Röle geri-beslemeli kontrol sistemi kullanarak bir süreci ifade eden transfer fonksiyonun bilinmeyen parametrelerinin belirlenmesi son zamanlarda oldukça popüler olmuştur. Yüksek dereceli gerçek süreç transfer fonksiyonları, genellikle, birinci dereceden veya ikinci dereceden kararlı, kararsız ve integratör içeren model transfer fonksiyonları cinsinden modellenir. Bu tür model transfer fonksiyonlarının röle geribeslemeli kontrol sistemleri kullanılarak elde edilmesine yönelik literatürde çok sayıda yayın bulunabilir. Ancak, süreç kontrolünde, bazen süreçler ters cevaplı bir karakteristik gösterebilir. Bu durumda model transfer fonksiyonu sıfır içerecek şekilde seçilmelidir. Artan model parametre sayısının nedeniyle literatürdeki yaklaşımlar ile modelleme işleminde bir takım sıkıntılar ortaya çıkmaktadır. Ayrıca, literatürde ters cevaplı süreçler için, röle geri-beslemeli kontrol sistemi ile modelleme için önerilen çalışma çok azdır. Bu yüzden, bu bildiri de genetik algoritma ile röle geri-beslemeli kontrol sisteminde ters cevapların modellenmesi verilecektir. Elde edilen modellerin uygunluğu, gerçek ve model transfer fonksiyonların frekans cevap karakteristikleri ve sistemin çıkışında elde edilen osilasyonlar karşılaştırılarak denenmiştir.

In a control system the controller parameters have to be chosen so that the system behaves in the desired way. There are two approaches to find proper values of the controller parameters. The first approach is to assume a mathematical model of the process and then find the controller parameters based on the assumed model. The second approach is to choose some controller parameters, observe the behavior of the feedback loop and modify the controller parameters until the desired behavior is achieved. Model-based controller design is becoming more popular in engineering research studies. Many advanced control strategies incorporate various aspects of the internal model principle, which requires a model of the system. Some proportionalintegral-derivative (PID) controllers also include an implicit process model in their design. For some controller design approaches, such as a Smith predictor scheme, a process model is a requirement. Therefore, being able to obtain an accurate process model is an important task. Recently, the relay feedback control (Aström and Hagglund, 1984) has been widely used for the identification of an assumed model. The method was originally proposed for autotuning of a process by using limit cycle information, Kc and ωc, directly, but later was also suggested for use in for parameter estimation of a plant transfer function (Luyben, 1987). There are several reasons behind the success of the relay feedback method. First, the relay feedback method, as normally used, gives important information about the process frequency response at the critical gain and frequency, which are the essential data required for controller design. Second, the relay feedback method is performed under closed loop control. If appropriate values of the relay parameters are chosen, the process may be kept in the linear region where the frequency response is of interest. Third, the relay feedback method eliminates the need for a careful choice of frequency. Finally, the method is so simple that operators understand how it works. In the literature, the use relay feedback method for open loop stable, unstable and integrating processes can be found. However, in practice it is possible to encounter processes with inverse response as well. There are only a few studies considering the use of relay feedback control system for identification of such processes. Also, the numbers of unknown coefficients in model transfer function of inverse response processes are increased; hence identification approaches existing in literature may become ineffective. Therefore, to overcome the difficulty in identifying model parameters of processes with inverse response, this paper a genetic based identification method using relay feedback control system for inverse response processes is given. Obtained model transfer function and the real process transfer function frequency response characteristics and limit cycle oscillations are compared to illustrate the effectiveness of the proposed identification method.