Artificial Intelligence and Machine Learning for Ethylene Plant Operation

Ethylene plants represent a large capital investment and have many operational factors that can be controlled and or accounted for. In this talk we present how advances in machine learning, including so-called deep learning, can be used to give real-time guidance and control. We cover the basics of these methods and give concrete examples of techniques such as long-term, short-term memory neural nets and other artificial intelligence (AI) techniques. We then will present details of how this can be applied to ethylene production and other industrial processes. The overall goal is to build a virtual foreman that takes all of the knowledge from operational history to make predictions that optimize operation based on production goals while minimizing downtime. Some possibilities for the system include predicting component failure and scheduling maintenance, as well as controlling the system to optimize output products as a function of feedstock, ambient conditions, and system state. Such as system is now possible without an investment in a large supercomputer, as would have been the case even 5 years ago. With the right system in place, AI can be a competitive advantage for an ethylene producer. Moreover, because machine learning improves over time, the investment in AI will appreciate rather than depreciate.