PFAB: A Generic Software Framework for Predictive Fabrication in Construction

As digital fabrication systems in construction become more sensorized and controllable, they expose growing opportunities for data-driven process understanding that remain largely underutilized.Consequently, additive manufacturing workflows continue to rely on expert intuition and trial-and-error, which limits their industrial adoption in dynamic environments, such as construction sites.This paper presents PFAB, an open-source Python-based framework that enables predictive fabrication workflows by structuring data, coordinating learning tasks, and guiding parameter selection.PFAB formalizes fabrication as a data-driven process that generates predictive models through active learning methods and supports performance-based optimization of process parameters.Its modular, schema-first design separates framework-level orchestration from fabrication-specific logic, allowing the same predictive workflow to be applied across fabrication systems.The framework is demonstrated through Learning-by-Printing, an exemplary implementation for extrusion-based additive manufacturing, and is complemented by an executable reference implementation illustrating end-to-end usage.