| Budget Amount *help |
¥5,200,000 (Direct Cost: ¥5,200,000)
Fiscal Year 1995: ¥2,000,000 (Direct Cost: ¥2,000,000)
Fiscal Year 1994: ¥3,200,000 (Direct Cost: ¥3,200,000)
|
|---|
| Research Abstract |
Two-and Three-dimensional position/movement detection system using a multi-dimensional magnetic sensor and a neural network was realized.
First, position and movement detection for parallel moving object and bending arms in a plane was achieved. As a magnetic source, permanent magnet was used for obtaining the static magnetic field and a coil was used for the alternating magnetic field. The two-dimensional semiconductor Hall cell was used for detection of the static magnetic field and the two-dimensional coil was used for the alternating magnetic field. As a signal processing part, an analog neural network was used. In the experiments, the detection error of less than 5% was achieved for area of around 100 mm square. Also, the optimization of the structure of the network was studied.
Next, based on the two-dimensional results, the system was expanded to three-dimensional. The neural network was optimized in order to directly treat the three analog signals. As a result, three-dimensional position detection was achieved with the error of less than 5%. The detection with signal processing takes several tens milliseconds and it can be used for real time processing.
In this research, we found that the analog neural network, especially three-inputs network, is practically useful for such applications, whereas the limitation of the application of neural networks is not clear until now. From the result that the performance of the detection is not depend on the shape and spatial resolution of the magnetic source and detector, this system opens the novel application of the movement control of the micro-machines, etc.
|
