A carregar...
Publicação
Implementação de uma interface estatística para aplicação de data science a dados de controlo de qualidade de rolhas de cortiça
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 6.41 MB | Adobe PDF |
Orientador(es)
Resumo(s)
O processo de fabrico de rolhas de cortiça, quer sejam rolhas naturais, quer sejam rolhas técnicas, apresenta uma variabilidade natural associada. A aplicação de ferramentas estatísticas é assim fundamental para avaliar o desempenho do processo, ou os resultados da implementação de ações de melhoria, ou, simplesmente, resultados que permitam avaliar métricas estatísticas que caracterizam as diferentes etapas do processo de fabrico. A utilização correta destas ferramentas requer conhecimentos específicos de estatística.
O objetivo principal deste trabalho foi a construção de uma interface estatística que possa guiar o utilizador na utilização das técnicas estatísticas mais adequadas aos objetivos pretendidos.
Foi construída uma interface estatística que inclui as ferramentas de data science: ferramentas de visualização de dados, ferramentas de inferência estatística que utilizam regras heurísticas para guiar o utilizador na direção de utilização de estatística paramétrica ou não paramétrica; ferramentas de filtragem de dados, ferramentas de redução de dimensionalidade dos dados (PCA) e ferramentas de modelação linear (PLS).
Foram estudados 3 casos que mostram as potencialidades da ferramenta desenvolvida. O primeiro caso de estudo avalia o desempenho de rolhas naturais com tratamento tendo-se concluído que a humidade apresenta variação ao longo do período de armazenamento independente do tipo de embalagem e que as rolhas com este tratamento apresentam menores forças de extração. O segundo caso avalia o impacto, em rolhas naturais engarrafadas e rolhas naturais não engarrafadas, do tempo de armazenamento e do tipo de tratamento nas variáveis relacionadas com a utilização da rolha, tendo-se concluído que não existem diferenças estatisticamente significativas entre tratamentos e entre períodos de armazenamento no momento de torção das rolhas nas duas situações e para rolhas não engarrafadas, também, não existem diferenças estatisticamente significativas entre períodos de armazenamento e entre tratamento nos valores de progressão capilar e relaxação. Já para rolhas engarrafadas conclui-se que a progressão capilar e a humidade apresentam diferenças estatisticamente significativas entre tratamentos e entre períodos de armazenamento. O terceiro caso de estudo avalia o impacto e capacidade preditiva das variáveis de processo sobre a absorção em rolhas microaglomeradas de cortiça, tendo-se concluído que as variáveis que têm maior correlação são o tipo de cola, humidade do granulado, massa volúmica aparente, temperatura de forno frio, massa de granulado, taxa de compressão e temperatura ambiente, de acordo com o modelo PCA. Quanto às variáveis que melhor preveem a absorção destas rolhas, de acordo com o modelo PLS são a humidade do granulado, a humidade relativa, massa de cola, percentagem de cola, teor de água, e o tipo de cola.
The process of manufacturing cork stoppers, whether they are natural or technical stoppers, has an associated natural variability. The application of statistical tools is therefore essential to evaluate the performance of the process, or the results of the implementation of improvement actions, or, simply, results that allow the evaluation of statistical metrics that characterize the different stages of the manufacturing process. The correct use of these tools requires specific knowledge of statistics. The main objective of this work is the construction of a statistical interface that can guide the user in the use of the most adequate statistical techniques for the intended purposes. A statistical interface was built that includes data science tools: data visualization tools, statistical inference tools that use heuristic rules to guide the user in the right direction in the use of parametric or non-parametric statistics; data filtering tools, data dimensionality reduction tools (PCA) and linear modeling tools (PLS). Three cases were studied that show the potential of the developed tool. The first case evaluates the performance of natural cork stoppers with treatment, concluding that humidity varies throughout the storage period, regardless of the type of packaging, and that cork stoppers with this treatment have lower extraction forces. The second case assesses the impact, in bottled natural stoppers and natural non-bottled stoppers, of the storage time and type of treatment on variables related to the use of the stopper, having concluded that there are no statistically significant differences between treatments and between storage times of stoppers in both situations and for non-bottled stoppers, also, there are no statistically significant differences between storage time and between treatment in capillary progression and relaxation values. As for bottled stoppers, it is concluded that capillary progression and humidity show statistically significant differences between treatments and between storage periods. The third case study assesses the impact and predictive capacity of process variables on the absorption of micro-agglomerated cork stoppers, and it was concluded that the variables with the highest correlation are the type of glue, granulate moisture, apparent volumic mass, cold oven temperature, granulate mass, compression ratio and ambient temperature, according to the PCA model. As for the variables that best predict the absorption of these stoppers, according to the PLS model, they are the moisture of the granulate, the relative humidity, glue mass, glue percentage, water content, and type of glue.
The process of manufacturing cork stoppers, whether they are natural or technical stoppers, has an associated natural variability. The application of statistical tools is therefore essential to evaluate the performance of the process, or the results of the implementation of improvement actions, or, simply, results that allow the evaluation of statistical metrics that characterize the different stages of the manufacturing process. The correct use of these tools requires specific knowledge of statistics. The main objective of this work is the construction of a statistical interface that can guide the user in the use of the most adequate statistical techniques for the intended purposes. A statistical interface was built that includes data science tools: data visualization tools, statistical inference tools that use heuristic rules to guide the user in the right direction in the use of parametric or non-parametric statistics; data filtering tools, data dimensionality reduction tools (PCA) and linear modeling tools (PLS). Three cases were studied that show the potential of the developed tool. The first case evaluates the performance of natural cork stoppers with treatment, concluding that humidity varies throughout the storage period, regardless of the type of packaging, and that cork stoppers with this treatment have lower extraction forces. The second case assesses the impact, in bottled natural stoppers and natural non-bottled stoppers, of the storage time and type of treatment on variables related to the use of the stopper, having concluded that there are no statistically significant differences between treatments and between storage times of stoppers in both situations and for non-bottled stoppers, also, there are no statistically significant differences between storage time and between treatment in capillary progression and relaxation values. As for bottled stoppers, it is concluded that capillary progression and humidity show statistically significant differences between treatments and between storage periods. The third case study assesses the impact and predictive capacity of process variables on the absorption of micro-agglomerated cork stoppers, and it was concluded that the variables with the highest correlation are the type of glue, granulate moisture, apparent volumic mass, cold oven temperature, granulate mass, compression ratio and ambient temperature, according to the PCA model. As for the variables that best predict the absorption of these stoppers, according to the PLS model, they are the moisture of the granulate, the relative humidity, glue mass, glue percentage, water content, and type of glue.
Descrição
Palavras-chave
cortiça rolha data science estatística