Tech Lead Advanced Structures and Composites Center
The rapid adoption of artificial intelligence in advanced manufacturing demands not only high-quality data but also robust data pipelines that can transform raw information into actionable insights. Building on the Material Process Property Warehouse (MPPW), this work presents an AI-integrated data and model lifecycle framework that supports and orchestrates the journey from data ingestion through model training and deployment. The current implementation features a resilient data pipeline that integrates heterogeneous manufacturing data, including process parameters, sensor streams, material properties, and inspection results, into a structured data warehouse, ensuring seamless access, traceability, and data integrity. Machine learning models are being trained directly from carefully curated subsets of this warehouse. This approach is designed to support a range of ML applications, including anomaly detection and large language model (LLM)-based querying of additive manufacturing data within the warehouse. The LLM query interface is used as a representative example to demonstrate the end-to-end data-to-inference pipeline with reinforcement feedback.
Once trained, models are deployed in a parallel inference layer integrated with the data infrastructure, enabling real-time inference and decision support without disrupting ongoing operations. Notably, the pipeline incorporates a reinforcement feedback loop that captures user interactions, expert corrections, and system performance metrics to refine model weights, supporting continual improvement and adaptability to evolving process dynamics. This feedback-driven AI–data ecosystem is designed to transition manufacturing data systems from passive repositories toward active intelligence engines. By unifying secure data management, scalable model training, and adaptive deployment, the MPPW framework demonstrates a working pipeline for ML model integration and lifecycle management. The LLM-based query model serves as a practical example of this closed-loop operation, while additional models, including image analysis-based deformation detection, are being incorporated as part of ongoing development using the same framework. The result is a progressively evolving, provenance-rich environment aimed at accelerating rapid prototyping and data-driven process development in additive manufacturing.
Learning Objectives:
Upon completion, participants will be able to describe the architecture and data pipeline design that enables AI and machine learning integration within advanced manufacturing data warehouses.
Upon completion, participants will be able to evaluate how user feedback and adaptive retraining loops can reinforce and improve AI models for continuous learning in manufacturing environments.
Upon completion, participants will be able to understand how the additive manufacturing digital thread connects design, process, and material data to ensure full traceability and certification readiness.
