Manufacturing contributes $2.9 trillion to U.S. gross domestic product, and improving its cyberinfrastructure resilience is essential for economic competitiveness and national security. Modern manufacturing has evolved into a data-intensive scientific cyberinfrastructure in which interconnected sensors, learning-enabled controllers, and human-in-the-loop decision systems drive production, yet the research community lacks open, FAIR-compliant security datasets that capture the tightly coupled cyber-physical-human interactions characteristic of realistic manufacturing environments. This project builds and operates a publicly accessible, annotated security dataset for manufacturing cyberinfrastructure by integrating a fully instrumented assembly testbed at the University of Georgia Innovation Factory with a high-fidelity digital twin simulator. The platform generates time-synchronized, multimodal traces organized by the Purdue Enterprise Reference Architecture, spanning network logs, physical process data such as torque curves and vibration signatures, and de-identified operator metadata collected under benign, fault, and adversarial conditions. Key innovations include a hybrid generation environment that uses a digital twin for deterministic replay and safe exploration of rare or safety-constrained events; a three-dimensional attack surface characterization encompassing cyber, physical, and human layers; and integrity-verified data lineage pipelines that ensure dataset auth