Maintenance of process control by statistical analysis of product optical spectrum

• US 5,862,060 A
• Filed: 11/22/1996
• Issued: 01/19/1999
• Est. Priority Date: 11/22/1996
• Status: Expired due to Term

• Current Assignee: UOP Llc (Honeywell International Inc.)
• Original Assignee: UOP Llc (Honeywell International Inc.)
• Inventors: Murray, Richard C. Jr.
• Primary Examiner(s): KEMPER, MELANIE A
• Application Number:
  US08/755,425
• Time in Patent Office: 788 Days
• Field of Search: 364/500, 364/496, 364/498, 364/550, 364/468.28, 364/554, 364/528.01, 364/528.06, 364/528.07, 356/300, 356/301, 356/302, 356/303, 250/339.12, 250/339.07, 250/339.09, 702/22, 702/23, 702/30, 702/31
• US Class Current: 700/266
• CPC Class Codes: G01N 21/359 using near infrared light
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04

Abstract

Control of many chemical processes is optimized using compositional data, rather than physical properties, as the means of control. Using near infrared spectroscopy as the analytical tool affords the opportunity for on-line measurements in real time. A calibration set of NIR spectra bounding the acceptable process space within which a particular property is to be controlled is obtained by conventional means. Applying a multivariant statistical method to the calibration set affords a method of identifying the most important characteristics of the set which governs the controlled property, and is inherently related to the composition of the process product. Control then is effected by using only a small number (2-4) of such characteristics, which can be determined quickly and simply from the measured spectra, as the control criteria to be applied to the process as a whole. The result is a very effective way of controlling a complex process using only 2-4 non-obvious criteria objectively determined from a calibration set and which can be applied in real time and virtually continuously resulting in a well-controlled process whose product composition is (ideally) invariant.

Claims

1. A method of controlling a chemical process comprising:

a) obtaining the spectrum of each member in a set of calibration samples produced in said chemical process, said set of calibration samples being composed of members each of which are representative product streams of said chemical process, said set of calibration samples bounding a standard product stream of the chemical process;

b) determining by principal component analysis a number, not more than four, of factors which can be used in combination with a score of that factor to express each spectrum in the calibration sample set;

c) determining the scores of each factor for the absorption spectrum of the standard stream;

d) determining the scores of each factor in the absorption spectrum of a test stream and determining their difference relative to the standard stream; and

e) using the difference in the scores of each control the process.

2. The method of claim 1 where the spectrum is obtained in the optical range of the electromagnetic spectrum.

3. The method of claim 1 where the spectrum is a near infrared spectrum.

4. The method of claim 1 where the spectrum is a nuclear magnetic resonance spectrum.