Deep Learning-based Precision Control for Six-axis Compliant Nanopositioner
Refereed conference paper presented and published in conference proceedings


全文

其它資訊
摘要In this work, we present a new control method for compliant mechanisms based on deep learning models that achieve 100 nm precision in multiple axes with low-lost strain sensor arrays. In our setup, the strain sensor array is applied to the flexing components on a custom-designed compliant six-axis nanopositioner such that it simultaneously measures the mechanical strain (i.e., displacement) and mode shapes for real-time control, i.e., the deep model can learn, predict, and control the motion of the nanopositioner as a time sequence or 2-D images. During the training stage, capacitance sensors are used as references to minimize position errors; the number and location of sensor arrays can be optimized via the deep learning model. In the experiments, we demonstrate that the deep model developed based on the reinforcement learning method [1] can fully replace the classical PID control, realizing 100 nm precision without tuning any control parameters.
出版社接受日期01.08.2019
著者Xiangbo Liu, Chenglin Li, Hongsheng Li, Xiaogang Wang, Shih-Chi Chen
會議名稱Annual Meeting of the American Society for Precision Engineering
會議開始日28.10.2019
會議完結日01.11.2019
會議地點Pittsburgh, Pennsylvania
會議國家/地區美國
會議論文集題名Proceedings of the 34th ASPE Annual Meeting
出版年份2019
月份10
出版社ASPE
出版地USA
頁次139 - 144
國際標準書號978-188770678-0
語言美式英語
關鍵詞deep learning, precision control

上次更新時間 2021-15-10 於 23:39