Salomon: Un nuevo enfoque para la mejora de procesos de negocio mediante la produccion inteligente basada en modelos predictivos de control hibridos y autoadaptativos.

摘要

The manufacturing process is defined as any activity that getting as its input a set of raw materials is able to modify them in order to achieve a new product. This kind of activities, as simple as they may look, is the main reasons of the growth and evolution of our society, as we know it.;In its beginnings, the ancient manufacturing process was a magic surrounded activity. Nowadays, it still happens the same. Besides, other requirements, such as quality measures or environmental guidelines, complicate the whole process. In addition, we cannot forget that we are living in a globalised market and every little improvement to any of the multiple processes deployed at any plant, could become a competitive boost.;In light of this background, the scientific community has spent many years developing new methods for supervising and controlling a manufacturing plant—one example is the Supervisory, Control And Data Acquisition (SCADA). Other systems are capable of being two steps ahead, trying to foresee what will happen in the plant. Those are commonly referred as Model Predictive Control systems. Although there are commercial and academic solutions, the experts have identified a number of limitations such as the inability to work with a multivariable system, the difficulty to adapt the solution to changes in the manufacturing process and the predictive models that do not adapt to the nature of the business process.;In order to meet the needs that companies have already identified in terms of quality, cost reduction and eco-manufacturing, as well as the ability to obtain a competitive boost, we have designed, developed, evaluated and implemented a system that is able to readjust the manufacturing process, working with the constraints previously defined.;Accordingly, we formulated the following hypothesis 'It is possible to model the business flow of a manufacturing process as a knowledge cloud through the creation of an hybrid (linear and nonlinear) model predictive control, based on current machine learning techniques, keeping it updated and carrying the manufacturing control to optimise some of its critical parameters.' .;For its validation, we have developed the following tasks with the aim of creating the complete system defined above. Hence, we have focused our efforts towards completing the following tasks: (i) developing statistical classifiers based on machine learning techniques and thinking the way of joining them to improve the prediction system, (ii) defining and designing the method that will keep up to date the models, (iii) determining which actions have to be carried out and, once decided, informing operators or other control systems already incorporated in the plant and, finally, (iv) evaluating and comparing the developed solution.;Nowadays, manufacturing processes are fully integrated into society. Therefore, improvements in the state of the art, as detailed here, do not only affect those people directly involved in production plants. In other words, the result of a manufacturing process, a manufactured product, can be part of a much larger system than any of us may end up using it.