Adaptive disturbance rejection using AI

The main objective presented in this paper is to develop and implement the cascade control structure for the separation process and the enrichments of the
¹³
C isotope, being a time-consuming industrial process. This classical control structure is compared with an adaptive control method obtained using artificial intelligence methods. The classical cascade control strategy consists of a PI controller in the internal loop tuned using the modulus criterion and a PID controller in the external loop. This structure can reject disturbances that occur both on the input of monoethanolamine and on the isotope concentration itself, by implementing two control elements consisting of a PD controller with filter, each. The separation process’s settling time with enrichment of the
¹³
C isotope is reduced by implementing the adaptive cascade control structure: for this purpose, all the four controllers are discretized using an interval of independent sampling times for each corresponding case, resulting in a range of coefficients for each controller; these coefficients are learnt using neural networks, facilitating the conversion back to the continuous time, leading so to the possibility to minimize the process settling time.