基於深度學習與線性規劃之刮銅與移線CAM編輯自動化

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/89891

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/89891

题名:	基於深度學習與線性規劃之刮銅與移線CAM編輯自動化
作者:	顧詠文;Ku, Yung-Wen
贡献者:	資訊工程學系
关键词:	電路板;可製造性設計;電腦輔助設計;深度學習;機器人流程自動化;線性規劃
日期:	2022-08-04
上传时间:	2022-10-04 12:03:42 (UTC+8)
出版者:	國立中央大學
摘要:	PCB廠在接到上游客戶所給的原始設計時，必須根據自己的生產規範，在不影響PCB原始功能的條件下，以CAM(Computer-Aided Manufacturing)軟體進行編修。此步驟是為了避免製造過程的瑕疵產生。其中一種常見的規範，是線路之間的最小間距限制。一片PCB的設計圖上存在許多圓形的Pad，以及線狀的Line物件，這些物件代表著線路的配置，以導通之後要焊接在PCB上的各種電子元件。工程師常須削切Pad的外型(此動作又稱刮銅)，或者移動Line的位置，來取出電路之間的間距，避免生產精細度不足，造成兩邊相連而導通的情況。這種圖形化的編修過程常是長時間、瑣碎、單調而重複性高的，工程師平均需花3至5小時以上處理一片PCB的編輯與修正。然而，這些過程非常仰賴工程師人工地判斷及繪製，因此不易以機器自動化取代。本論文以CAM軟體—Genesis2000為平台，針對兩種常見的間距問題，提出自動化的解決方案。對於Pad物件間間距不足的情況，我們使用影像辨識的深度學習模型VGG16，判斷軟體螢幕畫面中Pad物件配置情況，預測刮銅的長度與方向。至於Line物件與其他物件間距不足的情況，我們蒐集配置上與該物件相關的所有物件資訊，以線性不等式描述這些物件的間距關係，以及線性函數來代表欲優化的目標，將整個問題轉換為線性規劃問題。 ;When the PCB factory receives the original design from the upstream customer, it must be edited with CAM (Computer-Aided Manufacturing) software to meet the production specifications of the factory, without affecting the original function of the PCB. This step is to avoid defects in the manufacturing process. One of the common specifications is the limit of minimum spacing between circuits. On a PCB, there are many circular pads and line objects. These objects represent the configuration of electronic circuits, conducting various electronic components soldered on the PCB. Engineers often have to cut the shape of the pads, or move the lines to make space between the circuits, to prevent short circuit. This graphical editing process is often long, trivial, monotonous and highly repetitive. Engineers spend an average of 3 to 5 hours on editing a piece of PCB. However, these processes heavily rely on manual judgment and operation by engineers. Therefore, it’s not easy to replaced human by machine automation. Based on the CAM software Genesis2000, this paper proposes an automatic solution for two common spacing problems. For the case of insufficient spacing between pads, we use the deep learning model VGG16 for image recognition to judge the configuration of pad objects in the software screen, and predict the length and direction of the cutting operation. For the case of insufficient spacing between pad and line, we collect all the related object information, describe the distance relationship of these objects with linear inequalities, and use a linear objective function as the target we want to optimize. Namely, we convert the entire problem to a linear programming problem.
显示于类别:	[資訊工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	37	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....