Fiber-reinforced polymer composites are critical in aerospace, automotive, and wind energy applications, yet subsurface defects, including delamination, barely visible impact damage, and porosity, can reduce compressive residual strength by 40%–60% without surface indication. Infrared thermography provides non-contact, full-field inspection, but parameter specification at untested configurations remains reliant on experimental benchmarks, with no unified framework separating thermophysically fundamental detection limits from hardware-remediable constraints. This systematic review synthesizes 90 peer-reviewed studies (2012–2025) following PRISMA 2020 guidelines and derives a three-condition detectability envelope: a defect at depth
with lateral dimension
is detectable when the thermal diffusion length (
), the empirical aspect ratio (
,
), and the hardware energy (
) conditions are simultaneously satisfied. Conditions one and two are governed by material physics; only condition three is hardware-remediable. To the authors’ knowledge, this separation has not previously been stated in unified form for composite NDE. Predicted detection depths agreed within
20% of observed means for CFRP and thermoplastic systems. Pooled mean detection depths in CFRP were 5.2 ± 1.3 mm for pulsed thermography and 8.4 ± 2.1 mm for lock-in thermography; standard deviations reflect between-study heterogeneity rather than a controlled comparison. The framework explains defect-type-specific AI gains: convolutional neural networks improve accuracy by 31.1% for barely visible impact damage but only 8.7% for flat-bottom holes, consistent with selective condition-three noise suppression. Five research priorities are derived: thermoplastic-specific validation, field probability-of-detection data, passive and stimulated SHM-oriented thermography, multi-modal sensor fusion, and explainable AI for certification.