An Adaptive Approach to Speed Up Computer-Aided Observability Analysis

CARBALLIDO J.A; PONZONI I.; BRIGNOLE N.B

Rio das Pedras, Brasil

Congreso; ENPROMER 2005; 2005

Abstract. Observability Analysis (OA) is a crucial stage in the design and revamp of process-plant instrumentation. In this article we present a Multi-Objective Genetic Algorithm (MOGA) that finds an adequate initial sensor configuration for structural OA algorithms. The resulting selection is most convenient in terms of purchase and installation costs, as well as measurement reliability and exploitation of the mathematical model for the estimation of state variables. The MOGA implements the following features: binary representation for the individuals’ chromosomes, classic one-point crossover and bit-wise mutation, roulette wheel selection, fixed-size population, elitism, a non-pareto fitness function (with an aggregation approach) and a genotypic termination criterion. The algorithmic behavior was first evaluated through the analysis of several randomly generated examples. Then, the MOGA was applied to a model of a small-size industrial plant. Its efficacy was assessed by contrasting the OA algorithm’s performance with and without MOGA initialization. The genetic algorithm proved to be advantageous because it led to a significant reduction in the number of iterations required by the rigorous OA algorithm. Since several objectives, including engineering rules of thumb, mathematical features, sensor reliability and cost factors, are taken into account by means of the MOGA, the resulting solution outrivals standard initialization procedures. The prototype presented in this article will serve as a sound basis for the development of the definitive MOGA module, whose implementation will support actual large-size industrial plant models.