Because of the inherent complexity of biological systems, a unique model cannot describe the dynamic behavior of the system. On the other hand, available mathematical models lose their reliability by slight changes in operating conditions of the system among which one can mention substrate concentration, pH, temperature, etc. Based on this, designing a controller based on mathematical models cannot assure the pursuit of control goals. Therefore, model-independent controller design methods like Fuzzy Logic Controllers (FLC’s) and Neuromorphic (Artificial Neural Network Based) controllers are preferred. Since the complexity of biological systems can restrict knowledge of their behavior, FLC approaches may not be useful. Here it is where a neuromorphic controller shows up and plays a vital role in overcoming these limitations. This approach is model-independent and does not require expert knowledge about the system. Furthermore, by providing sufficient sets of data, a neural network can be trained to imitate the dynamic behavior of the system. Besides, it is possible to train a neural network to act as a controller (Mandic and Chambers, ۲۰۰۱).To validate the above claims about ANNs superiority, a batch bioreactor consists of an aerobic bacterium, Bacillus thuringiensis (Bt), was chosen as a case study. Bt’s aerobic life necessitates an adequate oxygenlevel in the bioreactor. This microorganism produces δ-endotoxins, which is a pesticide (Rómoli et al., ۲۰۱۶). It is worth mentioning that the
dissolved oxygen concentration also affects the bacterium’s productivity. So, designing a precise controller for
dissolved oxygen concentration is vital. Different mathematical models represent the dynamic behavior of this system, but by using a neural network, the mentioned limitations will be mitigated. To prepare appropriate sets of data to identify and control the system, the mathematical model in (Isaza et al., ۲۰۱۶) was used. Note that the availability of a mathematical model is not necessary. As soon as the experimental data and a pilot or industrial setup is ready, the ANNbased methods could be implemented based on their operating data and independent of a mathematical model.