Infrared thermography for boundary layer transition measurements

Infrared thermography has been successfully adopted in the field of flow diagnostics over the last decades. Detecting the laminar–turbulent boundary layer transition through variations in the convective heat transfer is one of the primary applications due to its impact on the aerodynamic performance. Recent developments in fast–response infrared cameras allow unsteady measurement of fast–moving surfaces and moving transition positions, which must consider the thermal responsiveness of the surface material. Experimental results on moving boundary layer transition positions are highly valuable in the design and optimization of airfoils or rotor blades in unsteady applications, for example regarding helicopter main rotors in forward flight. This review article summarizes recent developments in steady and unsteady infrared thermography, particularly focusing on the development of differential infrared thermography (DIT). The new methods have also led to advances in the analysis of unmoving boundary layer transition for static airfoil test cases which were previously difficult to analyze using single–image methods.


Introduction and scope
The laminar-turbulent transition of the boundary layer (BL) is one of the key factors when optimizing the aerodynamic performance of vehicles, referring to large drag penalties due to different skin friction coefficients, or referring to the impact of the BL parameters on the stall resistance. The measurement, prediction, and manipulation of the BL was tackled in numerous studies concentrating on steady aerodynamics with a stationary transition position, as found on fixed-wing aircraft, road vehicles, trains, etc Unsteady inflow conditions result in complex aerodynamics and a moving transition location, as for example seen on helicopter main rotor blades in forward flight. The Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. superposition of rotational and freestream velocities, and the consequent cyclic swashplate input to achieve moment trim, produce periodic variations of both the inflow magnitude and inflow direction at a given radial cross-section. It is known from simulations in steady hovering conditions that modeling of the BL transition is crucial for a correct prediction of the rotor power requirement, for example see Egolf et al [1]. Therefore, several contributions were made to integrate transition models into computational fluid dynamics over the last decades (Beaumier and Houdeville [2], Beaumier et al [3], Zografakis et al [4], Coder [5]). More recently, Heister [6] implemented several empirical criteria (Tollmien-Schlichting and crossflow instabilities, bypass mechanisms, and attachment line contamination) into an URANS framework provided by the DLR TAU Code. A similar approach was chosen by Richez et al [7] for the ONERA elsa code. Both studies performed successful comparisons to experimental data of the 7A/7AD rotor taken from integral torque measurements or point wise hot-film sensors applied within the GOAHEAD project [8]. However, a conclusive validation of numerical tools would benefit from spatially well-resolved experimental data including three-dimensional effects, which is apparent from the predicted transition distributions shown in figure 1 for a trimmed rotor at an advance ratio of 0.33.
For turbomachinery aerodynamics, an unsteady inflow of the compressor or turbine blades in a rotor-stator layout is caused by the periodic wake of preceding stages, which also results in moving BL transition [9, 10]. The transition not only affects the friction drag, but also the thermal loads of the blades. It is particularly difficult to access this type of flow with sensors and measurement equipment, hence, many studies simplified the setup by investigating the influence of periodic inflow perturbations on flat plates [11,12], curved surfaces [13], or stationary blade cascades [14]. In addition, early fundamental research of moving BL transition on flat plates [15,16] has currently been reviewed for fixed-wing aircraft applications [17], assuming that low-frequency atmospheric turbulence may affect the performance of natural laminar-flow airfoils.
The blades of horizontal-axis wind turbines can easily be subject to unsteady aerodynamics due to wind shear effects, yaw angles, interactions with the tower, or aeroelasticity [18,19]. It can be expected that these unsteady effects become increasingly important in the future, given that the laminar flow length is a crucial parameter of wind turbine airfoil design [20]. Another recent publication by Thiessen and Schülein [21] concentrated on moving transition positions due to the effect of forward flight velocity on a fixed-pitch propeller for unmanned aerial vehicles.
Steady BL transition over a wetted surface can be measured by techniques which are well-known and understood. There are several experimental methods available, with infrared thermography (IRT) being one of the most convenient and most popular methods. IRT is non-intrusive and produces spatially resolved results in a two-dimensional measurement domain.
No instrumentation of the model is required, but in some cases, an insulating surface coating must be applied depending on the base material. Pioneering work as described by Quast [22], Carlomagno et al [23], or de Luca et al [24] developed the underlying principle based on the Reynolds analogy. A small difference between the surface temperature T w and the fluid temperature T ∞ yields a convective heat fluxq depending on the local skin friction coefficient C f , with the external flow velocity U ∞ , and the fluid's thermal conductivity λ and kinematic viscosity ν. The surface temperature T w additionally depends on the conductive and radiative terms, but a qualitative evaluation of the surface temperature distribution T w is usually sufficient to differentiate between the laminar and turbulent regions of the BL. This also means that a temperature calibration of the camera, depending the surface emissivity and the heat flux sensor as described by Astarita and Carlomagno [25], is not mandatory. The current work will partly use the raw infrared signal in arbitrary camera units ('counts') as synonym for the surface temperature. In recent years IRT has become a standard tool for the detection of static transition positions. It was applied in wind tunnel testing of airfoils [26,27] or fixed-wing models [28], supersonic transition measurements [29,30], and in-flight measurements [31,32]. The current paper reviews recent developments of unsteady infrared thermography applied to moving boundary layer transition measurements in unsteady inflow conditions. In particular, the development of the differential infrared thermography (DIT) at the German Aerospace Center (DLR) in Götingen is covered in chapter 4 and illustrated with examples and applications in the field of helicopter rotor aerodynamics. The background to these activities is given by a summary of other unsteady but non-infrared methods in chapter 2, and a summary of steady infrared thermography in chapter 3 which is the basis for unsteady measurements. Detailed considerations of infrared radiation physics or camera technologies, see [25,33], or infrared applications on other aerodynamic topics, such as heat flux measurements for super-and hypersonic applications [34], are beyond the current scope.

Hot-film sensors
The hot-film is a flush-mounted surface sensor that uses the principle of constant temperature anemometry similar to the well-known hot-wires. It is a standard method for measuring the wall shear stress over the Reynolds analogy for heat transfer, also see equation (1), in a boundary layer [35,36]. It can be used to detect dynamically moving boundary layer transition [37], but can also distinguish between transition, stagnation point motion, flow separation, and shock formation [38]. The state of the literature for transition measurements on static airfoils with non-moving transition is particularly good, see for example [39]. It should be noted that much of this literature refers to the unsteady turbulent structures within the BL, see for example [40], rather than a dynamically moving boundary layer position as experienced by the pitching airfoil or the helicopter rotor in forward flight.
The pitching airfoil has been investigated for low-  [46]. Hotfilm sensors have also been extensively used for the investigation of dynamically moving BL transition on laminar airfoils [47,48], particularly in relation to aeroelastic flutter. Lorber and Carta [49] did extensive investigations of the BL transition and flow separation on a pitching airfoil, but the transition investigations involved only two sensors due to the low chordwise resolution. For the rotating system, Sémézis [50] performed a hot-film investigation on the 7AD rotor and investigations within the GOAHEAD project [8]. The data was also used for CFD validation [6]. A similar BL transition measurement using hot-films on an instrumented full-scale wind turbine blade was done by Schaffarczyk et al [51].
The evaluation of hot-film data presents several difficulties, which have increasingly been identified and at least partially solved in recent years. The first major problem is that for a dynamically moving BL transition, the transition intermittency is no longer seen as a transient stepping function. For a stationary BL transition region, a series of turbulent spots is generated and propagates past each sensor, and the fraction of time in which the sensor is in the laminar or turbulent state is the experimentally defined intermittency. For the moving boundary layer, an increased RMS signal is present, but the intermittency signal never appears, leaving the transition position to be assessed through the soft step attained between laminar and turbulent flow [52]. The step assessment also lends itself to fully automated data processing (through data skew), which solves the second problem of dealing with the large amount of data required to include the sensor cut-off frequencies of around 60 kHz [37,52].
An additional problem of hot-film sensors is that they involve an intrusion into the flow. Even if the sensor carrier foil is carefully kept to the same contour as the airfoil, the different surface roughness can result in a difference in the transition position, see figure 2. For the same reason, the edges of the foil and the localized heating spots of the sensors may cause disturbances. These influences can be reduced through good design and careful construction, but they can never be completely avoided. Similarly, Mertens et al [53], noted that the regions of a pitching airfoil with pressure sensors had measurably different boundary layer transition for some test cases, either due to the local surface heating, or the pressure taps.
For impulsive, particularly hypersonic test facilities, the hot-film gauges can be operated in an unheated mode, and are called thin-film gauges, see for example [54]. In this case the 'cold' resistance of the sensor is related to the heat transfer from the wall. The gauges have reaction times of the order of microseconds, but to the authors' knowledge were not yet used for dynamically moving boundary layer transition. An alternative to hot-films is to place hot-wires in the BL flow, and indeed these offer advantages for some flow topologies, especially for wake investigations [55] or for a rotating axisymmetric cone [56].

Pressure sensors and microphones
The σC P -method analyzes pressure signals from fastresponse surface pressure transducers [57]. If the pressure sensor signals are evaluated for an airfoil's periodic pitch motion, then the differences between different cycles will be low in the laminar flow regions, medium in turbulent regions, and a maximum near the transition position. This is because slight differences in the wind tunnel operating conditions or the pitch mechanism result in small cycle-to-cycle variations of the moving transition position. Hence, in a certain phase of the cycle, the transition position is in front of a given pressure sensor for some cycles but behind the sensor for the other cycles. This increases the cycle-to-cycle standard deviation of the pressure, σC p , since the transition affects the BL displacement thickness. A sample analysis of a pressure transducer mounted to the suction surface of a rotor blade at 19% chord is shown in figure 3. Due to the chosen cyclic pitch amplitude, both the BL transition and the BL separation inducing dynamic stall leave a footprint in the standard deviation of C p .
The σC p -technique does not require a very high temporal resolution of the pressure sensor as long as the overall aerodynamics is properly captured. This is due to the fact that no boundary layer modes are identified, but only the motion of the transition position which is a broadband signal. The method has been demonstrated for pitching airfoils at both high and low inflow velocities [57,59], and for pitching finite wings in both swept and unswept configurations [60,61]. An application to a rotor with cyclic pitch is also possible [62,63] but the sensor integration requires additional effort, for example by means of a telemetry system. Gao et al [64] and Wei et al [65] proposed an approach which is similar to the σC p -idea but was developed independently. In their case, it was shown that using a sliding window analysis rather than the cycle-to-cycle deviation yields transition results with a comparable quality. Another evaluation strategy in literature is the M-TERA intermittency approach. It was originally formulated for velocity fluctuations [66], but recently adapted to surface pressure fluctuations and transient inflow conditions [17]. The method must be seen as requiring additional validation.
Microphones with a much broader frequency response than the subsurface pressure sensors can also be used to detect BL transition, see [67,68]. Microphones allow an analysis of boundary layer frequencies or modes of interest in addition to  the mere transition location, but with additional difficulties in analyzing the acoustics.

Temperature-sensitive paint
Temperature-sensitive paint (TSP) is used in many of the same situations where infrared thermography can also be applied [69]. Under cryogenic conditions, TSP proved to be beneficial [70] in contrast to IRT due to the challenges of capturing the reduced radiated energy in the infrared spectrum [71]. The surface preparation with the delicate paint layer requires additional effort, and it offers the disadvantage that the surface contour may be altered. The surface roughness of the TSP can be very low if polished [72]. The data acquisition using optical cameras allows a higher resolution compared to infrared cameras, which enables a precise image dewarping and a tracking of small-scale aerodynamic features. TSP can be used for very short time exposures, see [73,74]. The use of difference-image techniques allows the use of TSP for the detection of dynamically moving BL transition, as applied on a slowly moving model during a pitch sweep test [75], but yet to be demonstrated on dynamically oscillating models.

Direct skin friction measurements
Direct skin friction measurements provide the most desirable data for any boundary layer problem. A possible approach is using sunk piezoelectric balances on which a small surface ('movable wall element') is mounted. The acquisition of a moving boundary layer transition seems feasible, since small-floating mass, fast-response gauges are possible [76]. However the practical difficulty of achieving the precise tolerances and low noise for this measurement mean that not much data is available [77], and have not yet been used for moving boundary layer transition. The sensors also have a temperature and acceleration sensitivity which must be compensated. Another method of skin friction measurement is given by shear-sensitive liquid crystals [78], but these have also not yet been used for dynamically moving boundary layer transition. fast-response infrared techniques to be discussed in chapter 4 are already included for the sake of completeness.

Infrared thermography (IRT) for steady transition positions
In comparison to other methods IRT enables convenient and two-dimensional measurements of the BL transition location. Therefore, IRT is particularly suitable for full-scale in-flight measurements, as demonstrated by Horstmann et al [79] or Crawford et al [31, 32] on general aviation aircraft. With a view to unsteady measurements in the following chapter, this section concentrates on applications with a stationary transition location due to steady inflow but fast-moving, particularly rotating surfaces. The basic IRT setup consists of an infrared camera, a strategy to achieve a fluid-surface temperature difference as required by equation (1), and, if needed, a surface preparation. The temperature difference can be achieved through: • Surface heating by external radiation flux sources: general purpose spotlights [63,80] The required temperature difference is often reported to be on the order of several Kelvin but it depends on the setup and the resulting signal-to-noise ratio, and no general recommendations can be made. The transition position itself may be affected if the heating is too strong, see Joseph et al [26] and Costantini et al [87] for further recommendations.
An insulating and dull coating can be applied to the surface in order to reduce the thermal conductivity and to maximize the infrared emissivity [25], which is mandatory for metallic surfaces but optional for carbon fiber reinforced plastics. Fiducial markers with differing radiation properties, for example silver conductive paint, can be used to register and dewarp the infrared images.
The inherent motion blur in the infrared images must be considered if a fixed ground-based camera is used to observe fast-moving rotor blades. To the authors' knowledge, no infrared camera placement within the rotating frame has yet been realized. Hub-mounted rotating cameras for the visible light spectrum have been demonstrated [88,89], requiring a miniaturized camera size, a high tolerance towards centrifugal forces, and a high resolution to perform image dewarp under oblique viewing angles.
Current infrared cameras are based on either thermal detectors, which sense the incident energy flux, or quantum detectors, which absorb incoming photons [25,33]. The former principle can, for example, be realized using mircobolometer arrays, which is the cost-effective standard choice for many cameras applied to steady-state BL measurements. The required image exposure times are in the range of several milliseconds [90] and, therefore, unsuitable to 'freeze' the motion of rotor blades. Quantum detectors are used in high-speed infrared cameras, featuring both a high repetition rate (f > 100 Hz) and small image exposure times (∆t < 100µs). The sensor resolution is currently between about 0.2 Mpx and 0.5 Mpx, which is at least one order of magnitude smaller than today's visible-light cameras. The signal-to-noise ratio of the infrared images depends on the surface emissions, the camera exposure time, and its noise-equivalent temperature difference (NETD), which is typically below 50 mK. The majority of the studies discussed in the following sections were conducted using either mercury cadmium telluride (MCT) or strained layer superlattice (SLS) sensors, both sensitive in the long-wave infrared band (LWIR), or indium antimonide (InSb) sensors, sensitive in the mid-wave infrared band (MWIR). In practice LWIR can be advantageous over MWIR since carbon fiber surfaces are opaque in the long-wave regime and require no additional surface treatment.
Reichstein et al [68] successfully performed BL transition measurements on a 2 MW-wind turbine spinning at up to 9 rpm, combining ground-based infrared cameras of both MCT and InSb types with blade-mounted microphones. The viewing distances of the cameras were 80 m and 140 m, with image exposure times of about 100 µs. It is noted that many large-scale measurements are limited by the scarce availability and high costs of telescopic lenses for infrared cameras. Reichstein et al only observed small laminar flow lengths on the blade's suction side during nominal operation conditions of the wind turbine, but larger laminar lengths occurred during the transient spin-up of the rotor due to lower Reynolds numbers. A similar investigation was conducted by Dollinger et al [91]. Other wind turbine-related research focusing on infrared on-site monitoring of structural blade defects found BL transition wedges due to leading edge erosion as a by-product [92].
The rotational frequency of rotors or propellers for aircraft applications is large in comparison to wind turbines, particularly when investigating sub-scale models. The corresponding motion blur can be reduced or avoided by optical blade tracking, as an additional measure complementing short-exposure cameras. Rotating mirrors are suitable for this purpose if the rotor hub position is stationary in a laboratory or wind-tunnel environment. Several options for different mirror setups were discussed by Raffel and Heineck [93]. The easiest choice is to place the mirror axis collinear to the rotor axis, as applied by Overmeyer et al [94]. The image now follows the rotor blade if the mirror frequency, f mirror , is half the rotor frequency, f rotor . An off-axis configuration as shown in figure 4 can be useful when the mirror-camera-assembly may not obstruct the rotor shaft and the rotor inflow or outflow.
The infrared images are derotated over the entire blade span if the mirror axis intersects the rotor axis at the hub including an angle β, and the frequencies of mirror and rotor are related by: which is a generalization of the on-axis case (β = 0). The chosen β should not be too large in order to minimize the image distortion due to an oblique viewing angle. Additionally, the periodicity between mirror and rotor must be considered. For example, β = arccos 2 . This means that repetitive images of the blade at the same azimuth angle and with the same viewing geometry can be taken every third revolution of the rotor corresponding to every second revolution of the mirror. This choice is a good compromise between the different requirements, and was for example chosen by Weiss et al [63]. Heineck et al [95] applied IRT and mirror-based derotation to the lower side of a two-bladed rotor with a radius of R = 0.9 m, spinning at f = 15 Hz in the hover chamber at NASA Ames. The large duration of image exposure, 4.5 ms, covered an azimuthal blade motion of ∆Ψ = 24 • , but the rotating mirror facilitated sharp infrared images, see figure 5. The blade was radiation-heated by a 1 kW-lamp in the lower image. Hence, the turbulent BL in the rear section of the blade is colder due to the stronger heat convection, and it appears in darker coloring. Fiducial markers used for image registration were placed in regular intervals near the leading edge, resulting in turbulent wedges. The interpretation of the infrared images is identical to static airfoil tests since the aerodynamics is steady in the rotating frame. The upper image is without heating. The transition-related coloring is inverted in comparison to the lower image, since the blade was slightly colder than the heating-up environment, but the pattern has a much lower contrast. This basic setup was further improved in subsequent measurement campaigns, but rotating mirrors were partly omitted, since newer camera generations offered an improved signalto-noise ratio at small exposure times. It is noted that measurements of unsteady aerodynamics still benefit from mirror derotation, as shown in the next chapter. Setups with two IR cameras examining both upper and lower blade surfaces were applied by Richter et al [85] during ground tests of a prototype full-scale helicopter rotor, and Overmeyer and Martin [83] on a Mach-scaled model helicopter rotor.
Richter and Schülein [82] measured the BL transition on both large-scale model rotors operated on whirl towers, and full-scale helicopter rotors during ground-run and hovering test cases. The infrared camera was mounted obliquely above the rotor plane viewing the blade's upper surface, see figure 6. The image exposure time of down to 20 µs was small enough to avoid motion blur while ensuring a sufficient Figure 6. Infrared images of a ground-run test of an Airbus EC135 rotor at 75% of the nominal rotor frequency, Richter and Schülein [82].
signal-to-noise ratio to acquire the steady BL transition position. A large number of turbulent wedges is noted due to contamination or surface defects at the leading edge.

Principle, pitching airfoil measurements, and optimization
The effects of unsteady inflow conditions on infrared BL transition measurements of a wetted surface can be divided into three categories: (i) Inviscid aerodynamics govern the time-dependency between the inflow conditions and the surface pressure distribution. For example considering pitch-oscillating airfoils, Theodorsen's theory [96] predicts a hysteresis between the pitch angle and the lift force due to differential velocities induced by shed vorticity in the wake. (ii) Viscous effects may introduce an additional hysteresis between an airfoil's pressure distribution and the boundary layer's transition position. This effect was shown by Richter et al [45] for a pitching airfoil, and confirmed by Gardner et al [61]. (iii) The thermal responsiveness, comprising the thermal diffusivity and the specific heat, connects the boundary layer's instantaneous convective heat transfer with the surface temperature as acquired by an infrared camera.
The 'true' aerodynamic transition position only includes effects (i) and (ii), whereas effect (iii) is an error which must be eliminated or reduced by the infrared measurement technique.
When the aerodynamic unsteadiness is sufficiently slower than the thermal responsiveness, steady-state evaluation methods can still be applied. This was shown by Szewczyk et al [97], observing the BL transition on a glider wing during a stall-and-recovery maneuver lasting about eight seconds. Simon et al [81] discuss the frequency response of infrared transition measurements for several surface materials and heat sources, with response times in the range of several seconds to minutes.
A possible evaluation strategy to overcome thermal hysteresis effects (iii) is differential infrared thermography (DIT), which was proposed by Raffel and Merz [98] and Raffel et al [99]. The DIT principle is introduced in figure 7, with a more detailed description following in the next sections. The infrared images, left and center, were taken successively during the pitch up-motion of a rotor blade. The aerodynamicrelated features are gradual variations of the infrared intensity in chordwise direction, and a prominent turbulent wedge caused by a transition dot. The differential image in figure 7, right, clearly reveals the transition moving forward between both time instants, as seen by a dark band representing decreasing temperatures due to increased cooling. The spanwise variation of the transition is due to three-dimensional effects close to the blade tip.
The difference between steady and unsteady infrared measurements can be seen in figure 8. Figure 8(a), top, shows the chordwise infrared signal of an airfoil's quasi twodimensional upper surface at static pitch angles, α, of 4.0 • and 4.5 • . The airfoil was radiation-heated so that the surface temperature, T w , was slightly above flow temperature. The transition region is located at about 0.25 < x/c < 0.35, where the surface temperature strongly decreases due to an increasing convective heat transfer. The steepest gradient dT/dx represents 50% intermittence [80]. Subtracting both static temperature distributions results in the differential signal ∆T shown in figure 8(a), bottom. The transition moves forward with increasing α, creating a negative peak at x tr ≈ 0.28 which can be attributed to 50% intermittence at the mean pitch of (4.0 • + 4.5 • )/2 = 4.25 • . Interpreting ∆T is unambiguous, since secondary effects related to uneven heating, to the external flow, or to inhomogeneities in the airfoil's material composition are canceled out. However, it was observed that reflections from the environment of the model must be carefully avoided, since the reflection's position on the model surface may shift during pitch angle variations, creating apparent and erroneous differential signals. Reflections can be prevented by a careful selection of the camera position and by using anti-reflective background material such as cloth curtains.
Infrared measurements at the same pitch angles but taken during a sinusoidal pitch motion are shown in figure 8(b). The thermal inertia of the model's CFRP surface is not negligible compared to the pitch frequency of 2 Hz. Hence, the distributions at α = 4.0 • and α = 4.5 • are very similar, and the steepest temperature gradient does no longer coincide with the instantaneous transition position. A similar observation is reported by Ikami et al [100] using temperature sensitive paint on a pitching airfoil. However, the peak of the differential signal ∆T is still a valid indicator for x tr , even though the peak level is about five times smaller compared to the static measurement.
Since the unsteady measurements were taken during the pitch upstroke, the transition position of x tr ≈ 0.295 is delayed in downstream direction due to hysteresis in comparison to the static value of x tr ≈ 0.28. It is noted that early DIT publications [99] proposed to not only interpret the DIT peak position but also the width of the peak, connecting the peak's base points to the start and end of the intermittency region. However, it is difficult to identify the base points in the presence of random measurement noise, and aerothermal simulations of pitching airfoils [101] later disproved the concept.
Hoesslin et al [102] proposed a concept similar to DIT with a view to turbomachinery applications. In their case, the surface temperature's decline rate after an impulsive flash is measured using high-repetition rate infrared images, which also visualizes the differing heat convection before and after BL transition. The concept should also enable short-term measurements, but so far, only steady airfoil results were published.
The basic principle of DIT was validated and refined in several measurement campaigns for quasi two-dimensional and pitching airfoils. Richter et al [46] compared DIT to hotfilm data and to the σC p -method [103] based on fast-response surface pressure transducers. The results cover the DSA-9A helicopter airfoil tested in the Transonic Wind Tunnel Göttingen, see figure 9(a), at M = 0.3, Re = 1.8 · 10 6 and different pitch frequencies up to 8.8 Hz. The infrared camera was a FLIR SC 7750-L LWIR MCT, sensitive in the spectral range of 8.0 − 9.4 µm, and combined with a 50 mm focal lengthlens. The native sensor size was cropped to 640 × 310 px, enabling an acquisition frequency of 190 Hz at an integration time of 100 µs. Images of the model's untreated CFRP surface were taken through a germanium window in the tunnel's side wall, see figure 9(b). The hot-film sensors are visible as bright spots and can be used for image registration, hence, no fiducial markers were needed. The model was heated with a 2 kW-spotlight mounted outside of the opposite side wall, and differential images of subsequent time steps were calculated. The DIT signal was averaged in spanwise direction, and the peaks were detected similar to figure 8(b). Figure 9(c) shows the DIT transition data (green dots) as a function of the phase t/T for a sinusoidal pitch motion (dashed black line) taken during multiple pitch cycles. Each data point represents the location x tr of a differential peak, with the peak being either positive or negative depending on whether the transition moves forward or backward. During motion reversal, see the gray areas in figure 9(c), the DIT results are unreliable due to a diminishing signal strength. This problem will treated in the next sections in more detail. Post-processing can be applied to remove outlier and to calculate the phaseaveraged transition location. The three experimental methods yield similar results, see figure 9(d), with deviations below 10% of the chord length for this setup. The split between upand downstroke due to hysteresis effects is largest for DIT, which is a result of the additional measurement lag (iii) due to the surface's thermal responsiveness. The other methods are supposed to cover only the 'true' aerodynamics (i) and (ii).
Wolf et al [80] revisited the sinusoidally pitching airfoil model of Richter et al [46] in the low-speed '1-meter' wind tunnel at DLR Göttingen, using the same MCT infrared camera and a similar radiation heating with a flux rate in the range of 500 − 1500 W/m 2 over the suction surface. The inflow Mach and Reynolds numbers were lowered to M = 0.14 and Re = 1 · 10 6 , increasing the BL's laminar length at a given lift coefficient. The low-speed tunnel enabled cost-effective parameter studies and optimizations of the DIT method. A sketch the setup is in figure 10, which can be taken as a blueprint for similar pitching-airfoil studies [59,98,99].  The main focus of Wolf et al's investigation was set on the influence of the separation distance between two infrared images to be subtracted, which can be expressed as a time difference ∆t, or, assuming a periodic pitching motion with frequency f, as a phase difference ∆tf. It was shown that it is favorable to detune the acquisition frequency of the infrared camera from the aerodynamic periodicity f with its multiples. Now combining the infrared images of a large number of acquired pitch cycles provides a high resolution of the phase angle, and allows to optimize the phase difference ∆tf. In this example, a camera frequency of 99.98 Hz yields a phase resolution of ∆tf = 2 · 10 −4 . Since the transition-related peaks of ∆T are usually well below 1 K, this approach requires removing the long-term temperature drift, which even in a laboratory environment cannot be avoided, and which may vary with the spatial coordinates due to the model's internal structure or inhomogeneous heating. This is shown in figure 11, in which the black line is a sample unfiltered DIT signal as a function of the chordwise coordinate. The underlying phase difference is small, ∆tf = 0.05 corresponding to 5% of the pitch period, but both images were taken from different pitch cycles with a wall-clock difference of about 40 s. The red line is the high-pass filtered DIT signal, using a cut-off time twice the length of the pitching period. The transition-related peak at x/c ≈ 0.15 (green arrow) is visible in both distributions, but it is corrupted and slightly biased by temperature drift in case of the unfiltered signal. In contrast, the filtered signal shows a clear peak in the transition region and is nearly zero elsewhere.
The airfoil's DIT transition analysis for sinusoidal pitch oscillations with an amplitude of 7 • and a reduced frequency of k = πfc/U ∞ = 0.075 (f = 4 Hz) is shown in figure 12. The green and blue dots correspond to separations of ∆tf = 0.005 and ∆tf = 0.05, respectively. The smaller separation captures the general motion of the transition well, as seen in figure 12(a) (top). The transition position on the upper surface is close to the leading edge at high pitch angles around tf = 0.5 but close to the trailing edge at low pitch angles around tf = 0 and tf = 1. It is noted that the shape of the transition motion is not sinusoidal, since the relation between α and x tr is not linear and depends on the airfoil shape. However, the random scatter of the small separation ∆tf is large since the subtracted infrared images are very close, and the resulting DIT peak values shown in figure 12(b) are barely above the camera's noise level. The ratio between peak values and underlying measurement noise is crucial for the achievable spatial and temporal resolution of the DIT analysis. The random scatter is reduced for large DIT image separation, see figure 12(a) (bottom). This corresponds to an increased peak signal strength as in figure  12(b), with the negative or positive sign of the peak depending on whether the motion is towards the leading or trailing edge.  Despite the large separation, the differential signal diminishes towards the frontmost and backmost positions where the movement of the transition is slower. Particularly at tf = 0.5, the same bifurcation as in figure 9(c) is observed. This bifurcation can be attributed to a DIT double-peak structure, see the red detail in figure 12(a) (bottom), with a coexistence of both Figure 11. DIT temperature drift correction using a high-pass filter [80].
positive and negative peaks. Assuming that the peak search algorithm detects the maximum peak value regardless of its sign, the results will randomly switch between both states, creating the bifurcation. A detailed analysis in the next section will show that the true transition position is in between. At tf = 0.2, the large separation of ∆tf = 0.05 produces another gap in the transition data. This can be attributed to a broad negative-negative DIT double-peak structure, see the black detail in figure 12(a) (bottom). At this chordwise position the transition motion is fast and the intermittency regions in both infrared images no longer overlap, hence, the single DIT peak starts to split up into two separate peaks. This raises the question of the optimal infrared image separation ∆tf, with figure 13 showing the influence of ∆tf on the parasitic measurement hysteresis (iii). This hysteresis is defined as the pitch difference, ∆α, between the up-and downstroke for a given transition position. The true aerodynamic hysteresis as seen by the σC p -method was subtracted, so ∆α = 0 means an error-free DIT measurement. The hysteresis depends on multiple factors such as the pitch motion, the airfoil's geometry and surface material, the infrared camera's signal-to-noise ratio, etc. For a variation of the transition position, figure 13(a), and a variation of the reduced pitch frequency k, figure 13(b), three universal effects can be identified: In region a) the measurement hysteresis decreases with decreasing phase separation, since the subtracted infrared signals successively converge on the instantaneous state of the flow. The minimum measurement hysteresis, limited by the thermal responsiveness of the airfoil's surface, is reached in region b). A further reduction of the separation in region c) yields an increasing scatter and measurement uncertainty since the diminishing temperature difference between to images approach the camera's noise limit.
Gardner et al [101] conducted a 2D URANS simulation of Richter et al's [46] and Wolf et al's [80] experiment, including aerodynamic hysteresis effects. The predicted instantaneous skin friction coefficient C f was coupled to a 1D simulation of the unsteady heat transfer in the airfoil's wall-normal direction, allowing the evaluation of synthetic temperature distributions and derived DIT results. The study generally confirmed the validity of the DIT method, but also provided new insight into problems connected to the thermal responsiveness. Figure 14 shows the predicted chordwise differential temperature distributions, ∆T, before and after the reversal of the pitch motion at the minimum pitch angle, corresponding to a phase of ft = 0.
The sign of the peak expectedly changes from positive to negative. When the motion of the transition stops in its most rearward position at about x/c = 0.55, the ∆T-signal should be zero for a surface material with infinite responsiveness and no thermal lag. When introducing a finite thermal responsiveness, both positive and negative peaks are coexistent, which is particularly visible for the dark blue line at ft = 0.021. This simulation result has a strong qualitative resemblance to the experimental differential signal shown in the red detail in figure 12(a) (bottom). The 'double-peak'-structure yields systematic errors, which explain the bifurcation of the DIT data points observed in the gray areas of figures 9(c) and 12(a) (bottom). The true aerodynamic transition positions marked by dots in figure 14 are between the positive and negative peaks, and they do not coincide with the detected 'phantom' peaks marked by the dashed black locus curve.
Gardner et al [101] further investigated the effects of different surface materials, assuming that the DIT measurements benefit from both a low thermal diffusivity and a low specific heat. The results shown in figure 15 are referenced to the epoxy matrix material used for the CFRP model in Richter et al [46]. Even the best insulators, expanded polystyrene and cork, offer only moderate reductions in terms of DIT thermal delay representing the hysteresis (top graph).  On the other hand, the peak strength of ∆T can be significantly increased by a factor of 10 and more (bottom graph). This result motivates the application of surface coatings, as it was also shown that the unsteady temperature variations are restricted to a very thin surface layer with a thickness below 0.5 mm. On the other hand, the coating's application, its mechanical strength, and its surface roughness have to be considered.
Based on these observations and using the same experimental setup [80] Mertens et al [53,104] proposed that the image separation should be optimized with respect to a quality indicator which accounts for data scattering. This approach is particularly suitable for sinusoidal pitch motions, in which the speed of the transition motion and the corresponding differential infrared signal strongly varies over the pitch cycle.
Mertens et al [53] also proposed local infrared thermography (LIT) as a variant of DIT which adds further insight to the thermal responsiveness. A high phase resolution of the infrared images allows the analysis of the temperature T at a given position x/c over the entire pitch cycle. This is shown in figure 16, left, in which the temperature signal at x/c = 0.31 was low-pass filtered with sliding average window sized ∆tf = 0.02 (2% of the pitch cycle) to reduce noise. The convective cooling effect of the turbulent BL is clearly seen by means of a decreasing temperature between about tf = 0.24 and tf = 0.78. The transition passes slightly after the temperature peaks, see the square symbols in figure 16 (left), but before the peaks of the numerically derived temperature gradient dT/d(ft), see the circular symbols in figure 16 (right). It is noted that analyzing this local temperature gradient is equivalent to the DIT approach with a very small image separation ∆tf. LIT reduces the data scatter noted in the DIT results of figure 13 for ∆tf → 0 by applying the aforementioned sliding window filter.
Using optimized evaluation procedures and infrared data with a good signal-to-noise ratio, it is possible to extend the differential analysis to produce two-dimensional results down to the resolution of a single pixel [53,104]. This particularly avoids taking the spanwise average of airfoils as in [46,80]. Figure 17 shows the two-dimensional LIT analysis during the upstroke of a sinusoidal pitch motion, considering the airfoil's suction side with the flow from left to right. The detected transition area is marked in blue color and superimposed onto the original infrared image data. For an increasing pitch angle, left to right, the transition moves towards the leading edge. The smallest pitch angle, figure 17 (left), shows three-dimensional effects by means of a slightly tilted BL transition line interrupted by three transition wedges marked by green triangles. The top and bottom wedges result from an increased surface roughness due to the painted fiducial markers appearing as black dots in the images. The central wedge is caused by surface pressure taps which are not seen by the infrared camera. In this chordwise area small perturbations have a strong influence on the transition position due to the flat pressure distribution. Towards the leading edge, the transition is mainly caused by the strong pressure gradient dp/dx downstream of the suction peak. Hence, the influence of surface defects is smaller, and the transition line becomes more two-dimensional as in figure  17 (right).

Rotor measurements
Measurements of unsteady BL transition locations on rotating or fast-moving aerodynamic surfaces is one of the most challenging tasks for infrared thermography, since most of the difficulties described in the earlier sections add up.
The first successful application of DIT to a subscale rotor was demonstrated at the rotor test stand Göttingen (RTG) by Raffel et al [99], providing data on moving BL transition for unsteady conditions complementing earlier steady-state test cases [74,105]. The RTG, see figure 18, enables an azimuth-varying blade pitch through a swashplate similar to helicopters, but it has a horizontal axis and does not simulate edge-wise inflow as in helicopter forward flight. The RTG is optimized for the synchronized application of blade-mounted sensors, such as pressure transducers, and external optical camera systems.
A similar test with an improved setup and rotor frequencies up to 23.6 Hz was conducted by Weiss et al [63]. The application of a high-speed infrared camera and a rotating mirror allowed the acquisition of both DIT measurement images to be subtracted during a single rotor revolution with a constant azimuthal separation of ∆Ψ = 18 • , corresponding to a phase separation of ∆tf = 0.05 and an image frequency of 472 Hz. The camera was a FLIR X8500sc SLS LWIR, which is sensitive in the spectral range of 7.5 − 12 µm. The image integration time was set to 57 µs, which is lower in comparison to the MCT camera used in [80,99] due to a superior signal-to-noise ratio (NETD: 20 mK versus 35 mK). The spatial resolution was 2 mm/px using a 50 mm focal length-lens. The radiative heat flux provided by halogen lamps was about 400 − 500 W/m 2 , which is on a similar level as in [80]. DIT transition results for the suction side of the rotor blade and at a radial station of r/R = 0.8 are shown as black dots in figure 19. The unsteady inflow conditions were produced by a cyclic swashplate setting with a pitch amplitude of about 6 • , and the data is similar to pitching-airfoil data shown in the preceding section. A comparison to steady-state IRT results at the corresponding constant pitch angles, marked by blue squares   in figure 19, reveals the aerodynamic hysteresis by means of a shift towards larger phases tf, that is, towards the right side.
The combined results at the radial stations between r/R = 0.4 and 1.0 can be visualized in a polar coordinates, forming the transition map shown in figure 20 (left). The experimental data is compared to a numerical DLR TAU simulation of the experiment in figure 20, right. The simulation was set up based on the best practices described by Kaufmann et al [106], and the overall agreement of the BL transition position x tr is very good. The minimum and maximum pitch angles are at the top and bottom of the rotor plane maps, respectively. Note that the transition pattern, with light colors indicating x tr close to the trailing edge and dark colors indicating x tr close to the leading edge, is roughly symmetric but slightly rotated in counterclockwise direction representing aerodynamic hysteresis effects.
The model-scale measurements provide a rough estimate for the operating range of DIT using current infrared camera technology. Figure 21 shows the DIT thermal hysteresis (iii) as an excess pitch ∆Θ over the true aerodynamic hysteresis, corresponding to the definition used in figure 13. The data of references [46,63,80] is plotted as a function of the airfoil's or blade's pitch rate. For decreasing pitch rates below about 15 • /s, see the detail in the right figure, the thermal hysteresis monotonically decreases towards zero, representing an errorfree transition measurement for static conditions. For larger and technically relevant pitch rates the thermal hysteresis is bounded by about 1 • to 1.5 • , neglecting a single outlier at 2.3 • [63]. Nevertheless, the DIT principle was demonstrated even for very high pitch rates up to almost 900 • /s. It is noted that the stated thermal pitch hysteresis ∆Θ cannot be generally converted into a bias of the detected transition position, since the corresponding transition motion highly depends on the airfoil shape.
Overmeyer et al [94] conducted the first BL transition measurement of a trimmed rotor in forward flight conditions, applying DIT to the lower side of the three-bladed NASA PSP model rotor with a radius of 1.  Thermal hysteresis ∆Θ as function of the pitch rate dΘ/dt, data from Richter et al [46], Wolf et al [80], and Weiss et al [63], overview (left) and detail (right).
sinusoidal variation of the blade's inflow velocity and, therefore, the stagnation temperature as a function of the azimuth angle. On the advancing side of the rotor plane, see figure  22(a), the external flow is warmer than the blade, and the turbulent areas heat up faster than the laminar areas. The BL is almost fully turbulent in the inboard sections of the blade, whereas large laminar pockets at the tip are interrupted by turbulent wedges resulting from artificial transition dots. On the retreating side, see figure 22(b), the blade is cooled by the flow, and this effect is amplified in turbulent areas. The BL on the lower surface is fully laminar, again except for the forced turbulent wedges near the blade tip. Since both laminar and turbulent regions appear as coherent areas with nearly uniform temperature changes, the differential images resemble individual infrared images for steady-state flow conditions. In agreement with the aerothermal simulations and experimental findings by Gardner et al [107], the transition can therefore be detected by means of the first edge (gradient) seen from the trailing edge, which avoids differential signals due to suction peakrelated changes close to the leading edge. This is a deviation from DIT with continuous blade heating, in which only the moving laminar/turbulent interface appears as a differential peak signal. For Overmeyer et al's experiment an additional constant blade heating was disregarded, since a superposition with the inevitable periodic temperature change partly led to a decreased signal-to-noise ratio of the DIT images. Based on these findings Gardner et al [107] conducted in-flight infrared measurements of an Airbus EC 135 helicopter at a forward flight speed of 80 kn, taking images of the rotor's upper surface from a second chase helicopter as seen in figure 23. The distance between both helicopters was 50 − 100 m limited by safety considerations. The same high-speed SLS camera as in [63] was used. The frame rate was set to 294 Hz, corresponding to a phase difference of ∆tf = 0.022, or an image-to-image blade motion of ∆Ψ = 8 • .
The infrared images were acquired with a cropped sensor size of 768 × 800 px and an integration time of 50 µs, which sufficiently suppresses motion blur even without a rotating mirror. A telescopic lens with a focal length of f = 200 mm was applied to the camera. The maximum blade pitch rate due to cyclic swashplate inputs was estimated using trim calculations to about 150 • /s, which is well within the DIT range  [94] demonstrated in figure 21. However, no artificial blade heating was available, and the measurements solely relied on the azimuthal variation of the stagnation temperature. The blade surface temperature increases in radial direction and proportional to Ωr 2 , see figure 24 (left), reflecting the average stagnation temperature of the inflow. The variation of the surface temperature over the rotor plane's azimuth angle is small, but by tendency, the blade is heated on the advancing side (azimuth 0 • to 180 • ) and cooled on the retreating side (azimuth 180 • to 360 • ) due to the sinusoidal change of the stagnation temperature. The peak-to-peak difference between local stagnation temperate and average blade temperature is about 20 K in the midspan region, see the black line in figure 24 (right). The corresponding reaction of the instantaneous surface temperature is in the sub-Kelvin regime and phase-delayed due to the limited thermal responsiveness of the blade surface at the rotor frequency of f = 6.6 Hz. However, the infrared-measured data (red dots) generally follows the numerical prediction of the surface temperature (green line), but the variations approach the camera's noise limit.
Nevertheless, Gardner et al [107] successfully applied DIT, and identified a region between leading edge and about 20% chord which shows a differential signal on the advancing side of the rotor plane. Unfortunately, the DIT result could not be unambiguously connected to BL transition, since this area was also covered by an erosion tape. It was known from hover tests that the step between the tape's trailing edge and the blade surface triggers transition, but the DIT is also affected by the different thermal responses of both surface materials, with both transition and responsiveness effects adding up in differential images. Even though the general DIT principle was demonstrated, future in-flight measurements will benefit from rotor blades with a larger amount of laminar flow over the blade and a better optical resolution of the images, which in this study was limited by the minimum safety distance between both helicopters and the available lens optics for infrared cameras.

DIT-based stall detection
Some of the experiments in the preceding section showed as a by-product that DIT is also an efficient tool to detect BL separation. Airfoil testing by Gartenberg et al [108] already noted that flow separation affects the time-averaged convective heat transfer, with a low convection in laminar separation bubbles due to recirculating flow, but a high convection in fully separated regions due to vortex shedding. This enables an evaluation similar to steady-state IRT, as also shown later by Montelpare and Ricci [109]. The separation detection can now be extended to fast-moving surfaces and short time intervals using modern infrared cameras and the DIT method, which in this case exploits the unsteadiness of the flow separation. Figure 25(a) shows two differential infrared images of an airfoil's upper surface. Instantaneous images at different times but at the same pitch angle were subtracted. This yields no relevant difference apart from camera noise for attached flow (top) but strong variations for fully separated flow (bottom). Gardner et al [59] interpreted this pattern as the footprint of large-scale coherent and three-dimensional flow structures shed within the separated region. Therefore, flow separation can be detected by means of a high standard deviation within the DIT images, either in spatial or temporal direction. The spatial standard deviation, σDIT, within a window close to    [110] the leading edge of the airfoil in deep-stall pitch oscillations is shown as red line in figure 25(b). Two distinct DIT peaks at about t/T = 0.4 and t/T = 0.8 are related to flow separation and flow reattachment, as a comparison to the blue line of the surface pressure-based σC p -method shows. Apart from both σDIT-peaks there is a low level of thermal unsteadiness during attached conditions and a medium level of unsteadiness during fully separated flow. It is noted that a quantitative relation between σC p and σDIT is neither meaningful nor required for separation detection, hence, the quantities are shown in arbitrary units, eliminating the need for extensive sensor and camera calibrations.
Raffel et al [110] utilized the setup at the rotor test stand Göttingen, see figures 18 and 20, but increased the collective pitch setting so that dynamic stall was initiated during the downstroke motion of the blade. The stall map shown in figure 26 was acquired using the DIT principle and successfully compared to point-wise dynamic pressure transducers.

Conlusions
The main conclusions of recent developments in steady and unsteady infrared thermography for boundary layer (BL) transition detection can be summarized as follows: (i) Infrared thermography (IRT) is an easy-to-use standard technique measuring the spatial distribution of the nonmoving laminar-turbulent BL transition. It can be adapted to rotor applications with steady inflow conditions. (ii) Recent developments in high-speed, short-exposure infrared cameras with photon detectors permit the acquisition of instantaneous unsteady infrared images. (iii) The interpretation of instantaneous infrared images must consider the thermal responsiveness of the surface material, which is generally non-negligible in aerodynamic applications.
(iv) Differential techniques can be used in standardized image processing methods such as differential infrared thermography (DIT) to overcome or reduce the thermal history effects of the surface. (v) DIT was applied to several pitching-airfoil tests and rotor tests, showing a good agreement but an additional measurement-related lag in comparison to point-wise fast-response data provided by hot-films or pressure transducers. DIT is also a useful tool for an unambiguous interpretation of steady transition positions. (vi) The separation distance between the two subtracted infrared images is an important parameter of differential methods. The optimal DIT separation will depend on both aerodynamics and experimental setup. An aposteriori optimization is often possible under laboratory conditions when carefully choosing the acquisition frequency. For the current studies, a separation of 1% to 2% of the aerodynamic cycle provided good results and can be taken as a rule-of-thumb suggestion. (vii) Short-exposure infrared images can additionally be used to detect other aerodynamic phenomena such as flow separation. (viii) Unsteady infrared thermography is a very promising candidate for full-scale measurements, for example on helicopter main rotors or wind turbine blades during operation. First applications showed promising results, but further improvements in both camera technology and rotor blade performance are needed to validate the principle.