Systematic review of multiscale thermal prediction models for FDM/FFF additive manufacturing: Toward digital twin integration with physics-informed machine learning

Fused filament fabrication/fused deposition modeling (FFF/FDM) is a widely used additive manufacturing (AM) technique valued for its affordability, safety, and greater accessibility compared to metal-based AM processes. Nevertheless, the fundamental complexities of multi-scale thermal phenomena in these processes have major effects on part quality, residual stresses, microstructural evolution, and interlayer bonding. The inherently transient and spatially heterogeneous thermal history introduced by localized heating and cooling poses immense challenges for process control and predictive modeling. Traditional numerical modeling for simulating these phenomena is computationally demanding, mesh-dependent, and not well-suited for multi-scale, real-time applications. Purely data-driven machine learning (ML) models are highly flexible but do not satisfy the constraints of physics, require extensive training data, and fail under sparse conditions. This work presents the first review of the thermal history of FFF/FDM within a multiscale modeling framework, focusing on the integration of digital twin (DT) approaches and physics-informed machine learning (PIML). The objective is first-time-right production, adaptive optimization, and real-time thermal control of autonomous defect detection in FFF/FDM. Some of the challenges currently faced are low availability of high-fidelity labeled thermal datasets; nonexistent standardized integration protocols for the sensors; and a lack of robust interoperability across modeling, control, and manufacturing systems. In the future, tackling these issues through standardized benchmarking datasets, interpretable PIML architecture for AM applications, and large-scale pilot demonstrations in industry-relevant settings should be the target of research. Advances in these areas will transform FFF/FDM into a new scientific domain that merges physics, AI, and advanced cyber-physical infrastructure to enable predictable, high-performance manufacturing for mission-critical applications.