In this thesis, we address the problem of temperature control in a heat exchanger used in the pharmaceutical industry. The existing control system is based on traditional methods and has many possibilities for improvement. The methodology includes hardware and software, where the key methods are predictive control and the use of a transparent machine learning technique - a decision tree, which is used to generate a predictive model for predictive control purposes. The results demonstrate how to learn a predictive model using the collected data and machine learning algorithm and transparently integrate the trained predictive model into a control system based on a programmable logic controller.
|
