On plastics and chemometrics

I just got back from three days in beautiful downtown Houston, attending a meeting of licensees of ChevronPhillips Chemical's polyethylene plastic process. Every two years, they all get together to discuss common problems and to hear about all the latest research results from ChevronPhillips. A little background: polyethylene is the plastic used to make milk bottles (and a lot of other things), and ChevronPhillips, a 50/50 joint venture of ChevronTexaco and ConocoPhillips, is one of the world's largest producers of the stuff.

One of my consulting jobs is basically to install an on-line monitoring system for a new polyethylene plant that CPChem is licensing to another company. This system uses a process called Raman spectroscopy. Basically, in Raman spectroscopy you shine a laser light on the material of interest. Most of the light just bounces off, but a small percentage (maybe one photon in a million) interacts with the material by being absorbed for a few nanoseconds, then is re-emitted (scattered) at a different frequency than it came in at. The difference in frequency corresponds to characteristic vibrations of the atom-to-atom bonds in the molecules. The spectrometer captures those scattered photons and displays their frequencies and the intensity of each frequency. What I do is analyze that data to figure out what's really happening to the molecule.

Sometimes the analysis is pretty easy. For example, with the polyethylene process, you can see the starting materials at the beginning of the reaction, then watch the signal for the finished plastic grow in and the starting materials diminish as the reaction proceeds. This is a particularly handy thing to be able to do, since the reaction is done inside big metal tubes; without something on-line in real time like Raman, you can only analyze what you have at the beginning and end of the reaction. Alternatively, you can take samples out of the reactor and analyze them, but that can be difficult to do, and the analyses are often slow (up to a couple hours), while Raman analysis takes only about 90 seconds. The difference between on-line analysis and conventional analysis has been likened to the difference between a biopsy and an autopsy.

One of the really interesting things that can be done with Raman and some other types of spectroscopy is to determine properties for which the instrument does not receive any signals! I know that sounds a little strange, but let me give an example. If you take a Raman or infrared spectrum of gasoline, it's possible to see peaks that represent different chemical groups that make up the chemical compounds of which gasoline is made: aromatics, paraffins, olefins, and so on. However, there are some properties you'd like to know that are only indirectly related to those compounds. For example, the octane number of gasoline is crucial; octane measures the tendency of gasoline to 'knock,' and is used to differentiate different grades at the service station. Unfortunately, though the octane number of a gasoline sample obviously depends on the components used to make up that gasoline sample, there are no peaks that by themselves give the octane number. Octane can be determined in the laboratory using an engine that compares a given gasoline to two reference compounds. The analysis is difficult, requiring highly-trained technicians, and takes about two hours.

What can be done with that problem is a very elegant mathematical process that falls in the area where chemistry and statistics overlap, an area called chemometrics. To determine octane where no octane peaks are present, one first obtains a bunch of samples (20-50 for gasoline) of gasolines with different octane numbers. The spectra of those gasolines and the corresponding octane numbers are entered into a mathematical matrix, which is then manipulated by a chemometrics computer program (with a little help from me, the operator!) The result is an equation, called the correlation vector, that relates the height of each point in the spectrum to the octane of the gasoline. The correlation vector is simply a series of numbers, one number for each point in the spectrum. To get the octane number of an unknown gasoline, you simply run the spectrum and get the height of the response at each point. You then multiply each height by the correlation vector number for that point on the spectrum and add up all of these products; the result is the predicted octane number.

With polyethylene, one property of importance is the density of the plastic. Different uses for polyethylene require polymers of differing density, so it's important to know that number. During polymer synthesis, the producer wants to make the required material, so it's important for him to know quickly if he's making the right material or if the process is having problems. Raman spectroscopy coupled with chemometrics allows prediction of polymer density while the polymer is still being made, even though (as for gasoline octane) there is no single peak in the Raman spectrum that gives the density. It's also possible to determine the viscosity of the melted polymer before it's ever melted using Raman spectroscopy.

I'm working on a variety of these things for the new plant. By the way, ChevronPhillips has patent coverage on some of these processes, so much of what I'm doing is only possible because it's for a CPChem licensee. In fact, I couldn't even talk about some of these things if they hadn't already been published in the open literature.

Back to Houston. I got to spend several days listening to people talk about problems with pumps, replacing little parts of this and that, problems with how to prevent important holes from getting plugged up, and how to keep important walls from getting holes in them. And then we got to hear about Raman process monitoring!

I was very pleased about the positive reaction to a colleague's paper on the subject, and am even more pleased that the future may bring more opportunities for me to work in the area.

Even though I'm officially retired, there are so many fun and interesting things to do, how could I possibly just sit around or play golf?