Advanced manufacturing processes involve multi-component materials and complex thermo-chemical reactions that result in a large volume of data, often generated at a high rate. Ideally, operators would get near real-time feedback/data on material properties and process parameters, thus enabling them to discover new mechanisms and phenomena more quickly and make adjustments that would improve efficiency and effectiveness. However, current approaches deliver feedback/data “post mortem,” long after the manufacturing process is complete.
Argonne National Laboratory researchers are using machine learning to analyze data and optimize manufacturing processes in near real-time (in minutes). As a proof of concept, the team is applying this methodology to the flame spray pyrolysis (FSP) process. FSP is a gas phase combustion synthesis method enabling the production of a range of materials. Applications include catalysts and battery cathode materials.
When applied to FSP, Argonne’s approach considers processing parameters such as composition and gas-flow rates. Insights gained from experimentation, computational fluid dynamics, and thermodynamics serve as input for machine learning algorithms, which probe the design/process parameter space for optimum baseline parameters. Based on the machine learning findings, processing parameters are updated to yield better results.
This work is providing us a clearer picture of manufacturing processes, thus making them more efficient and effective.