Data-driven Methodologies for System Identification

System Identification aims to understand the behavior and evolution of physical processes or industrial systems, which are often nonlinear. In the context of modern industrial automation, this task is essential for control, fault diagnosis, monitoring, and performance optimization. These activities require accurate mathematical models capable of describing the nonlinear relationships between input and output. The problem that arises is that determining such models can be complex or not fully possible. In recent years, system identification has addressed this issue through data-driven approaches, based on the collection of data from observable quantities of the system—particularly its inputs and outputs. Such data can be used to train neural networks so that they can learn patterns and relationships useful for constructing an analytical model of the system. This thesis covers two data-driven approaches for the identification of nonlinear systems using neural networks. The first approach is based on an ensemble learning framework, where multiple autoencoder-type neural networks are combined to improve prediction of system's evolution. Through a dedicated procedure, the best autoencoders, according to a score based on performance and diversity respect to other autoencoders, are selected from the ensemble, enhancing the overall performance. The approach was tested on nonlinear benchmarks, showing improved identification performance compared to individual models. The second approach focuses on the modeling of a thermal process consisting of the cooling dynamics of a thin bread produced in an industrial bakery. The dynamics are identified using a neural network trained with thermal images of the bread, achieving high prediction accuracy and consistency with thermodynamic principles through the integration of physical knowledge into the training process. Together, these studies contribute to the broader development of data-driven approaches, supporting the development of automation in line with the principles of Industry 5.0, with potential applications in small and medium-sized enterprises located in regions where the technological transformation is currently underway.