| DC 欄位 |
值 |
語言 |
| DC.contributor | 光電科學與工程學系 | zh_TW |
| DC.creator | 賴昇倍 | zh_TW |
| DC.creator | Sheng-Pei Lai | en_US |
| dc.date.accessioned | 2022-7-27T07:39:07Z | |
| dc.date.available | 2022-7-27T07:39:07Z | |
| dc.date.issued | 2022 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=106256001 | |
| dc.contributor.department | 光電科學與工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 攝影是大多數行動裝置的必備功能，關鍵的塑膠鏡頭零件在配對的生產過程中，最耗費人力資源；現有產品約有10~15% 在公差上下限邊緣或超出公差，裝配後累積公差很容易就會超出標準範圍，目前處理方法是靠專業技師以人工、像差互補方式，將零件排列、配對找出最佳組合，再實際小批量生產進行確認，耗時、費工、程序複雜。
本論文主要提出利用像差理論以及機器學習預測工具，研究新的生產配對、驗證方式以達到流程精簡目標：
首先架構出一個利用零件尺寸進行ZEMAX光學MTF結果的模擬，進行實際生產預測；(1)利用零件生產的歷史數據判斷生產公差範圍。(2)將範圍公差隨機帶入ZEMAX 模擬軟體產生批量數據庫，作為回歸模型基礎。(3)根據模型基礎進行擬和程度分析修正成與ZEMAX 輸出一致的裝配資料。(4)持續收集實際生產數據與零件尺寸對應MTF數據進行二次修正；其中修正方式，可以進行線性修正，或是對預測模型再次回歸訓練；完成零件尺寸與生產結果相對的預測資料。
完成的光學預測結果，直接教導電腦搭配出實際生產想要的MTF組合；光學理論所謂的搭配其實就是在進行鏡頭像差的抵消；基於此概念，我們將兩個主要影響鏡頭成像的像場彎曲(Field Curvature)與傾斜像差(Tilt)兩種像差進行搭配抵消。電腦程序是先將零件進行洗牌搭配預測其品質，並設置門檻值，通過門檻則跳出迴圈；可透過AI學習，優化搭配速度與精準度，再進行生產流程安排、實際驗證，最後將總體結果與傳統生產方法做比較，產生效益分析；結果顯示此方法可將14天的選配工時縮短到1天，且隨資料庫的累積學習，生產效率與品質越來越好，此為光學工廠智能化的重要基礎。 | zh_TW |
| dc.description.abstract | This article is discussed about a how can AI tools help improve the effect of the optical lens production manufacture efficiency. Training AI model to predict optical lens image quality at element level and design a performance optimize loop by quick simulation. Let factory could raped improve lens production quality and go into rand up stage. Shorten product design lead time.
It’s can also provide a good way to monitor production line. Base on IoT tools maturity. It’s easy to set check point in production pipeline. After we get enough raw data. We can feed datasets to AI model then let AI sent any signal if it detects any trend in factory. Use them to act early, avoid yield loss.
In the machine learning process, we use the optical lens design tool ZEMAX to simulate and establish a basic MTF evaluation model as well as to compare with the real production lens set and match the MTF evaluation quality.
Based on experimentation results we can see that the AI method can increase the effect triple to classical production method. It’s a large potential way to boost the optical industry going forward. | en_US |
| DC.subject | 人工智能模型 | zh_TW |
| DC.subject | 離焦曲線 | zh_TW |
| DC.subject | 像平面 | zh_TW |
| DC.subject | 決策樹 | zh_TW |
| DC.subject | 配對表 | zh_TW |
| DC.subject | AI model | en_US |
| DC.subject | Through focus | en_US |
| DC.subject | Image plane | en_US |
| DC.subject | Decision tree | en_US |
| DC.subject | Pairing table | en_US |
| DC.title | 機器學習工具應用於光學鏡頭工廠生產策略協作方法之研究 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Research on the Application of Machine Learning Tools to the Production Strategy Collaboration Method of Optical Lens Factory | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |