(155c) Practical Machine Learning for Process Control, Monitoring and Diagnostics

Process monitoring and control has historically had a rich influx of data centric approaches. The advent of Big Data and Machine Learning/AI has promised much to enhance current capabilities. In many cases, these approaches have struggled to take off because (a) they are too complex and impractical to maintain in industry and/or (b) they have failed to include the most important ingredient during design and execution – the (domain) subject matter expert (SME) who understands the process. Utilizing SME knowledge and feedback is critical for the success of any advanced analytics framework in industry.

In this talk, requirements for a practical, industrially implementable, machine learning framework are discussed. Three examples are discussed. The first application highlights the capability of a practical machine learning framework to simplify APC maintenance and troubleshooting when used in conjunction with domain expertise. Such an approach can also be leveraged during controller and inferential modeling to identify important features and embed known, domain specific relationship constraints. The second example demonstrates how machine learning methods can be used to simplify APC/RTO LP/QP tuning utilizing economic objectives and operational preferences. The third example highlights how practical machine learning can play an important part in gathering insights from vast volumes of data. Data used to gather insights can be from diverse sources – process data, alarms, events are some common examples.