Desacoplamento de sistemas de controle multivariáveis por ICA com modificação do branqueamento.

Abstract

The use of sensors in process control systems is of vital importance for the proper operation and monitoring of industrial plants. In turn, process signals may have interference from other sources and, in some cases, it is not possible to observe directly the individual signals of the sources. In view of this, signal processing and separation techniques are used in order to extract the information from the sources contained in mixed signals. The main signal separation techniques are associated with the Independent Component Analysis (ICA), which has undergone significant evolution since its creation in the 1980s. Such evolution also had the contribution of the Principal Component Analysis (PCA) and the development of computational processing power. However, these techniques have two basic glitches: deviation of amplitude and phase change, which limit their use in control systems. Therefore, this research aims to present a solution to the problem of amplitude in ICA techniques for use in decoupling reduction in multivariate systems. The proposed correction, based on the stage of whitening ICA algorithms, which generated the technique MOD-ICA, was used as an alternative to breaking the correlation between variables in multivariate systems. Such technique was used for projecting and obtaining controlling pairs in an industrial plant of anhydrous ethanol production modeled on the Aspen Dynamics platform. In the case study proposed in this research, a significant reduction in the conditional numbers of the proposed controlling pairs was observed, and the separating matrix was used as a parameter of decoupling for the control system. As a result, the proposed MOD-ICA technique can be used as a tool for generating control systems, and its separating matrix can be considered as a model for decoupling reduction, which results in a more robust control system for process variation.