Dissertations / Theses on the topic 'Roughness prediction'

Thesis (Ph. D.)--Civil and Environmental Engineering, Georgia Institute of Technology, 2007.<br>Dr. J. David Frost, Committee Chair ; Dr. G. Wayne Clough, Committee Co-Chair ; Dr. William F. Marcuson III, Committee Member ; Dr. Paul W. Mayne, Committee Member ; Dr. Susan Burns, Committee Member.

The wind power prediction plays a vital role in a wind power project both during the planning and operational phase of a project. A time series based wind power prediction model is introduced and the simulations are run for different case studies. The prediction model works based on the input from 1) nearby representative wind measuring station 2) Global average wind speed value from Meteorological Institute Uppsala University mesoscale model (MIUU) 3) Power curve of the wind turbine. The measured wind data is normalized to minimize the variation in the wind speed and multiplied with the MIUU to get a distributed wind speed. The distributed wind speed is then used to interpolate the wind power with the help of the power curve of the wind turbine. The interpolated wind power is then compared with the Actual Production Data (APD) to validate the prediction model. The simulation results show that the model works fairly predicting the Annual Energy Production (AEP) on monthly averages for all sites but the model could not follow the APD trend on all cases. The sensitivity analysis shows that the variation in production does not depend on ’the variation in roughness class’ nor ’the difference in distance between the measuring station and the wind farm’. The thesis has been concluded from the results that the model works fairly predicting the AEP for all cases within the variation bounds. The accuracy of the model has been validated only for monthly averages since the APD was available only on monthly averages. But the accuracy could be increased based on future work, to assess the Power law exponent (a) parameter for different terrain and validate the model for different time scales provided if the APD is available on different time scales.

Irregularities in pavement profiles that exceed standard thresholds are usually rectified using a Diamond Grinding Process. Diamond Grinding is a method of Concrete Pavement Rehabilitation that involves the use of grinding wheels mounted on a machine that scraps off the top surface of the pavement to smooth irregularities. Profile Analysis Software like ProVAL© offers simulation modules that allow users to investigate various grinding strategies and prepare a corrective action plan for the pavement. The major drawback with the current Smoothness Assurance Module© (SAM) in ProVAL© is that it provides numerous grind locations which are both redundant and not feasible in the field. This problem can be overcome by providing a constrained grinding model in which a cost function is minimized; the resulting grinding strategy satisfies requirements at the least possible cost. Another drawback with SAM exists in the built-in grinder models that do not factor in the effect of speed and depth of cut on the grinding head. High speeds or deep cuts will result in the grinding head riding out the cut and likely worsening the roughness. A constrained grinding strategy algorithm with grinder models that factor in speed and depth of cut that results in cost effective grinding with better prediction of post grind surfaces through simulation is developed in this work. The outcome of the developed algorithm is compared to ProVAL's© SAM results.<br>Master of Science

The surface quality is identified by surface roughness parameters. The average surface roughness (Ra) is used in this study, as it is the most commonly used roughness parameter in the industry. A particular curved cavity of a forging die is selected for the experimental study. Different milling methods are tested. The comparison studies are conducted between 3-axes and 5-axes milling, linear and circular tool path strategies and down and up milling. According to the experimental study, appropriate method is determined for the milling of a particular curved cavity of a forging die. The experimental analysis based on design of experiments (DOE) has been employed by considering cutting speed, feed per tooth and stepover parameters. Multiple linear regression technique is used by which a mathematical formula has been developed to predict the Ra values for milling parameters. The results of the mathematical formula are controlled by conducting test experiments and good correlations are observed between the results of the formula and the results of test experiments.

Pavement roughness has been found to have an effect on the coefficient of friction measured with the Locked-Wheel Skid Tester (LWT) with measured friction decreasing as the long wave roughness of the pavement increases. However, the current pavement friction standardization model adopted by the American Society for Testing and Materials (ASTM), to compute the International Friction Index (IFI), does not account for this effect. In other words, it had been previously assumed that the IFI's speed constant (SP), which defines the gradient of the pavement friction versus speed relationship, is an invariant for any pavement with a given mean profile depth (MPD), regardless of its roughness. This study was conducted to quantify the effect of pavement roughness on the IFI's speed constant. The first phase of this study consisted of theoretical modeling of the LWT using a two-degree of freedom vibration system. The model parameters were calibrated to match the measured natural frequencies of the LWT. The calibrated model was able to predict the normal load variation during actual LWT tests to a reasonable accuracy. Furthermore, by assuming a previously developed skid number (SN) versus normal load relationship, even the friction profile of the LWT during an actual test was predicted reasonably accurately. Because the skid number (SN) versus normal load relationship had been developed previously using rigorous protocol, a new method that is more practical and convenient was prescribed in this work. This study concluded that higher pavement long-wave roughness decreases the value of the SP compared to a pavement with identical MPD but lower roughness. Finally, the magnitude of the loss of friction was found to be governed by the non-linear skid number versus normal load characteristics of a pavement.

<p>Disturbances introduced in wall-bounded .ows can grow andlead to transition from laminar to turbulent .ow. In order toreduce losses or enhance mixing in energy systems, afundamental understanding of the .ow stability is important. Inlow disturbance environments, the typical path to transition isan exponential growth of modal waves. On the other hand, inlarge disturbance environments, such as in the presence of highlevels of free-stream turbulence or surface roughness,algebraic growth of non-modal streaks can lead to transition.In the present work, the stability of wall-bounded .ows isinvestigated by means of linear stability equations valid bothfor the exponential and algebraic growth scenario. Anadjoint-based optimization technique is used to optimize thealgebraic growth of streaks. The exponential growth of waves ismaximized in the sense that the envelope of the most ampli.edeigenmode is calculated. Two wall-bounded .ows areinvestigated, the Falkner–Skan boundary layer subject tofavorable, adverse and zero pressure gradients and the Blasiuswall jet. For the Falkner–Skan boundary layer, theoptimization is carried out over the initial streamwiselocation as well as the spanwise wave number and the angularfrequency. Furthermore, a uni.ed transition-prediction methodbased on available experimental data is suggested. The Blasiuswall jet is matched to the measured .ow in an experimentalwall-jet facility. Linear stability analysis with respect tothe growth of two-dimensional waves and streamwise streaks areperformed and compared to the experiments. The nonlinearinteraction of introduced waves and streaks and the .owstructures preceding the .ow breakdown are investigated bymeans of direct numerical simulations.</p><p>Descriptors: Boundary layer, wall jet, algebraic growth,exponential growth, lift-up e.ect, streamwise streaks,Tollmien-Schlichting waves, free-stream turbulence, roughnesselement, transition prediction, Parabolized StabilityEquations, Direct Numerical Simulation.</p>

<p>A 6082 aluminium alloy has been characterized with regard to the influence of surface roughness on fatigue strength.</p><p>Fatigue life testing of smooth specimens was used to establish reference curves for the material in extruded and forged T6 condition. The extruded material was found to have better fatigue strength than the forged material, although the cyclic stress-strain response was similar for both. The forged material was tested in T5, T6 and T7 tempers, showing no significant difference in fatigue strength.</p><p>Surface roughness was created by circumferential grinding of cylindrical test specimens, and the surface topography was measured using a white light interferometry microscope. The measurements proved to be accurate, although errors were observed for certain surface features. Residual stresses were quantified by X-ray diffraction. Compressive residual stresses of around 150 MPa were found in both rough and smooth specimens. Load cycling did not significantly alter the surface residual stresses.</p><p>Stress solutions ahead of all major surface grooves were found using a linear elastic material model. Estimates of cyclic stresses and strains were calculated in the notch roots using different Neuber corrections of the linear solution. The results were compared to finite element analysis employing a bilinear kinematic hardening model. A generalized version of the Neuber correction was found to be within 20% of the nonlinear finite element results.</p><p>Several empirical models for the notch sensitivity factor were investigated. These were found to be unable to describe the notch influence on fatigue life and initiation life. In order to follow this approach, it was recommended that different test specimens should be used where the short fatigue crack growth could be monitored.</p><p>It was shown that microstructural fracture mechanics theories could be used to estimate the fatigue limit of rough surfaces. In some cases, initiation from material defects or weaknesses would override the influence of surface geometry. In one specimen, the initiation appeared to have started as at a de-bonded grain, while in other cases, initiation was thought to have started at larger second phase particles embedded in notch roots. Further work in this area should focus on statistical descriptions of surface roughness, inherent material defects, and their interaction.</p>

Les nouveaux bâtiments dans les zones urbaines sont divisés en zones commerciales et en zones habitables. Les magasins commerciaux se trouvent généralement au rez-de-chaussée, les résidences privées aux étages supérieurs. Cette utilisation a révélé des perturbations critiques dues au bruit généré par les chariots de livraison lorsque les bâtiments sont principalement occupés (par exemple tôt le matin). Ces chariots génèrent des vibrations à basse fréquence (moins de 100 Hz) qui se propagent facilement dans la structure du bâtiment et aux étages supérieurs, perturbant les habitants qui s'y trouvent. Si des travaux ont été réalisés pour étudier le bruit d'impact, peu de recherches ont été menées dans le domaine du bruit de roulement dans les bâtiments. Cette thèse présente un modèle original du bruit de roulement dans les bâtiments, prenant en compte les facteurs d'influence tels que la rugosité de la roue et du sol, les propriétés matérielles de la roue et du sol, la vitesse du chariot, et la charge sur le chariot. Les irrégularités discrètes, telles que les méplats des roues et les joints de sol, sont également prises en compte. Le modèle est capable de saisir les phénomènes physiques présents dans le contact de roulement à l'intérieur, ainsi que d'estimer l'avantage relatif du niveau sonore de l'ajout d'un revêtement de sol à un système de plancher donné. Le modèle peut être utilisé comme un outil pour étudier comment différents systèmes de revêtement de sol (y compris les systèmes multicouches) réagissent à l'excitation par roulement, dans le but de développer des solutions de construction à plusieurs étages qui sont mieux équipées pour combattre ce type de source de bruit<br>New buildings in urban areas are divided into commercial and residential areas. Commercial stores are generally located on the ground floor, private residences on the upper floors. This use has revealed critical disturbances due to noise generated by delivery carts when buildings are mainly occupied (e.g. early morning). These carts generate low frequency vibrations (less than 100 Hz) which easily propagate through the building structure and to the upper floors, disturbing the residents therein. While work has been done to study impact noise, little research has been done in the area of rolling noise in buildings. This thesis presents an original model for rolling noise in buildings: taking into account the influencing factors such as the roughness of the wheel and floor, the material properties of the wheel and floor, the speed of the trolley, and the load on the trolley. Discrete irregularities, such as wheel flats and floor joints, are also taken into account. The model is capable of capturing the physical phenomena present in the rolling contact indoors, as well as estimating the relative noise benefit of adding a floor covering to a given floor system. The model can be used as a tool to study how different flooring systems (including multi-layer systems) respond to rolling excitation, with the aim of developing multi-story building solutions that are better equipped to combat this type of noise source

Made available in DSpace on 2014-06-11T19:24:47Z (GMT). No. of bitstreams: 0 Previous issue date: 2005-01-19Bitstream added on 2014-06-13T18:21:15Z : No. of bitstreams: 1 franca_tv_me_bauru.pdf: 6693171 bytes, checksum: a1e62f81ea86d1eeed97e598cb2bb8f5 (MD5)<br>Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)<br>Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)<br>Atualmente, a fabricação é caracterizada pela sua complexidade, pluralidade de disciplinas e crescente demanda de novas ferramentas e técnicas para a solução de difíceis problemas. As redes neurais artificiais oferecem uma nova e diferente alternativa para investigar e analisar os desafiadores tópicos relacionados à manufatura. Desta forma, estudou-se neste trabalho os assuntos relacionados à aplicação das redes neurais na predição da circularidade e rugosidade da peça retificada pela análise de algumas variáveis de saída do processo. Foram empregados nos ensaios de usinagem: um fluido de corte (óleo emulsionável), um rebolo superabrasivo de CBN com ligante vitrificado e peças temperadas e revenidas de aço VC-131. Este trabalho também utilizou outras tecnologias de otimização do processo de retificação, tais como: a utilização de defletores aerodinâmicos para a quebra da camada de ar e a refrigeração otimizada por meio de um jato de fluido direcionado. Os ensaios de usinagem foram realizados para gerar a base de dados utilizada nos testes das redes neurais (ensaios computacionais). Fez-se portanto, diversos experimentos variando-se a velocidade de avanço, ou mergulho do rebolo na peça. As variáveis de saída analisadas que serviram de dados de entrada para a RNA foram: a força tangencial de corte (Ft), a energia específica de retificação (u), o desgaste diametral do rebolo, o parâmetro DPO e a emissão acústica (EA). A rugosidade e circularidade foram utilizadas para o treinamento das RNA s. Nos testes computacionais, foram analisadas duas bases de dados: a primeira referente às médias de todos os 40 ciclos de retificação, já a segunda utilizou todos os valores destes 40 ciclos. Ainda foram examinadas diferentes combinações de dados de entrada para verificar a influência do parâmetro DPO na predição. Os resultados...<br>Nowadays, the manufacturing is characterized by its complexity, plurality of subjects and increasing demand of new tools and techniques for the solution of difficult problems. Artificial neural nets propose a new and different alternative to investigate and analyze the challenging topics related to the manufacturing. The objective of this work is to study the use of artificial neural nets in the prediction of roundness and roughness of a ground workpiece. It was used a CBN wheel, emulsion oil and workpieces made of VC-131 steel. This work also used other technologies of grinding optimization, such as: the use of a coolant shoe to break the air curtain layer in addition and the high pressure fluid jet. Grinding tests had been carried through to generate the database used in the artificial neural nets (computational tests). Different feed rates were used in these experiments to generate outputs such as: tangential cutting force (Ft), specific energy of grinding (u), diametrical wear of the wheel, DPO parameter and acoustic emission (EA). The roughness and roundness were used to train the RNA's. In the computational tests, it was verify the influence of the DPO parameter in the prediction as well as two different databases. The results suggest that this parameter (DPO) was not able to substitute the tangential cutting force (Ft) and the acoustic emission (EA) in the prediction. Moreover, it was verify the need of an input that represents the dynamic stiffness of the machine-tool-workpiece system to improve the roundness prediction.

Le frottement à l’interface des surfaces influence les performances des éléments mécaniques. Le frottement a été étudié expérimentalement dans la plupart des études. Dans ce travail, le frottement est prédit à l'aide d'une simulation numérique dans des conditions de contact rugueux avec une lubrification élastohydrodynamique (EHL). La technique classique Multigrille fonctionne bien pour limiter le temps de calcul et les besoins en mémoire. Cependant, le choix de la grille grossière a une influence importante sur la robustesse du code et son efficacité pour résoudre le problème brut. Dans la première partie de ce travail, une méthode de construction de grille grossière proposée par Alcouffe et al. est implémenté dans le code EHL Multigrille indépendamment du temps. Ensuite ce solveur modifié est étendu aux cas transitoires pour résoudre le problème de contact avec rugosité. La courbe de frottement est généralement représentée en fonction du « ratio », le rapport entre l'épaisseur du film d'huile et la valeur moyenne quadratique de la rugosité de la surface. Cependant, ce paramètre est moins approprié pour tracer les variations de frottement dans des conditions de haute pression (régime élasto piézo-visqueux). Dans la deuxième partie de ce travail, le coefficient de frottement est calculé à l'aide du code EHL modifié pour de nombreuses conditions de fonctionnement ainsi que pour les paramètres d'ondulation de surface. Les résultats de la simulation montrent qu'il n'y a pas de courbe de frottement unique lorsque l'ancien paramètre « ratio »est utilisé. En se basant sur la théorie de la réduction d'amplitude, un nouveau paramètre de dimensionnement qui dépend des conditions de fonctionnement et des paramètres d'ondulation est trouvé, ce qui peut donner une courbe de frottement unique pour les situations de haute pression. Pour les surfaces rugueuses plus complexes, une méthode basée sur la densité spectrale de puissance (PSD) est proposée pour prédire les variations de frottement dans la troisième partie de ce travail. La rugosité artificielle de la surface est utilisée pour tester d’abord la méthode de prédiction rapide. Un bon accord est trouvé entre la simulation numérique complète et cette prédiction rapide. La méthode de prédiction rapide est ensuite appliquée pour analyser la variation de frottement de la rugosité de surface mesurée. Le nouveau paramètre d’échelle et l’augmentation du frottement prédite par la méthode PSD montrent une bonne précision technique pour une utilisation pratique<br>The friction of interfacial surfaces greatly influences the performance of mechanical elements. Friction has been investigated experimentally inmost studies. In this work, the friction is predicted by means of numerical simulation under an elastohydrodynamic lubrication (EHL) rough contact condition. The classical Multigrid technique performs well in limiting computing time and memory requirements. However, the coarse grid choice has an important influence on code robustness and code efficiency to solve the rough problem. In the first part of this work, a coarse grid construction method proposed by Alcouffe et al. is implemented in the current time-independent EHL Multi-Grid code. Then this modified solver is extended to transient cases to solve the rough contact problem. The friction curve is usually depicted as a function of “lambda ratio”, the ratio of oil film thickness to root-mean-square of the surface roughness. However this parameter is less suitable to plot friction variations under high pressure conditions (piezoviscous elastic regime). In the second part of this work, the friction coefficient is computed using themodified EHL code for many operating conditions as well as surface waviness parameters. Simulation results show that there is no single friction curve when the old parameter "lambda ratio" used. Based on the Amplitude Reduction Theory, a new scaling parameter depends on operating condition and waviness parameters is found, which can give a unified friction curve for high pressure situation. For more complex rough surfaces, a power spectral density (PSD) based method is proposed to predict friction variations in the third part of this work. The artificial surface roughness is employed to test the rapid prediction method firstly. Good agreement is found between the full numerical simulation and this rapid prediction. Then the rapid prediction method is applied to analyze the friction variation of measured surface roughness. Both the new scaling parameter and the friction increase predicted by the PSD method show good engineering accuracy for practical use

Resumo: Atualmente, a fabricação é caracterizada pela sua complexidade, pluralidade de disciplinas e crescente demanda de novas ferramentas e técnicas para a solução de difíceis problemas. As redes neurais artificiais oferecem uma nova e diferente alternativa para investigar e analisar os desafiadores tópicos relacionados à manufatura. Desta forma, estudou-se neste trabalho os assuntos relacionados à aplicação das redes neurais na predição da circularidade e rugosidade da peça retificada pela análise de algumas variáveis de saída do processo. Foram empregados nos ensaios de usinagem: um fluido de corte (óleo emulsionável), um rebolo superabrasivo de CBN com ligante vitrificado e peças temperadas e revenidas de aço VC-131. Este trabalho também utilizou outras tecnologias de otimização do processo de retificação, tais como: a utilização de defletores aerodinâmicos para a quebra da camada de ar e a refrigeração otimizada por meio de um jato de fluido direcionado. Os ensaios de usinagem foram realizados para gerar a base de dados utilizada nos testes das redes neurais (ensaios computacionais). Fez-se portanto, diversos experimentos variando-se a velocidade de avanço, ou mergulho do rebolo na peça. As variáveis de saída analisadas que serviram de dados de entrada para a RNA foram: a força tangencial de corte (Ft), a energia específica de retificação (u), o desgaste diametral do rebolo, o parâmetro DPO e a emissão acústica (EA). A rugosidade e circularidade foram utilizadas para o treinamento das RNA’s. Nos testes computacionais, foram analisadas duas bases de dados: a primeira referente às médias de todos os 40 ciclos de retificação, já a segunda utilizou todos os valores destes 40 ciclos. Ainda foram examinadas diferentes combinações de dados de entrada para verificar a influência do parâmetro DPO na predição. Os resultados... (Resumo completo, clicar acesso eletrônico abaixo)<br>Abstract: Nowadays, the manufacturing is characterized by its complexity, plurality of subjects and increasing demand of new tools and techniques for the solution of difficult problems. Artificial neural nets propose a new and different alternative to investigate and analyze the challenging topics related to the manufacturing. The objective of this work is to study the use of artificial neural nets in the prediction of roundness and roughness of a ground workpiece. It was used a CBN wheel, emulsion oil and workpieces made of VC-131 steel. This work also used other technologies of grinding optimization, such as: the use of a coolant shoe to break the air curtain layer in addition and the high pressure fluid jet. Grinding tests had been carried through to generate the database used in the artificial neural nets (computational tests). Different feed rates were used in these experiments to generate outputs such as: tangential cutting force (Ft), specific energy of grinding (u), diametrical wear of the wheel, DPO parameter and acoustic emission (EA). The roughness and roundness were used to train the RNA's. In the computational tests, it was verify the influence of the DPO parameter in the prediction as well as two different databases. The results suggest that this parameter (DPO) was not able to substitute the tangential cutting force (Ft) and the acoustic emission (EA) in the prediction. Moreover, it was verify the need of an input that represents the dynamic stiffness of the machine-tool-workpiece system to improve the roundness prediction.<br>Orientador: Paulo Roberto de Aguiar<br>Coorientador: Eduardo Carlos Bianchi<br>Banca: Leonardo Roberto da Silva<br>Banca: Rodrigo Eduardo Catai<br>Mestre

Le toilage appartient à la famille des procédés de finition – superfinition abrasives. Il permet d’obtenir des surfaces ayant une très bonne rugosité. En outre, il s’agit d’une très bonne alternative pour la production à grande échelle, car c’est un procédé très efficace et stable. Malheureusement, le toilage reste un procédé difficile à optimiser et nécessite de nombreux essais avant de trouver les conditions optimales. La modélisation numérique du procédé de toilage est donc une excellente alternative aux optimisations empiriques qui prennent beaucoup de temps. Ce travail propose des méthodes de modélisation pour prédire la rugosité générée par le procédé de toilage. Après une campagne expérimentale visant à comprendre l’effet des paramètres du toilage sur l’intégrité de surface (rugosité et contraintes résiduelles), l’objectif de la thèse était de construire un modèle capable de prédire les rugosités générées par le procédé de toilage. Un nouveau modèle 3D cinématique a été développé. Le modèle se base sur la description cinématique du procédé de toilage et l’utilisation d’une mesure de toile réelle. Il consiste à simuler le rayage multi passe d’une toile sur une surface. La trajectoire de rayage de la toile est déterminée par la cinématique de toilage et l’interaction entre la toile abrasive et la surface de pièce usinée est supposée parfaite (opérations booléennes). Une première comparaison des rugosités prédites par le modèle et des rugosités expérimentales permet d’identifier les pistes d’amélioration du modèle pour une prédiction de rugosité plus réaliste. Pour tenir compte de la flexibilité du système galet-toile, un traitement numérique des topographies de toile a été proposé. Ce traitement permet d’aligner les grains à la même hauteur. Deux méthodes d’alignement de grains ont été explorées et comparées. L’effet de ces deux méthodes d’alignement sur les résultats de prédiction de rugosité a été étudié. Ceci a été suivi par une étude de sensibilité du modèle vis-à-vis des vitesses cinématiques. Cette étude de sensibilité a permis de conduire à des simplifications du modèle. Ces simplifications permettent de réduire le temps de simulation de 12 heures à moins de 3 minutes. Ainsi, le modèle 2D (adaptation du modèle 3D) a été développé. L’idée du modèle consiste à négliger l’effet du mouvement d’oscillation et de facto de simuler des rayages monodirectionnels. L’effet des paramètres de toilage (taille de grains, force de toilage et dureté de galet) a été étudié. Ensuite, une discussion sur la sensibilité du modèle vis-à-vis des dispersions de toile et des propriétés mécaniques de la pièce a été abordé. Les résultats de simulation permettent d’observer les mêmes tendances expérimentales, par contre les rugosités prédites sont plus faibles que les rugosités expérimentales. Ces observations ouvrent vers des pistes d’amélioration du modèle, à travers l’amélioration de la compréhension et la modélisation de l’indentation entre la toile et la pièce durant le procédé de toilage. Enfin, ce travail de thèse traite de la caractérisation de la tenue en fatigue (expérimentalement) en flexion rotative des éprouvettes obtenues par tournage dur et par tournage dur + toilage<br>The belt finishing belongs to the family of abrasive finishing processes. It allows obtaining surfaces with a very good roughness. Moreover, it is a very good alternative for large-scale production, as it is a very efficient and stable process. Unfortunately, belt finishing remains a difficult process to optimize and requires many trials before finding the optimal conditions. Therefore, numerical modeling of the belt finishing process is an excellent alternative to time-consuming empirical optimizations. This work proposes modeling methods to predict the roughness generated by the belt finishing process. After an experimental campaign aimed at understanding the effect of belt finishing parameters on surface integrity (roughness and residual stresses), the objective of the thesis was to build a model capable of predicting the roughness generated by the belt finishing process. A new 3D kinematic model was developed. The model is based on the kinematic description of the belt finishing process and the use of a real abrasive belt measurement. It consists in simulating the multi-pass scratching of a belt on a surface. The scratching trajectory of the abrasive belt is determined by the kinematics of the process, and the interaction between the abrasive belt and the machined surface is assumed perfect (Boolean operations). A first comparison of the roughness predicted by the model and the experimental roughness allows us to identify ways to improve the model for a more realistic roughness prediction. In order to take into account the flexibility of the roller-abrasive belt, a numerical treatment of the abrasive belt topographies was proposed. This treatment allows aligning the grains at the same height. Two grain alignment methods were explored and compared. The effect of these two alignment methods on the roughness prediction results was studied. This was followed by a sensitivity study of the model with respect to kinematic velocities. This sensitivity study led to simplifications of the model. These simplifications allow reducing the simulation time from 12 hours to less than 3 minutes. Thus, the 2D model (adaptation of the 3D model) was developed. The idea of the model is to neglect the effect of the oscillation movement, then to simulate unidirectional scratching. The effect of belt finishing parameters (grain size, toiling force and pebble hardness) was studied. Then, a discussion of the sensitivity of the model with respect to the abrasive belt dispersions and the mechanical properties of the part was addressed. The simulation results show the same experimental trends, but the predicted roughness is lower than the experimental roughness. These observations open the way to improvement of the model, through the improvement of the understanding and the modeling of the indentation between the belt and the part during the process of belt finishing. Finally, this thesis deals with the characterization of the fatigue strength (experimentally) in rotational bending of specimens obtained by hard turning and hard turning + sheet metal forming

[ES] La calidad superficial en las piezas mecanizadas depende del acabado superficial, resultado de las marcas dejadas por la herramienta durante el proceso de corte. Las aproximaciones teóricas tradicionales indican que estas marcas están relacionadas con los parámetros de corte (velocidad de corte, avance, profundidad de corte...), el tipo de máquina, el material de la pieza, la geometría de la herramienta, etc. Pero no todos los tipos de mecanizado y selección de materiales pueden dar un resultado ambiguo. Hoy en día, de manera progresiva, se están utilizando las técnicas de fresado de Alta Velocidad sobre materiales de difícil mecanizado cada vez más. El fresado de Alta Velocidad implica a un considerable número de parámetros del proceso que pueden afectar a la formación topográfica 3D de la superficie. La hipótesis de que los parámetros de rugosidad superficial dependen de las huellas dejadas por la herramienta, determinadas por las condiciones de trabajo y las propiedades del entorno, condujo al desarrollo de una metodología de investigación personalizada. Este trabajo de investigación muestra como la combinación de los parámetros, inclinación del eje de la herramienta, deflexión geométrica de la herramienta y comportamiento vibracional del entorno, influencian sobre el parámetro de rugosidad superficial 3D, Sz. El modelo general fue dividido en varias partes, donde se ha descrito la influencia de parámetros del proceso adicionales, siendo incluidos en el modelo general propuesto. El proceso incremental seguido permite al autor desarrollar un modelo matemático general, paso a paso, testeando y añadiendo los componentes que más afectan a la formación de la topografía de la superficie. En la primera parte de la investigación se seleccionó un proceso de fresado con herramientas de punta plana. Primero, se analiza la geometría de la herramienta, combinada con múltiples avances, para distinguir los principales parámetros que afectan a la rugosidad superficial. Se introduce un modelo de predicción con un componente básico para la altura de la rugosidad, obtenida por la geometría de la herramienta de corte. A continuación, se llevan a cabo experimentos más específicamente diseñados, variando parámetros tecnológicos. Esto empieza con el análisis de la inclinación del eje de la herramienta contra la mesa de fresado. Los especímenes de análisis son muestras con cuatro recorridos de corte rectos con corte en sentido contrario. Las trayectorias lineales con diferentes direcciones dan la oportunidad de analizar la inclinación del husillo de fresado en la máquina. Un análisis visual reveló diferencias entre direcciones de corte opuestas, así como marcas dejadas por el filo posterior de la herramienta. Considerando las desviaciones de las marcas de corte observadas en las imágenes de rugosidad superficial obtenidas a partir de las medidas, se introdujo un análisis sobre el comportamiento dinámico del equipo y de la herramienta de corte. Las vibraciones producen desviaciones en la mesa de fresado y en la herramienta de corte. Estas desviaciones fueron detectadas e incluidas en el modelo matemático para completar la precisión en la predicción del modelo. Finalmente, el modelo de predicción del parámetro de rugosidad Sz fue comprobado con un mayor número de parámetros del proceso. Los valores de Sz medidos y predichos, fueron comparados y analizados estadísticamente. Los resultados revelaron una mayor desviación de la rugosidad predicha en las muestras fabricadas con diferentes máquinas y con diferentes avances. Importantes conclusiones sobre la precisión del equipo de fabricación han sido extraídas y de ellas se desprende que la huella de la herramienta de corte está directamente relacionada con los parámetros de la topografía de la superficie. Además, la influencia de la huella está afectada por la geometría de la herramienta de corte, la rigidez de la herramienta y la precisión del equipo. La geometría de la herramienta conforma la base del parámetro Sz, desviación de la altura de la superficie. Las conclusiones alcanzadas son la base para recomendaciones prácticas, aplicables en la industria.<br>[CA] La qualitat superficial en les peces mecanitzades depèn de l'acabat superficial, resultat de les marques deixades per l'eina durant el procés de tall. Les aproximacions teòriques tradicionals indiquen que aquestes marques estan relacionades amb els paràmetres de tall (velocitat de tall, avanç, profunditat de tall...), el tipus de màquina, el material de la peça, la geometria de l'eina, etc. Però no tots els tipus de mecanitzat i selecció de materials poden donar un resultat ambigu. Avui en dia, de manera progressiva, s'estan utilitzant les tècniques de fresat d'Alta Velocitat sobre materials de difícil mecanització cada vegada més. El fresat d'Alta Velocitat implica un considerable nombre de paràmetres del procés que poden afectar la formació topogràfica 3D de la superfície. La hipòtesi que els paràmetres de rugositat superficial depenen de les empremtes deixades per l'eina, determinades per les condicions de treball i les propietats de l'entorn, va conduir al desenvolupament d'una metodologia d'investigació personalitzada. Aquest treball de recerca mostra com la combinació dels paràmetres, inclinació de l'eix de l'eina, deflexió geomètrica de l'eina i comportament vibracional de l'entorn, influencien sobre el paràmetre de rugositat superficial 3D, Sz. El model general va ser dividit en diverses parts, on s'ha descrit la influència de paràmetres addicionals del procés, sent inclosos en el model general proposat. El procés incremental seguit permet a l'autor desenvolupar un model matemàtic general, pas a pas, testejant i afegint els components que més afecten a la formació de la topografia de la superfície. En la primera part de la investigació es va seleccionar un procés de fresat amb eines de punta plana. Primer, s'analitza la geometria de l'eina, combinada amb múltiples avanços, per distingir els principals paràmetres que afecten la rugositat superficial. S'introdueix un model de predicció amb un component bàsic per a l'altura de la rugositat, obtinguda a través de la geometria de l'eina de tall. A continuació, es duen a terme experiments més específicament dissenyats, variant paràmetres tecnològics. Això comença amb l'anàlisi de la inclinació de l'eix de l'eina contra la taula de fresat. Els espècimens d'anàlisi són mostres amb quatre recorreguts de tall rectes amb tall en sentit contrari. Les trajectòries lineals amb diferents direccions donen l'oportunitat d'analitzar la inclinació del fus de fresat en la màquina. Una anàlisi visual revelà diferències entre direccions de tall oposades, així com marques deixades pel tall posterior de l'eina. Considerant les desviacions de les marques de tall observades en les imatges de rugositat superficial obtingudes a partir de les mesures, es va introduir una anàlisi sobre el comportament dinàmic de l'equip i de l'eina de tall. Les vibracions produeixen desviacions en la taula de fresat i en l'eina de tall. Aquestes desviacions van ser detectades i incloses en el model matemàtic per completar la precisió en la predicció de el model. Finalment, el model de predicció de el paràmetre de rugositat Sz va ser comprovat amb un major nombre de paràmetres del procés. Els valors de Sz mesurats i predits, van ser comparats i analitzats estadísticament. Els resultats van revelar una major desviació de la rugositat predita en les mostres fabricades amb diferents màquines i amb diferents avanços. Importants conclusions sobre la precisió de l'equip de fabricació han estat extretes i d'elles es desprèn que l'empremta de l'eina de tall està directament relacionada amb els paràmetres de la topografia de la superfície. A més, la influència de la empremta està afectada per la geometria de l'eina de tall, la rigidesa de l'eina i la precisió de l'equip. La geometria de l'eina conforma la base del paràmetre Sz, desviació de l'altura de la superfície. Les conclusions assolides són la base per recomanacions pràctiques, aplicables en la indústria.<br>[EN] Surface quality of machined parts highly depends on the surface texture that reflects the marks, left by the tool during the cutting process. The traditional theoretical approaches indicate that these marks are related to the cutting parameters (cutting speed, feed, depths of cut...), the machining type, the part material, the tool geometry, etc. But, different machining type and material selection can give a variable result. In nowadays, more progressively, High Speed milling techniques have been applied on hard-to-cut materials more and more extensively. High-speed milling involves a considerable number of process parameters that may affect the 3D surface topography formation. The hypothesis that surface topography parameters depends on the traces left by the tool, determined by working conditions and environmental properties, led to the development of a custom research methodology. This research work shows how the parameters combination, tool axis inclination, tool geometric deflection, cutting tool geometry and environment vibrational behavior, influence on 3D surface topography parameter Sz. The general model was divided in multiple parts, where additional process parameters influence has been described and included in general model proposed. The incremental process followed allows the author to develop a general mathematical model, step by step, testing and adding the components that affect surface topography formation the most. In the first part of the research a milling procedure with flat end milling tools was selected. First, tool geometry, combined with multiple cutting feed rates, is analyzed to distinguish the main parameters that affect surface topography. A prediction model is introduced with a basic topography height component, performed by cutting tool geometry. Next, specifically designed experiments were conducted, varying technological parameters. That starts with cutting tool axis inclination against the milling table analysis. The specimens of analysis are samples with 4 contrary aimed straight cutting paths. Linear paths in different directions give a chance to analyze milling machine spindle axis topography, as well as marks left from cutting tool back cutting edge. Considering the deviations of cutting marks observed in the images of the surface topography obtained through the measurements, the milling equipment and cutting tool dynamical behavior analysis were introduced. Vibrations produce deviations in the milling table and cutting tool. These deviations were detected and included in the mathematical model to complete the prediction model accuracy. Finally, the prediction model of the topography parameter SZ was tested with increased number of process parameters. Measured and predicted SZ values were compared and analyzed statistically. Results revealed high predicted topography deviation on samples manufactured with different machines and with different feed rates. Relevant conclusions about the manufacturing equipment accuracy have been drawn and they state that cutting tool's footprint is directly related with surface topography parameters. Besides, footprint influence is affected by cutting tool geometry, tool stiffness and equipment accuracy.<br>Logins, A. (2021). High speed milling technological regimes, process condition and technological equipment condition influence on surface quality parameters of difficult to cut materials [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/164122<br>TESIS

The fully rough form of the law of the wall is commonly used to quantify velocity profiles and associated bed shear stresses in fluvial, aeolian, and coastal environments. A key parameter in this law is the roughness length, <i>z</i>₀. Here we propose a predictive formula for <i>z</i>₀ that uses the amplitude and slope of each wavelength of microtopography within a discrete-Fourier-transform-based approach. Computational fluid dynamics (CFD) modeling is used to quantify the effective <i>z</i>₀ value of sinusoidal microtopography as a function of the amplitude and slope. The effective <i>z</i>₀ value of landscapes with multi-scale roughness is then given by the sum of contributions from each Fourier mode of the microtopography. Predictions of the equation are tested against <i>z</i>₀ values measured in  ∼ 10⁵ wind-velocity profiles from southwestern US playa surfaces. Our equation is capable of predicting <i>z</i>₀ values to 50 % accuracy, on average, across a 4 order of magnitude range. We also use our results to provide an alternative formula that, while somewhat less accurate than the one obtained from a full multi-scale analysis, has an advantage of being simpler and easier to apply.

碩士<br>國立中正大學<br>通訊工程研究所<br>107<br>In precision machining industry, computer numerical control (CNC) machine tools play an important role. In order to improve the quality and accuracy of the workpiece, the requirements for machining precision are getting higher. The selection and optimization of machining parameters are one of the main factors that affect the machining precision. Machining parameters are usually determined by the trial-and-error method. This method not only consumes time and manpower, but also results in just acceptable but not optimized parameter values. In order to solve this problem, this thesis uses neural network based machine learning approach to build a machining precision prediction model. The prediction model takes machining parameters as features and surface roughness and contour error as labels, and then uses the measurement results of machining experiments to train classification and regression neural networks respectively. The constructed prediction model can predict the influence of different machining parameters on surface roughness and contour error, and assist the machining operator to select the appropriate machining parameters.

碩士<br>中原大學<br>工業與系統工程研究所<br>99<br>Intelligent control theory has been studied in modern research and widely applied in various fields. With the rapid technological advances, however, intelligent control system becomes more complex and is difficult for researchers to define accurately. As a result, data mining is proposed to solve problems in the relevant areas. In this study, the neural algorithm as the main theory background, Data mining is assisted into neural prediction system, the data analysis and identify each one of the relationship, When the target can be determined by making the relationship between the intensity of each data to distinguish reliable and unreliable information, The study assumed the use of reliable information as neural prediction system for the training set of data, through the neural network training and testing to predict the desired results. As data mining has reliable information about the data analysis into the characteristics, and use information to establish its relationships, then this relationships to identify trends, this method has the data into the ability of future trends, therefore the data mining activities for a variety of area, but data mining models in the construction process will be a major problem encountered：How the rules between each data set up? this is the data mining process will build the difficulties, most of the previous solution to employ experts, expert's experience as a basis for analysis, this study used data mining of Bayesian theory, hire experts to solve problems, analyze data using Bayesian theory and experimental results verify its reliability. finally, the data mining algorithms integrated with neural algorithms, develop a comprehensive system of intelligent prediction. For the demonstrate that the proposed method of reliability and accuracy, in this study, combined with the development of data mining algorithms and neural projections into the instance to the surface roughness, Build a surface roughness prediction system, and the return of the past as a basis for fuzzy prediction system and compare the accuracy of the forecasts, Finally, the use of hypothesis testing comparing the two kinds of intelligent forecasting system of the significant differences, to verify the accuracy of forecasting system

碩士<br>國立中興大學<br>機械工程學系所<br>107<br>In this study, the spindle and vise vibrations as well as the spindle current were measured synchronously during the process of milling Inconel 718. The surface roughness (represented by Ra) of workpiece was investigated by determining the correlation among the Ra, the signals of vibration, the cutting parameters, and the current signals, under the different combinations of cutting parameters. The prediction models of workpiece surface roughness were built through the Elman neural network. In the experiment, the features of signals were extracted through the Empirical Mode Decomposition (EMD), envelope analysis, fast Fourier transform(FFT), and the determination of root-mean-square, kurtosis, skewness, and multiscale entropy. The Pearson correlation analysis was utilized to select the features that have high correlation with the Ra value. The Elman neural network model is then trained by the selected features and employed for predicting the workpiece surface roughness. The surface roughness prediction model was employed to optimize the cutting parameters according to the constraints. In this study, the feed rate is maximized under the constraints of certain Ra values in the optimization process. The optimal combination of cutting parameters were obtained through the process of genetic algorithm and the particle swarm algorithm. The optimized cutting parameters were validated by the experiment result. The result of using different signal features and different optimization algorithms are also compared and discussed.

碩士<br>淡江大學<br>土木工程學系碩士班<br>95<br>Performance predictive models have been used in various pavement design, evaluation, rehabilitation, and network management activities. As pavement design evolves from traditional empirically based methods toward mechanistic-empirical, the equivalent single axle load (ESAL) concept used for traffic loads estimation is no longer adopted in the recommended Mechanistic-Empirical Pavement Design Guide. The success of the new design guide considerably depends upon the accuracy of pavement performance predictions. Thus, this study will first investigate its goodness of fit and strive to develop improved performance prediction models for pavement roughness using the Long-Term Pavement Performance (LTPP) database (http://www.datapave.com or LTPP DataPave Online). Exploratory data analysis (EDA) of the response variables indicated that the normality assumption with random errors and constant variance using conventional regression techniques might not be appropriate for prediction modeling. Therefore, generalized linear model (GLM) and general additive model (GAM) along with Poisson distribution were adopted in the subsequent analysis. Box-Cox power transformation technique, visual graphical techniques, as well as the systematic statistical and engineering approach proposed by Lee were frequently adopted during the prediction modeling process. By keeping only those parameters with significant effects and reasonable physical interpretations in the model, various tentative performance prediction models were developed. The goodness of the model fit was further examined through the significant testing and various sensitivity analyses of pertinent explanatory parameters. The tentatively proposed predictive models appeared to reasonably agree with the pavement performance data, although their further enhancements are possible and recommended.

碩士<br>國立虎尾科技大學<br>機械與機電工程研究所<br>97<br>In this dissertation, high performance in IGA arises mainly from intelligent crossover based on orthogonal experimental designs. Using OED with both orthogonal array (OA) and factor analysis, it can analyze the effect of several factors simultaneously. Accurate estimation of surface roughness of workpieces in turning operations play an important role in the manufacturing industry. It is used to assess the workpiece in the end milling process of performance criterion. For example：surface seals, ball bearing, gear, cam and journal. Surface roughness has very great impact for the equipment. In this paper, Intelligent Genetic Algorithm(IGA) with Fuzzy Neural Network(FNN) is used to model and predict the workpiece surface roughness for the end milling process. IGA is powerful by using the Orthogonal Experimental Design’s algorithm. It can effectively reason to near-optimal solutions. In this paper, the model of the FNN uses previously researcher’s 125 training data and 18 validation data. There are 50 parameters to be optimized. Experimental results show that IGA with FNN model can improve the accuracy of modeling and prediction, and outperforms the ANFIS methods by MATLAB and reported recently in the literature. We use MATLAB and built C++ complied function in our application for improving computation time.

In any machining process, apart from obtaining the accurate dimensions, achieving a good surface quality is also of utmost importance. A machining process involves many process parameters which directly or indirectly influence the surface quality of the product. Surface roughness and waviness in turning process are caused due to various parameters of which feed, speed, depth of cut are important ones. A precise knowledge of these optimum parameters would facilitate reduce the machining costs and improve product quality. Extensive study has been conducted in the past to optimize the process parameters in any machining process to have the best product. Current investigation on turning process is a Taguchi optimization technique applied on the most effective process parameters i.e. feed, cutting speed and depth of cut while machining SS 304 as the work piece with brazed cutting tool. Main effect plots are generated and analyzed to find out the relationship between them. The details of experimentation and analysis are given in the following context.

碩士<br>國立臺灣大學<br>機械工程學研究所<br>101<br>Grinding processing is the common processing method in the precision processing, which plays the most important part in the final stage of the processing because it relates to surface accuracy and scale quality. In general grinding, excess grinding force will rough workpiece&apos;&apos; surface, and which has to be measured after grinding. In order to judge the trends of grinding force and surface roughness efficiently, this study will cooperate with SkyMars, establishing the model of grinding force and surface roughness to estimate the quality of workpiece. This study applies different grinding depth, workpiece&apos;&apos;s speed, wheel&apos;&apos;s tangential velocity, wheel abrasive size, and abrasive type to explore the relationship between the grinding force and surface roughness. Further, this study investigates whether the range of grinding process leads abnormal phenomena such as plowing force or burning. Eventually, applying grinding verification to predict error value on grinding force model and surface roughness Ra value model is needed. According to the consequence of grinding force experiment, grinding force increases depend on increasing grinding depth, workpiece&apos;&apos;s speed and wheel abrasive; while decreasing when grinding wheel&apos;&apos;s tangential speed increases. However, grinding depth has an effect on grinding force more than workpiece&apos;&apos;s speed, wheel&apos;&apos;s tangential velocity, wheel abrasive size, and abrasive type. Based on the consequence of surface roughness Ra value experiment, surface roughness Ra value and grinding force are in direct proportion, and wheel abrasive size has great effect on surface roughness Ra value. The result shows that reducing grinding depth and workpiece&apos;&apos;s speed or gaining wheel&apos;&apos;s tangential velocity speed decrease grinding force and surface roughness Ra value. However, if the product of grinding depth and workpiece&apos;&apos;s speed are much less than wheel&apos;&apos;s tangential velocity, grinding force is hard to be predicted, and surface roughness Ra value decreases depend on increasing workpiece&apos;&apos;s speed. According to the result, the study predicts that is working region in grinding grain grit number #60~#120, depth of cut is between 5μm~10μm in grinding, wheel&apos;&apos;s tangential velocity speed is between 450m/min~1800m/min, the ratio of wheel&apos;&apos;s tangential velocity speed and workpiece&apos;&apos;s speed are less than 100. Grinding force and surface roughness Ra value can be measured in predicable processing area through formula. The result shows that the maximal error of predicable grinding force is 19%, and the maximal error of predicable surface roughness Ra value is 8.86%. Further, grinding force and surface roughness Ra value are in direct proportion. During processing, distance method is the efficient way to inform operators whether keeping processing if they are aware of abnormal phenomena such as plowing force or burning.

碩士<br>中原大學<br>工業與系統工程研究所<br>105<br>In nowadays product property of diverse and few and short life time, the need of the quality becomes more and more strict. Milling in a common way in processing, and the roughness will affect the damage level of cutting tool and the rate of manufacturing plan, to grasp the roughness will affect the quality and in order to get the accuracy of roughness lots of scholar wish to use predict, monitor the process variation and decrease the cost on the other hand. Under the different process environment and plan, there are lots of element that cannot be grasped. Therefore the sensing technology cam help to monitor the variation in the process, analysis the process effect that cause by external element. This allows to decrease the defective rate and improve the accuracy of prediction. Traditional prediction needs lot of data to model which does not fit to the present process, moreover, it needs to modify under different process. Therefore, we use Gary theory to predict and chose the strongest sensing data through the use of Grey Relational Analysis and use moving range control chart to filter the noise and put the final data in GM(1,N) and find the trend property through Accumulated Generating Operating (AGO) for prediction system. In the research we use Computer Numerical Control (CNC) milling, it is able to use less data for predicting roughness and are able to model in real time also through the setup of control boundary it will be able to prevent noise in the forecast model which effect the prediction accuracy. In order to verify the accuracy and feasibility of this research, we use two different process parameter apply few data in the prediction system and predict the roughness. The result shows the accuracy are 97.56% and 98.02% which are able to verify the accuracy and feasibility.

碩士<br>中原大學<br>工業與系統工程研究所<br>98<br>Intelligent control theory has been studied in modern research and widely applied in various fields. With the rapid technological advances, however, intelligent control system becomes more complex and is difficult for researchers to define accurately. As a result, fuzzy theory is proposed to solve problems in the relevant areas. Although the fuzzy theory can be used to solve complex issues and make accurate definitions, two main issues occur in the process of the building fuzzy systems including defining appropriate membership functions in the fuzzy IF- THEN rule bank and searching the best combined number pairs in the fuzzy membership functions. In order to handle these two issues, the current study adopted regression analysis to define the fuzzy IF-TEHN rule bank, and membership functions of Taguchi Method to search the best combined number pairs in the fuzzy system. Then the two are combined to build an effective and accurate fuzzy prediction system. The fuzzy prediction system proposed in the study was used to predict surface roughness for verifying its effectiveness and accuracy. The method composed of fuzzy regression and Taguchi Method was developed, and was proved to accurately predict surface roughness and improve predictions of the original fuzzy system. Finally, the study testified that the method composed of fuzzy regression and Taguchi Method has a better and more accurate prediction compared to regression analysis.

碩士<br>國立勤益科技大學<br>機械工程系<br>106<br>High speed and high precision machining has become the most important technology in manufacturing industry. The surface roughness of high precision components is regarded as the important characteristics of the product quality. On the other hand, regenerative chatter occurring in high speed machining could damage the machined surface and restricts the process efficiency and the longevity of cutting tools. To avoid chatter and increase machining precision, most of the engineers have selected the appropriate cutting conditions according the machining stability lobes diagram, but that cannot guarantee the surface quality in good conditions. In order to obtain better surface roughness, the selection of the cutting parameters is a prerequisite. This study was therefore aimed to investigate the influence of the machining conditions on the surface roughness. In study, the stability lobes diagram of a specific cutter was calculated based on the tool end frequency response functions, which was measured by the vibration test conducted on the milling machine. Basically, the stability lobes implied the cutting parameters (spindle speed and cutting depth) for stable machining. Next, a series of machining experiments were conducted. The surface roughness of workpieces was examined by means of the white light interferometer. 　　According to the machining tests, machined surface with or without chattering were marked on the lobes diagram for verification of the machining conditions. On the other respect, the ANOVA analysis reveals that the surface roughness show a positive correlation with machining conditions. Using multivariable regression analysis, the mathematical model describing the relationship between surface roughness (Ra) and cutting depth (a), feed rate (f) and spindle speed (s). The predicted roughness is shown to agree well with the measured roughness, an average percentage of errors of 17%. The average percentage of errors of the tool vibrations between the measurement and the predictions of exponential model is about 7.39%. Also, the tool vibration under various machining conditions are found to have a positive influence on the surface roughness (r=0.78). As a summary of this study, the stability lobes diagram was verified experimentally, which can help to identify the machining conditions without chattering in machining. Besides, a mathematical model was successfully developed to predict the surface roughness and vibration level of the tool under different cutting condition.

博士<br>國立高雄第一科技大學<br>工程科技研究所<br>95<br>Side milling is a machining process in end milling operations that utilizes the peripheral cutting edge of an end mill to perform broad surface machining upon the vertical wall of workpieces. It is a common manufacturing process used in molding and mechanical components. The corresponding performance is closely related to several cutting parameters, such as cutting speed, feed per tooth, axial depth of cut, radial depth of cut, overhang length, and flank wear of peripheral cutting edge. Conventionally, a cutting condition selected by operators is mostly based on their experiences and is not an optimal and economical solution for best performance. Therefore, it is imperative to construct a prediction model that can effectively evaluate processing results and offer a solution for optimizing cutting conditions. This research consists of two parts. First, the polynomial network is adopted to construct a prediction model for surface roughness. Second, the grey-relational analysis combined with two-stage experimental design, principle component analysis, and fuzzy logic is proposed to achieve an optimal design of cutting parameters. A series of experiments are organized in a ) 3 2 ( 12 11 36 × L orthogonal array and performed on a B8 machine center. The analysis of variance is used to convert the results into F values for each cutting parameters with which number of input parameters vi of polynomial network and thereby a prediction model for various definitions of surface roughness can be developed using an abductive modeling technique. Finally, a set of experimental data is utilized to test all the developed prediction models. The results show that the more input parameters the polynomial network is implemented, the higher capability of prediction for surface roughness it can achieve (i.e. the predicted values are closer to the experimental data). Recently, the grey-relational analysis has been widely applied in the optimal design of cutting parameters with multiple performance characteristics. In this study, a grey relational analysis is applied to a set of two –stage experiments designed to determine the cutting parameters for optimizing the side milling process with multiple performance characteristics. The cutting parameters under consideration are cutting speed, feed per tooth, axial depth of cut, radial depth of cut, overhang length, and flank wear of peripheral cutting edge. In this study, the L36 and L9 orthogonal arrays were introduced for the two-stage experimental designs and trials. Lower-the-better was used as a quality characteristic to evaluate the experimental results. It was found that applying the grey relational analysis with a two-round experimental design strategy is simple, effective and efficient in developing an optimal cutting parameters combination. The results of the confirmation test also show that this new approach can greatly improve the cutting performance of side milling process. Accordingly, the optimal combination of cutting parameters obtained using two-stage experiment approach can be closer to the ideal one. In order to objectively reflect the corresponding weighting value of each performance characteristic, the grey-relational analysis is specially integrated with the principle component analysis to deal with the optimization problems with multiple performance characteristics. A grey relational grade obtained from the grey relational analysis is used as a performance index to determine the optimal combination of cutting parameters. The principle component analysis is used to calculate the corresponding weighting value of each performance characteristic while applying grey-relational analysis. The confirmation test shows that the predicted values of grey relational grade obtained based on the optimal combination of cutting parameters is significantly close vii to the experiment values. Thus, this support the proposed application of the additive model. Furthermore, the repeatability of experiment with this combination is excellent. Side milling can be classified as heavy cutting and finishing cutting based on their types of processes. This paper presents an optimal cutting parameter design of heavy cutting in side milling for SUS304 stainless steel. The selected cutting parameters are spindle speed, feed per tooth, axial depth of cut, and radial depth of cut, while the considered performance characteristics are tool life and metal removal rate. The orthogonal array with grey-fuzzy logics is applied to optimize the side milling process with multiple performance characteristics. A grey-fuzzy reasoning grade obtained from the grey- fuzzy logics analysis is used as a performance index to determine the optimal cutting parameters. The results indicate that this optimization algorithm can effectively and swiftly acquire an optimal combination of cutting parameters. Hence, it is believed that this optimal result can be applied to practical processes to effectively reduce manufacturing cost and greatly enhance manufacturing efficiency.

碩士<br>中原大學<br>工業與系統工程研究所<br>104<br>Because of progressions of industries as well as innovations of products, the requirement for CNC (Computer Numerical Control) machining process of computer assisted manufacturing and demands for accuracy of dimensions is growing up, which makes studies about CNC milling process essential topics when it comes to automatic industries. After CNC machining, the influence of surface roughness affects the quality profoundly, and many savants wish to employ approaches of prediction to curtail the time consumed by setting parameters and obtain ideal surface roughness that not only cut down extra costs but also provide quality improvements. However, many of them still stick to utilizing sensors to accomplish those predictions, but installations of that equipment are far more arduous than that we can imagine. These methods are rare in real practice in that high costs are required. In this study, adding the factor of hardness to collect data without any sensor is presented, and higher accuracy as well as effectiveness, comparing to conditions without factor of hardness, are achieved by employing BPNN (Back-Propagation Neural Network) for prediction of surface roughness. Elements inputted into BPNN lead to reasonable predictions, but the number of them affects the accuracies of predictions. This study analyzes the correlations between input elements and hardness using grey relation analysis before training for BPNN. After removing unnecessary elements, training for BPNN and predicting, this study proves that ideal prediction can be obtained by utilizing BPNN with deletion of peripheral factors. For not only certifying but also analyzing the reliability and accuracy of the approach, this research employs correlation analysis, Back-Propagation Neural Network and t-test of hypothesis testing before establishing a surface roughness prediction system. These two systems are put to the t-test, which includes experimental group with hardness factors and control group without them. Base on the results of tests mentioned above, the prediction system with hardness factors possesses higher reliability and accuracy than those without. Keyword: hardness, surface roughness, BPNN(Back-Propagation Neural Network), prediction system.

碩士<br>中原大學<br>工業與系統工程研究所<br>101<br>With the invention of the Computer Numerical Control (CNC) and development of the material technology, the engineering of advanced manufacturing is greatly improved. These advantages accelerate the development of manufacturing industry, which becomes a stable foundation of the automation. However, CNC nowadays are widely used in different kinds of industries, which mainly focus on how to minimize the cost and maximize the production and profit. These strategies play important roles in reaching entrepreneur’s goals and visions. At this point, a right decision making at a right time and the reduction of waste are the main benchmarks of many companies. To achieve the benchmarks, quality management is the key factor. However, the inspection of quality control always takes time. To shorten the process time, the idea of “In-process Quality Monitoring System” has been applied and developed. In CNC operations, drilling is one of the most basic and common operations. However, there are few researches studying the quality measurement of this part. Presently, the manufacturing industry conducts off-line inspection to examine the surface roughness of drilling. The off-line method needed a lot of time with high cost. The surface roughness can be effectively controlled if the influencing factors can be precisely acquired. With a new in-process prediction system, the inspection cost is reduced and so the time is shortened as well. To fit the “In-process 100% inspection” system in the drilling operations, the purpose of this research is to combine the Sensing Technology and the CNC in-process prediction system of surface roughness. This system inputs the machining parameters and the signal from force sensor as the factors. A neural network is applied to construct the prediction model of the system. Then we compare the accuracy of the system with the other prediction system without sensing technology. With repetitive training and testing, the system can reaches the idea of total quality measurement which can assist the entrepreneurs to reduce the cost and shorten the lead time. The result indicates that the related influencing factors under Back Propagation Network (BPN) training prove that the cutting signal from the force sensor can be used to effectively predict the surface roughness in drilling operations. This study uses Taguchi method to find the optimal set of the network variables for BPN training, which allows the operator to immediately response via the signal from the sensor.

碩士<br>國立中興大學<br>機械工程學系所<br>105<br>This study primarily investigates the correlation among the cutting parameters, the surface roughness level of S45C steel through the milling process and the vibration signals that are recorded synchronously. With different combinations of cutting parameters, such as: feed rate of per cut, cutting depth and clamping torgue of vise, the different levels of surface roughness are predicted by using the artificial neural network (ANN). The vibrations are measured by the accelerometers which are mounted on the spindle and the vise. The features of vibration signals are extracted through utilizing the envelope analysis, RMS (root-mean-square), kurtosis, skewness, fast Fourier transform (FFT) and frequency normalization. The features of higher priority are selected based on the analysis of correlation and then collected as the input layer parameters of ANN for surface roughness prediction. The prediction accuracy and results of using different classes of input features are also disscussed and compared.

碩士<br>中原大學<br>工業工程研究所<br>100<br>With the development of the electronics industry, the demands of computer numerical control (CNC) machine for milling operations are increased. Accompanying with the requirements of product quality, CNC milling technology becomes an important issue of various automation industries. Controlling of the surface roughness has a significant impact to the quality in CNC milling operations. In order to achieve accurate required surface roughness, many researchers hope to take advantages of the predictable manner, and to reduce the additional cost and improve the quality of the product. The purpose of this study is to develop a new grey relation neural network (GRANN) model to predict the surface roughness. In the neural network, Input factors are required in order to get the predicted value, but if there are too many or less input factors, both of them may affect the accuracy of the prediction system. Therefore, in this study, a grey relation analysis is applied before the training of neural network to analyze the correlation between the input factors and the predicted target and furthermore to remove unnecessary input factor. With the significant input factors found by grey relation, a neural network would be implemented to develop the GRANN prediction system. The experiment proved that neural network training can achieve the ideal predictions after using gray relational analysis sieve factor. To prove the reliability and accuracy of the method, this study uses the GRANN model to develop the surface roughness prediction system in milling operations, and this system compares with the accuracy of the system which doesn’t use grey relation analysis. Finally, A two-sample t-test is applied to verify if the GRANN prediction system can perform better than the traditional prediction system.

碩士<br>國立虎尾科技大學<br>機械與機電工程研究所<br>97<br>Genetic algorithm used in engineering widely, including automatic control, system optimization design in recent years. Intelligent Genetic Algorithm(IGA) with Fuzzy Neural Network(FNN) is used to model and predict the workpiece surface roughness for the end milling process. IGA is powerful by using the Orthogonal Experimental Design’s algorithm. It can effectively reason to near-optimal solutions. Surface roughness is very important in the manufacturing industry. It is used to assess the workpiece in the end milling process of performance criterion. For example：surface seals, ball bearing, cam, gear and journal. Surface roughness has very great impact for the equipment. Megakaryocytes (Mks) are an considerably rare cell population that are very important in myeloid cells, and produced by Hematopoietic stem cells(HSCs) through complex development processes. In this paper, the model of the FNN uses previously researcher’s 48 training data and 24 validation data. There are 126 parameters to be optimized. Experimental results show that IGA with FNN model can improve the accuracy for modeling and prediction of surface roughness, and outperforms the ANFIS methods by MATLAB and reported recently in the literature. In this paper, IGA with FNN is used to model and predict the HSCs production. Experimental results show that IGA with FNN model can improve the accuracy for modeling and prediction of HSCs, and outperforms the ANFIS methods by MATLAB and reported recently in the literature.

碩士<br>國立虎尾科技大學<br>自動化工程研究所<br>99<br>In recent years, modeling and forecasting is widely used in many places, such as stock index, the projected seasonal waves, motor controller application object tracking ... and so on. In this paper, IGA and adaptive Neuro-Fuzzy Inference System (FNN) method as a basis to predict the distribution coefficients of volatile organic compounds for medicinal chemistry and environmental chemistry of volatile organic compounds in terms of the distribution coefficient can be used to study organic compounds in in vivo migration behavior and the environment, so this is an important issue. The current clinical drugs used mostly organic compounds, the partition coefficient month is more of drug free transfer in vivo can be successfully passed through the membrane to the bacteria, can play a role in treatment of disease, so research is the system partition coefficient into a major key drugs.We used to IGA and FNN modeling and forecasting in order to enhance its pharmaceutical success rate. As in the manufacturing sector for the importance of surface roughness, which is used to assess the quality of the workpiece on the side milling criteria. For example, the surface seal, ball bearings, gears, cams or journal and other applications, we apply the IGA to optimize the surface roughness and FNN modeling and forecasting. The results indicate that, to the structure of FNN with the IGA to search for the best FNN model parameters can be more efficient to predict the error of the smaller surface roughness and the distribution coefficient of organic compounds.

碩士<br>中原大學<br>工業工程研究所<br>100<br>With competitions between enterprises, it is importance and necessary to control costs and quality of products. Therefore, the development of prediction system is tending to become a significant role in last decades. The models of the system including model simulation, statistical prediction, regression analysis and various soft computing have been generally used in the prediction system. its accuracy is also increasing gradually. However, the major component of building the models is a large number of data. Under the circumstance, the prediction system is difficult to be implemented under the small batch and new producing lines situations. The purpose of this study is to develop a real-time adaptive learning prediction system, which can be utilized in a small sample, such as production shortfall in the early stages of producing. Moreover, this model can be adjusted in accordance with different specifications of operation conditions. This research is to develop a gray prediction system associated with gray relational analysis, which is use select the input factors of the system. Furthermore, a gray prediction theory would work as framework to build a real-time adaptive learning prediction system with a small number of data. The gray prediction implemented the result of accumulated generating operation, the calculation of gray values, effect data and establishes information anticipation with a small-scale model through the trend of characteristics of. This study applied the model to predict the surface roughness in milling operations. The correlation between cutting forces and surface roughness has been analyzed. With proper cutting force as input factors, a gray real time adaptive learning surface roughness prediction system in milling operations is built. The system does not need any training data in advance to construct the prediction model. It would only use limited data from trial production or set up to create the real-time prediction model for each production batch. To evaluate reliability and accuracy of the system, the model would be compared with a traditional neural network prediction model of a large sample. The result of comparison between these two models shows no significant difference. The time and cost of collecting and data and training model are significantly reduced.

碩士<br>國立中央大學<br>資訊工程學系<br>107<br>Industry 4.0 Smart Manufacturing is a hot topic recently. Today's global manufacturing industry is committed to smart manufacturing through industrial Internet of Things, big data analytics, and Cyber Physical System (CPS) technologies to improve product performance and product quality by saving production time and cost. This paper explores Virtual Metrology (VM) research to predict product quality before or after the production process has not been completed, without the need of product measurement. Specifically, this paper focuses on the Surface Roughness prediction of wire-cut EDM machines, and uses the 2nd order regression and the deep neural network method to predict the surface roughness of the product through production parameters before product processing. In addition, this thesis uses Markov Chain and Deep Neural Network (DNN) method to predict the surface roughness of the product through production parameters and time series data of the production process after the product is processed. In order to deal with the time series data with different lengths, this paper uses the Markov chain extraction feature to normalize the length and then predict the product quality through the neural network. We use the full factor experimental method to collect experimental data to verify the prediction accuracy of the proposed method. The experimental results show that the proposed prediction method has good mean absolute error and error rate.

碩士<br>大同大學<br>機械工程研究所<br>89<br>In this study the Grey System Theory is applied to predict the surface roughness (Ra) and redress cycle of the wheel in surface grinding process. The Ra value of workpieces varied with grinding feedrate and depth were collected in experiments. And then, the data were used to create the roughness prediction model. Furthermore, the result of the prediction will used to forecast the redress cycle time of grinding wheel.Finally, the prediction accuracy were verified by experiments to prove the proposed method is useable . Experimental results show that the method proposed here is able to get good prediction result under a threshold of 1.6 Ra value. But, if the grinding depth is smaller than 0.015mm, the grinding wheel will get into a self-redress cycle and the chatter mark can be found on the surface of workpieces。Besides, the effect of the number of original sequence numbers on the prediction accuracy was also studied in this paper by experimental approaches.；And, the results show that the created prediction model can forecasts the redress cycle of grinding wheel and keep the prediction error under 5% by using just 5 initial Ra value only.

碩士<br>國立中興大學<br>機械工程學系所<br>103<br>Technical products have characteristics such as short life cycle, high surface quality request and large quantity. Therefore, manufacturing factories gradually tend to use general types machine tool equipped with intelligent management, intelligent design and monitoring systems. Recently, with technological development, there are more and more indirect sensors which are normally used, and computer efficiency is higher than the past. The obstacles of hardware are gradually removed. Accordingly, in the field of manufacturing, many researchers use artificial neural networks (ANNs) to predict the surface roughness. They use milling experiment data for training ANNs, and get the relationship between inputs and outputs. In most of cases, well trained ANNs can predict workpiece surface roughness effectively. The prediction system can achieve the intelligentization of rising process quality. However, when their systems got the prediction of workpiece surface quality, the data did not be further used. This research project will investigate the use of ANNs with sensor fusion to construct an effective surface quality prediction system by making use of force and acoustic emission signals. The prediction system was then optimized under the constraint that the workpiece surface roughness must be lower than the requested surface roughness. The optimization system determined the best parameter combination by maximizing the material removal rate (MRR), and then relaying this information to the machine tool controller. The optimization system experiments show that the maximum error of is about 11%, and the Mean absolute percentage error is about 5.2%, and each optimization operation takes around 110 sec. The performance of system is excellent. The optimization system can be modify and setup on pc-based CNC controller. Let manufacturing industry field can achieve the intelligentization of rising process efficiency.

碩士<br>中原大學<br>工業與系統工程研究所<br>101<br>An artificial neural network (ANN) was developed and has been widely applied for years. One of the most applications is the supervised back propagation neural network (BPN). A traditional BPN adopted the steepest descent method to train and adjust weight values between neurons. It has the following shortcomings: (1) apt to convergence to local minima (2) slowly updated weight values and had long learning time (3) it is possible to become divergence results if the learning speed value increases too much and too fast. This study implements regrouping particle swarm optimization (RPSO) method to replace the steepest descent method at the beginning of training for BPN to achieve global search. It can prevent the weight values from local minima. And then use the Levenberg Marquardt (LM) method in posterior part to achieve a better result for BPN. Finally, this study applied the new model to develop a surface roughness monitoring system in milling operations and used Taguchi Method to obtain a set of training parameters from both ANN and RPSO to configure the parameters. We obtain the best parameter setting to construct Surface Roughness forecasting system of BPN based hybrid regrouping particle swarm optimization and LM (RPSOLM-BPN). To prove the effectiveness, accuracy and stability of this study, we compare the accuracy and stability of the developed prediction results between RPSOLM-BPN and LM of Back-propagation neural networks (LM-BPN). A t-test and F-test are used to compare the differences between these two prediction systems. To verify the accuracy and stability of the prediction system, the optimization of the number of training data has been conducted. The results show that RPSOLM-BPN prediction system which is proposed in this study has more accurate and stable results than that of the LM-BPN. Finally, in the study of optimization of training data set, there is no significant difference between 80 samples and 150 samples at significance level α = 0.05. As a result, training data can reduced to 80 samples, it can reach to the same prediction results as 150 training data samples.

碩士<br>國立高雄第一科技大學<br>機械與自動化工程所<br>95<br>In this paper , genetic-based learning algorithm method of adaptive-network based fuzzy inference system(ANFIS) was used to predict the work piece surface roughness. Spindle speed , feed rate and depth of cut are the input variable , and the final output is the surface roughness . The premise part of the ANFIS is applied by Gauss membership function , and the consequent part is applied by the Takagi-Sugeno (TS) fuzzy model to inference. Taguchi-genetic algorithm was used to train the parameters of both the Gauss membership function in the premise part and the function in consequent part. A total of 48 sets of experimental date were used for training. After the training, another 24 sets dates were used to check out how correct the results are. Finally, we compared the prediction accuracy of surface roughness by Genetic-based learning Algorithm method of adaptive-network based fuzzy inference system and the general ANFIS which include triangular and trapezoidal membership function . The comparison indicates that the adoption of genetic-based learning algorithm method get a better performance.

碩士<br>國立中興大學<br>機械工程學系所<br>107<br>The surface quality of the grinding process is affected by many factors, such as grinding wheel speed, grinding depth, feed rate, machine vibration and grinding wheel wear. The workpiece surface quality and the grinding wheel state are assessed based on the experience of operators by observing the produced sparks and noise in the traditional machining. This makes it difficult to control the grinding quality levels, and thus results in the increase of time and material costs. Therefore, it is crucial to predict the surface roughness of workpiece and to monitor the wear state of the grinding wheel through the intelligent machining technology. In this study, SUS304 was continuously processed by the surface grinder, and the vibration accelerations of the grinding wheel and the magnetic platform were measured synchronously. With different grinding parameters, such as grinding depth, feed rate, and grinding wheel speed, the envelope analysis and time / frequency domain signal processing techniques were applied to extract the signal characteristics. The relationship between the Ra values and the signal characteristics was investigated. The selected signal characteristics were casted into the Long Short-Term Memory Network to predict the surface roughness and the remaining life of the grinding wheel.

碩士<br>大葉大學<br>機械工程研究所碩士班<br>95<br>Manufacturing of silicon wafer starts with growth of silicon ingots. Typical processes to turn a silicon ingot to silicon wafers include slicing, edge profiling, grinding, lapping, polishing, etching etc. Grinding lapping and polishing are used to flatten the rough surface of wafer caused by slicing. Lapping and polishing processed for flatting wafers. In this study models for predicting the roughness of silicon wafers are developed. The prediction models are based on the grinding parameters and grit size of the grinding wheel. Box-Benkhen experiment design is used to collect the require data. Try to find the relationship of cutting degrees and surface roughness. And collect the require data of cutting degrees. Power function is used to build the regression function. Finally find the relationship of all control factor and surface roughness.

碩士<br>中原大學<br>工業與系統工程研究所<br>98<br>Every industry hopes to reduce the cost, the waste and promote the productivity and efficiency to achieve the goal and vision. For achieving this goal, quality control is the most important role in manufacturing processing. Therefore, how to reach quality control efficiency must using some kinds of method to reduce the fail rate and promote the productivity, ex: Quality 7 Tools, 6 sigma and QCC etc. However, using this methods to achieve quality control, we usually through workpiece measurement. In the manufacturing industry, the surface roughness is an important index to evaluate product quality. Surface roughness direct impacting surface wear resistance, fatigue strength, reliability and leak-proof quality, thermal conductivity and drive precision and so on, so the surface roughness is an important index assessing the surface conditions, reflecting the quality, especially in the machinery processing has become an essential quality requirements. In the past, has many of constructing a prediction decision-making system research. But how to promote the decision-making system precision is researchers hope. This research using combination of Regression and Neural Network to develop a regression neural network prediction model, the purpose is predicting the surface roughness efficiency, reducing the variation of data and promote system precision.

碩士<br>靜宜大學<br>資訊管理學系<br>107<br>The application of the intelligent monitoring system could effectively improve the surface roughness during the surface milling of the product. Based on ease of installation and cost effectiveness, the external accelerator collected vibration signals generated during the production process on the processing machine to predict the surface roughness of the product. More accurate predictions could be obtained by combining deep learning prediction models. In this study, three models of FFT-DNN, FFT-LSTM and 1D-CNN were used to investigate training model and prediction performance. Feature extraction plays an important role for training and result prediction. Before the training and prediction results, the original vibration signal is extracted by FFT and 1D-CNN one-dimensional convolution filter to remove bad quality signals such as noise. The results show that the LSTM model with temporal modeling ability presents good predictability for higher Ra values. The 1D-CNN with better feature extraction capability exhibits highly accurate prediction performance and can achieve highly accurate levels in the low to medium Ra range. Based on the current experimental results, vibration signals combined with deep learning prediction model under sufficient data set training can be applied to predict the surface roughness of the product in milling process.