Ultra-clean Bellows Pump Flowrate Prediction Based on Multi-sensor Information Fusion with Noncontact Magnetic Tagged Bellow Deformation Derivation
Huimin Shen,Xiaowei Gu,Yaogang Chen,Liang Hu
DOI: https://doi.org/10.1109/tim.2024.3481536
IF: 5.6
2024-01-01
IEEE Transactions on Instrumentation and Measurement
Abstract:The bellows pump (BsP) has been widely used for ultra-clean fluid transportation in the semiconductor manufacturing process (SCMP). To ensure stable pump output and a high yield rate, maintaining the efficiency and safety of the entire SCMP system is crucial. Thus, flow pulsation compensation is vital. Advanced prediction of pump flow provides key information for timely pulsation compensation and effective monitoring of working conditions. However, the ultra-clean requirement brings great challenge for it. This paper proposes a flowrate prediction method based on multi-sensor information fusion processed by a developed neural network, combining the derived bellow deformation and the instantaneous flowrate. Taking advantage of the nonferromagnetic characteristic of the BsP component composition, the in-situ bellow deformation can be contactlessly derived by solving an inverse magnetic problem about the marked rectangular permanent magnet (rPM). Experimental systems have been constructed for both inverse magnetic solution and flowrate prediction validations. The experimental results show that the absolute rPM marker localization errors range in [0.094, 0.409] mm with a mean value of 0.207 mm and [0.094,0.409] rad with a mean value of 0.046 rad; the absolute errors of the predicted flowrate range in [0.02, 0.09] L/min for flowrate of [3.12, 5.14] L/min. As a typical application, the predicted flowrate enables flowrate fluctuations control in the BsP, which will directly influence the wafer cleaning yield and residues of particles. Simulations based on Simulink were carried out utilizing the experimentally derived predicted flowrate. The results show that the maximum flowrate fluctuation can be reduced from 3.0 L/min to 0.3 L/min for flowrate of 3.1 L/min, addressing the application of the proposed flowrate prediction method.