工程塑膠小孔製程之田口類神經網路啄鑽參數模型

CCUR > College of Natural Science and Engineering > College of Engineering > Department of Mechanical Engineering > thesis > Item 987654321/48447

Please use this identifier to cite or link to this item: https://irlib.pccu.edu.tw/handle/987654321/48447

Title:	工程塑膠小孔製程之田口類神經網路啄鑽參數模型 Peck-Drilling Parameter Model of a Small-Hole Process of Engineering Plastics Using Taguchi-based Neural Networks Method
Authors:	吳興垚
Contributors:	機械工程學系數位機電碩士班
Keywords:	工程塑膠鑽削切削推力田口方法倒傳遞類神經網路
Date:	2020
Issue Date:	2020-08-20 10:12:45 (UTC+8)
Abstract:	工程塑膠材料具高熱膨脹係數特性，鑽削製程中會因切削熱因素，而產生縮孔現象，特別在小孔製程中，鑽削參數與切削特性需詳細地討論，才能獲得穩定的孔品質。本研究執行工程塑膠PEEK之小孔鑽削實驗研究，探討製程參數對軸向切削推力(axial thrust force)與孔品質特性之影響。實驗採啄鑽工法並施以切削劑加工，分析的啄鑽參數有：主軸轉速、進給率、啄鑽深度，以三水準因子之L27直交表進行全因子實驗，推論出最佳參數水準條件。分析的鑽削案例有二：∅0.5 mm與∅1.0 mm。在參數模型方面，本研究應用田口直交表具因子正交性的實驗組合，建立田口基類神經網路模型。本文使用倒傳遞類神經網路(BPNN)，設定三項啄鑽參數與切削推力為輸入層神經元變數，並以兩項孔品質特性-孔徑縮量與真圓度為網路輸出層神經元變數。透過逐步增加網路訓練例的方法，經過三階段的網路訓練，獲得可以精確模擬網路訓練例的網路模型，並具備其它未參與網路訓練案例的預測功能。 Engineering plastic has a high thermal expansion coefficient, during the drilling process, a hole shrinkage is always generated due to the cause of cutting heat. Especially for a small-hole fabrication, drilling parameters and cutting characteristics should be discussed thoughtfully to obtain the stable hole quality. This study conducted small-hole drilling experiments of engineering plastic PEEK, to investigate the effects of drilling parameters on the axial thrust force and hole characteristics. The experimental hole was machined by peck-drilling method and operating with coolant. Peck-drilling parameters analyzed included the spindle speed, feed rate, and depth of peck drilling. The three-level full-factorial experiment, L27, was carried out to derive the optimal factorial level combination. Two cases of 0.5 and 1.0 mm diameter were performed in this study. As to the parameter modeling, this paper established a Taguchi-based neural network model based on the parameter combinations with factorial orthogonality of Taguchi’s orthogonal array using the Back-Propagation neural network (BPNN). Four variables were set as the input layer neurons including the three peck-drilling parameters and the axial thrust force, and two characteristic of the hole’s shrinkage in diameter and roundness were designated as the variables of the output layer neurons. Through the method of increasing the training examples progressively, after three-stage training process, a neural network was developed, which can precisely simulate the patterns of network’s training sets and provide the predict function for the cases not involved in the network training.
Appears in Collections:	[Department of Mechanical Engineering ] thesis

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	177	View/Open

View Licence

Loading...