The Nearest-in-Control Neighbor Concept Applied to Projection Monitoring Techniques

CEDEÑO MARCO ; ZURITA NOELIA; RODRIGUEZ LEANDRO; SANCHEZ MABEL

Bahía Blanca

Congreso; IX Congreso Argentino de Ingeniería Química; 2017

PLAPIQUI - CONICET

It is a common issue in the last years to find a strategy that handles adequately a great amount of process data provided by the chemical industries and to convert the raw data into useful information. There have been an incremental developed of strategies devoted to the detection stage of the process monitoring, giving solutions in the case of detecting a fault for not only the linear process set but for the nonlinear process also. It is not the case for the identification stage due to only some few strategies were developed. In this work it is presented the PCA and the Kernel PCA strategies to face the detection stage, while in the identification stage a solution is offered by taking the NICN-concept to methods that handle variable projections onto the latent variable space. This is accomplished by solving an optimization problem to find the nearest neighbor in control given a pre-defined metric on the latent space. The results are obtained over some faults simulations of the Tennessee Eastman Process and compared by specialized indices in order to highlight the best strategy.