Title: Selection of variables using FDA for the state identification of an Anaerobic UASB-UAF hybrid Pilot Plant, fed with winery effluents.
1Selection of variables using FDA for the state
identification of an Anaerobic UASB-UAF hybrid
Pilot Plant, fed with winery effluents.
- M. Castellano1, G. Ruiz 2, W. González1, E. Roca3
and J.M. Lema3 - 1Dep. of Statistics and O.R. University of
Santiago de Compostela, Spain - 2School of Biochemical Engineering. Catholic
University of Valparaiso, Chile - 3Dep. of Chemical Engineering. School of
Engineering. University of Santiago de
Compostela, Spain
IV International Specialized Conference on
Sustainable Viniculture Winery Wastes and
Ecology Impact Management Viña del Mar Chile,
November 2006
Winery2006
2This is about...
- The Anaerobic Wastewater Treatment
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
3This is about...
- The Anerobic Wastewater Treatment
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
4The Anerobic Wastewater Treatment
- The treatment characteristics
- Requires low energy Generates low sludges.
The problem Variations over Influent
properties and composition
Monitoring Diagnosis and Control System (MDC)
FOR Stable Operation Conditions
5The Anerobic Wastewater Treatment
- The solution
- Monitoring Diagnosis and Control System (MDC)
- early and automatic detection of perturbations
- (overload, presence of toxic, inhibitory
compounds, suddenly changes in pH)
6This is about...
- The Problem
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
7The Monitoring Control Variables
- Selection Criteria
- Low response delay
- High sensibility
- Low cost of both, sensor itself and its
operation-maintenance requirements.
- Previously
- Gas flow rate and H2/CH4 in the gas phase
- H2/CO in the gas phase
- H2 in the gas phase
- Gas flow rate and CH4 in the gas phase
- Alkalinities (total and partial) in the liquid
phase - pH in the liquid phase and gas flow rate
8The Monitoring Control Variables
- The statistical analysis
- Functional Discriminant Analysis (FDA)
- Classification
- Select the minimum number of variables for
process state identification purpose.
9This is about...
- The Problem
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
10Discrimination Statistical Techniques
- Functional Discriminant Analysis (FDA)
- Simple Statistical Classification Tool
- Linear Transformation of process variables
- Requires A priori knowledge about groups
- Objectives
- Minimize the missclassification error
- Minimize variance into each group
- Maximize variance between groups
11Discrimination Statistical Techniques
Weight
Mens mean
Women
Womens mean
Height
12Discrimination Statistical Techniques
- Other techniques of classification
- Consider more sophisticated functions lead to
more sophisticated classification techniques.
Some of the more popular and useful - Quadratic discrimination
- Non parametric density estimation functions
- Neural networks
13This is about...
- The Problem
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
14Application of FDA
- Selection of Variable using FDA
- FDA assigns data to different groups.
- The FDA classification is tested using all the
possible combinations of the variables in order
to select the best ones, so the most useful
variables for MDC.
15This is about...
- The Problem
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
16Experimentation
- The pilot plant and its instrumentation
- A UASB-UAF pilot plant fed with diluted wine. 26
variables were used to follow the process.
Measurement devices - feed and recycling flow meters
- pH meter
- inflow and reactor Pt100
- gas flow meter
- infrared gas analyser (CH4 and CO)
- gas hydrogen analyser
- TOC/TIC combustion analyser
- Other parameters were calculated methane and
hydrogen flow rate (Q CH4) (QH2) and organic
loading rate (OLR).
17Experimentation
- The experimental conditions
- Stable conditions at an OLR of 5 kg COD/m3d.
- Three consecutive increases of the OLR until SS.
- The duration of each state was around 5 days,
- (HRT was in the range of 0.6 to 1.5 d.)
18Experimentation
- The experimental conditions
19This is about...
- The Problem
- The Monitoring Control Variables
- Discrimination Statistical Techniques
- Application of FDA
- Experimentation
- Results and Conclusions
20Results and Conclusions
- Selection of Variable using FDA
- Classification analysis was made using 1
variable, all the combination of 2 variables and
so on.
21Results and Conclusions
- Selection of Variable using FDA
- 137 of the combination of 2 variables achieve a
100 of goodness classification.
The solution is not unique, so another criteria
should be used to select the variables for
monitoring
22Results and Conclusions
- Other criteria
- Constant temperature, influent pH and
recirculation flow rate. - Specific substance determinations in the liquid
phase are rare in industrial application - Qgas and P highly are correlated
- High cost of the on line equipment for TIC/TOC on
line measurement - Variables in the liquid phase are supposed to
present higher response time than the gas phase
variables
23Results and Conclusions
- The selected variables were QH2, H2, Qg, QCH4 ,
CH4
24Results and Conclusions
- Not subjective technique to select the variables
that should be used for an MDC system was
developed. - Not only one group of variables that must be
selected, but many combinations can achieve same
performance. - Economical and technical criteria have been
considered. - Gas phase variables obtain good results, even if
only one variable is selected (H2)
25For more information...
- María Castellano Méndez
- mcaste_at_usc.es
- Dep. of Statistics and O.R. University of
Santiago de Compostela, Spain
Gonzalo Ruiz Filippi gonzalo.ruiz_at_ucv.cl School
of Biochemical Engineering. Catholic University
of Valparaiso, Chile