Machine learning framework for petrochemical process industry applications
Year: 2018
Type: dissertation
Abstract: Machine learning has many potentially useful applications in process industry, for example in process monitoring and control. Continuously accumulating process data and the recent development in software and hardware that enable more advanced machine learning, are fulfilling the prerequisites of developing and deploying process automation integrated machine learning applications which improve existing functionalities or even implement artificial intelligence. \n \nIn this master's thesis, a framework is designed and implemented on a proof-of-concept level, to enable easy acquisition of process data to be used with modern machine learning libraries, and to also enable scalable online deployment of the trained models. The literature part of the thesis concentrates on studying the current state and approaches for digital advisory systems for process operators, as a potential application to be developed on the machine learning framework. \n \nThe literature study shows that the approaches for process operators' decision support tools have shifted from rule-based and knowledge-based methods to machine learning. However, no standard methods can be concluded, and most of the use cases are quite application-specific. \n \nIn the developed machine learning framework, both commercial software and open source components with permissive licenses are used. Data is acquired over OPC UA and then processed in Python, which is currently almost the de facto standard language in data analytics. Microservice architecture with containerization is used in the online deployment, and in a qualitative evaluation, it proved to be a versatile and functional solution.
Source: Aaltodoc (Aalto University)
Author: Toni Oleander
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2951849673