(393h) Why Plant Operations Are Unstable after All the Design and How Data Science Can Help

Modern chemical plant operations involve a great deal of energy and material integration to be efficient. However, even though the plants were designed with sound principles, minor deviations or uncertainties in real-world operations can break the assumptions and lead to drastically undesirable operation behaviors, resulting in waste of energy, waste of raw material, loss of productivity, and compromised quality.

In this talk we demonstrate with many industrial examples how plant operations are oscillatory persistently without an effective way to identify the causes. We then present the use of data analytics and machine learning methods for troubleshooting and root cause diagnosis. Plant troubleshooting is different from process monitoring where a normal batch of data is assumed available or given. Plant troubleshooting deals with data that contain existing problems to be diagnosed. Dynamic component analysis and causality analysis of time series data are proposed for feature detection and root cause analysis. Data visualization is made possible by using the dynamic dimension reduction property of the methods. Industrial applications are used to demonstrate the effectiveness of the proposed methods.