Lots of data, no one looks at it – controlling a manufacturing process involves more than monitoring process parameters in isolation.  There are interactions among pressure, temperature, flow, level and chemistry that define the overall plant performance.  These interactions are further affected by factors such as the source or quality of the ingredients, position in the life cycle of any catalytic or cleanup beds, and corrosion or fouling of reaction vessels and heat exchange units.  Not only do all these factors have local impact, but in multistage processes each successive stage will be to some extent dependent on parameters in use in previous (sometimes even subsequent) zones.

The technology used in evaluating plant data as described in this overview is referred to as pattern recognition, data mining or (when applied to chemical data) chemometrics, and is based on multivariate statistics.  The purpose is to extract process-relevant patterns of association among the typically highly-correlated control measurements, using these patterns to build a fingerprint of the system as a whole.  The procedure automatically builds a reference set of plant profiles.  We use this previous plant experience to decrease operating cost, forecast production yield, monitor process consistency, and schedule maintenance.

We have access to a lot of data generated by sensors on the process line.  Plant personnel have little time to spend examining these data closely; they typically trend only a few key parameters.  Beyond simple trending, this is an information-rich data set.  Typically, we assemble the raw process data for an exploratory data analysis to quantify multivariate trends and groupings. Let us help you build a better working knowledge of your plant.