| Project/Area Number |
05555079
|
|---|
| Research Category |
Grant-in-Aid for Developmental Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
Intelligent mechanics/Mechanical systems
|
|---|
| Research Institution | TOKYO NATIONAL COLLEGE OF TECHNOLOGY |
|---|
Principal Investigator |
KOSAKA Toshifumi TOKYO NATIONAL COLLEGE OF TECHNOLOGY,Associate Professor, 情報工学科, 助教授 (60153524)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
SHINKAI Takashi ASMO Co., Ltd., Production Engineering, 生産技術部
KOMATSU Tomohiro ASMO Co., Ltd., Production Engineering, Chief, 生産技術部, 主任
YAMADA Shinji YAMANASHI UNIVERSITY,Faculty of Engineering, Professor, 工学部, 教授 (90020403)
KITAMURA Toshiya YAMANASHI UNIVERSITY,Faculty of Engineering, Research Associate, 工学部, 助手 (80224971)
MATSUBAYASHI Katsushi TOKYO NATIONAL COLLEGE OF TECHNOLOGY,Research Associate, 機械工学科, 助手 (80239061)
|
|---|
| Project Period (FY) |
1993 – 1994
|
| Project Status |
Completed (Fiscal Year 1994)
|
| Budget Amount *help |
¥1,600,000 (Direct Cost: ¥1,600,000)
Fiscal Year 1994: ¥400,000 (Direct Cost: ¥400,000)
Fiscal Year 1993: ¥1,200,000 (Direct Cost: ¥1,200,000)
|
|---|
| Keywords | Neural Computer / Neural Network / motor / abnormal sound / inspection / noise |
|---|
| Research Abstract |
The automatic classification of sounds of small motors was examined with Neural Computer. The motor emitting comfortable sound is judged normal, even if the sound level is high. The motor emitting noisy sound is judged abnormal, even if the sound level is very low. Skilled workers inspect sounds of motors abnormal or not. In addition they classify factors of abnormal noise with its timere. As the inspection is sensuous and anyone cannot make definite criterion of judgment, it has been impossible to inspect motors automatically.
The classification system with Neural Computer for practical use was developed. Auto correct algorithm on learning and fast convergence algorithm were applied in the Neural Computer program. Neural computer learning characteristics were investigated with total error and energy. Weight reversal output was applied to know the status of Neural Computer after learning. Motors in remote control door mirrors were examined and their power spectra were learned. After lea
… More
rning, unlearned data were inputted to Neural Computer and they were classified factors of abnormal noise correctly with high percentage. At the classification whether the motor had abnormal sound or not, very few abnormal sound data were recognized as normal sound.
In the factory of motors, at the end of assembling line of geared motors, Neural Computer classification system was applied. The motor is part of air conditioner of car. One of the abnormal sounds is emitted from gears. It is difficult to classify the sound with comparing power spectra. The average value, maximum value, fluctuation width and the number of abnormal peak extracted from 8 outputs (time domain) from 1/3 octave band filter were employed as data. After learning of these data, more than 99% of unlearned data were classified correctly for each category. This classification system has been worked at the end of assembly line since September 1994. As single board computer with Neural Network program has been developed, system has been small and it has reduced the cost. Then, this classification system is applied to other assembly lines. Less
|
