Hybridization of Machine Learning for Advanced Manufacturing

  1. Redondo Guevara, Raquel
Dirigida per:
  1. Emilio Santiago Corchado Rodríguez Director/a
  2. Álvaro Herrero Cosío Director
  3. Javier Sedano Franco Director/a

Universitat de defensa: Universidad de Salamanca

Fecha de defensa: 09 de de desembre de 2020

Tribunal:
  1. Hilde Pérez García President/a
  2. Enrique Antonio de la Cal Marín Secretari/ària
  3. Dragan Simic Vocal

Tipus: Tesi

Teseo: 644545 DIALNET

Resum

In the industry context, nowadays the terms "Advanced Manufacturing", "Industry 4.0" and "Smart Factory" are becoming real. Industrial companies seek to be more competitive, whether in costs, time, consumption of raw materials, energy, etc. Companies seek to be efficient in all areas and also to be sustainable. The future of many companies depends on their degree of adaptation to changes and their innovation capacity. Consumers are increasingly demanding, looking for personalized and specific products with high quality, at low cost and non-polluting. In order to meet these requirements, industrial companies implement technological innovations. These technological innovations include the previously mentioned Advanced Manufacturing and Machine Learning (ML). The present research work is focused in these fields, hybrid intelligent solutions have been designed and applied, combining several ML techniques in order to solve problems in the field of manufacturing industry. Intelligent techniques have been applied, including Artificial Neural Networks (ANN), multiobjective genetic algorithms, projection methods for the reduction of dimensionality, clustering techniques, etc. In order to obtain the mathematical model that best represents the real system under study, Systems Identification techniques have also been used. Several techniques have been hybridized with the purpose of building more robust and reliable solutions. By combining specific ML techniques, more complex systems are created with a greater representation / solution capacity. To solve such problems, these systems use data and knowledge about them. The proposed solutions search for solving ad wide spectrum of real-world complex problems, handling aspects such as uncertainty, lack of precision, high dimensionality, etc. This doctoral thesis includes several real case studies, in which various ML techniques have been applied to different problems in the field of manufacturing industry. The real case studies of the industry that are addressed, with four different data sets, correspond to: • Dental high-precision milling process. • Data analysis for predictive maintenance purposes in a company from the automotive sector. The different developed hybridizations of ML techniques have been applied and validated with real and original data sets, in collaboration with industrial companies or milling centres, allowing to solve current and complex problems. Thus, the work carried out has not only had a theoretical approach, but it has also been applied in an empirical way, allowing industrial companies to improve their processes, reducing costs and time, polluting less, etc. The satisfactory results that have been achieved indicate the helpfulness and contribution that ML techniques can make in the field of Advanced Manufacturing.