Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
CAD/CAM系統之常用銑刀切削參數資料庫與案例探討;
No full text[[abstract]]近年來各國紛紛提出新一代的製造產業政策，例如最廣為人知的工業4.0 (Industry 4.0)，顯示了製造業逐漸回歸各國的核心產業，而CAD/CAM系統在製造業中扮演著非常重要的輔助角色，因此若能透過建立一套製造業常用銑刀切削參數優化系統將能提升製造業產品開發速度與製造週期。本研究利用CUTPRO軟體進行敲擊實驗得到穩態切削圖與模擬切削力，透過穩態切削圖可得知轉速與切深的關係，藉以判斷刀具商手冊建議之切削條件是否會造成顫振問題，接著使用CATIA軟體進行加工路徑規劃，再將規劃後的加工路徑匯至MACHPRO軟體進行加工參數優化，最後將優化後的加工程式應用於機台上實際進行切削驗證，確認本研究之可行性且可應用於產業界。本論文中透過上述實驗流程中的穩態切削圖提供轉速與切削深度設定依據，讓操作者在使用CAD/CAM系統能較容易選擇合適的加工參數，並利用MACHPRO軟體進行切削參數優化，縮短加工時間，提升加工效益。
In recent years, many industrialized countries have proposed different manufacturing-related policies, for instance industrial 4.0, it shows manufacturing is gradually becoming the core industry in developed countries. Especially, CAD/CAM system play an important role to aid manufacturing. If we establish a cutting parameter optimization system for milling cutter, it will enhance the speed of product development and shorten the manufacturing cycle.This study utilized CUTPRO software to obtain the chatter stability lobes diagram and the simulated cutting force. The chatter stability lobes diagram help us observe the relationship between the spindle speed and the depth of cut in order to judge whether cutting parameters offered by the tool manufacturer's manual will cause chatter or not. Then, CATIA software was used to plan the machining toolpath. Next, MACHPRO software was adopted to analyze the machining toolpath and to optimize machining parameters. Finally, the optimized machining program was applied on the machine tool and has validated the feasibility and potential application to the machining industry of this study.In this thesis, the chatter stability lobes diagram provide guidelines in setting appropriate spindle speed and cutting depth; this can enable a CAD/CAM engineer to select suitable machining parameters in CAD/CAM system and optimize cutting parameters via MACHPRO software easier. It can effectively shorten machining time and improve processing efficiency
進階有限元素接觸分析在生物力學之應用;Advanced finite element contact analyses to Biomechanics Applications
No full text[[abstract]]有限元素接觸分析是在生醫工程用中是技術複雜的領域，因為生物接觸體具有復雜的幾何形狀與經歷大變形或受其他因素的影響。研究針對熱塑性矯正器接觸力傳遞在牙周韌帶（PDL）引起的初始應力分佈、摩托車頭盔接觸力傳遞對於減少頭部受傷的風險評估，進行動靜態數值研究。靜態有限元分析（FEA）模擬，以25歲女性患者牙齒模型，進行檢查熱塑性矯正器在彈性恢復力傳遞過程中PDL應力響應的影響。結果表明不同厚度的熱塑性矯正產生的PDL應力，提供使牙齒移動數值方法。動態有限元分析（FEA）模型為三種通風設計頭盔和三種傳統頭盔在騎士頭部安全和舒適度之間評估平衡。結果表明在所有的頭盔中，半罩頭盔加速度最大。通風槽的頭盔設計，確保頭部保護和在炎熱氣候中的舒適度，提供安全與舒適的解決方案。
This article the application of computational biomechanics to the simulation of contact mechanics as relevant to the study of thermo-plastic and helmet. A numerical investigation is performed into the in-itial stress dis-tribution induced within the periodontal ligament (PDL) by thermo-plastic appliances with different thicknesses. Based on the plaster model of a 25-year-old female patient. The results presented in this study provide a useful insight into as a result of the compressive and tensile stresses induced by thermo-plastic appliances of different thicknesses. Motorcycle helmets are essential for reducing the risk of head in-juries in the event of an impact. However, during the design of hel-mets, a compromise must be made between user safety and user com-fort. FEA models were constructed for both a prototype helmet design, and three traditional helmet designs. The half-face helmet re-sulted in the greatest headform acceleration. the results suggest that the pro-posed open-face helmet design with ventilation slots provides a prom-ising solution for ensuring both user protection and user com-fort in warm climates
整合動態加減速與命令整形之插補器設計;Novel interpolation design using input shaping technique and dynamic acceleration/deceleration algorithms
No full text[[abstract]]為了考慮振動與輪廓誤差在工具機上的影響，本論文考慮了基底、馬達、導螺桿與平台之間的關係，建立完整工具機傳動系統之動態模型。我們將動態模型中的共振頻率與阻尼做為插補器的參數，設計full order modified input shaping with zero vibration (FMISZV) 演算法來抑制各軸振動。此方法雖然可以降低振動，但由於後插補命令所產生的誤差，以致轉角輪廓誤差加大。為了考慮此輪廓誤差，我們發展了一套integrated dynamic acceleration/deceleration interpolation (IDAD)演算法以結合既有的動態模型，並將前加減速與後加減速整併於插補器內，藉由調整轉角速度，滿足轉角輪廓誤差容忍精度。最後，模擬實驗結果，驗證本論文提出演算法的可行性。FMISZV 不僅抑制主要之低頻共振，由於其具有低通濾波器的效應，因此也可以避免引發高頻振動。架構在FMISZV之上的IDAD，更能有效降低轉角輪廓誤差。此篇論文的主要貢獻在於設計了一套整合動態模型的插補器，比起傳統的插補器，能提供更佳的動態性能。
In order to consider the influence of vibration and contour error on the computer numerical control (CNC) machine tool, this dissertation considers the relationship between the bed, the motor, the lead screw, and the platform and establishes a more complete dynamic model of the feed drive system. We utilize the resonant frequency and damping ratio obtained from the dynamic model as the parameters of the interpolation to design the full order modified input shaping with zero vibration (FMISZV) algorithm to suppress the vibration of each axis. Although this method can reduce the vibration, but due to the acceleration/deceleration after interpolation (ADAI) method, the dynamic corner contour error increases. In order to consider this dynamic corner contour error, we developed an integrated dynamic acceleration/deceleration interpolation (IDAD) algorithm to combine the developed dynamic model, acceleration/deceleration before interpolation (ADBI), and ADAI in the interpolation, by adjusting the connective feedrate between two linear line (G01) blocks to meet the given dynamic contour error at the corner. Finally, we perform the simulations and experiments to verify the feasibility of the proposed algorithms. It is shown that the FMISZV not only suppresses the main low-frequency resonance but also avoids causing high-frequency vibration because it has the effect of low-pass filter. By integrating the IDAD and FMISZV, it is more effective in reducing the dynamic corner contour error. The main contribution of this dissertation is to design the algorithm which could integrate the dynamic model with the interpolation. As compared to the traditional interpolation, the proposed methodology can provide better dynamic performances
半競爭風險資料下最大概似估計法與經驗概似比值檢定法;Maximum Likelihood Estimation and Empirical Likelihood Ratio Test under Semi-competing Risks Data
No full text[[abstract]]這篇論文主要討論半競爭風險資料下的非終端事件時間的存活函數雙樣本檢定的問題。針對非終端事件時間的存活函數估計，我們建構了經驗概似函數，再利用PSO演算法得到最大概似估計值。針對檢定問題，我們提出最大概似比值檢定法。從模擬研究中可以發現我們的方法表現不錯。最後我們利用此篇論文的方法分析真實資料。
This thesis considers two sample testing problem of the survival function of the non-terminal event time under semi-competing risks data. For the survival function estimation of the non-terminal event time, we construct the empirical likelihood function, then maximize it by the PSO algorithm to obtain the MLE. For the testing problem, we develop the empirical likelihood ratio test. From simulation studies, it shows the performance of the proposed approaches is well. Finally , we analyze a real data example for illustration
基於Retinex 演算法之影像強化技術及其效能優化;Performance Optimization of Retinex–Based Image Enhancement Algorithm
No full text[[abstract]]近年來影像強化演算法已經被廣泛的應用在各影像處理實務中，而在各種強化方式中，Retinex被認為是最能有效保持圖片細節的演算法。雖然有此優勢，但為了維持影像細節和環境光的平衡，傳統Retinex演算法最為人詬病的地方即是運算耗時的問題。此外，和大部分影像強化演算法一樣，傳統的Retinex演算法無法適應不均勻環境光的場景，在分析與計算的過程中容易發生過度強化等問題。為了改善以上問題，本論文提出以下幾點貢獻，首先，針對演算法最耗時的光場估計部分提出一優化演算法，可大幅的減少時間消耗；此外，我們還提出具可適性的影像恢復處理方式，此方式可以自動偵測圖片最亮和最暗的部分，根據偵測結果調整核心演算法以避免過度強化。最後，為了使本論文提出之演算法能有更快的處理速度，我們在最後一個章節提出硬體加速考量，希望藉由軟硬體協同設計來使演算法效能有最佳優化結果。實驗結果顯示，我們的演算法不僅可以增強影像中的細節、維持影像光源的自然性，並且可以適應各式光源場景，不會有過度強化的問題出現；而運算時間也大大降低。
Image enhancement plays an important role in digital image processing and has been applied to fields of science and engineering. Among all of the image enhancement algorithms, Retinex-based algorithms are considered most effective in maintaining details of images. Despite the advantages, Retinex-based algorithms are generally more complex and consume considerably more time to maintain the balance between image detail and ambient illumination. And like most image enhancement algorithms, conventional Retinex-based algorithms are prone to over-enhancement when processing images with non-uniform illumination. To deal with aforementioned issues, various approaches are proposed and explored in the thesis. First, an optimized illumination estimation, which is the most time-consuming part of conventional Retinex-based algorithms, is proposed to reduce the time consumption. Furthermore, we propose an adaptive restoration process, which autonomously detects the brightest and darkest parts of the image and make modifications to the core algorithm accordingly to avoid over-enhancement. We also explored the possibility of hardware acceleration in the final chapter. It is hoped that the optimal performance can be achieved by the hardware and software co-design. Experimental results show that proposed Retinex-based algorithm can enhance details of images while preserving perceived naturalness without over-enhancement. And the execution time is greatly reduced
基於單攝影機影像處理之前車距離估測;Distance Estimation of Front Vehicle Based on Single Image
No full text[[abstract]]先進駕駛輔助系統 (Advanced driver assistance system, ADAS) 是近年來各大車廠亟欲開發的一項技術，常見的輔助系統包括車道偏移警示 (LDW) ，前車碰撞警示 (FCW) 以及主動式車距控制巡航系統 (ACC) 等應用。先進駕駛輔助系統的目的，是在駕駛可能發生危險情況下，發出警示與輔助駕駛，避免意外發生以達到降低交通事故的發生率。 本論文針對車輛行駛的當前車道，利用車輛底部陰影特徵與車輛底部具強烈邊緣特性來偵測前方車輛，最後提出基於車道寬度與基於車輛底部陰影位置的演算法來計算與前車距離。首先依據偵測出來的車道線定義感興趣區域，對區域內進行車輛偵測。由於車底陰影在日間之行車影像上是一個穩定的特徵，並不會受到車輛顏色影響，且車底陰影的顏色灰階值通常小於當前車道顏色灰階值。因此我們統計車道區域範圍內的顏色灰階值，當作車底陰影的參考閥值，同時利用不同於偵測車道線的邊緣閥值，以較低的邊緣閥值偵測車輛邊緣。在偵測車輛時，以車輛邊緣特徵為主、車底陰影特徵為輔，以邊緣像素比例以及暗點像素比例判斷前方車輛底部位置。本論文最後統計車輛偵測在不同情境下的偵測結果，本系統在前方有車輛時的正確偵測率可達 95% 以上。
The Advanced driver assistance system (ADAS) is a technology that the most automobile manufactures has been developed in recent years. The Lane Departure Warning System, the Forward Collision Warning System, and the Adaptive Cruise Control System are the common assistant System. In our thesis, we use the characteristic of the shadow of vehicle and the strong edge of the vehicle to detect the vehicle on current lane. Finally, we propose an algorithm to calculate the distance of front vehicle based on the width of the lane and the position of the shadow of vehicle. At first, we detect the feature of vehicle that according to the ROI, which is depended on the detected lane. Since the shadow of vehicle is a stable feature in daytime’s image, that is not affected by the color of vehicle’s body, and the gray level of the vehicle’s shadow is usually smaller than the gray level of the lane. Therefore, we calculate the gray level of the ROI as the referential thresholding, that use to detect the shadow of vehicle. In the meantime, we detect the edge of vehicle using the lower thresholding, which is different from the thresholding that detected to the lane. When detects the vehicle, the feature of vehicle’s edge is the first consideration, and the shadow of vehicle is the secondary. And we use the ratio of the edge’s pixel and the ratio of the dark pixel to get the position of the vehicle. In the end of the thesis, we statistics the ratio of accuracy of the detect vehicle under different situations, has up to 95%
運用飛輪實作主動平衡機器人;Development of an Active Balancing Robot with Reaction Wheels
No full text[[abstract]]本論文將透過在機器上置入飛輪，經由控制飛輪的轉動產生力矩來產生反作用力達到平衡。首先將著重於硬體架構的設計，從天寶導航的SketchUp軟體開始製作，繪製3D模擬圖；接著透過MakerBot出產的Replicator 3D印表機來產生塑膠材質的模型；最後將處理器、感測器、DAC、無刷馬達及驅動器買齊，與工廠製作出的鋁合金材質飛輪一起安裝後，完成硬體方面的架設。接著再經由撰寫程式碼，透過實作來驗證理論的可行性。
The purpose of this master thesis is in order to design a robot by installing two wheels and controlling the rotation of the wheels which produce torque. I focus on the hardware design at first with producing my 3D plan with SketchUp from Trimble Navigation and then produce the prototype in plastic with a 3D printer, “Replicator”, from MakerBot Industries. At last, I put a processor, a sensor, two DACs, two brushless motors with drivers, and two aluminum wheels on the robot to finish the hardware part. After that, we verify our theory by software
基於傳統方法和卷積神經網絡的心房顫動和充血性心力衰竭識別之比較;Comparison of Atrial Fibrillation and Congestive Heart Failure Recognization Based on Traditional Method and Convolutional Neural Network
No full text[[abstract]]本研究針對兩種疾病:心房纖維顫動(Atrial Fibrillation)、心力衰竭(Heart Failure)，探討相對於正常心律（Normal Sinus Rhythm）的疾病辨識方法。本研究提出兩個藉由生理訊號辨識疾病之方法，並對這兩種方法比較與探討。由於正常人與患有心臟疾病的人在心率變異上有極顯著的差異，因此一方面我們可擷取大量心率變異性特徵並且搭配分類器來識別有無病症。另一方面，最近深度學習廣泛應用在數位訊號分析領域，本研究也探討深度學習適用於生理時間序列的可行性。 本研究提出以心電圖(Electrocardiogram)上RR間隔為主要分析的訊號。以傳統方法包含特徵擷取、特徵萃取、分類。而所擷取的特徵有:時間統計特徵、頻域特徵、熵特徵、Poincare、高階頻譜特徵，特徵萃取則是使用線性鑑別分析(Linear Discriminant Analysis)，分類器選擇支持向量機(Support Vector Machine)、類神經網絡(Neural Network)。此外，在分三類的項目中，提出使用兩階段式分類法能更加提升分類結果的正確率。 本研究另一個系統採用卷積神經網絡(Convolutional Neural Network)，以RR間隔和其短時間傅立葉轉換(Short Time Fourier Transform)為對象，企圖從中提取特徵並分類。本篇使用Leave-one-out 以及Two-fold 做驗證，深入比較兩種系統的優劣和效能。 研究結果顯示，使用卷積神經網絡之系統優於傳統系統，在AF和NSR的分類以Leave-one-out和Two-fold驗證出最佳正確率分別為97.44%、98.08%，CHF和NSR分類的最佳正確率分別為93.10%、95.69%，兩階段分三類則分別可達到為91.00%、92.50%。
This study focus on two diseases: atrial fibrillation (AF) and congestive heart failure (CHF), and try to develop a disease identification system to identify them from normal sinus rhythm. This study presents two methods that identify disease based on physiological signals. The performance of these two systems are compared. There is a significant difference in heart rate variability between people who is normal and who has heart disease. On the other hand, we can extract a variety of characteristics and use a classifier to identify the presence of disease. On the other hand, deep learning is widely used in the field of digital signals analysis, this study also explores the feasibility of applying deep learning to physiological time series. This study uses the RR interval extracted from the electrocardiogram (ECG) as the main signal for analysis. The traditional method includes feature extraction, feature discrimination, and classification. We employ time statistical, frequency domain, Entropy, Poincare, High-order spectral features and use the linear discriminant analysis (LDA) to increase the discrimination capability. Selecting support vector machine (SVM) and neural network (NN) are employed the classifiers. In addition, in the classification of three categories, we proposed a two-stage classification method to further improve the accuracy of classification results. In this study, we also propose another system using convolutional neural network (CNN). The RR interval and its short time fourier transform (STFT) are ued as inputs and their features are extracted for classification. The verification methods include leave-one-out and two-fold cross-validation. The advantages and disadvantages of both systems are compared. The results show that the system using convolution neural network is superior to the traditional system. The classification of AF and NSR get accuracies of 97.44% and 98.08%, using leave-one-out and two-fold, respectively. The best result for CHF and NSR classification, accuracy rates are 93.10% and 95.69%, respectively. Moreover, using two-stage scheme to classify three classes can achieve accuracy rates of up to 91.00% and 92.50% ,respectively
視覺機器手臂控制的伺服系統研究;A Development of Eye-on-hand Robotic Arm System
No full text[[abstract]]關於醫療工程的研究，近年來是相當熱門的領域之一，機器手臂的發明帶來許多幫助，多為應用於工業與醫療領域。在醫療方面，微創手術是個偉大的發明，使醫療進步一大步，也因為精細度高，所以導致設備昂貴。我們希望機器手臂能普及化，在偏遠地區也可以享用，進可以代替手術副手的工作，如：固定關節傷口、清理血液，分擔醫療工作量，避免太過操勞。也可以幫助救援活動，在救護車上幫忙患者做初步檢查，協助醫護人員救護，促使執行效率更好，搶黃金救援時間。本論文，我們利用微型電腦(Raspberry Pi)來控制整個機器手臂，透過攝影機偵測到目標物位置資料，經過正向運動學演算，並傳輸訊號控制馬達運動。在機器手臂上裝4組重量輕且體積小的機電感測器，透過姿態資料，取得機器手臂的姿態，再透過無線網路連接上電腦，所以感測器姿態以無線網路傳輸至電腦，在電腦上執行Vpython，呈現機器手臂的同步姿態模擬。我們不僅有系統的控制，還有模擬端的參考，審視回饋的功能，藉此降低失誤對病患造成傷害。
In recent years, research in medical engineering has become quite popular. In the medical area, the greatest invention is minimally invasive surgery. It’s a huge progress for medical engineering. But the procedure is expensive because it requires high accuracy. In order to develop a low cost robotic arm to help doctors do work such as sewing wounds, or cleaning blood off of patients. By using the robotic arm, doctors can perform their job more effectively; they can save time doing preliminary checks on patients. In the thesis, Raspberry Pi is used to control the whole robotic arm. Using a webcam to detect a target, and then the Raspberry Pi for image processing to control the servo motors. The controller calculates the robotic arm’s attitude from the visual data and then send it to a computer by Wi-Fi. Finally, the synchronous simulation of the robotic arm’s attitude is presented on screen by Vpython. It not only depends on the control system but also reference for synchronous simulation. It gains feedback by placing the servo motor and sensor together. Thus the robot can avoid making mistakes on patients
基於四元樹切割和聚焦分析之多聚焦影像融合;Multi-focus image fusion based on quadtree segmentation and focus analysis
No full text[[abstract]]隨著科技的發展，數位相機除了扮演著重要角色，其性能也有顯著的提升。然而，在拍攝影像期間有著眾多因素會影響影像清晰度，例如：焦距、光圈、鏡頭的種類……等等。因此，我們常藉著拍攝多張影像，各自取出每一張影像有對焦 (in focus)的區域再將其融合為一張全聚焦 (all in focus)影像。本論文提出一個全聚焦影像合成的演算法，係以四元樹 (quadtree)分割為基礎，共分為兩大部分，第一部分為前處理，藉由計算各個子區塊峰度 (Kurtosis)來得到初始的融合影像以及初始深度圖；第二部分為後處理，用來修正初始深度圖上的錯誤深度，再利用修改過後的深度圖優化初始的融合影像。根據實驗結果，本論文提出的演算法所產生出的合成影像有不錯的品質。此外，本論文共用了四種客觀品質評估指標來評斷融合影像的品質。其結果顯示本論文提出的演算法所產生之融合影像在四個評估指標下，大部分是優於市售軟體的。
When we take pictures, there are so many factor may influence the clarity of photos such as aperture and shutter. In this paper, we concern about doing multi-focus image fusion based on quadtree segmentation. The multi-focus image sets are captured by adjusting the positions of the imaging plane step by step. In this way, the objects at different depths will have their best focus at different images. Our target is to get the all-focus image and estimate the corresponding depth image for this multi-focus image set. First, we utilize kurtosis to determine whether a block should be subject to further splitting. After all the blocks are segmented completely, the final region definitions are used to perform WTA (Winner-take-all) for choosing image pixels of best focus from the image set. Depth image then corresponds to the label image by which image pixels of best focus are chosen. At last, the experimental results show that the algorithm we proposed has better performances compared to commercial software