Pulsed thermography for composite NDT: the physics, the defect envelope, and the 3 to 5 mm depth limit

Pulsed thermography (PT), also called flash thermography, is an active infrared method for non-destructive testing of carbon fiber reinforced polymer (CFRP) composites. A short, high-energy heat pulse is deposited on one surface of a part, and an infrared camera records how that surface cools. The camera does not see the defect. It sees the surface temperature consequence of how a buried feature perturbs heat flow, and the inspection image is computed from the cooling curve afterward, by signal processing that is part of the measurement rather than an accessory to it.

The crucial properties to understand before reading any PT image: detection range in standard CFRP sits in the first 3 to 5 mm and is a rule of thumb rather than a material constant; tight vertical surface-breaking cracks are a documented blind spot for physical reasons that do not depend on the equipment; and the method excels at the defect families that interrupt through-thickness heat flow over a meaningful lateral cross-section, which covers most of what matters for delaminations, disbonds, and barely visible impact damage (BVID).

This reference walks through what the method actually measures, the physics that generates a defect image, what PT detects well and where it goes blind, the data-reduction pipeline that turns raw thermal video into an inspection image, the standards under which PT operates as a certified aerospace methodology, and where the field is moving.

What pulsed thermography actually measures#

PT is an active infrared method. A short heat pulse, usually from a xenon flash tube, is deposited on the inspection surface, and an IR camera records the evolving surface temperature field as the surface cools. NASA Langley describes flash infrared thermography as a large-area rapid-inspection technology for composite structures. Recent systematic reviews continue to classify active thermography as a preferred non-contact method for large-area and complex-geometry composite inspection.

The conceptual point worth holding onto is that the camera does not see the defect. Heat diffuses away from the flashed surface; if it encounters a region that impedes through-thickness conduction (a delamination, a disbond, a void, an air-filled flat-bottom hole, an inclusion), the surface above that region cools differently from adjacent sound material. Practical inspection terms formalize this as a thermal signal (the temperature difference between a defect pixel and a sound pixel) and a corresponding thermal contrast metric. Defect equals thermal-contrast evolution. That is the core mental model.

That model works best for planar or laterally extended features that interrupt heat flow across the laminate or across an interface. Consequently, PT has long been used for delaminations, disbonds, BVID, inclusions, porosity, voids, and face-sheet to core disbonds in sandwich structures. NASA Langley work on graphite-epoxy and honeycomb structures explicitly reports sensitivity to voids, inclusions, delaminations, porosity, impact damage, face-sheet delamination, and face-sheet to core disbond.

The fundamental limit follows directly from the physics. Thermography is a diffusion technique, not a wave-reflection technique like ultrasound and not a line-of-sight density-imaging technique like X-ray CT. Diffusion is forgiving of curvature and fast over large areas. It is also lossy: the defect signature attenuates strongly with depth.

The physics that create a defect image#

Transient heat conduction and the surface cooling law#

A standard industrial PT system discharges several thousand joules, typically 4,800 to 9,600 J, from xenon flash tubes inside a reflective hood over roughly 2 to 15 ms. The optical pulse deposits energy at the surface, and the temperature field evolves according to the Fourier heat equation. The flash is idealized for modeling as an instantaneous Dirac impulse uniformly absorbed across the front plane at time zero.

For a homogeneous, semi-infinite solid subjected to a uniform instantaneous surface heat impulse of energy density $Q$ (J/m²), the temperature field $T(x,t)$ obeys the one-dimensional Fourier equation:

$\frac{\partial^2 T}{\partial x^2} = \frac{1}{\alpha}\frac{\partial T}{\partial t}$

where $x$ is depth, $t$ is elapsed cooling time, and $\alpha$ is thermal diffusivity, the ratio of thermal conductivity $k$ to volumetric heat capacity $\rho c$. Solving under adiabatic surface conditions gives the surface temperature decay as a power law:

$\Delta T_{\text{surf}}(t) = \frac{Q}{e\sqrt{\pi t}}$

where $e = \sqrt{k\rho c}$ is the thermal effusivity. On a double-logarithmic scale this cooling path is a straight line of constant slope minus 0.5:

$\ln(\Delta T_{\text{surf}}(t)) = \ln\!\left(\frac{Q}{e\sqrt{\pi}}\right) - \tfrac{1}{2}\ln(t)$

This minus 0.5 log-log slope is the baseline reference for sound, defect-free material. Most modern processing methods are built on detecting deviations from it.

Thermal impedance and interface reflection#

When the propagating transient thermal disturbance reaches a subsurface discontinuity (an air-filled delamination, a dry fiber bundle, a foreign inclusion), it meets a sharp change in thermal impedance. Air has an extremely low thermal conductivity ($k_{\text{air}} \approx 0.026$ W/m·K) compared with epoxy ($k_{\text{epoxy}} \approx 0.2$ W/m·K) and carbon fibers ($k_{\text{carbon}} \approx 10$ to $100$ W/m·K, strongly orientation-dependent). The defect behaves as a thermal-resistance plane.

The disturbance partially reflects at this boundary, governed by a thermal reflection coefficient:

$R = \frac{e_1 - e_2}{e_1 + e_2}$

Because $e_2$ (air) is far smaller than $e_1$ (CFRP), $R$ approaches 1. Near-complete reflection. Heat is effectively trapped between the defect plane and the front surface, and the surface directly above the defect cools more slowly than surrounding sound material and departs from the minus 0.5 slope. The successive round-trip thermal paths within this layer (spanning $2d, 4d, 6d, \ldots$ for a defect at depth $d$) sustain the slower decay.

Time dependence and lateral diffusion#

Defect visibility is fundamentally time-dependent. Sound and defective regions each evolve with their own transient signature, and the useful observables include the temperature difference, the maximum temperature difference, the maximum contrast, and the times at which those maxima occur.

Early in time the response is approximately one-dimensional, dominated by through-thickness flow. As time increases, heat spreads laterally: at the lateral edges (tips) of a buried defect a high horizontal gradient develops, and trapped heat bleeds sideways into cooler surrounding material. This lateral conduction is the dominant mechanism that limits the peak amplitude of the thermal contrast and accelerates its later decay. Edges soften, and small deep defects become progressively harder to distinguish from background texture and non-uniform heating.

The rate of lateral dissipation scales with the defect aspect ratio $D/d$ (lateral diameter over depth). A low aspect ratio (a small, deep defect) is the worst case. Lateral diffusion dampens the surface signature before a visible contrast can accumulate.

Anisotropy and surface condition in CFRP#

Heat flow in real CFRP is multidimensional and strongly anisotropic. Carbon fibers conduct well along their longitudinal axes while the matrix is insulating, so in-plane conductivity (aligned with fiber orientation in each ply) substantially exceeds through-thickness conductivity. Global thermal diffusivity depends heavily on fiber volume fraction. Diffuse measurements show it can range from roughly $1.5\times10^{-7}$ to $2.2\times10^{-7}$ m²/s as fiber volume content shifts from about 13 percent to 28 percent. High in-plane diffusivity accelerates lateral dissipation, blurs spatial defect boundaries, and makes quantitative sizing more complex.

Surface condition matters too. NASA Langley's graphite-epoxy experiments showed that near-surface stitching patterns, fiber-related texture, emissivity variation, and camera vignetting can mask real subsurface flaws in raw frames. Background subtraction, averaging, and phase processing improved detectability substantially, but some material-property inhomogeneity from stitches could not be fully removed in post-processing.

The defect detection envelope#

What PT sees well#

The short answer is expansive near-surface or interfacial defects that present a high cross-section perpendicular to the through-thickness heat flux. Canonical win cases include interlaminar delaminations, disbonds, BVID with an underlying delamination field, skin-to-core disbonds in Nomex or foam-core sandwich panels, void-like features and porosity regions, and dry resin-starved zones. The technique is also effective at spotting foreign object debris (FOD) such as misplaced backing tape or protective film introduced during layup, and at characterizing void content because composite diffusivity is sensitive to both micro-void volume fraction and pore shape.

These capabilities span diverse aerospace material systems: carbon fiber with bismaleimide, epoxy, phenolic, or polyimide thermoset matrices, as well as high-temperature thermoplastics such as PEEK and PEKK. Reported applications range from ultra-thin carbon skins (e.g., TeXtreme 100 fabrics with ~0.1 mm nominal ply thickness bonded with structural adhesives such as HexBond EA9394) up to massive multi-layer carbon-carbon brake disks.

The 3 to 5 mm depth limit and why it is a rule of thumb#

For depth, the practical guideline is the first few millimeters, not all the way through the part. A widely cited working figure is a 3 to 5 mm reliable-detection regime in standard CFRP, with systematic reviews tabulating IR thermography for composites in roughly the 1 to 5 mm range.

The physical mechanism is the strong decay of the transient thermal disturbance with depth. The thermal diffusion length $\mu$ (the depth at which amplitude falls to about 37 percent of its surface value) is:

$\mu = \sqrt{\frac{\alpha}{\pi f}}$

where $f$ is the equivalent frequency of the transient response. The low transverse diffusivity of CFRP ($\alpha_{\text{transverse}} \approx 1.5\times10^{-7}$ to $4.5\times10^{-7}$ m²/s) acts as a low-pass filter, rapidly damping high-frequency components as they go deeper. Reaching a depth of about 5 mm and returning to the surface takes a long transit time (scaling as $t \propto d^2/\alpha$), during which lateral diffusion has ample opportunity to bleed off the trapped heat. The contrast of a defect deeper than ~5 mm is therefore almost entirely dissipated before it reaches the surface, sinking below camera noise.

Treat the 3 to 5 mm value as a rule of thumb rather than a material constant. Detectability depends on defect depth, defect diameter, defect thermal resistance, surface finish, non-uniform heating, camera sensitivity, observation time, and algorithm choice, all interacting. The classical NASA result reaching 5.3 mm was obtained on favorable coupons with significant processing, while broader reviews summarize practical ranges nearer the first few millimeters. There is no universal hard cutoff.

Deep, small, or thermally weak defects#

The first category PT handles poorly is the deep, small, thermally-weak defect. As depth increases, maximum contrast falls and time-to-peak rises. Reviews note limited penetration in thick composites and difficulty with defects in cores or inner resin layers; optical thermography has been shown to struggle with defects embedded deeper in sandwich cores and inner resin layers.

The tight vertical-crack blind spot#

The second category is the classic tight, vertical (transverse) crack blind spot. Uniform flash excitation drives a heat flux that is essentially perpendicular to the surface, downward, parallel to the plane of a vertical crack. A narrow surface-breaking crack running vertically into the part presents little projected cross-section to block this out-of-plane flux, so it offers almost no thermal resistance to the primary heat flow. Heat flows uninterrupted alongside the crack walls. Little or no surface contrast is created. Consequently perpendicular matrix microcracks, vertical fiber breaks, and tightly closed kissing bonds are effectively undetectable with standard PT.

This is a well-established physical conclusion. Conventional wide-area pulsed and lock-in methods are mainly sensitive to features that interrupt heat flow across the thickness or along the disturbance's propagation path. Li, Almond, and Rees noted bluntly that PT is not suitable for important surface-breaking cracks with micron-scale openings in metals, and later grating-thermography work generalized the reason: for a vertical crack the propagating signal is effectively parallel to the crack and therefore insensitive. They also showed that a tight crack becomes visible only when a localized heat source creates lateral flow the crack can perturb. This is precisely why specialized laser-spot or laser-line thermography or vibrothermography is used when the genuine target is a tight crack rather than a delamination-type flaw.

A fair CFRP-specific nuance: matrix cracking can be observed under certain loading or differential setups, but it is not the easiest defect family for ordinary one-sided flash inspection. NASA Langley fatigue-loading studies detected matrix cracking and delaminations using a combination of passive thermography, difference flash thermography, and controlled loading, underscoring that crack-like damage often needs a tailored excitation or loading condition rather than a single snapshot flash.

From raw thermal video to an inspection image#

A pulsed-thermography inspection is essentially a data-reduction pipeline. The flash and camera produce a thermal movie. The useful NDT image is computed afterward. The typical chain is: acquire raw 3D thermal video, normalize temperature, enhance (TSR, PCT, PPT, and contrast methods such as DAC and MDAC), interpret and quantify.

Raw frames at a single fixed time are often hard to interpret because stitch patterns, surface marks, emissivity variation, and non-uniform background dominate. In NASA's graphite-epoxy work, even straightforward time averaging and background subtraction were needed to make all five embedded defects localizable.

Thermographic Signal Reconstruction (TSR) is the de facto industry standard. Each pixel's normalized cooling path is transformed to a double-logarithmic scale and fit with a least-squares polynomial of degree $n$ (typically 6 to 9, with 7 or 8 often optimal for CFRP). The first and second logarithmic derivatives are then computed analytically, and they are largely insensitive to flash non-uniformity and local emissivity variation. For an ideal flat-bottom hole or backwall at depth $L$, the second derivative peaks at a characteristic time $t^* = L^2 / (\alpha_{ez}\pi)$, so locating the peak frame yields depth.

Principal Component Thermography (PCT) is a spatial-temporal SVD reduction. EOF1 typically captures the largest variance share (often greater than 90 percent), dominated by global illumination non-uniformity; EOF2 through EOF4 isolate the transient cooling dynamics and cleanly separate subsurface structure and defects. This concentrates subtle defect signatures into a few high-contrast low-noise synthetic images and removes the need to scan hundreds of raw frames.

Pulsed Phase Thermography (PPT) moves the response into the frequency domain via a pixel-wise DFT. Phase images resist uneven heating, surface reflections, local geometry variation, and emissivity fluctuation. The blind-frequency method correlates the frequency at which a defect's phase contrast disappears to its physical depth.

Modified Differential Absolute Contrast (MDAC) incorporates sample thickness and diffusivity through thermal-quadrupole theory in Laplace space, modeling the front-face temperature of a finite plate analytically. That prevents divergence of the sound-reference curve and extends valid contrast computation to much longer times than classical DAC's semi-infinite assumption permits.

The central lesson here is that no single algorithm always wins. The processing choice changes detectability, apparent defect boundaries, and practical usefulness. Post-processing is not an accessory to the physics. It is part of the measurement system.

Standardization and the aerospace lineage#

PT is standardized for aerospace use:

• ASTM E2582 (Standard Practice for Infrared Flash Thermography of Composite Panels and Repair Patches Used in Aerospace Applications) is the primary standard for aerospace Tier 1 suppliers and MRO facilities. It defines reference standards (flat-bottom-hole calibration blocks at specified depths), baseline system-performance thresholds, spatial-uniformity criteria, and signal-processing steps to size and locate delaminations, voids, inclusions, and fluid ingress.
• EN 16714 is the European framework for active thermographic procedures in three parts (general principles, equipment, terms and definitions), requiring written procedures approved by an ISO 9712 Level 3 thermographer.
• ISO 9712, EN 4179, and NAS 410 govern thermographic testing (TT) personnel certification: Level 1 (acquisition), Level 2 (interpretation and procedure drafting), Level 3 (method approval and training).

The transition of PT from a qualitative laboratory concept to a quantitative, certified engineering practice was driven substantially by research at NASA Langley Research Center. NASA scientists, notably William P. Winfree and Joseph N. Zalameda, pioneered modeling and inversion algorithms tailored for carbon fiber laminates. A key enabler was the Thermal Quadrupole method, which solves the 3D heat-diffusion equations in Laplace space using analytical transfer matrices rather than finite-element meshing, enabling quantitative inversion (surface thermal histories to physical maps of depth, area, and contact resistance) with at least an order-of-magnitude reduction in computation time.

NASA program work made the evolution visible. The X-33 launch-vehicle program is cited as a case where thermal NDT proved valuable because it scanned faster and more cheaply than ultrasonics, could be used in the factory area, avoided water-immersion issues, and even found defects missed by ultrasonics. By the 2010s and 2020s, NASA Langley was applying flash thermography to large composite cylinders, honeycomb structures, and on-orbit inspection concepts alongside large-area ultrasonics, CT, terahertz, and automated recognition software.

In manufacturing quality assurance, PT is frequently mounted on multi-axis robotic gantries that step-scan large carbon fiber structures (fuselage sections, rudders, elevators, wing skins). It is heavily used to screen core-to-skin adhesion in sandwich panels so adhesive voids are caught before assembly. In fleet maintenance, PT is a primary diagnostic for in-service aircraft; it maps BVID from bird strikes, runway debris, or ground-equipment collisions, flags water or hydraulic fluid ingress in honeycomb panels, and validates the bondline integrity of composite repair patches.

Where PT sits in the broader NDT portfolio#

PT is neither a niche curiosity nor a universal replacement for other NDT methods. In the reviewed composite-NDT literature, ultrasonics remains the most researched and applied family (about 45 percent of reviewed papers), phased-array ultrasonics about 20 percent, and infrared thermography about 25 percent. PT has strong mindshare and real industrial use; UT and PAUT still dominate where depth resolution and internal characterization are mission-critical.

Conventional pulse-echo ultrasound penetrates more than 50 mm in fully consolidated CFRP with excellent depth resolution but has a near-surface dead zone (the first ~0.5 mm) where PT is strongest. The two methods are complementary on depth coverage. Ultrasound requires liquid couplant and contact; PT is fully non-contact with large stand-off possible. X-ray CT gives full 3D internal visualization but with long scan times, high cost, demanding segmentation, and sample-size limits. Shearography is a full-field non-contact method that measures deformation gradients under load and locates spatial boundaries of skin-to-core disbonds but cannot determine defect depth.

The strongest current signal is integration. Multimodal studies fuse PAUT and PT images precisely because each compensates for the other's weak spots: PT covers area fast and non-contact, UT and PAUT supply the sharper through-thickness story. A 2024 study demonstrated complementary fusion of PAUT and pulsed-thermography images on a composite structure. The field is moving toward method combinations and multimodal-fusion architectures, not winner-take-all selection.

What this means#

PT is best understood as a fast transient heat-flow measurement whose strength is near-surface, laterally extended, thermally resistive damage, and whose limitations are set by diffusion physics. The 3 to 5 mm reliable detection regime in standard CFRP is a working guideline that depends on defect geometry, surface condition, camera noise, and processing choice. Tight vertical cracks are a physical blind spot, not an equipment shortfall.

For an inspection workflow, the right question is rarely "PT or UT?" It is "which method or method combination is appropriate to the defect family and geometry I am looking for, and what does the standard say about how that method should be deployed?" Presidio Composites operates pulsed thermography NDT and treats the method as one tool in a broader portfolio; the inspection report names the method used, the depth regime it covers, and where the defect family suggested a multimodal follow-up by laser-spot thermography, vibrothermography, fluorescent dye penetrant, or referral for PAUT.