Quantitative InfraRed Thermography Journal

Explainable, interpretable, and physics-guided AI for infrared thermography

Infrared thermography (IRT) is increasingly enhanced by AI/ML for forward- and inverse-thermal modeling, diagnostics, and prognostics. Industrial deployment, however, hinges on certification and compliance workflows (e.g., safety cases and qualification), where explainability and interpretability are essential to justify decisions and satisfy standards bodies and regulators. This Special Issue (SI) advances physics-guided learning, explainable and interpretable ML (XAI/iML), and trustworthy modeling. Models may be informed by physics via analytical formulations, hard/soft constraints, and numerical simulators (e.g., FEM), moving IRT toward explainable or even interpretable frameworks.

Topics include but are not limited to:

• Physics‑guided learning for IRT: physics‑informed neural networks (PINNs) for inverse heat conduction, parameter estimation, and damage detection; hard/soft physical constraints; coupling to numerical models (e.g., FEM); surrogate and operator learning (e.g., neural operators) for PDE‑governed thermal processes.
• eXplainable AI and interpretable ML (XAI/iML): feature‑importance maps; concept‑ and exemplar-based explanations tied to known thermographic patterns; inherently interpretable models with sparse/low‑rank priors; physics‑consistency checks (energy balance, monotonicity, conservation); human‑in‑the‑loop decision support for inspectors.
• Trustworthy AI and certification: uncertainty quantification (aleatoric/epistemic), calibration and coverage; out‑of‑distribution detection and stress testing; verification and validation (V&V) for AI‑enabled IRT pipelines; documentation (model/data cards, risk and validation reports), and traceability (versioned data/code/configs, seeds, environment) aligned with AI risk management guidance.
• AI‑enhanced thermographic modalities and processing for active IRT: AI‑driven excitation design and adaptive control for active IRT with different excitation types, including optical (flash or laser), mechanical (vibro/ultrasonic), and inductive (induction/eddy‑current), under thermal‑safety constraints; learned demodulation and denoising for step or square/long pulse (SHT or SPT/LPT), pulse (PT), lock‑in (LIT), and modulated waves; interpretable, low‑dose alternatives to PPT/TSR/PCT with physics‑based regularization.
• Passive and solar‑loading IRT with physics‑guided AI: models of daily solar‑heating and cooling cycles that couple radiative, conductive and convective heat transfer with meteorological inputs (solar irradiance, wind, ambient temperature, humidity); physics-guided inversion for defect/moisture/void detection under uncontrolled excitation; uncertainty quantification across weather and view-geometry variability; explainable detection of artifacts from reflections, shadows, and emissivity changes; scheduling and acquisition policies (time-of-day/sky-condition selection) learned with physics-in-the-loop; protocols for reproducibility (weather logs, view/illumination geometry, emissivity management, reference patches/coatings).
• Emissivity and reflectivity mitigation: physics‑guided, explainable temperature–emissivity separation with Planck/Kirchhoff constraints; modeling and compensation of reflected radiance (e.g., view/illumination geometry); XAI to detect and explain reflection‑driven artifacts and spurious correlations; uncertainty‑aware, constraint‑regularized inference and calibration with reference patches/coatings.

These methods apply across aerospace structures (e.g., composite laminates and bonded repairs); electrical/electronic components and assemblies (e.g., PCBs, power modules, batteries); additively manufactured parts (polymers and metals); civil infrastructure (concrete, masonry, pavements); and cultural heritage assets. Field deployments with environmental compensation and emissivity management are welcome. Benchmark datasets with ground truth and uncertainty reporting, as well as reproducibility metrics and reporting guidelines, are especially encouraged.