Detection and Diagnosis of Plant-wide Oscillations: An Application Study

1
Detection and Diagnosis of Plant-wide
Oscillations An Application Study

• Vinay Kariwala
• M.A.A. Shoukat Choudhury, Sirish L. Shah,
• J. Fraser Forbes, Edward S. Meadows
• Department of Chemical and Materials Engineering
• University of Alberta

Hisato Douke, Haruo Takada Mitsubishi Chemical
Corporation, Mizushima, Japan
2
Outline

• Problem Description
• Detection
• Theory (Autocorrelation function)
• Application Results
• Diagnosis
• Theory (Valve Stiction)
• Application Results
• Future Directions

3
Problem Description
Condenser
Feed
Reflux Drum
Top Product
Stripper
Side Stripper
Bottom Product
Oscillations in Condenser Level
Distillation Column
4
Problem Description

• Condenser Level
• Oscillations with Large amplitude
• Back-off from Optimal operating point
• Economic Potential
• 1 increase in set points 20M Yen/year
• Previous attempts
• PID tuning, MPC model
• Not successful

5
Plant-wide Oscillation Detection
6
Scope of Analysis
7
Data Description

• Data Set 2880 samples, 1 min. data,
• Variables 45 Tags
• 15 Controller Outputs (MV)
• 15 SISO control loops
• 5 cascade control loops
• 2 DMCs

8
Detection Philosophy

• Which variables are oscillating?
• Which variables have common oscillations?
• Important to find
• All variables with common oscillations
• Root cause likely to lie within this set

9
Detection by Visual Inspection

• Multiple oscillations destroy Regularity
• Noise overshadows Oscillations

Fourier Transform
Time trends
Power Spectrum
Presence of Oscillation Peak in Spectra Period
and Regularity Difficult to Judge
10
Detection using ACF
Time Trend
Auto Correlation Function
Power Spectrum
Effect of Noise Reduced
ACF oscillates at same frequency as
signal Regularity of oscillations Zero
Crossings of ACF
11
Detection using ACF
ACF
Zero Crossings
Period of Oscillation
Oscillation regular if
12
Clustering using ACF
Oscillation considered significant if
(Power in selected band)/(Power in entire
spectrum) gt

• Two signals same frequency oscillation if

Ref Thornhill et al., JPC, 2003
13
Multiple Oscillations
Fourier Transform
Two peaks in Spectra Use Band pass
filters Calculate ACF for each filtered signal
14
Detection Algorithm

• Remove Non-stationary trends

Repeat if more than one oscillations present in
every filter range OR stop
15
Detection Results

• Low frequency range
• 158 min./cycle 27 tags
• 137 min./cycle 10 tags
• Medium frequency range
• 62 min./cycle 11 tags
• 75 min./cycle 23 tags
• 86 min./cycle 5 tags
• High frequency range
• 43 min./cycle 5 tags
• 25 min./cycle 1 tag
• 4 min./cycle 1 tag

Condenser Level
16
Low frequency detections
158 samples/cycle
137 samples/cycle
PV
PV
OP
OP
17
Summary of Detection

• Low frequency oscillations
• 158 minute/cycle
• 26 tags other than condenser level
• Plant wide nature of oscillations revealed
• Root cause should lie in this set

18
Diagnosis of Oscillations
19
Possible Reasons

• Poorly tuned Controller
• External disturbances
• Process induced oscillations
• Valve Problems
• MPC model mismatch

20
Definition of Stiction
valve output (mv)
valve input (op)
21
Test of Nonlinearity
Central Idea Nonlinear interactions between
different frequencies
Bispectrum
DFT
Normalized Bispectrum squared Bicoherence
22
Linear and nonlinear Signal
23
Test of Non-linearity (contd)
Non-Gaussianity Index and Nonlinearity Index
Critical Values of bic2crit is determined at 95
or 99 confidence interval of the squared
bicoherence
Gaussian Linear
Non-Gaussian Linear
Non-Gaussian Nonlinear
Frequency independent
Frequency dependent
24
Flow Control Loop in a Refinery
Loop is Nonlinear
NGI 0.02 and NLI 0.55
25
Pattern of Stiction in PV-OP Plot
PV
PV
OP
OP
26
Quantification of Apparent Stiction
4
x 10
1.145
1.14
1.135
1.13
a
b
1.125
PV
?
P
Q
1.12
1.115
1.11
1.105
38.1
38.2
38.3
38.4
38.5
38.6
38.7
38.8
38.9
OP
Apparent Stiction
0.35
27
Nonlinearity Analysis
28
Stiction Quantification
FC5
PC1
TC2
No Stiction
0.5
1.25
29
Research Directions

• ACF based Detection Algorithm
• False Detection, Premature Termination
• Stiction Quantification
• Assumption of linear disturbance
• Path Analysis
• Oscillation Propagation
• Model Predictive Controller
• Oscillations due to model mismatch

30
Acknowledgements

• NSERC
• Dr. Nina Thornhill, UK
• Ebara San, Amano San,
• Oonodera San
• Computer Process Control group