Common Data Model to Rapidly Certify AM Parts with Reduced Inspection Leveraging AI / ML

D. Reed, J. Shah, W. Sobol, T. Kirk, A. Kitt
United States

Keywords: common data model, additive manufacturing, interoperability


The ability to effectively leverage additive manufacturing (AM) technologies hinges on quickly identifying, prioritizing, and communicating essential information required to manufacture parts. Capturing the right data during the manufacturing process may allow organizations to qualify structural parts for use straight from the manufacturing process, without the need for repetitive evaluation or lot acceptance testing. Currently the ability to generate, identify, prioritize, disseminate, reuse, and qualify AM data is challenging and time consuming. MxD, AmericaMakes, and LIFT—three DoD manufacturing innovation institutes—have executed a collaborative DoD-funded program to reduce cost and time required for AM qualification. The Collaboration Team used a multi-faceted approach which considered data sharing, ontologies, data element mapping, security, and identification & prioritization of data types. This presentation summarizes the MxD sub-team’s development of a minimum-viable common data model and the introduction of FAIR data ecosystem principles into a cybersecure framework for transcribing data across production phases and platforms. This is one foundational element in a multi-year planned collaboration towards the vision of parts which are “born certified” through hybrid MBE and analytical approaches. A stepwise approach was utilized to establish the infrastructure required for digital qualification processes: 1. Identify the minimum data requirements for current-state and advanced process qualification, developing a common data model to facilitate creation of broad data sets that can be used to identify sources of variation; 2. Establish a secure data repository and data transfer protocol linking the process history of a material to part quality; 3. Establish a roadmap for the use of machine learning methods on recorded data to rapidly correlate process variation to final quality; 4. Identify opportunities to supplant redundant physical verification & validation testing with ICME models and provide recommendations for ICME toolset improvements required to support this digital qualification; 5. Implement AM equipment health monitoring, i.e., direct data measurement of equipment key health parameters such as gas flow and laser quality to establish acceptable system performance limits that reduce the cost of scaling across a fleet of identical AM systems; 6. Test assertions and assumptions by manufacturing a statistically significant quantity of parts to assess how AM process and post-process variables affect final quality. This presentation summarizes conclusions against steps one through three and initial results from AI analytics through steps 5 and 6. Key feedback from multidisciplinary subject matter experts formed the basis of a minimum data set for validation and verification. A review of available standards and ontologies led to the selection of the additive manufacturing common data dictionary and data model (AM-CDD and AM-CDM) as baselines which were extended to organize inbound data into logical and interoperable units. A secure data repository was built to house and transmit AM data. The ability of the secure data repository to translate CDM-format data into ICME-ready inputs was demonstrated and data from the build was analyzed via AI/ML to demonstrate correlations via experimentally designed production-relevant process variation.