Multivariate detection of abnormal conditions in a process plant

• US 7,937,164 B2
• Filed: 09/28/2007
• Issued: 05/03/2011
• Est. Priority Date: 09/29/2006
• Status: Expired due to Fees

• Current Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Original Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Inventors: Hamad, Ahmad A., Samardzija, Nikola
• Primary Examiner(s): Patel; Ramesh B
• Assistant Examiner(s): CHANG, SUNRAY
• Application Number:
  US11/864,529
• Publication Number:
  US 20080082302A1
• Time in Patent Office: 1,313 Days
• Field of Search: None
• US Class Current: 700/28
• CPC Class Codes: G05B 23/021 adopting a different treatment of each operating region or a different mode of the monitored system, e.g. transient modes; different operating configurations of monitored system
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: G05B 23/0245 based on a qualitative model, e.g. rule based; if-then decisions
• CPC Class Codes: Y10S 715/965 for process control and configuration

Abstract

Methods and systems to detect abnormal operations in a process of a process plant include collecting on-line process data. The collected on-line process data is generated from a plurality of dependent and independent process variables of the process, such as a coker heater. A plurality of multivariate statistical models of the operation of the process are generated using corresponding sets of the process data. Each model is a measure of the operation of the process when the process is on-line at different times, and at least one model is a measure of the operation of the process when the process is on-line and operating normally. The models are executed to generate outputs corresponding to loading value metrics of a corresponding dependent process variable, and the loading value metrics are utilized to detect abnormal operations of the process.

Claims

1. A system for facilitating detection of an abnormal operation of a process in a process plant, the system comprising:

a data collection tool adapted to collect on-line process data from a process within the process plant, wherein the collected on-line process data is representative of an operation of the process when the process is on-line and wherein the collected on-line process data is generated from a plurality of process variables of the process including one or more independent process variables that affect a behavior of the process and one or more dependent process variables dependent upon at least one of the one or more independent process variables, wherein the one or more dependent process variables describe the behavior of the process, and wherein the collected on-line process data comprises a plurality of observations of a plurality of process variables and wherein the plurality of observations of a set of collected on-line process data comprises a first data space having a plurality of dimensions;

an analysis tool comprising a multivariate statistical analysis engine adapted to generate a first representation of the operation of the process within a data space having fewer dimensions than the first data space based on a first set of the collected on-line process data generated from the plurality of process variables of the process, wherein the first representation of the operation of the process is adapted to be executed to generate a first outcome related to a dependent process variable, wherein the first outcome is a first loading value metric of the dependent process variable for the first set of collected on-line process data, and wherein the multivariate statistical analysis engine adapted to generate a second representation of the operation of the process within a data space having fewer dimensions than the first data space based on a second set of the collected on-line process data generated from the plurality of process variables of the process, wherein the second representation of the operation of the process is adapted to be executed to generate a second outcome related to the dependent process variable, wherein the second outcome is a second loading value metric of the dependent process variable for the second set of collected on-line process data; and

a monitoring tool adapted to determine changes in the process based on a comparison of the first and second outcomes related to the dependent process variable.

2. The system of claim 1, wherein the monitoring tool is adapted to analyze a change between the first and second outcomes related to the dependent process variable to detect an abnormal operation of the process.

3. The system of claim 1, wherein the first and second outcomes related to the dependent process variable comprise two of a plurality of outcomes related to the dependent process variable generated from corresponding representations of the operation of the process based on different sets of the collected on-line process data generated from the plurality of process variables of the process, wherein the monitoring tool is adapted to monitor a trend in the plurality of outcomes to detect an abnormal operation of the process.

4. The system of claim 1, wherein the first set of the collected on-line process data is representative of a normal operation of the process when the process is on-line and operating normally, wherein the first outcome comprises a baseline outcome related to the dependent process variable, and wherein the second set of the collected on-line process data is representative of a monitored operation of the process when the process is on-line and operating.

5. The system of claim 4, wherein the monitoring tool is adapted to compare the second outcome to the baseline outcome to detect an abnormal operation of the process.

6. The system of claim 1, wherein the multivariate statistical analysis engine is adapted to perform a principal component analysis on the first set and second set of the collected on-line process data to generate corresponding first and second loading matrices each representing the operation of the process, wherein the first and second outcomes related to the dependent variable each comprise a loading value metric of the dependent variable generated from a component of the corresponding loading matrix, wherein the component corresponds to loading values corresponding to the dependent variable.

7. A method of facilitating detection of an abnormal operation of a process in a process plant, the method comprising:

collecting on-line process data from a process within the process plant, wherein the collected on-line process data is representative of an operation of the process when the process is on-line and wherein the collected on-line process data is generated from a plurality of process variables of the process including one or more independent process variables that affect a behavior of the process and one or more dependent process variables dependent upon at least one of the one or more independent process variables, wherein the one or more dependent process variables describe the behavior of the process, and wherein the collected on-line process data comprises a plurality of observations of a plurality of process variables and wherein the plurality of observations of a set of collected on-line process data comprises a first data space having a plurality of dimensions;

generating a first multivariate statistical representation of the operation of the process within a data space having fewer dimensions than the first data space based on a first set of the collected on-line process data generated from the plurality of process variables of the process;

generating a first outcome related to a dependent process variable from the first multivariate statistical representation, wherein the first outcome is a first loading value metric of the dependent process variable for the first set of collected on-line process data;

generating a second multivariate statistical representation of the operation of the process within a data space having fewer dimensions than the first data space based on a second set of the collected on-line process data generated from the plurality of process variables of the process;

generating a second outcome related to a dependent process variable from the second multivariate statistical representation, wherein the second outcome is a second loading value metric of the dependent process variable for the second set of collected on-line process data; and

determining the presence of an abnormal operation based on a comparison of the first and second outcomes related to the dependent process variable.

8. The method of claim 7, wherein determining the presence of an abnormal operation comprises analyzing a change between the first and second outcomes related to the dependent process variable to detect an abnormal operation of the process.

9. The method of claim 7, wherein the first and second outcomes related to the dependent process variable comprise two of a plurality of outcomes related to the dependent process variable generated from corresponding representations of the operation of the process based on different sets of the collected on-line process data generated from the plurality of process variables of the process, wherein determining the presence of an abnormal operation comprises monitoring a trend in the plurality of outcomes to detect an abnormal operation of the process.

10. The method of claim 7, wherein the first set of the collected on-line process data is representative of a normal operation of the process when the process is on-line and operating normally, wherein the first outcome comprises a baseline outcome related to the dependent process variable, and wherein the second set of the collected on-line process data is representative of a monitored operation of the process when the process is on-line and operating.

11. The method of claim 10, wherein determining the presence of an abnormal operation comprises comparing the second outcome to the baseline outcome to detect an abnormal operation of the process.

12. The method of claim 7, wherein generating the first multivariate statistical representation comprises performing a principal component analysis on the first set of the collected on-line process data to generate a first loading matrix representing the operation of the process at a first time, and

wherein generating the second multivariate statistical representation comprises performing a principal component analysis on the second set of the collected on-line process data to generate a second loading matrices representing the operation of the process at a second time,

wherein the first and second outcomes related to the dependent variable each comprise a loading value metric of the dependent variable generated from a component of the corresponding loading matrix, wherein the component corresponds to loading values corresponding to the dependent variable.