以經驗模態分析疊紋二值圖振動量測人工智慧分類之研究;Research on Artificial Intelligence Classification for Vibration Measurement by Moir? Binary Graph Analysis Based on Empirical Mode

NCU Institutional Repository > 理學院 > 光電科學研究所 > 博碩士論文 > Item 987654321/85918

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/85918

題名:	以經驗模態分析疊紋二值圖振動量測人工智慧分類之研究;Research on Artificial Intelligence Classification for Vibration Measurement by Moir? Binary Graph Analysis Based on Empirical Mode
作者:	林竑帆;Lin, Hong-Fan
貢獻者:	光電科學與工程學系
關鍵詞:	疊紋;經驗模態分析;經驗模態分解;人工智慧;卷積神經網絡;卷積神經網路;二值化;Moir?;Empirical Mode Decomposition;EMD;AI;?Convolutional Neural Networks;CNN;Binarization
日期:	2021-07-09
上傳時間:	2021-12-07 11:42:13 (UTC+8)
出版者:	國立中央大學
摘要:	接觸式和非接觸式是獲得物體表面信息的兩種方法。前者需要與被測物接觸，這使得獲取信息的效率非常差，並且在接觸期間被測物或檢測端可能變形或損壞。導致結果不准確。後者不會接觸被測物，量測效率更高，本研究之目的為創立一量化生理指標的非接觸式系統。在這項研究中使用三角測量來估計距離可以追溯到公元前 600 多年。希臘哲學家泰勒斯利用相似性原理(交集定理)來測量自己與金字塔的影子長度以及高度。若以此為基準輔以Moir?疊紋藉此放大振動情形，可使量測更為容易。而本研究系統就是利用投影出的光柵與電腦後製上去的光柵產生之Moir?疊紋變化偵測微小顫動。投影機投射光柵，雖然方便操作且可隨被測物不同及時更換適合的光柵，但其投影的閃爍頻率卻會對所測出的數據產生(0.6hz與10hz)影響，所以過往所得到的量測結果為相對數值而非絕對數值，因此以二值化方式消除其影響，並用區域多點平均方式消除因光線角度產生陰影而量測數值經fft後為無任何頻率變化的無效點，最後使用經驗模態分析獲取振動之瞬時頻率彌補傅立葉轉換無法明確了解資料的瞬時頻率的劣勢，在處理非平穩及非線性數據上具有優勢。本研究之實驗以可調整振動頻率之體感振動器進行實驗，導入人工智慧卷積神經網絡(Convolutional Neural Networks，CNN)進行分辨不同頻率，本論文方法的平均準確率高達98.07%，原使方法在相同訓練權重、相同訓練樣本數下平均準確率僅為53.76%，本論文方法準確率提高了44.31%。 ;Contact and non-contact are two methods to obtain surface information. The former needs to be in contact with the measured object, which makes the efficiency of obtaining information very poor, and the measured object or the detection end may be deformed or damaged during the contact. Lead to inaccurate results. The latter does not touch the object to be measured, and the measurement efficiency is higher. The purpose of this research is to create a non-contact system that quantifies physiological indicators. The use of triangulation to estimate distance in this study dates back to more than 600 BC. The Greek philosopher Thales used the principle of similarity (intersection theorem) to measure the length and height of the shadow between himself and the pyramid. If the Moir? moir? is used as a benchmark to amplify the vibration, the measurement can be made easier. In this research system, the projected grating and the Moir? moir? produced by the computer-made grating are used to detect small vibrations. The projector projects the grating. Although it is easy to operate and can be changed in time according to the measured object, the flicker frequency of its projection will have an impact on the measured data (0.6hz and 10hz), so the amount obtained in the past The measurement result is a relative value rather than an absolute value, so the effect is eliminated by binarization, and the area multi-point averaging method is used to eliminate the shadow caused by the light angle. The measured value is an invalid point without any frequency change after fft, and finally used The empirical modal analysis obtains the instantaneous frequency of vibration to make up for the disadvantage that the Fourier transform cannot clearly understand the instantaneous frequency of the data, and has advantages in processing non-stationary and nonlinear data. The experiment of this research is carried out with a somatosensory vibrator with adjustable vibration frequency, and artificial intelligence convolutional neural network (CNN) is introduced to distinguish different frequencies, the average accuracy of the method in this paper is as high as 98.07%. The average accuracy of the original method under the same training weight and the same number of training samples is only 53.76%. The accuracy of the method in this paper is increased by 44.31%.
顯示於類別:	[光電科學研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	61	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....