Aucun titre de diapositive

1
TELEMonitoring and Advanced teleControl of high
yield wastewater treatment plants
Proposal number IST-2000-28156
Kick-off Meeting
INRIA, Sophia Antipolis, 13-14 September 2001
Workpackage 4 Supervision System
J-Ph. Steyer LBE-INRA Narbonne
2
WP4 will ? design the supervision system
? develop knowledge base management ?
develop data base management. In order to
provide assistance to human operator to help
them address the problems.
3
WP4 within TELEMAC
4
The three fault levels that must be tested
Level 1
Level 2
Level 3
Physical components
Biological components
Theoretical components
(pipes, pumps, regulators,..)
(bacterial populations)
(algorithms, strategies,)
Fault complexity
5
The process used in Narbonne
Influent Raw industrial distillery
vinasses Reactor Circular column Up-flow
fixed bed reactor - 3.5 m height, - 0.6 m
diameter, - 982 liters of total volume.
Media Cloisonyl - Specific surf. 180
m2/m3 - Volume 33.7 liters Total effective
volume 948 liters
6
Schematic layout of the plant
TOC 2.35
TOC analyzer
CH4/CO2 sensor
ultrafiltration membrane
H2 sensor
gas flowmeter
Titrimetric sensor (TA, PA, VFA, Bic)
Biogas
Infrared Spectrometer (COD, TOC, VFA, TA, PA)
1 m3 Up-Flow Fixed Bed Reactor
NaOH
Water
Heater
Raw industrial vinasses
Heat Exchange
Temperature
pH

Dilution System
Output
pH
NaOH
7
Reasons for Diagnosis (in 1998)
2) Increase of pH in the reactor
4) Back to normal
3) Saturated Signal
Time (h)
Time (h)
Complete stop of the process for one month and
new start-up (with new sludges) required
8
Integration of a diagnosis modulein the control
scheme (at the local level)
Process
Actuators
Sensors
Controller
Fault Detection
Fault Analysis
Automatic Learning
9
Diagnosis based on Process Knowledge
10
Structural Analysis
Formalisation of subprocess interactions
Flowrec
Flowin
Theater
Text
Heater
Treact
Pumpheater
Unmeasured Variables
Flowheater
Text
Measured Variables
Actuators
Temperature Sub-System
Treact
Flowrec
Pumprec
Graph of causal influences
Theater
Pumpin
Flowin
Control
Local Control
Pumpheater
Flowheater
Sensors
11
Closed Loop FDI (in 1999)
2) Automatic stop of the feed flow (safety rule)
1) Problem with pHin regulation in the feeding
3) No increase of the pH
temps (h)
temps (h)
Time required before being back to normal 24
hours ! (vs. 1 month)
12
Benefits from On-Line Advanced Sensors (in 2001)
CODinput ? 17 g/l
CODinput ? 8.5 g/l
CODinput ? 17 g/l
Water
60
Voluntary changes of Qin
40
Influent flow rate (l/h)
20
Automatic firing of a safety rule
0
14
Problem on the pH regulation in the input
11
pH in the input (UpH)
8
5
2
12
Increase of pH in the reactor
10
pH in the reactor (UpH)
8
6
400
Stop of microorganism activity
Back to normal
200
Biogas flow rate (l/h)
Re-start of microorganism activity
Time (h)
0
0
50
100
150
200
250
300
350
13
Benefits from On-Line Advanced Sensors
Problem "too big" to be solved by closed loop
FDI but ... thanks to on-line advanced
sensors, less than 1 week to recover (vs. 1
month)
14
Local and Remote Supervision in WP4
Remote supervision system
Local supervision system
Local fault detection, isolation and
decision rules
Global fault detection, isolation and
decision rules
15
Local and Remote Supervision in WP4
Remote supervision system
Local supervision system
Based on the knowledge of data and events a
posteriori knowledge
Based on the process a priori knowledge
16
Deadlines and Deliverables in WP4
Year 2
Year 3
Year 1
T6
t12
t18
T24
t30
t36
Specifications
WP 4.1
Supervision system
Fault detection and isolation
WP 4.2
Data base management
WP 4.3
Coherency check
WP 4.4
Decision support system
WP 4.5
Integration
WP 4.6
17
USC contribution to WP4
WP 4.1 Contribution to the specifications of the
Supervision System 0.5 MM (5.3) Professor
Juan Lema (20), Associate Professor Enrique
Roca (40) Ph.D. student Amaya Franco (40) WP
4.3 Data base and Knowledge base management
2.5MM (6.5) Specification and structure (1
MM). Type of events to be stored (1.5 MM).
Professor Juan Lema (20) Associate
Professor Enrique Roca (20) Ph D student
(30) Engineer Jorge Rodríguez (30) WP 4.4
Quality and coherence check for mathematical
calculation 4 MM (12) On line check of the
consystency of the advanced control modules with
respect to their objectives (2 MM). Development
of a module to detect a steady state of the
process (2 MM). Professor Juan Lema
(10) Associate Professor Enrique Roca
(10) Ph.D. student Amaya Franco (30),
Engineer Jorge Rodríguez (30) Technician
(20 ). WP 4.5. Supervision system integration
2 MM (13.5) Encoding and integration of the
developed algorithms (2 MM). Professor Juan
Lema (10) Associate Professor Enrique Roca
(10) Ph.D. student Amaya Franco (50),
Engineer Jorge Rodríguez (30). Total in
WP4 9 MM (68.8)
Together in WP4.1 ?
18
INRA contribution to WP4
WP4.1 Supervision System Specifications 1
MM (5.3) WP4.2 Fault detection and isolation 17
MM (22.5) WP4.4 Quality and coherency check for
mathematical calculation 6 MM (12) WP4.5 Decision
support system 7.5 MM (13.5) Total in WP4
31.5 MM (68.8) Human resources related to
WP4 PhD student Laurent Lardon (100 - 36
MM) Research Scientist Jean-Philippe Steyer (10
- 5 MM) But also Post-Doc Ana Punal (100
- 12 MM)
19
CCLRC contribution to WP4
WP4.1 Supervision System Specifications 1 MM
(5.3) WP4.3 Data base and knowledge base
management 4 MM (6.5) WP4.5 Decision support
system 3.1 MM (13.5) Total in WP4 7.1 MM
(68.8) Human resources related to WP4 Simon
Lambert and Theo Dimitrakos Information Science
Engineering Group in the Business Information
Technology Dpt of CCLRC
20
SPES contribution to WP4
WP4.1 Supervision System Specifications 0.5MM
(5.3) WP4.2 Fault detection and isolation
(learning) 4 MM (22.5) WP4.4 Quality and
coherency check for mathematical calculation 2 MM
(12) WP4.5 Decision support system 1.5 MM
(13.5) WP.4 6 Supervision system integration 7
MM (9) TOTAL in WP4 15 MM (68.8) Human
resources related to WP4 Paolo Ratini
(engineer) 30 Stefano Ughi (software
developer) 100
21
LEMAIRE contribution to WP4
WP4.1 Supervision System Specifications 1.5
MM (5.3) WP4.2 Fault detection and isolation 1.5
MM (22.5) WP4.5 Decision support system 1
MM (13.5) Total in WP4 4 MM (68.8) Human
resources related to WP4 Engineer Pierre
Lemaire (? - ? MM) Other ? (? - ? MM)
22
DOMECQ contribution to WP4
WP4.1 Supervision System Specifications 0.8
MM (5.3) WP4.5 Decision support system 0.4
MM (13.5) Total in WP4 1.2 MM
(68.8) Human resources related to WP4 ? (?
- ? MM)
23
To Start with WP4.1

• WP 4.1 Supervision system specification
• ? Specify the hardware and software structures
  of the supervision system,
• ? Define the responsibilities and interactions
  of the different approaches
• and techniques to be used,
• ? Specify the database and knowledge base
  structure,
• ? Define the adaptivity requirements of the
  supervision system
• in link with a modular process structure,
• ? Specify the components of the supervision
  system that will be implemented
• on the plants and those that will be located in
  the monitoring centre.

INRA 1 MM CCLRC 1 MM SPES 0.5 MM USC 0.5
MM LEMAIRE 1.5 MM DOMECQ UK 0.8
MM TOTAL 5.3 MM
Deliverable 4.1 Report on Supervision System
Specifications (T06)
24
Some open questions
Mainly Related to Data/Knowledge Base
Structure ? How to represent the structure and
functioning of the plants in a way which is
general and adaptable, but also allows useful
reasoning ? ? Will this data/knowledge base be
the central representation for the entire
supervision system, or will it have a more
restricted role ? ? Will there be a
methodology for constructing the knowledge base
when a new plant is introduced ? ? How the
different techniques for fault detection and
diagnosis will be integrated ? ? Can we build a
framework in which different techniques can be
fitted ? ? How will the system choose between
techniques ? ? When will it alert the operator
at the remote centre ? ? How to make the
supervision system generic and adaptable to
different plants and remote centres ? ?
Hardware and software architecture (overlaps with
WP5) ?
25
To Get Ready for WP4.2 (Fault detection and
isolation )

• After receiving the data from the plants, they
  will be analysed in order to characterise the
  faults and their indicators that can occur in
  the hardware and software sensors, in the
  actuators, in the models and in the controllers.
  The challenge is here to make this step modular
  and generic to be used and maintain at the
  industrial scale while handling the biological
  dimension as already underlined, we will have
  to develop a specific and generic formalism that
  is appropriate to the workpackage objectives.
• Specific attention will be also paid to the
  management of the time between two occurrences of
  faults since it is recognised to be of high
  importance (i.e., the persistency of a fault will
  have to be managed, a fault X that occur after a
  fault Y may have a different meaning than if it
  occurs after a fault Z, ).
• These aspects will then be handled within a
  qualitative causal model that will allow us to
  handle the structure of the plant in an easy way.
  Fuzzy logic methodology will be also part of this
  task but it will be strongly coupled with
  model-based and pattern recognition-based methods
  to made the FDI very modular and very general for
  anaerobic digestion processes.
• Another very important point will be to provide
  correct diagnosis even in case of partial or
  uncertain information. Indeed, if a plant is
  poorly instrumented, results from mathematical
  models could be used but not on the same level
  than on-line measurements. Uncertainty is an
  important factor that will require specific
  attention in the FDI strategy. In addition,
  questions like which is the best suited model
  from the collection developed in WP3 in order to
  detect a fault on one process with one specific
  configuration ? or even though there is no
  measurements available for this variable and no
  model can provide correct prediction of its
  evolution, can it be suspected since other
  variables are in faulty mode?will have to
  answered automatically by the FDI scheme.
• Last but not least, the development of a
  learning strategy (mainly based on artificial
  neural networks and on statistical data analysis)
  will be studied in order to easily introduce new
  faults in the FDI scheme and to facilitate the
  maintenance efforts at the industrial scale.

In red, we need your help !!! In green, we'll try
to do our best ...
INRA 17 MM, SPES 4 MM, LEMAIRE 1.5 MM