A carregar...
Publicação
Estudo e parametrização de uma linha de trituração utilizando algoritmos de Machine Learning
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 39.51 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A transformação digital, impulsionada pela Indústria 4.0, tem vindo a criar novas oportunidades para a otimização de processos produtivos através da análise de dados e da aplicação de algoritmos de Machine Learning. Esta dissertação apresenta o estudo e a parametrização de uma linha de trituração de cortiça, recorrendo a dados recolhidos em tempo real pelo sistema Supervisory Control and Data Acquisition (SCADA). O objetivo principal é identificar as variáveis que mais influenciam a produção útil e desenvolver modelos preditivos capazes não só de estimar a produção, mas também de indicar quais variáveis exercem maior impacto nos resultados. A metodologia adotada seguiu a abordagem Cross Industry Standard Process for Data Mining (CRISP-DM) para análise e modelação de dados. Foram aplicados
diferentes algoritmos de Machine Learning, incluindo Regressão Linear, Random Forest e eXtreme Gradient Boosting (XGBoost), com afinação de hiperparâmetros e validação cruzada k-fold. Os resultados mostraram que os modelos de ensemble, em particular o eXtreme Gradient Boosting, obtiveram melhor desempenho preditivo nas métricas Erro Absoluto Médi(MAE), Raiz do Erro Quadrático Médio (RMSE) e coeficiente de determinação (R2). Para interpretar o modelo, recorreu-se ao método Shapley Additive Explanations (SHAP), que permitiu identificar a influência global
e individual das variáveis mais relevantes. A análise evidenciou variáveis controláveis pelos operadores no processo produtivo com impacto significativo na produção, fornecendo informação útil para a melhoria do processo. Conclui-se que a integração de dados industriais com Machine Learning pode apoiar a tomada de decisão, reduzir desperdícios e aumentar a eficiência produtiva.
Digital transformation, driven by Industry 4.0, has been creating new opportunities for the optimization of production processes through data analysis and the application of Machine Learningalgorithms. This dissertation presents the study and parameterization of a cork grinding line, using data collected in real time by the Supervisory Control and Data Acquisition(SCADA) system. The main goal is to identify the variables that most influence useful production and to develop predictive models capable not only of estimating production but also of indicating which variables exert the greatest impact on the result. The methodology adopted followed Cross Industry Standard Process for Data Mining (CRISP-DM) for data analysis and modeling. Different Machine Learning algorithms were applied, including Linear Regression, Random Forest, and eXtreme Gradient Boosting (XGBoost), with hyperparameter tuning and k-fold cross-validation. The results showed that ensemble models, particularly eXtreme Gradient Boosting, achieved the best predictive performance in the metrics Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and coefficient of determination (R2). To interpret the model, the Shapley Additive Explanations (SHAP) method was used, which made it possible to identify the global and individual influence of the most relevant variables. In conclusion, the analysis highlighted variables controllable by operators in the production process that had a significant impact on output, providing useful information for process improvement. The integration of industrial data with Machine Learning can support decision-making, reduce waste, and increase production efficiency.
Digital transformation, driven by Industry 4.0, has been creating new opportunities for the optimization of production processes through data analysis and the application of Machine Learningalgorithms. This dissertation presents the study and parameterization of a cork grinding line, using data collected in real time by the Supervisory Control and Data Acquisition(SCADA) system. The main goal is to identify the variables that most influence useful production and to develop predictive models capable not only of estimating production but also of indicating which variables exert the greatest impact on the result. The methodology adopted followed Cross Industry Standard Process for Data Mining (CRISP-DM) for data analysis and modeling. Different Machine Learning algorithms were applied, including Linear Regression, Random Forest, and eXtreme Gradient Boosting (XGBoost), with hyperparameter tuning and k-fold cross-validation. The results showed that ensemble models, particularly eXtreme Gradient Boosting, achieved the best predictive performance in the metrics Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and coefficient of determination (R2). To interpret the model, the Shapley Additive Explanations (SHAP) method was used, which made it possible to identify the global and individual influence of the most relevant variables. In conclusion, the analysis highlighted variables controllable by operators in the production process that had a significant impact on output, providing useful information for process improvement. The integration of industrial data with Machine Learning can support decision-making, reduce waste, and increase production efficiency.
Descrição
Palavras-chave
Machine Learning SCADA cortiça otimização da produção SHAP Industry 4.0 Indústria 4.0 Cork Production optimization