Skip directly to search Skip directly to A to Z list Skip directly to page options Skip directly to site content

NIOSHTIC-2 Publications Search

Search Results

Application of neural networks for process fault diagnosis and safety.

Authors
Venkatasubramanian-V
Source
School of Chemical Engineering, Purdue University, West LaFayette, Indiana 1995 May:9 pages
Link
NIOSHTIC No.
00228695
Abstract
A final performance report on neural network fault detection and diagnosis was presented. The research goal was the development of neural networks for the design of chemical process hazard detection, prevention and safety systems. Only highlights of the findings were presented. Previously published supporting material were referenced. Linear activation function based neural nets, using hyperplanes in the input units space, were considered unsuitable for the description of fault space. The use of an ellipsoidal activation function for the hidden nodes generated bounded fault classes without leading to generalization problems. Several related problems involving the number, location and size of the ellipsoids were solved in referenced material. Network, training set and input space decompositions significantly reduced computational complexity and training time by using principal component analysis and hidden nodes specialization to reduce input space. The extraction of relevant qualitative information from noisy data was accomplished by the development of a hierarchical representation to model process trends. Classes of trends from noisy time series were identified using neural network techniques. Applicability of these techniques to process monitoring was tested using computer simulations of a continuous stirred tank reactor and a fluidizied catalytic cracking process system. At 5% to 10% noise levels, the neural net provided noise tolerant fault diagnosis. Noise tolerance was found to depend on the importance of the sensor monitored and the number of other sensors read by the network. The usefulness of a neural network as a diagnostic system for complex, potentially dangerous processes to aid operators in accident prevention was discussed. The author concludes that neural networks are superior to alternatives for many problems in process fault diagnosis.
Keywords
NIOSH-Grant; Control-technology; Analytical-models; Computer-models; Information-processing; Simulation-methods
Contact
Chemical Engineering Purdue University West Lafayette, IN 47907
Publication Date
19950502
Document Type
Final Grant Report
Funding Amount
161706
Funding Type
Grant
Fiscal Year
1995
NTIS Accession No.
NTIS Price
Identifying No.
Grant-Number-R01-OH-02740
NIOSH Division
OEP
Priority Area
Control-technology
Source Name
School of Chemical Engineering, Purdue University, West LaFayette, Indiana
State
IN
Performing Organization
Purdue University West Lafayette, West Lafayette, Indiana
TOP