An Integrated Simulation and Process Control Platform for Distributed Manufacturing Process Chains
pdf
Rapid and customized part realization imposes stringent demands on part attributes, e.g., mechanical properties, microstructure, surface finish, geometry, etc. However, they can very rarely be directly measured and/or controlled in a production process. Instead, measurements are taken of accessible and measurable primary process responses that are known to influence the part's attributes. These primary process responses are then controlled through the manipulation of a set of controllable process parameters. This ubiquitous strategy assumes that the proper control of the primary process responses will implicitly yield the desired part attributes. The current work aims to replace this implicit assumption by a model-based explicit evaluation of the part's attributes that uses newly established process models, available measurements of process responses and historical data from a continuously updated database. In effect, this approach implies a direct, instead of an implicit, control of the part's desired attributes and, as such, moves a step closer to a truly model-based process control and rapid part certification.
Progress to date on the four key problem areas addressed in the project can be summarized as: (1) The software code for the volumetric model and underlying data structure that implements a voxel data storage structure has been parallelized to accelerate specific stages of the computation, including voxel refinement, property interpolation, data query, and output generation. The voxel storage strategy, including grid refinement, has been integrated with the GAMMA (in-house code) thermal simulation output format and demonstrated for AM process simulation data; (2) In terms of modeling to map process parameters to desired part attributes a microstructure simulation capability has been developed using a cellular automaton (CA) model coupled to thermal histories predicted either by the GAMMA code or by more detailed thermo-fluid simulations. The CA software has also been parallelized, allowing very large scale (order one billion-cell) simulations of grain nucleation and growth. The CA simulation capability has been validated against experimental results for both single-track scans and multiple layer builds; (3) For reduced-order model development a stacked recurrent neural network (RNN) structure has been formulated for the instantaneous prediction of the process-structure-property-performance (PSPP) framework with high spatial and temporal resolutions. While the RNN model has been proven to work successfully for thermal-related features, the effectiveness of this methodology to predict microstructural features will be investigated in the following year; (4) For the verification of the developed platform work continued on upgrading the functional capabilities of the Additive Rapid Prototyping Instrument (ARPI), developed under an NSF MRI grant, by an active cooling rate controller, consisting of a secondary laser and a cryogenic source and an active base plate. Portions of a network adapter for real-time remote communication with the machine’s sensors and controller have been completed.
Submitted by Anonymous
on
Rapid and customized part realization imposes stringent demands on part attributes, e.g., mechanical properties, microstructure, surface finish, geometry, etc. However, they can very rarely be directly measured and/or controlled in a production process. Instead, measurements are taken of accessible and measurable primary process responses that are known to influence the part's attributes. These primary process responses are then controlled through the manipulation of a set of controllable process parameters. This ubiquitous strategy assumes that the proper control of the primary process responses will implicitly yield the desired part attributes. The current work aims to replace this implicit assumption by a model-based explicit evaluation of the part's attributes that uses newly established process models, available measurements of process responses and historical data from a continuously updated database. In effect, this approach implies a direct, instead of an implicit, control of the part's desired attributes and, as such, moves a step closer to a truly model-based process control and rapid part certification.
Progress to date on the four key problem areas addressed in the project can be summarized as: (1) The software code for the volumetric model and underlying data structure that implements a voxel data storage structure has been parallelized to accelerate specific stages of the computation, including voxel refinement, property interpolation, data query, and output generation. The voxel storage strategy, including grid refinement, has been integrated with the GAMMA (in-house code) thermal simulation output format and demonstrated for AM process simulation data; (2) In terms of modeling to map process parameters to desired part attributes a microstructure simulation capability has been developed using a cellular automaton (CA) model coupled to thermal histories predicted either by the GAMMA code or by more detailed thermo-fluid simulations. The CA software has also been parallelized, allowing very large scale (order one billion-cell) simulations of grain nucleation and growth. The CA simulation capability has been validated against experimental results for both single-track scans and multiple layer builds; (3) For reduced-order model development a stacked recurrent neural network (RNN) structure has been formulated for the instantaneous prediction of the process-structure-property-performance (PSPP) framework with high spatial and temporal resolutions. While the RNN model has been proven to work successfully for thermal-related features, the effectiveness of this methodology to predict microstructural features will be investigated in the following year; (4) For the verification of the developed platform work continued on upgrading the functional capabilities of the Additive Rapid Prototyping Instrument (ARPI), developed under an NSF MRI grant, by an active cooling rate controller, consisting of a secondary laser and a cryogenic source and an active base plate. Portions of a network adapter for real-time remote communication with the machine’s sensors and controller have been completed.