(155b) A Practical Application of Detection-Based Multiclass Classification of Faults: Tennessee Eastman Process

A practical
application of detection-based multiclass classification of faults: Tennessee
Eastman Process

Nour Basha, M. Ziyan Sheriff, Costas
Kravaris, Hazem Nounou, and Mohamed Nounou

10.0pt;font-family:" times new roman>Abstract

The swift detection and accurate
classification of faults is crucial for time-sensitive industrial processes in
order to meet minimum standards of quality and production and prevent
irreversible damage to valuable equipment. The problem of multiclass
classification is of great interest to literature nowadays, due to its heavy
reliance on advanced concepts of machine learning and data mining.

Model-free process monitoring
techniques, such as Principal Component Analysis (PCA), have proven to be an
excellent resource to researchers for applications involving the detection of
faults [1,2,3]. That is due to their ability to filter out noise, extract
correlations between variables and detect even small deviations from normal
operating conditions. In addition, sophisticated statistical methods, such as
the Generalized Likelihood Ratio (GLR), have frequently been used in tandem
with process monitoring techniques in order to significantly improve their
fault detection rate [4,5].

This work extends the application
of a combination of efficient process monitoring techniques towards the problem
of multiclass classification, namely the classification of faults found in the
benchmark Tennessee Eastman (TE) process [6,7]. Our proposed method also makes
use of established machine learning methodologies in order to map an array of fault
detection based decisions to a single classification decision per sample. The
precision of our proposed method is compared to the performances of multiple
deep learning techniques on the same dataset [8] in order to demonstrate a more
tangible argument of its efficacy and validate its performance.

10.0pt;font-family:" times new roman>References

line-height:normal">[1]    I. T.
Jolliffe, Principal Component Analysis, Springer, 2002.

line-height:normal">[2]    P. Sanguansat
(Ed.), Principal Component Analysis: Engineering Applications, InTech, 2012.

line-height:normal">[3]    P. Sanguansat
(Ed.), Principal Component Analysis: Multidisciplinary Applications, InTech, 2012.

line-height:normal">[4]    S. Wang, M. R.
Reynolds-Jr., A GLR control chart for monitoring the mean vector of a multivariate
normal process, Journal of Quality Technology 45 (2013) 18–33.

line-height:normal">[5]    M. Z.
Sheriff, M. N. Karim, H. Nounou, M. Nounou, Process monitoring using PCA-based
GLR methods: A comparative study, Journal of Computational Science 27
(2018) 227–246.

line-height:normal">[6]    J. J. Downs, E. F.
Vogel, A plant-wide industrial process control problem, Computers &
Chemical Engineering 17 (1993) 245–255.

line-height:normal">[7]    C. A. Rieth, B. D.
Amsel, R. Tran, M. B. Cook, Additional Tennessee Eastman Process Simulation
Data for Anomaly Detection Evaluation (2017). doi:10.7910/DVN/6C3JR1.

line-height:normal">[8]    S. Heo, J. H. Lee,
Fault detection and classification using artificial neural
networks, 10th IFAC Symposium on Advanced Control of Chemical Processes
51 (18) (2018) 470–475.