1. Bioinspired robots
2. Roll-to-roll printed electronics systems
2.1. Lateral Control
Roll-to-Roll printed electronic system manufactures electronic components using film forming methods on a flexible substrate (it is called web). The manufacturing process starts with unwinding of the web. Designed components are patterned on the web. The process ends with unwinding.
However, nowadays most of electronic components are very small in size and requires multi-layer design. So during the film forming processes, lateral error can occur as shown in the following figure. The error lowers the quality of the printed components. So the minimization of lateral error for multi-layer printing is an important issue. The goal of this work is reducing the lateral error under |50um| with web transfer speed at 15mpm.
Following figure shows a lateral control system using a web guide. It is composed of a web guide and two vision sensors. Camera1 measures lateral error using pattern recognition algorithm and the error is fed back to controller. The controller drives the web guide for compensation of the lateral error.
Following figure shows an experimental result. The first plot is lateral error input and the second one is after the lateral error compensation and measured with camera 2. The designed controller maintains lateral error under |50um| with web transfer speed at 15 mpm.
3. Human-Machine Interface
3.1. Robust Hand Posture Classification to Arm Posture Change
The goal of this work is developing a robust hand classifier to arm posture change.
Forearm perimeter sensor detects physical change of the arm during hand grasp. It employs a strain gauge pasted on a rubber pad and detects the strain of the rubber pad by the muscle contraction when the hand grasps.
Development of a robust hand posture classifier requires removal of misclassification generating factors muscle fatigue, sensor location and arm posture, for example. Arm posture can lower the performance of a classifier with physical change detection sensor. However, in human daily life, many motions include hand and arm posture change at the same time. So, if the arm posture is included in the feature vector, then it may improve the performance of the classifier.
An inertial measurement unit (IMU) can measure the orientation of the arm. It can be used as a feature of arm posture change.
The two classes were classified with k-nearest neighbor (k-NN) and support vector machine (SVM) classifier after a training session. To verify the effectiveness of the suggested method a classification without the arm orientation feature was compared with the results. The following plots shows the effectiveness of adding arm orientation feature.
For two subjects, it improves 5% of classification success rate.
국가
대한민국
소속기관
건국대학교 (학교)
연락처
02-450-3731 https://sites.google.com/site/sangyoonleeku/
책임자
이상윤 slee@konkuk.ac.kr