Data-physics framework embedding machine learning for controllable permeability design and optimization in additive manufacturing

• Data-Physics Driven Approach with Reduced Data Dependency Distinct from purely data-driven machine learning methods, this work combines physical knowledge and computational simulations to improve permeability prediction accuracy. Distinct from purely data-driven approaches, the hybrid data–physics framework achieves reliable permeability prediction with only 25 training cases, yielding a maximum absolute deviation of 8.53× 10⁻¹ 3 m² from the target values, while maintaining strong physical interpretability. • Machine Learning-Assisted Design for Controllable Permeability This study proposes a novel machine learning-assisted design method for controllable permeability, significantly reducing the extensive experimental data and iterative trials required in conventional processes. By integrating neural networks with finite element analysis to establish process-porosity-performance relationships, the method enhances the accuracy of permeability control and enables precise prediction of optimal process parameter ranges based on performance requirements, thereby improving design controllability. • First-Time-Right Fabrication of Gradient Porous Structures with Mechanical-Permeability Synergy This study achieved anisotropic permeability control without pre-defined 3D models, successfully fabricating porous materials with a gradient porosity distribution in a first-time-right manner. The optimized process demonstrates multifunctional synergy with concurrent high permeability (bubbling pressure: 2.16 kPa) and mechanical strength (yield strength: 196.77 MPa) under 350 W laser power and 1800 mm/s scanning speed. By circumventing conventional trial-and-error approaches, this paradigm enables programmable fabrication of functionally graded porous materials with application-specific performance customization in additive manufacturing. Additive manufacturing enables the fabrication of functionally graded porous structures through localized pore generation control in critical component regions. While pore morphology critically determines both permeability and mechanical performance, conventional design approaches heavily rely on empirical expertise and iterative trial-and-error processes. This study proposes a hybrid data-physics framework that integrates finite element analysis (FEA) and machine learning (ML) to establish process-porosity-permeability correlations, and achieves goal-oriented permeability optimization. Specifically, an artificial neural network (ANN) was constructed to accurately predict the nonlinear relationships between porosity and process parameters(Ⅰ). Simultaneously, FEA was employed to simulate the permeability behavior of porous materials under varying porosity conditions, elucidating the specific influence patterns of porosity on fluid flow characteristics(Ⅱ). Distinct from purely data-driven approaches, the hybrid data-physics framework achieves reliable permeability prediction with only 25 training cases, yielding a maximum absolute deviation of 8.53×10 −13 m² from the target values, while maintaining strong physical interpretability. By utilizing this model to accurately predict the required process parameter ranges based on performance requirements, porous structures with customized gradient porosity distributions were successfully fabricated first-time-right. This approach eliminates the need for iterative experimental parameter tuning required in conventional methods, significantly enhancing design and production efficiency. Notably, when the porosity was within the range of 17.71% to 22.33%, favorable compatibility between mechanical properties and permeability performance was demonstrated(Ⅲ). The optimized process exhibited exceptional multifunctional synergy: high permeability (bubbling pressure: 2.16 kPa) coupled with notable mechanical properties (yield strength: 196.77 MPa) at 350 W laser power and 1800 mm/s scanning speed. This study provides a design framework for the flexible control of pore morphology to engineer materials with tailored functional properties in additive manufacturing of porous structures.