Model-based Process Control for Transferred Arc Synthesis of Nanoparticles

• Applicants: Professor Dr.-Ing. Steven Xianchun Ding; Professor Dr.-Ing. Frank Einar Kruis
• Subject Area: Mechanical Process Engineering ; Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 504661005

Electrical discharges, including arc and spark discharges, offer a clean and scalable approach to metal and ceramic nanoparticle synthesis in the gasphase. These methods are characterized by inherent instability due to electrode erosion, causing severe process dynamics. Arc discharge typically yields higher production rate and adjustable primary particle size, while spark discharge is more stable but not usable for high throughput production. Therefore, spark discharge is suitable at laboratory scale and has been employed in the first funding period to develop, validate and incorporate methods to achieve real-time control. A special research domain of physics informed machine learning, called control theory informed machine learning (CTIML), is applied exploring integration of learning-based architectures into control theoretic structure, the stable kernel representation (SKR), which serves as state estimator and residual generator. A simple particle formation model, as well as quasi real-time measurements of key performance indicators (KPI) such as the primary particle size, number of primary particles per aggregate and the production rate were utilized to accurately describe the system's input-output relationship. KPIs are calculated from the combination of parallel mobility and aerodynamic measurements. Building on this foundation, the second funding period aims to extend autonomous control to the more complex arc discharge reactor and integrate it within a full process chain. The SKR model together with the stable image representation (SIR), an observer-based control system, forms an SIR-SKR prediction model, assuring numerical computational stability and reducing online optimization computation of the model predictive control (MPC) algorithm. The SKR-SIR-based MPC facilitates the predictive distribution control by enforcing a predefined target probabilistic distribution on the particle size and morphology via an additional term in the loss function in the CTIML framework. The formation of micron-sized particles caused by splashing effects is addressed by employing a cyclone separator with adjustable separation characteristics. Product quality will be monitored through real-time classification of plasma disturbances and particle morphology variance, where low-quality material is diverted. Finally, control over particle morphology will be extended beyond synthesis by regulating a sintering furnace through SIR-SKR-based MPC, where residence time and furnace temperature are adjusted in response to on-line morphology measurements, ensuring target particle structures are achieved independently of upstream process variations.

• DFG Programme: Priority Programmes
• Subproject of: SPP 2364: Autonomous processes in particle technology - Research and testing of concepts for model-based control of particulate processes