bow shock ahead of it. When the front of the bow shock intercepts the target, it reflects off the target and continues to reverberate between the target surface and the projectile surface, compressing a pocket of gas between them. This phenomenon was also reported in Ref. 3 during impacts in argon and in air, where the temperatures measured upon initial contact exceeded the equilibrium adiabatic temperature of the metals.

Second, while the projectile deformed against the target, brighter metallic particles emerged from the interface between the projectile and the target, and formed a corona around the bulk of the projectile [Figs. 5(b) and 6(b) in air, and Figs. 7(b) and 7(d) in argon and nitrogen, respectively]. These particles were the smallest, fastest moving, and brightest fragments produced during the early-impact stage. The speed of these particles was found to exceed the impact speed of the projectile in a companion study.¹⁸ In argon and nitrogen, the light emitted by these metal particles was of weaker intensity, despite the exposure time being twice as long as that used in air, and of shorter duration. Consequently, the intense light emission observed in all oxidizing atmospheres during the early-impact stage must have included the contribution of prompt chemical reactions between the metal particles and the oxygen in the surrounding atmosphere. Rapid oxidation initiated within at most hundreds of nanoseconds (250 ns including the exposure time and the uncertainty on the trigger) from first contact between the projectile and the target [Fig. 5(a)].

Signals from the 650 nm channel of the three-color spatially integrated pyrometer for Ti and Zr impacts are provided in  Appendix B. On the top are three trials conducted with Ti projectiles in $N 2$ and air, and on the bottom are three trials conducted with Zr projectiles in Ar, $N 2$⁠, and air. The signals obtained in air were saturated due to the settings of the pyrometer discussed in Sec. II. For all reactive metals studied, the signals rose sharply upon collision in all gaseous atmospheres. However, the amplitude achieved in air was substantially higher than in argon and nitrogen. This indirectly confirmed that an additional mechanism of energy release occurred in oxidizing atmospheres upon impact, consistent with the observation made using the imaging pyrometer. Furthermore, the signals obtained in air were significantly longer in duration, as reported in Idrici et al.,³ where this was attributed to combustion of the metal particles and secondary impacts producing more reacting fragments.

Third, it was observed that the tilt angle of the projectile relative to the target consistently determined the preferred direction of motion of the ejecta of particles. They preferentially emerged between the projectile and target surfaces that formed the smallest angle (or angles in some cases). For example, in Fig. 8, the Zr projectile intercepted the target on the left of its flat face, forming a small (acute) angle with the target. This resulted in most of the Zr particles exiting to the right. In Figs. 5 and 6, the deformation of each projectile primarily occurred on their curved surface, starting at a contact point near their left base. The ejecta of bright particles was symmetric on either side of the growing line of contact (or line of symmetry) between the curved face of the projectile and the target. The ejecta emerging between the surfaces that formed the largest angle appeared to have a smaller mass (Fig. 5, left). The only exception was with Zr projectiles, where a non-negligible amount of ejecta also emerged between the surfaces that formed the largest angle, as seen in the upper left corner of Fig. 6, frames (d) to (f).

Fourth, the number of bright metal particles increased as the projectile deformed. However, the fine fragmentation eventually ceased, and a “dark” front appeared, consisting of a region of cool metal particles and/or the bulk of the projectile that continued to deform and split into larger fragments [Fig. 5(e) and Fig. 6(d)]. Only the fine particles shed at the interface between the projectile and the target appeared to undergo oxidation during the early-impact stage. The bulk of the projectile and the larger fragments were not visibly reacting. However, evidence of oxidation was observed on the new surfaces of larger Zr⁶ and Ti fragments, indicated by their blue and purple discoloration (see  Appendix C).

Finally, at the impact velocities under study, the bright and fine metal particles shed at the interface appeared to comprise only a fraction of the initial projectile mass.

Sequences of images captured with the Shimadzu camera showing the impact of an Al and a Zr projectile in air are presented in Figs. 9 and 10, respectively. These images highlight some differences in the motion of fragments during the post-impact stage between the lighter metals which are expected to burn with the production of vapor species (such as Al and Mg) and the heavier metals expected to undergo heterogeneous combustion (such as Ti and Zr).

During the impact of Al and Mg projectiles, the particles shed at the interface formed either a continuous front (as in Fig. 9) or two fronts, depending on the tilt angle. These metal particles rapidly decelerated due to drag and remained near the target. In Fig. 9, image (c), larger fragments appeared on the left side of the bright zone and became more apparent by image (e).

In contrast, during the impact of Zr and Ti projectiles, multiple thin, independent jets of particles formed and rapidly moved away from the impact location in all directions, as shown in the later images in Fig. 10. The dispersal of the Ti and Zr particles over a larger volume is likely due to the higher inertia of the metal particles similar to the explosive dispersal of dense metal particles in heterogeneous explosives.

The next step was to estimate the temperatures achieved by the bright particles shed at the interface in the different environments. The temperature maps were obtained between $250 ns$ and $4.5μ s$ to avoid including the light emission from the compressed gas.

The stoichiometric adiabatic temperature at standard state, $T ad$⁠, was computed for each metal in each atmosphere using CEA,¹⁹ with results presented in Table II. The calculations assumed the fragments were initially in the solid phase at 293 K, and the atmosphere surrounding the particles at 101 kPa and 293 K, although this assumption is likely erroneous. As noted briefly above, the bright metal particles travel at supersonic speeds, driving bow shocks ahead of them that modify the flow field around them. Since the physical state of the metal particles and oxidizer influence $T ad$⁠, the values presented in Table II must be treated as reference values only.

Temperature maps in vacuum, argon, and nitrogen could not be obtained, as emission intensities fell below the linearity thresholds of one or more channels (see  Appendix A), despite using an exposure time of 400 ns. The light emitted by the ejecta, following thermalization of kinetic energy and oxidation with the trace amount of oxygen remaining after evacuating the test section, was too weak to render useful images. Nevertheless, the detection of these incandescent particles indicates that their temperature increased during the impact. This temperature rise is a necessary condition, in the presence of sufficient oxygen, to achieve reaction rates high enough to sustain chemical reactions. Hence, the particles that are hotter in these inert environments should be the same ones that undergo rapid oxidation when the oxygen concentration is sufficiently high.

Although nitrogen is not typically considered an inert atmosphere, it was included because its properties closely resemble those of air, making it a valuable atmosphere for comparison. Nitrogen can chemically react with Mg, Al, Ti, and Zr to form $M g 3 N 2$⁠, AlN, TiN, and ZrN, respectively. However, light emission from the ejecta remained weak even for Ti and Zr, which have $T ad$ values in nitrogen that are higher than in air (see Table II).

Figure 11 shows an example of temperature maps obtained at $1$ (left) and $3μ s$ (right) during the impact of a Mg projectile in Ar– $O 2$ 40%, using the 430, 570, and 630 nm channels of the HSFC-Pro. Images (a) were captured using the 630 nm channel, and images (b) show the remaining pixels in white after applying intensity thresholding. These white pixels were the only ones considered for temperature measurements. Images (c) present the corresponding temperature maps of condensed matter, derived from pixel triads using Eq. (A6) in  Appendix A. They show a range of uniformly distributed temperatures with no observable gradients. The temperature distributions were Gaussian, with a distinct temperature peak, as illustrated in image (e). A Gaussian fit (red curve) was applied to the data to extract the mean temperature (⁠ $T$⁠) and the standard deviation (⁠ $σ$⁠), representing the temperature range encompassing 68% of the data points. Pixels with temperatures within the range $( T ± σ ) mean$ are highlighted in images (d). The Gaussian temperature distribution was common to all impact tests. However, when intensity thresholding was performed inadequately, a bimodal distribution with an additional lower temperature peak was obtained. A similar sequence of images is shown in Fig. 12 for the impact of an Al projectile in air at $1$ (left) and $4.5μ s$ (right).

Results for aluminum trials in oxidizing atmospheres were plotted on a graph of temperature $( T ± σ ) mean$ vs time and shown in Fig. 13. Each atmosphere is labeled with a specific color, cyan for trials conducted in air, purple for Ar– $O 2$ 21%, and red for Ar– $O 2$ 40%. $T ad$ of Al in air and Ar– $O 2$ 40% were added for reference. The uncertainty in the temperature was estimated to be $±$100 K (see  Appendix A for details).

The temperatures obtained in all three oxidizing atmospheres overlapped significantly, even if there is a difference of 260 K between the equilibrium adiabatic temperature in air and Ar– $O 2$ 40% for Al. Furthermore, the temperature measurements were independent of time during the first few microseconds of the impact event, allowing the aggregation of temperature distributions from all trials conducted in a specific gas. Summing the Gaussian temperature distributions resulted in a new Gaussian distribution, from which the mean temperature and standard deviation for the ensemble of data sets were extracted $( T ± σ ) ens$⁠. They were (⁠ $3335±105$⁠) K in air, (⁠ $3357±118$⁠) K in Ar– $O 2$ 21%, and (⁠ $3389±114$⁠) K in Ar– $O 2$ 40%, indicating a modest increase in mean temperature with higher oxygen concentration and substantial overlap in the data. All the temperature measurements were slightly below $T ad$ in air.

A summary of the results for Mg impacts conducted in air (cyan) and in Ar– $O 2$ 40% (red) is shown in Fig. 14, including the $T ad$ in each atmosphere. Results for trials conducted with the 430, 570, and 630 nm narrow bandpass filters were labeled with solid circle symbols. In a few trials, the 430 nm filter was replaced with the 480 nm filter (for Mg, Ti, and Zr) in an attempt to increase the intensity of the images. These trials were labeled with solid triangular symbols (⁠ $△$⁠).

Starting with the results obtained with the 430 nm filter, similar to the Al trials, the temperatures estimated were independent of time during the first few microseconds of impact. Therefore, the mean temperature and standard deviation of each ensemble of data sets were (⁠ $2990±118$⁠) K in air and (⁠ $3100±90$⁠) K in Ar– $O 2$ 40%, a difference of 110 K between their means, close to the difference in their equilibrium temperature, but there is still significant overlap when accounting for the standard deviation. The temperature measurements were just below their respective $T ad$ values.

With the 480 nm narrow bandpass filter, the mean temperatures were higher by 300 and 200 K in air and Ar– $O 2$ 40%, respectively, and the standard deviations were higher by about 100 K from the results obtained using the 430 nm filter. However, when introducing the 480 nm narrow bandpass filter in a few Ti and Zr impact tests, the temperatures were essentially the same as with the 430 nm filter. This result can be explained using spectroscopy. The spectra captured during the impact of Ti and Zr projectiles were continuous, while the spectra captured during the impact of Mg projectiles contained atomic lines and molecular bands superimposed on the continuous background. Figure 15 shows a sample of emission spectra captured during the impact of a Mg projectile in air with an integration time of 4 ms. Mg atomic lines were emitted in the UV range (left) at 383.23 and 383.83 nm, and in the VIS range (right) at 516.73, 517.27, and 518.36 nm. Moreover, the MgO molecular band is present between 490 and 500 nm. The transmittance of the 480 nm narrow bandpass filter is at a maximum (⁠ $∼$60%) between 477.5 and 485 nm and decreases to nearly zero by 495 nm. The caveat is that even if the transmittance is minimal at 490 nm (⁠ $∼$2%), if MgO production begins during the first few microseconds of impact, it will increase the amount of blue light emission acquired by the blue channel of the imaging pyrometer, which will raise the condensed matter temperature measured. Indirectly, using the 480 nm narrow bandpass filter demonstrated that MgO production begins during the early-impact stage. Provided this conclusion is agreed upon, Mg in the gaseous phase must have also been produced rapidly after first contact between the projectile and the target.

Figure 16 shows emission spectra captured during the impact of an Al projectile. The Al doublet at 394.4 and 396.1 nm appeared in the UV range (left), and the strong blue-green $X 2 Σ +− B 2 Σ +$ AlO molecular band appeared in the VIS range (right). The conclusion drawn regarding the production of Mg and MgO gaseous species starting during the first few microseconds of impact are believed to also apply to Al and AlO species. Simpson et al.⁴ found that in a vacuum (⁠ $∼$1.33 kPa) with only trace amounts of oxygen available, Al vapor still reacted to produce AlO within $7μ s$ after the initial contact between the projectile and the target. Furthermore, Seidl et al.,⁹ using also time-resolved broadband emission spectroscopy, demonstrated that LaO is produced within $7μ s$ (exposure time) after perforation of a steel armor thin plate by ferrocerium (a pyrophoric alloy containing La at 23%) projectiles at velocities between 1100 and 1471 m/s. The intensity of the metal oxide band decreased directly after perforation and was of short duration, because light emission following perforation of a thin plate is less than following an impact on a rigid target.^7,9

The existence of species in the gaseous phase is consistent with the knowledge that Mg and Al burn in the vapor phase. An attempt was made to derive the gas temperature by fitting the molecular bands to a simulated spectral intensity distribution, with only some success fitting the $Δv=−1$ band of AlO $X 2 Σ +− B 2 Σ +$ to the Exomol line list.^21,22 Before fitting the band, the continuous background was subtracted. The estimated gas temperature ranged between 3377 and 3680 K. The lower limit agreed well with the results of Soo et al.²¹ for flat flames stabilized in a suspension of micrometer-sized Al particles, whereas the upper limit exceeded the volatilization temperature of $A l 2 O 3$⁠. Difficulties in subtracting the continuous background were believed to be responsible for the large range (about 300 K) in the estimated gas temperature.

Figure 17 is a sample of the temperature maps obtained during the impact of a Ti projectile in Ar– $O 2$ 40% at $1$ (left) and $3μ s$ (right), using the 430, 570, and 630 nm channels of the HSFC-Pro. Images (a) show the original images captured by the 630 nm channel. Images (b) display the remaining pixels in white after applying intensity thresholding. Images (c) present the corresponding temperature maps, with one of the temperature distribution shown in image (d). Once again, the temperature distributions were Gaussian with no obvious temperature gradients observed on the maps. By applying a Gaussian fit (red curve) to the data, the mean temperature and the standard deviation $( T ± σ ) mean$ were extracted. A similar sequence of images is shown in Fig. 18 for the impact of a zirconium projectile in Ar– $O 2$ 40% at $0.5$ (left) and $2μ s$ (right).

Results for titanium impacts conducted in air (cyan) and Ar– $O 2$ 40% (red) were plotted on a graph of temperature $( T ± σ ) mean$ vs time shown in Fig. 19 with the $T ad$ in air and Ar– $O 2$ 40% added for reference. The temperatures in both oxidizing atmospheres overlapped, despite the difference of 400 K between their equilibrium adiabatic temperatures. Furthermore, the temperature measurements showed no dependence on time during the early-impact stage. Thus, $( T ± σ ) ens$ were (⁠ $3257±130$⁠) K in air and (⁠ $3217±104$⁠) K in Ar– $O 2$ 40%. Both means coincided with $T ad$ in air.

Results for zirconium impacts conducted in air (cyan), Ar– $O 2$ 21% (purple), and Ar– $O 2$ 40% (red) are shown in Fig. 20. The $( T ± σ ) ens$ were (⁠ $3327±139$⁠) K in air, (⁠ $3344±154$⁠) K in Ar– $O 2$ 21%, and (⁠ $3326±145$⁠) K in Ar– $O 2$ 40%. There is again an overlap in the temperatures measured across all three oxidizing atmospheres. Surprisingly, the mean temperatures were significantly below $T ad$ in air, i.e., 3777 K.

The spectra captured during the impact of Ti and Zr projectiles had no atomic lines nor molecular bands, contrary to Mg and Al projectiles. Ti and Zr particles undergo surface combustion.

It was not obvious what condensed matter temperatures were expected during the first few microseconds of the high-velocity impact events. The equilibrium adiabatic temperatures in the different oxidizing atmospheres, which also corresponded to the temperatures measured during the post-impact stage in Idrici et al.,³ were used as the primary reference values to compare with the current experimental results. Temperature measurements and determination of emission species in the study of impact-initiated energy release in oxidizing atmospheres are rare and temperature measurements focused on the first few microseconds of impact are even more rare.^7–9 

In the following discussion, some of the conclusions from previous literature on individual metal particle ignition and dense metal flames will be cited, although the experimental conditions in the present study are different. In this study, the exposure time was set to 200 ns and temperatures were measured for less than $5μ s$ after initial contact of the projectile with the target, while in metal combustion studies exposure times are at least a few tens of microseconds. Furthermore, while the burning metal particles under investigation traveled at supersonic speeds relative to the atmosphere, comparable studies in metal combustion typically involve much lower relative speeds between the particles and the oxidizing flow.

A summary of the mean temperatures and standard deviations of the ensembles of data sets $( T ± σ ) ens$ for each reactive metal in every oxidizing atmosphere tested is presented in Table III and shown graphically in Fig. 21.

When Mg and Al particles burn in the diffusion-limited regime, they undergo homogeneous combustion with a vapor flame some distance from the surface of each particle, where combustion product metal oxides of nanometer or sub-micrometer size form. Figures 15 and 16 show the presence of gaseous species typical of vapor flames. The light emission captured by the imaging pyrometer originates from the surface of all the condensed matter species within its field of view. Some metal oxides, such as $A l 2 O 3$⁠, exist only in the condensed phase, while MgO, which was detected in the vapor phase (Fig. 15), can also exist in the condensed phase. Condensed phase metal oxides have high emissivity at high temperatures and are believed to contribute significantly to the continuum radiation recorded by each channel of the imaging pyrometer.^21–25 

The color temperatures measured for Mg were significantly above the boiling point of the metal particles (1363 K) and, on average, just below the melting point of the metal oxide MgO (3100 K). The temperatures measured during the Al trials were above both the boiling point of Al (2791 K) and the melting point of $A l 2 O 3$ (2345 K), but well below the volatilization point of $A l 2 O 3$ (4000 K). Thus, the color temperatures in both Mg and Al trials appear to have been strongly influenced by metal oxides in the condensed phase rather than the pure metal particle surface (see Table III and Fig. 21).

Densmore et al.⁷ measured peak color temperatures in air of 3200 and 2850 K for Ni/Al and Al-PTFE projectile, respectively, after they perforated a thin sheet of mild steel, and peak temperatures of 3600 and 3300 K, respectively, shortly after the fragments impacted an anvil. Measurements were conducted using a two-color imaging pyrometer. Provided that chemical reactions between Al particles and air oxygen were an important energy release mechanism in their high-velocity impact study, the temperatures measured in the present work during the early-impact stage of Al projectiles fall within their reported range for impacts on the anvil.

Given that the spectra captured during the impact of Ti and Zr projectiles had no atomic lines nor molecular bands, unlike Mg and Al, chemical reactions with oxygen occurred on the surface of their particles without a vapor flame. At these temperatures, the surface of the Ti and Zr particles would be at least partially molten, as their melting points are 1923 and 2128 K, respectively, which are well below the $( T ± σ ) ens$ measurements recorded in Table III. Furthermore, if any of the inherent thin metal oxide layer remained on the particle surfaces after impact, it would have been compromised, as the melting points of $T i 3 O 5$ and $Zr O 2$ are 2116 and 2988 K, respectively, also below the recorded $( T ± σ ) ens$⁠.

Temperature measurements of Zr ejecta had the largest discrepancies with the equilibrium adiabatic temperature in air. The temperature measurements did not deviate, within the scatter, from those for impact tests with Al projectiles (see Fig. 21), despite the 200 K difference between their $T ad$ values in air. Furthermore, in Idrici et al.,³ using the spatially-integrated pyrometer, the peak temperature of the Zr particles in air during the post-impact stage was in close agreement with $T ad$⁠. The discrepancies are partially accounted for by the calibration uncertainty of $±$100 K, the non-gray body behavior of the incandescent particles, and by the dependence of $T ad$ on the initial state of the metal particles and the gas surrounding them. However, the same errors apply to the temperatures measured for the other three metals.

It is likely that the lower combustion temperatures measured are a consequence of the reduced sensitivity of the imaging pyrometer at higher temperatures. As shown in Fig. 24 ( Appendix A), the pyrometer’s sensitivity diminishes above 3300 K, limiting the accuracy of measurements at very high temperatures. As discussed in Sec. III, the instrument plays a critical role in deducing the color temperature of the metal particles.^16,17,23,24 Zr has the highest $T ad$ in air and the largest discrepancy between experimental results and equilibrium temperature. In contrast, Mg and Ti have the lowest $T ad$ in air, below 3300 K, and their experimental results closely matched these values. Al was somewhat in the middle, so its temperature measurements may have been influenced by the imaging pyrometer as well.

To further substantiate the lower-than-expected temperature measurements during the impact of Zr projectiles, the results must be validated with another study. In particular, this will be achieved by shifting the sensitivity of the imaging pyrometer toward higher temperature values. This involves selecting narrow bandpass filters with lower wavelengths while maintaining at least a 150–200 nm difference between the minimum and maximum wavelengths. This shift toward the UV range will require a lens with higher transmittance in the lower wavelength range, which should improve the linearity of the sensors and, in turn, reduce the error associated with temperature measurements below 100 K.

Trials in inert atmospheres were conducted in part for the purpose of overcoming any temperature measurement deficiency of the imaging pyrometer. Trials in vacuum, argon, and nitrogen were effectively a go/no-go test with respect to the question of whether self-sustained chemical reactions of the metal particles in oxidizing atmospheres are initiated upon impact. These results confirmed that the temperatures measured in the presence of a sufficiently high oxygen concentration were due to rapid oxidation, rather than thermalization of kinetic energy.

Although temperature maps were strictly obtained 250 ns after initial contact between the projectile and the target, proving that chemical reactions initiated at least within less than 450 ns, images captured directly on impact (i.e., at 0 ns) showed that chemical reactions initiated before, within less than 250 ns. Due to this result, the word ignition was not used in this paper as it specifically refers to the transition from the kinetically-limited to the diffusion-limited regime of combustion.²⁶ This does not appear to be the case during a high-velocity impact event where the initial heating phase that is supposed to lead to thermal runaway is virtually non-existent, except perhaps for Zr which remains to be determined.

The presence of a temperature distribution in these impact tests was expected for three primary reasons. First, the light emission captured by the HSFC-Pro camera originates from the surface of various condensed matter species within its field of view. Second, the metal particles are ejected from the impact location at supersonic speeds relative to the surrounding gas, which may further influence the reaction rates and temperatures of the different surfaces in ways that are not currently known.

Third, although the size of the particles shed at the interface in this impact study is not known a priori, it is expected that the particles have a distribution of sizes. In a study conducted by Ning et al.,¹⁷ where individual iron particles (⁠ $25-55μ m$ diameter) were ignited with a laser in argon atmospheres with varying oxygen concentration, the peak temperatures were consistently higher for the larger particles when the oxygen concentration was above 21%.

The equilibrium adiabatic temperature was predicted to increase with oxygen concentration (see Table II). However, for each reactive metal under investigation, there was a significant overlap between the combustion temperatures measured at 21% and 40% oxygen concentration, indicating a weak dependence on oxygen concentration at least during the early-impact stage. In previous metal combustion studies, the oxygen content has, to some extent, influenced the measured temperature. In Ning et al.,¹⁷ an increase in the oxygen concentration from 21% to 30% led to an increase in the peak temperature of iron particles by 100 K on average, but between 30% and 51% there was virtually no difference. McRae et al.²⁷ measured an average temperature difference at two fuel lean concentrations of iron particles, in 30% and 40% oxygen in argon. However, there was still an important overlap between the results. Thus, the influence of oxygen concentration on the combustion temperature is not evident.

Furthermore, the microsecond timescale associated with the impact events in this study is perhaps not long enough for thermodynamic equilibrium to be achieved. Therefore, the lack of a temperature difference between the 21% and 40% $O 2$ concentration cases during the first few microseconds of impact was not unexpected.

All the impacts observed in the current investigation occurred with varying tilt angles between the projectile and the target. Although the tilt angle was somewhat random, the direction in which the smallest, brightest, and hottest metal fragments were ejected in relation to the tilt angle was not. The particles appeared as the projectile began to deform against the target and exited preferentially between the projectile and target faces that formed the smallest angle. Specifically with Zr projectiles, the material was also ejected in the direction with the largest angle, but in lesser quantities. Thus, the ejecta originated from material on and near the face of the projectile that was progressively colliding against the rigid target at supersonic speed. The ejecta must have been forced to exit rapidly before the two faces collapsed onto each other.

The observations related to the jetting phenomenon, i.e., hot and high-velocity ejecta, during a high-velocity impact, is reminiscent of the jetting phenomenon observed during explosion welding.²⁸ There are various geometries, but the most relevant for comparison here is the case of a flyer plate, made of a metal, positioned above a baseplate, also typically made of a metal, at an acute angle. An explosive material is distributed on top of the flyer plate and detonated at the point (or line) of contact between the two plates. As the detonation propagates down the length of the flyer plate, the high-pressure detonation products accelerate the flyer plate which impacts the baseplate progressively, one point at a time, at a specific angle and velocity. The material at each subsequent collision point undergoes a rise in pressure and deformation that are significant enough to bond the two surfaces together without melting. The extreme conditions at the contact point between the flyer and baseplates lead to hydrodynamic flow of the metals, forming a metal jet that moves away from the collision point.

The speed of the flyer plate and the collision angle are critical parameters in explosive welding as they influence the temperature achieved at the interface as well as the formation of the jet. For example, a more powerful explosive leads to a higher speed of the flyer plate and higher temperatures at the collision point due to the increased conversion of kinetic energy to thermal energy. The impact angle determines the shape and direction of the jet formation. A range of optimal angles exist below and above which no continuous jets are produced.²⁹ Within this range, a shallow angle produces a longer jet, whereas a steeper collision angle produces a shorter jet. The same concept is applied to shaped charge liners in terminal ballistics, with the metal jet formed being used to penetrate armor, albeit with a different configuration.

In this analogy, the projectile acts as the flyer plate, while the target serves as the baseplate. Rather than an explosive, the kinetic energy of the projectile drives the collision. Due to its tilt angle, the projectile initially impacts the target at a single point. As the projectile continues its motion toward the target, the collision point (or line of contact) shifts. This shift occurs more rapidly where the tilt angle is the smallest. As the projectile does not penetrate the rigid aluminum oxide target, but continues colliding against it, the material at the interface must be forced to exit through the free boundaries, or gap between the projectile and target. Consequently, the metal must be forced to exit more rapidly in the direction with the smallest tilt angle. In a companion study,¹⁸ particle jets were observed to move at speeds significantly exceeding the impact speed of the projectile. Similar to explosive welding, it is expected that a smaller tilt angle (above a threshold) and higher projectile speed will result in more and faster ejecta. However, this hypothesis was not tested in this study, as the size and velocity of the jets relative to the tilt angle were not systematically investigated. Where the tilt angle is the largest, the material does not have to exit as quickly, allowing it to deform and undergo fragmentation as predicted by Grady.^31,32 The scenario described above was observed in Fig. 8. As in explosive welding, the geology and planetary science communities hypothesized that jetting results from unstable flow at the moving point of contact between the projectile and the target, where an angle forms.^4,13,30

The analogy above provides a more complete explanation to the hypothesis suggested in Idrici et al.³ that the high temperatures measured in vacuum, between 2200 and 2600 K, were due to the particles being produced and heated through frictional effects, i.e., shear forces at the interface between the projectile and the target. In that study, the projectile’s tilt angle was not observed due to the lack of background lighting and the much lower frame rate of the high-speed Photron SA5 camera used.

The exact moment when hot ejecta production ceased was not clear. However, in the explosively-launched regime, ejecta production is likely to continue as long as the projectile moves toward the target (dependent on impact velocity) or until there is no remaining tilt angle between the projectile and the target.

Ablation has been suggested as the mechanism responsible for causing material jets formed during hypervelocity impacts to breakup into fine metal particles, which then vaporize.^4,13 Although Mg and Al atomic lines were observed in the UV and VIS emission spectra captured in oxidizing atmospheres (Figs. 15 and 16), from the images captured with the HSFC-Pro it is not clear if ablation contributed to a change in the phase of the ejecta for the impact velocities under study, which are considerably below that of the hypervelocity regime. The images captured of the metal particles show that they are already luminous in an inert environment and burning in oxidizing environments as they emerge from the interface between the projectile and the target. Furthermore, unlike the work of Simpson et al.,⁴ who obtained spectra for Al impacts at about 3 km/s in 1.33 kPa air, in the present experiments at lower impact speeds of 1.1–1.3 km/s, the Mg and Al impacts in vacuum produced insufficient light to record the emission spectra. Hence, no conclusive evidence was obtained to show that ablation during impacts in vacuum leads to the formation of gaseous metal species.

Since the projectile near the impact location is subject to stress levels above the material strength, hydrodynamic flow will occur. Thus, the jet of material that forms will be subject to the usual classical hydrodynamic instabilities, such as the shear-driven Kelvin–Helmholtz instability, which may lead to further fragmentation of the jet into fine particles.

The goal of this work was to obtain spatial temperature maps of the early-impact stage between small bulk reactive metal projectiles (Mg, Al, Ti, or Zr) and rigid $A l 2 O 3$ targets in oxidizing and inert atmospheres to obtain the corresponding temperature field of the fragments using three-color imaging pyrometry and determine if prompt chemical reactions with oxygen initiate during the early-impact stage. The metals were selected based on their mode of combustion. Al and Mg burn in the vapor phase, as indicated by the atomic lines and molecular bands in their spectra (Figs. 15 and 16), while Ti and Zr undergo heterogeneous combustion. In this study, it was found that a jetting mechanism, reminiscent of explosion welding, occurred during the high-velocity impacts, due to the tilt angle between the projectiles and the targets. Hot and high-velocity metal particles emerged preferentially between the faces of the projectile and target which formed the smallest angle upon collision. Only these particles were visibly hot in inert environments and reacted with oxygen in oxidizing atmospheres.

The metal particles were already burning as they were ejected from the interface, indicating that the chemical reactions with oxygen occurred promptly. The mean temperatures measured in air for Mg, Al, and Ti, were just below their respective stoichiometric adiabatic temperature. The only discrepancy was with the temperature measurements conducted during Zr impacts. This will be investigated further by modifying the imaging pyrometer to improve the accuracy at higher combustion temperatures.

Visually, the mass of fine luminous fragments shed at the interface represented only a fraction of the total mass of the projectile within the range of impact speed investigated. The jet formation was accompanied by the fragmentation of the bulk of the projectile into larger pieces that did not participate in the chemical reactions occurring during the early-impact stage. For the larger fragments to release their chemical energy, they must undergo secondary impacts with the test section walls at supersonic speeds.

A question remains regarding how much of the chemical energy of the metal projectile is released on a microsecond timescale upon impact. Rapid energy release during this stage will generate a local deposition of energy in the gas which will elevate its pressure. This topic is explored in a companion publication.¹⁸ 

The authors have no conflicts to disclose.

Dihia Idrici: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Project administration (lead); Software (lead); Supervision (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). Samuel Goroshin: Conceptualization (supporting). Jason Loiseau: Resources (equal). David L. Frost: Funding acquisition (lead); Resources (equal); Writing – review & editing (lead).

The data that support the findings of this study are available within the article.

 Appendixes A,  B, and  C provide a detailed description of the calibrations and data processing performed on the imaging pyrometer, signals recorded by the spatially integrated pyrometer, and a picture showcasing the discoloration on the surface of zirconium fragments.

The light frame in this study was either an image of the impact or an image of the calibration lamp. The dark frames and the bias frames were measured with no light on the sensor, with the cap on the lens. The dark frame must be recorded using the same digital camera ISO and exposure time settings as for the light frame. The flat-field frame can be captured at any ISO and exposure time settings. The bias frame recording must use the same ISO setting as the flat-field frame, but it must be taken with a very short exposure time.

Calibration to linear intensity increases random noise. To minimize the random noise, five to ten images were captured and averaged for each of the $D$⁠, $F$⁠, and $Bf$ frames. Sensor calibration was performed on the impact images but also on the images of the light emitted by a tungsten lamp which were used for spectral calibration (Appendix A 2).

The flat-field calibration [denominator of Eq. (A1)] is meant to investigate the Pixel-Response Non-Uniformity (PRNU). It was performed by using a Uniform Illumination Sphere–Light Source (UIS-LS) from StellarNet Inc., which consists of a $6$ in. integrating sphere with a $2$ in. diameter exit port. It comes with a tungsten filament bulb and a power control module with 8 levels to change the color temperature of the light output. The spectral output at each level is NIST (National Institute of Standards and Technology) traceable. Pixel-response non-uniformity can be a consequence of several factors, including the curvature of the lens which affects the intensity distribution, non-uniformity (dust and other imperfections) on the narrow bandpass filters, lens system, and beam splitters, as well as a damaged component in a channel. By including the complete optical setup in the flat-field calibration, those imperfections can be considered in the image calibration.

To uniformly illuminate the entire field of view, the lens was positioned against the $2$ in. port of the sphere with its focus set to infinity. Figure 22 shows color maps of the average normalized intensity distribution of the flat-field frame minus the bias frame (⁠ $F−Bf$⁠) for three of the channels. Ten frames of $F$ and $Bf$ were averaged for each channel. The pixel response of each channel was non-uniform as well as the response between channels.

The second step was spectral response calibration. Imaging pyrometry relies on a comparison of the light intensity captured by each channel at different wavelengths. However, for pairs of channels (including all the optical components), the ratio of their pixel count is most likely not equal to the true intensity ratio emitted by the incandescent object, because the spectral response of each channel is a function of wavelength. For example, the quantum efficiency of the S20 photocathode of an intensifier is maximum between 200 and 430 nm and decreases steadily until 900 nm. Furthermore, the transmittance of the lens also varies with wavelength. The Nikon AF-S Nikkor 300 mm lens reduced by nearly half the amount of blue light measured by the 430 nm channel compared to a Nikon AF DC-Nikkor 135 mm lens.

Here, $I E , m$ was acquired by positioning the calibration lamp far enough from the imaging pyrometer so that the details of the filament could not be imaged. Ideally, the light output would appear to come from a point source. First, an image of the lamp (turned off) was acquired by each channel to select a region of interest (ROI) corresponding to the circular output port. Second, at a given exposure time, an image of the light emitted by the hot tungsten filament, $L$⁠, and its corresponding dark frame, $D$⁠, were acquired by each channel. Five to ten images of each were captured and averaged to reduce random noise. Third, sensor calibration was performed on the average light frames using Eq. (A1). Fourth, the pixels contained within the ROI were summed or averaged to obtain $I E , m$ needed in Eq. (A2) to obtain the correction factor for each channel. The process was repeated at different exposure times (⁠ $2-150μ s$⁠).

The channels with the 570, 630, and 670 nm narrow bandpass filters responded linearly. Thus, the sensitivity ratios between these channels were nearly constant. However, the 430 nm channel behaved linearly with respect to the other channels only after approximately $25μ s$ (to the right of the red vertical dotted line). Thus, the largest discrepancies in sensitivity ratios involved the 430 nm channel. This is believed to be a consequence of the Nikon 300 mm lens blocking blue light preferentially, thus leading to a poorer quality of the images captured by the 430 nm channel at low exposure times. A minimum intensity, rather than a minimum exposure time, was believed to be required for the response of a channel to be linear. That minimum intensity was channel dependent, and it was set as a lower intensity threshold (or linearity limit) of the channel in the data processing.

The sensitivity ratios used to obtain the temperature maps for the 570, 630, and 670 nm channels were averaged over all exposure times. The sensitivity ratios involving the 430 nm channel were averaged after $25μ s$ (right of the red dotted line). Thus, another uncertainty in the temperature measurements was introduced by the variation in $σ m , 430$⁠. This uncertainty was estimated by using the lower, at $25μ s$⁠, and upper values of $σ m , 430$⁠. When using the lower limit, the mean temperature increased by less than 100 K, but when using the upper limit, the mean temperature decreased by less than 100 K. Thus, the uncertainty from using the average ratios was estimated to be at most $±$100 K. Note that the error from treating the metal particles and their metal oxides as gray bodies could not be quantified. With the tungsten lamp, for reference, the difference in temperature was less than twenty degrees when using a variable or constant emissivity. For now, it is assumed that the color temperature measurements have an uncertainty of $±$100 K.

To be able to detect variations in temperature, the ratios between the calibrated intensities (sensor and spectral) of two channels should show a strong dependence on temperature.³⁵  Figure 24 is a plot of the calibrated intensity ratio vs temperature between pairs of channels. Depending on the temperature range of interest, a specific triad of narrow bandpass filters should be selected. For this imaging pyrometer, the wider the difference in wavelengths, the greater the sensitivity to high temperature measurements.

Image correction, also called image registration, was performed on all impact images. The main disadvantage of using monochrome cameras with narrow bandpass filters as an imaging pyrometer, over a color camera with a triple-pass optical filter,^8,35 is the necessity and difficulty of aligning with precision all the images. In this experiment, each channel captured a slightly different image because the optical path followed by the light was slightly different as seen in Fig. 4. A software approach was adopted to correct for image distortion.³⁶ No attempt was made to fine-tune the position of the optical hardware inside the HSFC-Pro to avoid damaging the camera system.

Image registration was carried out by imaging the target with fiducial markers on it before the impact. MATLAB’s similarity transform function, from the image processing toolbox, was used to map the image produced by three of the cameras onto the image produced by the reference camera. The three resulting image transformation matrices were used to correct the impact images.

An additional consideration was internal reflection and reflection caused by the angle between the camera and the test section window. When the linearity limit was insufficient to remove the reflected images, an additional lower intensity threshold was applied.

Here, $I E , m$ corresponds to the intensity of a pixel of channel $m=1$⁠, 2, and 3.

Signals captured by the 650 nm channel of the three-color spatially integrated pyrometer are provided in Fig. 25 to highlight the differences in signal amplitudes across trials in argon, nitrogen, and air. On the top are three trials conducted with Ti projectiles in $N 2$ and air, and on the bottom are three trials conducted with Zr projectiles in Ar, $N 2$⁠, and air. The signals rose sharply upon collision in all gaseous atmospheres. However, signal amplitudes in air and other oxidizing atmospheres were significantly higher than those in argon and nitrogen, indicating an additional energy release mechanism during the early-impact stage. This observation was consistent with findings from the imaging pyrometer. The signals obtained in air were saturated due to the settings of the pyrometer discussed in Sec. II.

Larger Zr fragments recovered after an impact test in air (velocity between 1.1 and 1.3 km/s) exhibited a blue-purple discoloration on their new surfaces, indicating oxidation, as shown in Fig. 26.