NIOSHTIC-2 Publications Search

A syntactic pattern-recognition approach for process monitoring and fault diagnosis.

Rengaswamy-R; Venkatasubramanian-V
Eng Appl Artif Intell 1995 Feb; 8(1):35-51
A syntactic pattern recognition approach for processing data containing noise and its use in monitoring a fluidized catalytic cracking (FCCU) process was presented. Trend descriptive language was developed to describe the primitives, episodes, trends, and profile of noisy process data. The process data profile included information on the shape of the data curves, their duration or episode length, and the minimum and maximum data points within an episode. A set of nine primitives, or trend shapes, were identified. The advantages of neural networks over filters to deal with noise in pattern recognition was discussed. The neural network was used at a low level to identify primitives from noisy data, then an error correcting parser abstracted and used the resulting information. The incorporation of a post/processor error correcting code (ECC) was necessitated by large amounts of process signal noise and artificial discontinuities introduced by the discretization of time. The ECC functioned by searching through the primitives of a string to determine replacement primitives in cases of discontinuities. Each time a replacement primitive generated a new string, it was assessed a penalty. Strings that exceeded a prespecified penalty were excluded. The grammar used to perform error correction and explanation generation was discussed. The use of context information by the ECC to detect multiple faults occurring at different times was illustrated. The application of the described approach was presented for a FCCU process. Schematics of a FCCU process were provided, as well as lists of faults and sensor variables for the application of the approach. The authors conclude that the approach was capable of summarizing the important trends and potential consequences to provide a cause and effect explanation of process behavior to an operator.
NIOSH-Grant; Control-technology; Data-processing; Mathematical-models; Computer-models; Information-processing; Analytical-models; Author Keywords: Fault diagnosis; knowledge-based systems; neural networks; qualitative modeling; trend analysis; sensor monitoring
Chemical Engineering Purdue University West Lafayette, IN 47907
Publication Date
Document Type
Journal Article
Funding Amount
Funding Type
Fiscal Year
Identifying No.
Issue of Publication
Source Name
Engineering Applications and Artificial Intelligence
Performing Organization
Purdue University West Lafayette, West Lafayette, Indiana
Page last reviewed: May 5, 2020
Content source: National Institute for Occupational Safety and Health Education and Information Division