Article Preview
Top1. Introduction
Over the past decades and more specifically since the 70’s, the manufacturing industries have deployed significant resources in the improvement of their production method to reach a higher ambitious level than that of competitors. Indeed, in the background of globalization and trade liberalization, companies' survival depends on their capacities to provide a high quality/price rate for their products (Dawar & Frost, 1999). In the pursuit of this goal, industries have taken advantage of the rapid development of computers computing power (More’s law) (Schaller, 1997)that further increases their competitiveness and productivity by applying several strategic changes, such as automation and optimization of production lines generally composed of conveyors, electrical machines and sensors. This research focuses on the induction machine, which is the main element of automation systems. This machine is considered as the most often used motors in industries because of its ease of implementation and its sturdiness (Boldea & Nasar, 2010).
As a consequence, the optimal exploitation (optimization) of the automated production lines to increase the production efficiency is mainly based on an accurate knowledge of parameters of the asynchronous machine, such as resistances and inductances on the stator and the rotor. Indeed, accurate values of this machine allow computing the closed-loop, drive-fed motor (e.g. speed optimization), managing power consumption (e.g. saving 5% of power consumption) in the best way possible, designing an electrical installation as precisely as possible (e.g. optimal exploitation for the greatest installation lifespan), and predicting induction machine failures with utmost effectiveness (e.g. predictive maintenance optimization).
This expectation can be provided by predicting the behavior of the induction machine. In order to predict the asynchronous machine behavior, the induction motor parameters have to be extracted. There are two major approaches to obtain these parameters. The first one is based on off-line techniques (external measurements, optimization algorithms) (Babau, Boldea, Miller, & Muntean, 2007; P. Kumar, Dalal, & Singh, 2014; Salimin et al., 2013), and the second relies on on-line methods (Recursive Least Square, Extended Kalman Filter…) (Vieira, Azzolin, Gastaldini, & Gründling, 2010; Yazid, Bouhoune, Menaa, & Larabi, 2011). In this paper, is the authors’ focus is on the optimization algorithms, and more specifically on the Evolutionary Algorithms (EAs). These algorithms have proved their efficiency in the optimization and identification domain (Cantelli, D’Orta, Cattini, Sebastianelli, & Cedola, 2015; Shafiei & Binazadeh, 2015).
EAs are the most common algorithms; they are currently used in industries and academic researches. Indeed, they allow seeking solutions to complex optimization problems by using the stochastic methods and estimating a criterion called objective function (or fitness function). The main advantage of these algorithms is that they do not need to estimate the derivative of this function, unlike traditional optimization algorithms. Accordingly, EAs, in particular Genetic Algorithms (GAs) are very popular for identification processes.
EAs, more specifically, GAs, Multi-Objective GAs and Particle Swarm Optimization (PSO) are broadly applied to estimate parameters of induction motors. In different papers (Huynh & Dunnigan, 2010; Karimi, Choudhry, & Feliachi, 2007; Yousfi, Bouchemha, Bechouat, & Boukrouche, 2013), researchers assert that PSO are more efficient than GAs. Nevertheless, because of the lack of information (especially the searching boundaries) on the identification parameter and the number of tests achieved, the difference between GAs and PSO could be negligible.