- 學位論文
2D工程圖辨識學習與3D實體圖的自動生成
Deep Learning for 2D Drawings Recognition and Automatic Generation of 3D Solid Models
摘要
工程圖是產業界在產品研發設計過程中的主要參考依據之一。隨著電腦硬體和軟體的快速進步，工程圖的呈現方式已由傳統的手繪逐漸演變為使用電腦輔助設計（Computer Aided Design, CAD）軟體進行繪製。然而，目前技術仍需仰賴工程師運用CAD軟體將2D工程圖轉換為3D模型，這不僅耗時耗力，還有可能因工程師對工程圖的解讀有所疏漏，從而導致生產出的模型存在缺陷。 因此，本研究利用Darknet SDK（Software Development Kit, SDK）訓練YOLO（You Only Look Once）模型，以2D工程圖作為訓練資料，進行影像分割，以辨識不同特徵類型。本研究分別訓練了七個不同的特徵偵測模型，包括三視圖偵測（準確率100%）、圖形特徵偵測（準確率100%）、凹凸特徵偵測（準確率99.9%）、尺寸群組偵測（準確率99.8%）、尺寸線偵測（準確率99.1%）、理論尺寸偵測（準確率98.6%）以及字元偵測（準確率93.6%）。本研究藉由這七種預訓練模型，對2D圖紙中的尺寸、幾何圖形等特徵進行辨識。透過尺寸與物件邊長的配對，本研究將正確的模型尺寸參數導入資料庫，並利用Microsoft 的Entity Framework在ASP.NET應用程式中建立和管理資料模型，開發可視化介面，讓使用者可以輕鬆檢視工程圖辨識結果。此外，本研究結合CAD/CAM軟體Siemens NX，以及其提供的NX Open二次開發模組，將辨識出的2D工程圖特徵轉換為3D模型。這不僅可以降低2D轉3D模型過程中尺寸錯誤的風險，還能節省高達78%的3D繪圖時間，提高了2D工程圖辨識的準確性和模型轉換的效率。
並列摘要
Engineering drawings are one of the primary references in the industrial sector during the product development and design process. With the rapid advancement of computer hardware and software, the presentation of engineering drawings has evolved from traditional manual drawing to the use of Computer Aided Design (CAD) software. However, current technology still relies on engineers using CAD software to convert 2D engineering drawings into 3D models. This process is not only time-consuming and labor-intensive but also prone to errors due to potential misinterpretation of the drawings, resulting in flawed models. Therefore, this study utilizes the Darknet SDK (Software Development Kit) to train a YOLO (You Only Look Once) model using 2D engineering drawings as training data, performing image segmentation to recognize different types of features. Seven distinct feature detection models were separately trained, including three-view detection (100% accuracy), graphical feature detection (100% accuracy), concave-convex feature detection (99.9% accuracy), dimension group detection (99.8% accuracy), dimension line detection (99.1% accuracy), theoretical dimension detection (98.6% accuracy), and character detection (93.6% accuracy). Through these seven pre-trained models, this research identifies dimensions, geometric shapes, and other feature information within 2D drawings. By matching the dimensions with object lengths, the accuracy of numerical values is ensured and the correct values are imported into the database. Furthermore, this study combines the CAD/CAM software Siemens NX along with its NX Open development module to convert the recognized 2D engineering drawing features into 3D models. This approach not only reduces the risk of dimensional errors during the 2D-to-3D conversion process but also saves up to 78% of the time required for 3D drawing, enhancing the accuracy of 2D engineering drawing recognition and the efficiency of model conversion.