Optimization of a hypothetical bioreactor through machine learning

By R. Mazurek
An investigation was carried out to determine if a Keras TensorFlow neural network model could be used to optimize the yield of a simulated hypothetical bioreactor, and hence improve its financial performance. The results of the investigation demonstrated that a neural network can be used to better understand the individual process parameters which contribute to the overall yield of the simulated process. The model identified the most important optimization parameters and provided guidance as to where improvements could be obtained through more precise process control. 
It is becoming easier to find opportunities to improve the performance of industrial processes by uncovering relationships which exist in operating data. This is due to the expanding ability to process large amounts of data from a variety of sources, and the development of machine learning technology. 

These developments are part of the overall growth of IIOT (Industrial Internet of Things) which is receiving so much attention today. 
This white paper discusses how a model of an industrial process can be created using machine learning technology, and then used to optimize the process. While this approach can be used on a wide range of chemical, pharmaceutical, and consumer product production processes, a bioreactor was chosen for this simulation as they are typically non-linear in response to process parameters and hence difficult to model empirically. 

copyright 2019 Powell Simulation LLC