Model-based Process Control for Transferred Arc Synthesis of Nanoparticles

点击翻译按钮获取中文摘要

This project aims to establish an autonomous process control in transferred arc synthesis of metallic nanoparticles and subsequent processing steps. This process is characterized by relatively large fluctuations and is physically too complex to be modeled in detail. Therefore, the process control bases on a simplified model for the dynamics of the particle formation process (DPFP) taking place after the plasma region, having as control objectives the particle production rate, aggregate size and primary particle size which are therefore the key performance indicators (KPIs). The agglomerate dynamics is a function of evaporation rate and gas flow rate, as well as two parameters which cannot be directly measured or controlled but describe the temperature-history in the hot zone after the plasma. A process control is possible via regulation of the electric current and electrode distance, as well as the gas flow rates. Furthermore, the continuous removal of electrode material by plasma evaporation requires an adaptive control of the electrode distance. Information about the process state expressed in the KPIs is obtained from in-situ measurement of the arc characteristics (optical and electrical characteristics) as well as quasi-real-time online measurements of the particle mass concentration and the primary particle diameter and aggregate size. This is possible by determination of the effective density from two different equivalent particle diameters (electrical mobility and aerodynamic) with a time-resolution in the range of seconds. For autonomous operation, the system will be trained to recognize disturbances like ejection of unwanted large particles, plasma extinction and displacement of the arc. The control system will be designed in the unified framework of control and detection and bases on modeling of the couplings among the KPIs, hidden variables and control inputs based on the DPFP. This will be realized with the help of Machine Learning approaches. The DPFP-based predictive control will be designed so that it has a high robustness against model uncertainties and limited measurement performance. It will be adapted for the various operation regimes and contain recoverable control mechanisms against control performance degradation.