Data Driven Decision Support for Additive Manufacturing

Summary

“Digital disruption is a change in industry triggered by advances in information and communications technology. During the next few years, the technologies associated with this wave--including artificial intelligence, cloud computing, the Internet of Things, Industry 4.0, and data analytics--will advance and amplify one another’s impact.”   A digital disruption is burgeoning in AM, with sensors and measurements playing an increasingly important role in part production.  As several AM technologies reach production maturation in the coming years, the Data Driven Decision Support for Additive Manufacturing Project will develop new metrics, methods, guides, and best practices for informed decision making and parameter selection during the production process.  Novel, structured approaches for incorporating new data and a priori knowledge into design and process planning phases will facilitate design-to-product transformations by reducing uncertainties and increasing the probability of a successfully qualified part build.  As AM matures as a production technology so must communication capabilities between designer, manufacturer, and inspector.  New techniques for characterizing and measuring AM part geometry will provide designers with increased autonomy from manufacturing decisions.  Previous works in the Measurement Science for Additive Manufacturing program provide a foundation on which new methods and techniques to provide decision support to design and process planning will be further developed.

Description

Objective

To develop and deploy the metrics, models, and best practices for using product definition, advanced analytics, and machine learning methods in additive manufacturing design and process planning to reduce lead times and support first-part-correct goals.

Technical Idea

Future advancements in AM design and process planning will take advantage of the many new datasets resulting from process monitoring, material testing, and part evaluation.  AI and Machine learning techniques show great promise in fully realizing the potential, but how such methods are applied to different AM data sets are only beginning to be explored.  The project will explore different analytics, machine learning, and other AI techniques as means for maximizing how data is integrated into the data stream to create new feed forward and feedback opportunities.  As indicated by the “3DS” designation, data-driven decision support will be a focal point of project efforts.  Data-driven decision support will address 1) how newly generated data is incorporated, through feed forward and feedback, into AM decision making, 2) how newly generated data can be made to become part of the AM “knowledge pool,” feeding into, augmenting, and eventually becoming a priori knowledge, and 3) how new data and existing knowledge can be most effectively leveraged to make well-informed decisions.

Research Plan

New efforts in AM product definition and GD&T will aim to further communication capabilities for the manufacture and inspection of AM designs, particularly with addressing complex geometries such as lattice structure and complex properties such as the functional grading of materials.  These efforts will feed directly into the content of ASME Y14.46 Product Definition for Additive Manufacturing Draft Standard.  Currently released for trial use, incoming feedback from preliminary adoption of the draft standard will influence additional, yet to be finalized research and standardization activities, including additional efforts in characterizing technical data packages (TDP).  Several gaps identified in the AMSC efforts directly relate to these research activities, including: Gap D17: Contents of a TDP, Gap D18: New Dimensioning and Tolerancing Requirements, Gap D23 Documentation of New Functional and Complex Surface Features, and Gap D26: Design for Measurement of AM Features/Verifying the Designs of Features such as Lattices, etc.

To mature production capabilities, specifically how well designers and manufacturers are able to respond to variations in equipment manufacturing capabilities, new efforts in design rule fundamentals will be initiated.  The Principles of Design Rules work item will be expanded to focus on fundamental correlations between design, material, and process primitives. As a complement to current Powder Bed Fusion efforts, new design rule efforts will be needed to address other processes such as Directed Energy Deposition and for specific applications.  As new sectors, such as energy and automotive, become engaged in AM production processes, extensions into specific applications will be crucial.  Correlations identified by the 3DS project will be captured as customizable guidelines that are extendable to different equipment, processes, and applications. These efforts will directly address AMSC Gap D3 Process-Specific Design Guidelines, Gap D4: Application-Specific Design Guidelines, Gap D5: Support for Customizable Guidelines, and Gap D6: Software-encodable/Machine-readable Guidelines.

The project will establish best practices for adopting emerging analytics and AI techniques to support knowledge discovery in AM, including: identify design, process, and material fundamentals; identify and establish patterns in materials, process, and part datasets; analyze data in support of AM design support (e.g. support structures); and optimize processing conditions in fully-open AM systems.  Included in these analytic efforts, 3DS will leverage AM datasets and a priori modeling knowledge in both feed forward and feedback process planning methods, such as design rules or surrogate modeling.  Efforts in modeling will attempt to better understand process variability, taking advantage of high-fidelity modeling efforts such as AM Bench to continue model characterization and uncertainty quantification efforts.  Such efforts aim to ensure the credibility and accuracy of simulations and data-driven models, and will feed into ASME V&V 50 standardization efforts while addressing modeling challenges outlined in AMSC Gap D9: AM Simulation Benchmark Model/Part Requirement.

The success of an AI technique depends on two important factors: the availability of quality datasets and the identification of appropriate algorithms. In collaboration with the Data Integration and Management project, 3DS will investigate novel data structures, representations, and formats to best enable information and knowledge discovery throughout the AM production chain, from early design decisions to manufacturing plans to parameter selection.  The appropriateness of using available machine learning and related AI algorithms to support specific, AM-related decision-making will be investigated and best practices developed. A shared, open platform for manufacturing data analysis will be prototyped to facilitate algorithm selection for different smart manufacturing systems. The platform will connect multi-level operational technologies with selected AI toolsets while providing guidance for AI model/service discovery. Best practices in problem formulation and algorithm selection will be codified into new PMML standards and provide a novel approach to addressing numerous decision guidance-related AMSC gaps.