(78c) Visualization and Fault Detection in Radial Coordinates

Visualization and Fault Detection in Radial Coordinates
Ray Wang, Michael Baldea, and Thomas F. Edgar

McKetta Department of Chemical Engineering

The University of Texas at Austin, 1 University Station C0400, Austin, TX 78712

Mark Nixon, Willy Wojsznis, Ricardo Dunia

Emerson Process Management, Austin, TX

Email: mbaldea@che.utexas.edu

With the advancement of technology, many industries are devising new methods to produce and meet demand more efficiently. Data on various processes, be they chemical, manufacturing, or pharmaceutical, are collected using a variety of sensors. As processes increase in complexity to meet demand for innovative products, so does the number of sensors needed to keep track of the process. Furthermore, more precise control of the process may be needed, which leads to an increase in the sampling frequency of the sensors. Both of these trends lead to an increase in the volume of data collected.

As a consequence, process operating data have entered the “big data” age. While these data are readily available in databases or process historians, extract useful information for process analysis and decision support remains a difficult task.  An intuitive first step to understanding data is visualization; this is, however, challenging to accomplish for big process data using the customary 2D score plots.

To answer this challenge, in this paper we propose a variation of the parallel coordinate system [1], which we will refer to as 3D radial plots. 3D radial plots are based on using a closed-polygonal line to represent each data sample (as opposed to an open line used in parallel coordinates). The polygons representing data samples are “stacked” vertically according to the time data were acquired, which provides an explicit representation of time. Moreover, this representation allows for plotting of an arbitrary number of time-dependent variables on a single plot.

We exploit the geometric properties of the 3D radial plot framework and discuss several extensions, related to performing tasks such as fault detection and monitoring the evolution of a process during significant transitions between operating points. Further, we establish a connection between the proposed fault detection methodology and multivariate control charts.

The theoretical concepts are illustrated using industrial examples, amongst which, monitoring flooding in a distillation column and detecting surge events in a compressor.

References:

[1]          A. Inselberg, The plane with parallel coordinates, Vis. Comput. 1 (1985) 69–91.