Elena Pons Mata
Trabajo Fin de Grado del Grado en Ingeniería de Sonido e Imagen. Febrero 2016

The aim of this work is to develop a new system to detect welds between steel coils. This detection is carried out by a Kohonen Self-Organizing Map (SOM).

In order to achieve good results, a set of experiments have been carried out. On the other hand, we pay special attention to the selection and preprocessing of the input data parameters, which are supplied by an artificial vision system currently working in a real steelmaking processs. On the other hand, we take care of the training parameters, such as the evolution of the neighborhood radius and the learning rate, and of the criterio to assign each neuron to a particular cluster.

From these experiments we have performed a statistical analysis and found a sepecific Self Organizing Map, thar reachs a sensitivity of 0.975 and a precision of 0.638. This means that our system keeps a similar value for sensitivity with respect to the currently running system (0.975 vs 0.998), and improves precision in a 22.4% (0.638 vs 0.414).

As a secondary result of our work, we have algo developed an interface in order to organize the design and testing of the diferents experiments.