2D圖像深度學習在製造規劃的整合應用

Integrated Application of 2D-image Deep Learning on Manufacturing Planning

邱柏薰(BO-XUN QIU)

指導教授 ： 鍾文仁

中原大學/工學院/機械工程研究所/碩士(2020年)

https://doi.org/10.6840/cycu202000737

摘要

精密模具是由大量複雜且高精度之零件所組成，而目前這些零件的製程規劃需由經驗豐富的技術人員，依產品形狀、加工步驟與方法列出生產順序，而生產順序會因規劃人員而異，即同一產品可能因規劃者的不同而有不一樣的生產順序，也可能因此造成加工時間與品質有所不同，因此需要一套標準程序來決定生產順序；本研究使用深度學習與知識管理使系統來產生標準程序，減少因人員不同所產生的差異。
本研究使用TensorFlow建立具有FCN(Fully Convolutional Network)結構之類神經網路模型，以2D視圖作為訓練資料，用來對工程視圖進行圖像語意分割，以神經網路模型對視圖的辨識結果，能夠判斷出CNC銑床加工、放電加工與線切割3種主要零件加工方式，並用實際案例說明使用多種特徵資料訓練之特徵辨識模型，在特徵辨識能力與實用性上優於使用單一特徵資料訓練之特徵辨識模型。

關鍵字

製造規劃 ； FCN深度學習 ； 圖像語意分割 ； 因果關係

並列摘要

Precision molds are composed of numerous complex and high-precision parts, and the current process planning of these parts requires experienced technicians to list the production sequence according to the product shape, processing steps and methods, so the production sequence will vary depending on the planner. The same product may have different production sequences due to different planners, and may also result in different processing times and quality. Therefore, a set of standard procedures is required to determine the production sequence; this research uses deep learning and knowledge management to make the system produce standard procedures in order to reduce the difference caused by different personnel.
This research uses TensorFlow to build a neural network model with a Fully Convolutional Network(FCN) structure, and uses 2D views as programming data to segment the engineering views. The neural network model can be used to identify the results of the three main parts processing methods, CNC milling, electrical discharge machining and wire cutting, and used actual cases to illustrate that the feature recognition model trained with multiple feature data is better than the feature recognition trained with single feature data in feature recognition ability and practicality model.

並列關鍵字

manufacturing planning ； FCN deep learning ； image semantic segmentation ； causality

參考文獻

參考文獻

[1] Babic, B., Nesic, N. and Miljkovic, Z., 2008,"A Review of Automated Feature Recognition with Rule-based Pattern Recognition", Computers in Industry, 59(4), pp. 321-337.

[2] Li, W. D., Ong, S. K. and Nee, A. Y. C. 2010, "A Hybrid Method for Recognizing Interacting Machining Features", International Journal of Production Research, 41(9), pp. 1887-1908.

[3] Zhou, X., Qiu, Y., Hua, G., Wang, H. and Ruan, X., 2007, "A Feasible Approach to the Integration of CAD and CAPP", Computer-aided Design, 39(4), pp. 324-338.

[4] Kumar, N. and Garg, P., 2011, "Recognition of Distorted CAD Objects using Neural Networks", International Journal of Computer Applications, 14(8), pp. 18-22.