ABSTRACT ROBUST CONTROL CHARTS &Ccedil / etiny&uuml / rek, Aysun M. Sc., Department of Statistics Supervisor: Dr. BariS S&uuml / r&uuml / c&uuml / Co-Supervisor: Assoc. Prof. Dr. Birdal Senoglu December 2006, 82 pages Control charts are one of the most commonly used tools in statistical process control. A prominent feature of the statistical process control is the Shewhart control chart that depends on the assumption of normality. However, violations of underlying normality assumption are common in practice. For this reason, control charts for symmetric distributions for both long- and short-tailed distributions are constructed by using least squares estimators and the robust estimators -modified maximum likelihood, trim, MAD and wave. In order to evaluate the performance of the charts under the assumed distribution and investigate robustness properties, the probability of plotting outside the control limits is calculated via Monte Carlo simulation technique.

Total quality management (TQM) approach is often used to carry out company-wide continuous quality improvement plans in manufacturing and service industries. Similarly, TQM can also play a critical role for quality management in health care. Aiming to improve health care quality, experiences showed that major problems of non-patient care, patient records and vital signs monitoring are encountered. In this study, we aim to introduce TQM for quality improvement for intensive care unit (ICU) operations, including some solutions and the prototype of quality management. And vital signs monitoring at ICU is taken as an example of process. For quality improvement of non-patient care, Health Care Quality Development Life Cycle, including (1) quality requirement analysis, (2) quality specification review, (3) quality design, (4) quality implementation, (5) quality testing, (6) quality maintaining, and (7) quality validation, is discussed. The prototype of the first three phases for quality improvement at ICU is explored. Through quality requirement analysis, non-patient care quality at ICU is defined in areas of administration, facility and environment. For quality improvement of patient records maintaining, firstly, scope of health care information systems is categorized as administrative operational system, decision support system, clinical information system, and medical information system. According to this categorization and experience, some interesting result is found. For instance, the current applications of information systems for teaching hospitals in southern Taiwan surveyed are that most applications are administrative and clinical. And the essential information of patient records used in each information system is not complete or not easily accessed. Model of the patient record maintaining is introduced and the prototype design of patient records is recommended for quality improvement of patient records maintaining at ICU. To improve quality of vital signs monitoring is one essential requirement and specification for ICU quality improvement. Effective outcome measures of vital signs monitoring and early detecting of abnormal vital signs is considered important. For quality improvement of vital signs monitoring at ICU, heart rate graphs are taken as examples in our study through the heart rate graphs monitoring. Health professionals can understand the interactions of human autonomic nervous system. By use of digitizer, the computable heart rate data is acquired from each graph and grouped into mortality and near-to-normal cases. Then spectrum form of heart rate data, describing more about heart function, is used for statistical analysis. Several control chart methods have been experimented to detect small heart rate shifts from target, cumulative sum control chart (Cusum) is adopted in our study. The observable variable is the patient¡¦s heart rate, the purpose is to check the alarms pointed out by Cusum that could be partially be ascribed to changes of heart rate trend over time, and to a shift in the monitoring process mean. From summaries of nonconformities in the Cusum charts, mortality cases obviously have more nonconformities. It is obvious that Cusum control charts of mortality cases provide diagnostic information for vital signs monitoring process. In addition, Cusum charts may also inform ICU professionals that there is a small shift of patient heart rate, a continuously increasing or decreasing heart rate, and the adjustment of sympathetic nerve and parasympathetic nerve. In those cases, some special care is needed.

本論文針對防丟器的剖面進行追蹤分析。防丟器包含發射器及接收器，發射器會發射訊號，接收器會記錄RSSI (Receive Signal Strength Index)與發射點數，其中RSSI表示訊號的強度。在工程理論上，RSSI與距離具有函數關係；然而環境中的干擾及事件發生都會影響此函數關係，特別是事件發生會嚴重地改變此函數關係，因此論文主要目的在於區別事件是否發生。 　　所謂的剖面指的是變數之間的函數關係，而論文中的剖面追蹤是利用管制圖的概念，用管制圖來監控剖面的參數估計值。如果管制圖上的點子出界，則表示事件發生而導致失控。 　　本論文以腳踏車是否被偷為例，嘗試一些實驗後找出顯著影響的因子設計實驗，包含17種腳踏車未被偷之情境與18種腳踏車被偷情境；欲利用未被偷的實驗建立試驗管制圖，而以被偷之情境來追蹤，用以驗證管制圖之有效性。 　　論文中主要透過分析防丟器產生的RSSI與距離的剖面、距離與發射點數的剖面來探討事件是否發生。另外剖面追蹤其實是種事後追蹤的方法，為了能即時追蹤，本論文亦採用預測區間的方式，來追蹤事件是否發生。 　　本論文建議監控距離與發射點數的剖面，因該方法的表現最好，另外建議增加防丟器上能紀錄距離的功能，此方法會更加合適。 　　本論文提出的即時追蹤方式並沒有特別好，因此一個比較好的即時追蹤方法是未來值得研究的方向。 / The device of Babyfinder is designed to detect if an event occurs. The Babyfinder includes transceiver and receiver. The signal strength, Received Signal Strength Indicator (RSSI), generates once there are distances between transceiver and receiver. In wireless communication theory, the relationship between RSSI and distance should be expressed by the model that RSSI = a + b ln (distance) Nevertheless, some circumstance noises and user noises (or common causes), and/or events (special causes) may affect the variation of RSSI. Since the occurrence of events may change the functional relationship of RSSI and distance, to distinguish if the functional relationship is changed by the occurred events is the subject of this study. This study designs some events and noises experiments based on the real noise factors and special events. Two monitoring schemes are proposed to distinguish the occurred events and noise circumstance. One is the profile monitoring scheme, the other is the real time monitoring scheme. The two proposed approaches of profile monitoring scheme are considered to monitor the profile of RSSI and distance and that of distance and the number of transmitting points, respectively. The profile monitoring approach for distance and the number of transmitting points shows better performance. However, the profile monitoring is an after-event tracing approach. It cannot detect the occurred events in time. A better approach of real-time monitoring approach is worth to be proposed in the future study.

To determine the economic design of control charts and the specification limits with minimum cost are two separate issues in previous research areas. In this study, we proposed a method to determine the optimal design parameters of X control charts and the specification limits simultaneously from an economic viewpoint. We also consider two types of X control charts: one is the economic X control chart and the other is the economic statistical X control chart. We obtain the optimal results by minimizing the expected cost per unit time for the-larger-the-better quality characteristic with a Weibull distribution. We consider the asymmetric control limits because of the asymmetric feature of theWeibull distribution. Also, we are considering the difference between monitoring the process by using an economic statistical X control chart and conducting a complete inspection plan. Which way is better, process control or inspection plan? In our data analysis of the two types of X control chart, we find that the optimal expected cost per unit time with complete inspection is lower than without complete inspection. This is because the coefficient of Taguchi’s quadratic loss function we set is too small. And the analysis shows us the significant parameters for the optimal expected cost per unit time and design parameters. At last, in our numerical examples for two different types of X control chart, we find that the performance of the economic X control chart is as good as the economic statistical one. However, we suggest the producer use the economic statistical X control chart with a complete inspection plan to obtain a lower expected cost per unit time and larger power of the control chart.

Orientador: Marisa Masumi Beppu / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Quimica / Made available in DSpace on 2018-08-12T23:04:08Z (GMT). No. of bitstreams: 1 Tiba_RafaelMassao_M.pdf: 4156025 bytes, checksum: 6b7f2bf2c33d458475abf019c06c839f (MD5) Previous issue date: 2009 / Resumo: O objetivo do trabalho foi analisar a adequação de Gráficos de Controle (coloquialmente chamadas também de Cartas de Controle) para o controle da qualidade e do desempenho de processos químicos. Dentro deste tema, Cartas de Controle usualmente aplicadas na indústria foram testadas e analisadas para verificar o seu impacto e aplicabilidade com dados provenientes de processos transientes. Os resultados obtidos indicaram que a utilização inadvertida dessas ferramentas (inicialmente concebidas para a indústria mecânica) pode trazer impactos indesejáveis como ocorrência de alarmes falsos e atraso na detecção. Os dados provenientes de indústrias químicas, particularmente quando apresentam autocorrelação, devem ser analisados de maneira diferenciada. Voltado para essa necessidade, uma ferramenta inédita foi criada para o controle de qualidade quando há esse tipo de dados: a carta para coeficientes angulares, que utiliza as tendências lineares de pequenos trechos de dados para antecipar a detecção de causas especiais. Entre as vantagens dessa ferramenta estão facilidade de aplicação (a carta dispensa recursos computacionais específicos ou utilização de séries temporais), alto desempenho e robustez frente a ruídos nos dados. Os gráficos de controle apresentam a vantagem de dispensar o conhecimento prévio do processo, podendo tratá-lo como uma "caixa preta", abordagem muito conveniente para processos muito complexos, com grande número de etapas e variáveis, onde abordagem fenomenológica e fundamental através de modelos diferenciais matemáticos entre variáveis dependentes e independentes pode ser bastante laboriosa na prática do controle. Para verificar se um ajuste comum ao processo possa não ser suficiente para fazer o processo convergir novamente ao "set point" (e voltar à zona de controle), sugere-se o uso de uma ferramenta auxiliar, a Dinâmica de Sistemas. Ela seria usada na indicação da necessidade de intervenção especial em um processo. Com o seu método particular de criar modelos multivariados, a Dinâmica de Sistemas fornece uma maneira qualitativa e quantitativa de identificar as variáveis críticas para a qualidade do processo e para analisar seu comportamento dinâmico bem como as relações multivariadas existentes em uma planta química, consistindo em uma grande oportunidade para ser aplicada no nível operacional. Ela pode servir para mostrar se uma determinada intervenção especial faria o sistema se descontrolar ainda mais (fato comum de processos com excesso de controle, "over-controlled") ou se poderia servir para fazê-la convergir novamente a uma zona de controle / Abstract: The purpose of this study was to examine the suitability of using ordinary control charts such as Shewhart's and Averages chart to control the quality and performance of chemical processes. Within this theme, control charts commonly used in industry were tested and analyzed to determine its impact and applicability with data from transient processes. The obtained results indicate that the inadverted use of such tools (originally designed for mechanical engineering, which has discrete processes) can lead to undesirable effects such as the occurrence of false alarms and delay on special causes detection. Therefore, data originated at chemical industries, especially when autocorrelation is present, must be analyzed through special techniques. Towards this need, a new tool was created specially to this sort of data: The Slope Chart, which uses the linear trends of small subgroups of data to anticipate special causes detection. The main advantages of this tool are ease of application (this charts does not requires specific computer resources or time series modeling), higher performance and robustness against noise. Control Charts allow users to deal with the process as "Black Boxes", not requiring previous knowledge about the process for special cause detection. For a high complexity process, where there are a considerable number of steps and variables, this is an advantage provided that classic control would require phenomenological studies which might be laborious. Therefore, an auxiliary tool in the cases where an ordinary correction may not enough is proposed. Furthermore, it has been proposed the use of System Dynamics as an auxiliary tool to indicate whether or not it is necessary to act at the process. With its particular method of creating multivariate models, the System Dynamics serves the purpose of quality control, providing both qualitative and quantitative ways to see critical variables to the process quality and to analyze its dynamic behavior and the existing multivariate relations in chemical industries, consisting of a significant opportunity to be used at operational level. It can be useful to simulate whether a potential action would bring more variability to the process (which is usual in overcontrolled processes) or would be useful to correct the process / Mestrado / Engenharia de Processos / Mestre em Engenharia Química

La spectroscopie sur plasma induit par laser (LIBS) est une technique d'analyse élémentaire couplant l'ablation laser dans le cas des échantillons solides, à la spectroscopie d'émission atomique. Malgré de nombreux avantages qui en font une technique de plus en plus attractive, la spectroscopie LIBS n'est pas encore reconnue comme méthode d'analyse quantitative. En effet, les problèmes d'hétérogénéité des échantillons, d'effets matrice, d'auto-absorption des raies d'émission et surtout de manque de répétabilité participent à la dégradation des performances analytiques de la spectroscopie LIBS. Dans le but de faire évoluer la technique, ces travaux de thèse consistent à réaliser un cas particulier de démonstration de l'évaluation des performances analytiques d'un système LIBS de laboratoire en intégrant des notions de qualité. L'exemple porte ici sur l'analyse d'échantillons d'aciers certifiés. Une première étude concerne le déroulement de l'optimisation du système pour l'analyse quantitative. L'effet des différents paramètres expérimentaux sur le signal LIBS étant complexe, un protocole méthodique est indispensable. Une étude paramétrique est donc ici proposée en vue de déterminer les conditions expérimentales les plus propices à l'analyse quantitative. Une fois optimisée, la méthode LIBS est ensuite caractérisée grâce aux principes classiques de la validation de méthode. La justesse ainsi que la fidélité de la méthode sont évaluées dans des conditions de répétabilité et de précision intermédiaire. Cette dernière étude montre des résultats prometteurs pour la technique. L'application d'une carte de contrôle montre néanmoins un manque de stabilité du système de laboratoire et permet d'enclencher des actions correctives en vue d'améliorer ses performances analytiques. / Laser-Induced Breakdown Spectroscopy (LIBS) is an elemental analytical technique which combines laser ablation with atomic emission spectroscopy. LIBS spectroscopy has many advantages but is not recognized as a fully quantitative method. Indeed, the problem of samples' heterogeneity, matrix effects, self-absorption of emission lines and the lack of repeatability deteriorate the analytical performances of LIBS. In order to improve this technique, the work presented in this thesis includes an example of analytical performances evaluation with the use of quality notions of a laboratory LIBS system. The method is here specially applied to the analysis of certified steel samples. A first study deals with the optimization of the LIBS system for the quantitative analysis. As the effect of the different experimental parameters on LIBS signal is complex, a methodical protocol is necessary. Here, a parametric study is proposed to determine the experimental conditions suitable for quantitative analysis. Once optimized, the LIBS method is then characterized with basics of method validation. The trueness and the precision of the method are evaluated in conditions of repeatability and intermediate precision. This study shows promising results for LIBS technique. The application of a control chart reveals however an instability of the laboratory system and enables to introduce corrective actions to improve its analytical performances.

The diploma is focused on optimizing of rating during the process of painting by means of experimental methods and statistic analysis. Description of the study of the present state of painting process, result analysis and the project of optimal pre-set in practice.

This study develops a heuristic procedure for specifying parameters for a neural network configuration (learning rate, momentum, and the number of neurons in a single hidden layer) in Shewhart X-bar control chart applications. Also, this study examines the replicability of the neural network solution when the neural network is retrained several times with different initial weights.

Analysis and predictive modeling of massive datasets is an extremely significant problem that arises in many practical applications. The task of predictive modeling becomes even more challenging when data are imperfect or uncertain. The real data are frequently affected by outliers, uncertain labels, and uneven distribution of classes (imbalanced data). Such uncertainties create bias and make predictive modeling an even more difficult task. In the present work, we introduce a cost-sensitive learning method (CSL) to deal with the classification of imperfect data. Typically, most traditional approaches for classification demonstrate poor performance in an environment with imperfect data. We propose the use of CSL with Support Vector Machine, which is a well-known data mining algorithm. The results reveal that the proposed algorithm produces more accurate classifiers and is more robust with respect to imperfect data. Furthermore, we explore the best performance measures to tackle imperfect data along with addressing real problems in quality control and business analytics.

[pt] A probabilidade de alarme falso, alfa, dos gráficos de controle de processos depende dos seus limites de controle, que, por sua vez, dependem de estimativas dos parâmetros do processo. Esta tese apresenta inicialmente uma revisão dos principais trabalhos sobre o efeito dos erros de estimação dos parâmetros do processo sobre alfa quando se utiliza o gráfico de X e S individualmente e em conjunto. O desempenho dos gráficos é medido através de medidas de desempenho (número médio de amostras até o sinal, taxa de alarme falso, distribuição do número de amostras até o sinal, que, em geral, são variáveis aleatórias, função dos erros de estimação. Pesquisas recentes têm focado nas propriedades da distribuição condicional do número de amostras até o sinal, ou ainda, nas propriedades da distribuição da taxa de alarme-falso condicional. Esta tese adota esta abordagem condicional e analisa o efeito da estimação dos parâmetros do processo no desempenho conjunto dos gráficos de X e S em dois casos: Caso KU (Média conhecida – Variância desconhecida) e Caso UU (Média desconhecida – Variância desconhecida). A quase totalidade dos trabalhos anteriores considerou apenas um gráfico, isoladamente; sobre efeito da estimação dos parâmetros sobre o desempenho conjunto conhecemos apenas um trabalho, sobre gráficos de X e R, mas nenhum sobre gráficos de X e S. Os resultados da análise mostram que o desempenho dos gráficos pode ser muito afetado pela estimação de parâmetros e que o número de amostras iniciais requerido para garantir um desempenho desejado é muito maior que os números tradicionalmente recomendados na literatura normativa de controle estatístico de processo (livros texto e manuais). Esse número é, porém, menor que o máximo entre os números requeridos para os gráficos de X e de S individualmente. Questões a serem investigadas como desdobramento dessa pesquisa são também indicadas nas Considerações Finais e Recomendações. / [en] The false-alarm rate of control charts, alpha, depends on the control limits calculated, which depend, in turn, on the estimated process parameters. This dissertation initially presents a review of the main research articles about the effect of the estimation errors of the process parameters upon alpha when X and S charts are used separately and together. The charts performance is evaluated through performance measures (average run-length, false-alarm rate, run-length distribution, etc), which are, in general, random variables, function of the estimation errors. Recent researches focused on the properties of the conditional run-length, or still (in the case of Shewhart charts) on the properties of the conditional false-alarm rate distribution. This dissertation adopts this conditional approach and investigates the effect of parameter estimation on the joint behavior of X and S charts in two cases: KU Case (Known mean – Unknown variance) and UU Case (Unknown mean - Unknown variance). Almost all previous works considered just only one chart separately – just only one joint performance work is known by the author, one about the effect of the estimation errors of the process parameters upon X e R joint performance. The results show that the charts performance can be severely affected by the parameter estimation and the number of initial samples required to ensure the desirable performance is greater than the numbers of initial samples recommended by traditional statistical process control reference texts (books and manuals). This number is, however, smaller than the maximum between the numbers of samples required by the X and the S charts separately. Additional issues for follow-up research are recommended in the concluding section.