| Abstract Scope |
Identifying suitable process parameters in laser powder bed fusion (LPBF) remains a major challenge. Many different combinations of process conditions can produce parts with similar geometry but highly variable microstructures, residual stresses, and mechanical properties. Existing methods for process parameter selection rely on trial-and-error experiments or models, making it difficult to efficiently determine the process conditions required to achieve targeted product attributes. This study presents a physics-based inverse modeling framework that directly predicts feasible LPBF process conditions (inputs) from desired product attributes (outputs). The framework integrates physics-based mechanistic modeling, neural networks, and genetic-algorithm-based inverse optimization within a closed-loop system. Mechanistic modeling helps generate physics-consistent information on melt-pool geometry and temperature distribution. Neural networks enable rapid prediction and the generation of large synthetic datasets trained on model data. The genetic algorithm efficiently explores multiple feasible combinations of process parameters, rather than converging on a single non-unique solution, resulting in a computationally efficient, stable, and physically interpretable inverse design framework. The developed framework accurately captures the relationships between process parameters and melt-pool behavior, showing strong agreement with experimental and predicted data. The developed inverse framework successfully identifies multiple physically consistent combinations of laser power, scan speed, and other process conditions required to achieve targeted melt pool geometry. The physics-informed mechanistic and GA–NN models accurately capture melt-pool behavior, temperature fields, and solidification characteristics across different LPBF conditions. By combining mechanistic modeling, machine learning, and inverse optimization, the framework provides a computationally efficient and physically interpretable approach for process optimization. Overall, this work establishes a scalable pathway for physics-guided, defect-aware, and application-driven inverse design in metal additive manufacturing. |