| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.83 MB | Adobe PDF |
Advisor(s)
Abstract(s)
A indĂșstria 4.0 trouxe uma modernização ao sistema produtivo atravĂ©s da integração em rede das
entidades constituintes que, aliado à evolução das tecnologias de informação permitiu um aumento
da produtividade, da qualidade do produto, otimização dos custos produtivos e personalização do
produto Ă s necessidades do cliente. Apesar da complexidade do pensamento humano, a
inteligĂȘncia artificial tenta replicĂĄ-lo em algoritmos, criando modelos capazes de processar bases
de dados com um elevado volume de informação, gerando informação valiosa para a tomada de
decisĂŁo, como Ă© o caso do Machine Learning e o Deep Learning.
Na presente dissertação criou-se um modelo recorrendo å ferramenta open source Knime que com
base num conjunto de dados fornecidos pela Bosch, parametrizou o modelo com diversas técnicas
de pré-processamento e de seleção de recursos, permitindo desenvolver um modelo final para
previsão de falhas internas dos produtos na linha de produção da Bosch.
No sentido de selecionar o modelo mais eficiente, preciso e adequado ao problema em questĂŁo,
recorreu-se ao Logistic Regression, Decisison Tree, Random Forest, XGBoost e por Ășltimo o Naive
Bayes. De forma a diminuir o tempo de aprendizagem do modelo, o poder computacional
necessĂĄrio e a possibilidade de ocorrer o overfitting analisou-se a possibilidade de serem aplicadas
as seguintes técnicas de redução de dimensionalidade Correlation filter, Low Variance filter e o PCA.
O estudo demonstra que a eficåcia dos modelos melhorou com a aplicação das técnicas acima
descritas destacando-se o Logistic Regression aliado å técnica de seleção de recursos PCA, que
obteve um Recall de 100% e uma precisĂŁo e accuracy, ambas com 99.43%.
Industry 4.0 has brought an improvement into the production system through the entities network integration, which combined with the evolution of information technology has enabled an increase in productivity, product quality, production costs optimization and product customization to customer needs. Despite the complexity of human thought, artificial intelligence tries to replicate it in algorithms, creating models capable of database processing with a high information volume, generating valuable data for decision-making, as it is the case of Machine Learning and Deep Learning algorithms. In this dissertation it was created a model using the open-source tool Knime that, based on a dataset provided by Bosch, applied several pre-processing tools and feature selection techniques into the model, allowing the development of an algorithm for internal product failure prediction at Bosch production line. In order to select the most efficient, accurate and suitable model for the case study, it was analyzed different algorithms as Logistic Regression, Decision Tree, Random Forest, XGBoost and finally Naive Bayes. Regarding the decrease of the model learning time, computational power required and the possibility of overfitting, were analyzed the following dimensionality reduction techniques: Correlation filter, Low Variance filter and PCA. The study shows that the model accuracy improved using the techniques described above, mainly the Logistic Regression combined with the resource selection technique PCA, which obtained a Recall of 100% and a precision and accuracy, both with 99.43%.
Industry 4.0 has brought an improvement into the production system through the entities network integration, which combined with the evolution of information technology has enabled an increase in productivity, product quality, production costs optimization and product customization to customer needs. Despite the complexity of human thought, artificial intelligence tries to replicate it in algorithms, creating models capable of database processing with a high information volume, generating valuable data for decision-making, as it is the case of Machine Learning and Deep Learning algorithms. In this dissertation it was created a model using the open-source tool Knime that, based on a dataset provided by Bosch, applied several pre-processing tools and feature selection techniques into the model, allowing the development of an algorithm for internal product failure prediction at Bosch production line. In order to select the most efficient, accurate and suitable model for the case study, it was analyzed different algorithms as Logistic Regression, Decision Tree, Random Forest, XGBoost and finally Naive Bayes. Regarding the decrease of the model learning time, computational power required and the possibility of overfitting, were analyzed the following dimensionality reduction techniques: Correlation filter, Low Variance filter and PCA. The study shows that the model accuracy improved using the techniques described above, mainly the Logistic Regression combined with the resource selection technique PCA, which obtained a Recall of 100% and a precision and accuracy, both with 99.43%.
Description
Keywords
Industry 4.0 Artificial Intelligence Machine Learning Deep Learning Metaheuristic Algorithms