Powitec Power en BHEL 2011-11-22

Powitec Intelligent Technologies – Germany Page 1

Powitec

ChallengesChallenges

Solution

References

Discussion


Presentation Powitec

Challenges

Powitec‘s solution

References

Discussion

Advanced Combustion Advanced Combustion

OptimisationOptimisationin Large in Large Scaled ThermalScaled Thermal Power PlantsPower Plants

process optimized. value delivered.


Powitec


Solution

References

Discussion

� Founded in 2001

� Powitec offers optimisation solutions for:- Cement and Lime Industry (rotary kilns)- Fossil fired Power Plants- Waste-to-Energy Combustion Plants

� 45 employees experienced in complex industrial combustion processes

� Expertise in Neural Network & Digital Image Processing

� Partnership with Technical University Ilmenau, Germany FACULTY OF COMPUTER SCIENCE AND AUTOMATION: Neuroinformatics and Cognitive Robotics Lab

� Honoured in 2010 by Federal Minister of Environment with German Innovation Award Climate and Environment for Outstanding and Sustainable Technology

� More than 100 references: E.ON, Vattenfall, Lafarge, BuzziUnicem, Holcim, HeidelbergCement, Cemex, Dyckerhoff, CRH, Carmeuse, Cimpor, Titan, Lhoist in Europe, Africa, Asia and North America

Powitec Intelligent Technologies GmbHProcess Optimisation with Intelligent Technologies

Essen

Ilmenau


Powitec


Solution

References

Discussion

Dr. Röttgen: German Environment Minister, F. Wintrich: Technical Director Powitec, B. Beyer: Commercial Director Powitec

Dr. Schnappauf: CEO BDI, Prof. Töpfer: Former German Environment Minister and Director UN Environmental Program

German Innovation Award German Innovation Award

Climate and Environment Climate and Environment

for Outstanding and Sustainable Technologyfor Outstanding and Sustainable Technology


Powitec


Solution

References

Discussion

Online-flame- and boiler wall charac-

teristics by using different sensors+ automatic mutual information generation

PowitecPowitec‘‘s modular solutions modular solution

Flame and Boiler AnalysisFlame and Boiler AnalysisOnline- mill- and coal- characteristics

by frequency analysis+ automatic mutual information generation

Coal System AnalysisCoal System Analysis

Adaptive validation and analysis with innovative neural nets of all sensors

and process data describing the process.

Multi-Correlation (MiMo, Cross…)

Online-CFD

Open-Loop-Analysis and -Indication

AutoAuto--Analyse Sensor and Process DataAnalyse Sensor and Process Data

Self learning adaptive boiler optimisation with innovative neural nets

=> active learning to permanently optimize the air-/fuel-ratio

Closed-Loop-Control

AutoAuto--Optimise the ProcessOptimise the Process


Powitec


Solution

References

Discussion

Do you need ClosedDo you need Closed--Loop boiler control?Loop boiler control?

• Does every operator/every shift run the mill-burner-boiler

–system the same way?

• Are experienced, intuitive operators readily available these days?

• Do your operators try to find the best available settings

and trims every 1-5 minutes?

• Is the mill-burner-boiler –system capacity permanently

and completely utilised at its technical optimum?

• When things are fixed or adjusted in your plant do they

stay fixed and adjusted?

If you answered YES to all, you don`t need closed-loop boiler control...


Powitec


Solution

References

Discussion

ClosedClosed--Loop ControlLoop Control --> PiT Navigator> PiT Navigator

• Continuous on-line data (signals) from the process

via DCS,

• the possibility to influence the process via automated

actuators,

• a complex multi-dimensional process with significant

reaction times,

and controls this process

to a new optimum.

Auto Optimizer “PiT Navigator” is based upon

PiT Navigator is applicable for the opti-misation of stand-alone single aggre-

gates as well as for an overall, integrated process optimisation approach!


Powitec


Solution

References

Discussion

ClosedClosed--Loop ControlLoop Control --> PiT Navigator> PiT Navigator

• extensively uses existing process data,

• automatically selects and extracts features by relevance ranking,

• automatically generates process models (process description)

• uses the process models for optimising,

• and sends set-point corrections to the existing control system.

Auto-Optimization Solution PiT Navigator

=> Intelligent software for a continuous, optimized control

Thus PiT Navigator is• a Multi dimensional optimizer (like a group of experts),

• generically learning (like a human),

• and adapting automatically to changing process situations (like a human).

This benefits in

• stabilized and optimized process operation,

• increased production at decreased specific energy,

• fully automated operation without the need for human interventions.


Powitec


Solution

References

Discussion

N: Non-Linear

M: Model Computer models describing the process

P: PredictiveDecisive process results (NOx, CO, FCaO…) predicted

C: Control Permanent (24/7) Closed Loop Control

NMPC NMPC –– NonNon--Linear Model Predictive ControlLinear Model Predictive Control

Linear Non-Linearvs.


Powitec


Solution

References

Discussion

PiT Navigator: PiT Navigator: Advantages (1)Advantages (1)

• SELF-LEARNING: Mathematic statistical models do not rely

on expert knowledge; they learn from existing process data automatically

• ADAPTIVE: PiT Navigator process models learn

continuously by themselves to extend the knowledge from new process situations

• PERMANENT: Models are self-optimising 24 hours 7

days a week, even at already good, but still improvable process situations

• FLEXIBLE: Easy changes in optimisation targets

without reprogramming orre-parameterisation n of software


Powitec


Solution

References

Discussion

PiT Navigator: Advantages (2)

• FAST: Commissioning of NMPC software within 2-3

weeks on-site and with 5-10 man days customer involvement only

• COMPLETE: Vibration information and optical information

are analysed as well as non-linear correlations between all process data

• STRAIGHTFORWARD: Total cost of ownership is low, as no

permanent manual adaptation and re-programming is required: Self-training!


Powitec


Solution

References

Discussion

Powitec Patents:Powitec Patents:

80 Intern

ational P

atents

15 patents in applic

ation phase


Powitec


Solution

References

Discussion

Data acquisition from Process Control SystemData acquisition from Process Control System

Fuel amount, lab values

Feederamount, speed, height

Flue gas recirculation (mills)amount, pressure, temp.

mills (1 to n)

momentum, temperatures, amount

etc.

Mill airsamount, pressure, temp.

Burners (1 to n)

flap pos., flame detectors

De-Asher (boiler-/fly-/filter-ashes)

Amount, speed, height

AshAmount, lab values

Steam/fresh wateramount, pressure, temp.

Flue gasafter combustion chamber before/after heating

surfaces, amount, pressure, temp.,

CO, O2, NOx, SO2

Induced drought fanamount, pressure, temp., Delta p, Pel

Flue gas recirculation (boiler)amount, pressure, temp., Delta p, Pel

Heating surfacesBoiler wall

CO, O2, NOx,

Burn out airsamount, pressure, temp., flap pos.

Burner airsamount, pressure, temp., flap pos.

Forced draft fanamount, pressure, temp.,

delta p, Pel, flap pos.

Airamount, temperature, moisture

If data is available !!!


Powitec


Solution

References

Discussion

Process Optimisation systems: StatusProcess Optimisation systems: Status

Expert Systems Operator

Process

Interlocking

Actuating variablesMeasurements

Characteristics curves

Controller

PIDControl

System


Powitec


Solution

References

Discussion

Advanced Combustion ControlAdvanced Combustion Control

� Intensive usage of existing process data

�Automatic feature selection and extraction (relevance ranking)

�Automatic model generation (regression, neuronal networks, probabilistic nets, Gray-Box-Models)

�Use of the process models for optimising

�Set point integration into the DCS/PCS as correction values

(old system stays unchanged and ready for operation)

Expert System/ Database

Operator

Process

Interlocking

ActuatorsMeasure-ments

Characteristics curves

Controller

PID –controller

Relevance ranking

Process models +controller models

Optimiser

Data

Correctio

n

Management

Targ

ets

Control System

Data


Powitec


Solution

References

Discussion

1Proleit

2Oracle 9i

1Oracle 10g

1XTC

2Teleperm XP4RK 512 TCP

5Teleperm M7RK 512

1Symphony/Maestro19Profibus DP

3S5 Redundant1PMSX

1S51

OPC DA

Redundant

1Procontrol P40OPC DA

2Procontrol1Emerson14Modbus RTU

26PCS 741Siemens0Modbus +

1Infi 901KH-Automation6etmix

1Delta V5FLS1Analog

1AC 870P17ABB11Adlink

DCS TypeDCS MakerProcess

Connection

Type & amount of realised Process Connections

(as of June 2010)


Powitec


Solution

References

Discussion

Realised process connectionsRealised process connections

Procontrol P

Teleperm XP

Teleperm M

800xA

Symphony Maestro

FLS

PCS 7

Profibus DPMaster

CIF10 / CP341 Modbus Master

87TP01 87TS01Modbus Master

CM104

CP 441RK512

Intellution Inc. OPC Data Access 2.0

Server for iFix

AC870P

S7

S5

Contronic S

PowitecSystem Server

Analog

PMSX

HilscherCIF50DPS

Modbus RTU/TCP

Client

RK512 3964rRK512 TCP

Analog

OPC DA 2.05aClient

PMSX Client

PMSX pro

Ethernet


Powitec


Solution

References

Discussion

Integration of the set-point correction into the DCS (example)

Communication Interface between DCS

and PiT Navigator

Detailed process information (instrument

signals and DCS channels) with high quality

and in time.

The following interface options could be

provided by Powitec:

Direct link via RK512/3964R protocol

The process signals from the DCS, SCADA

or plant control system are connected via a

communication processor (installed by the

customer) to the PiT Navigator system. No

additional cost.

File exchange / Profibus / Modbus RTU

The process signals from the DCS, SCADA

or plant control system are cyclically written

into a file that contains finally the valid

information at a defined point in time. This

e.g. is possible with a OPC server with

scripting function. No additional cost.

OPC client

The PiT Navigator will be added by an

interface that is linked as an OPC client to an

existing OPC server. Additional costs 2.000

EUR (OPC client software licence).

DCS

combustioncontrolsystem

+

PiT System-Server

X

Profibus DP Master(e.g. Siemens S7 416DP)

Profibus Client(Hilscher CIF 50 DPS)

PID

) (

F

security layere.g. min/max airflow

air pipe

PiT Navigator

input:- setpoint airflow- actual airflow

output:- delta setpoint airflow

delta airflow

Profibus DP

setpoint airflow

actual airflow

t

1Powitec online


Powitec


Solution

References

Discussion

Tools for Process OptimizingTools for Process Optimizing

Relevancy / Sensitivity analysis:

� Which variables interact?Approach: � Data driven, no expert knowledge necessary

� Calculation of information content between variables� Maximisation of information content of extracted features:

feature

extractor

target

feature

DCS

mutual

information

Sensors


Powitec


Solution

References

Discussion


Automatic feature selection:

� Data driven, no expert knowledge necessary

� Calculation of the information content between target and measured variables

� Calculation of the redundancy between measured variables

� Elimination of redundant or not informative variables

x x'

Selection


Powitec


Solution

References

Discussion


Process modelling:

� Usage of existing process data + new sensor data

� statistic data driven process description

Approach:

� Neuronal Networks

� Probabilistic Models

� Gray-Box-Models

Application:

� Model represents the quantitative relations between all relevant values variable x

variable yvar

iable

p


Powitec


Solution

References

Discussion


Probabilistic Models

� description with probability distribution

� explicit modelling of fuzziness


Powitec


Solution

References

Discussion

Working on 3 different time scales

• Short term scale

– Control of short termed process changes like CO, NOx

– Rule based & PID …

• Middle term scale

– Predictive control of process values with high dead times like kiln torque, free lime etc.

– Neural Net Based

– Controlling on basis of current and predicted values

• Long term scale

– Following the long term targets like production increase, energy efficiency

– According to scoring


Powitec


Solution

References

Discussion

Unique features

� Distribution of the reward to the single agents in a multi-agent system

� Including expert knowledge in a learning system without constraining the learning process

� Applicability of each learning system on different applications

� Correct behaviour in critic states of a plant

+

Reward

If case x and control y

� bad result


Powitec


Solution

References

Discussion

Overview of current controller used by Powitec:

PID„classic“ Controllers:� PID- Controller� Rule-Based-Controller

Self learning Controller:� neuronal Controller:

� Pnav� CoSyNE� ADHDP ( Action Dependent Heuristic Dynamic Programming )

� Bayes System� Bayes-Controller

� Reinforcement learning� Bayesian Fitted Q-Iteration

� Mixed Controller� Neural Fitted Q-Iteration

Finite Elements Model

R?

+

+

Reward

Reward


Powitec


Solution

References

Discussion


Optimisation:

� Process model reflect quantitative relationships between

all process targets and parameters

� „inquiring“ the process model about the optimal setting

Possible optimisation potentials:

� Temperature balance

� Emissions (CIA, NOx, CO)

� Efficiency


Powitec


Solution

References

Discussion

Generic Modelling

• Self learning

• Transferable: Not depending on a specific process

• Structural designed with different decision and knowledge levels (like companies)

• Emergence: New properties or structures following the interplay of its elements

• Integration of Linear + Fuzzy + PiD + AI

• Feature generation instead of single data storage

…to achieve that the

• PLS receives a “human-like” brain

• same programmed structure is usable in a large variety of plants and processes


Powitec


Solution

References

Discussion

Strong changes in coal quality?

• Powitec recommends installation of

PiT VibraSensors

• Vibration Sensors at mill, classifier and PF-pipes

gather additional information

• From automatic feature extraction information

about the current coal cluster


Powitec


Solution

References

Discussion

PiT Vibrasensor

• 1-2 sensors per mill, 1 per classifier, 1 per pipe

• Piezo-ceramic sensor for

industrial use following military specification

• Positioning decided following trials,

fixed by magnets at locations <160°C

• Sampling with 10kHz, 16bit

• Data processing by analysis of spectral

characteristics (concentrations), statistic

verification and analysis of correlations

(Mutual Information Optimisation)

• Single server for acoustic analysis

Online vibration analysisOnline vibration analysis

Millcasing

Sensor position


Powitec


Solution

References

Discussion

Online Prediction of Coal Quality

Volatiles

Water

Ash

Heating Value

COAL MODEL

ADAPTIVITY

Process Data

Lab. Data (offline) via:- PHUMI- or DB-query

Model Retrainingand Testing

OnlinePrediction

PROCESSNAVIGATOR

Adaptive Coal Modelsare used as “Soft-Sensors”to provide additional online information for the PiT Navigator

Sensor Data


Powitec


Solution

References

Discussion

Online Coalmodel WATER (Hard Coal, Fenne)red=Prediction, green=Train Data, blue=Test Data. (all in %)


Powitec


Solution

References

Discussion

Online Coal Model VOLATILES

(Hard Coal, MKV Fenne)

red=Prediction, green=Train Data, blue=Test Data. (all in %)


Powitec


Solution

References

Discussion

PiT Navigator: Typical Project Steps

1. Analysis of mills, PC pipes, burner, boiler – system2. Analysis of PLC status and connection possibilities3. Common definition of interface, targets, actuating variables ramps,

borders, alarming4. Data:

– Reading connection to PLC

– Data validation and pre-processing

– Data analysis towards correlations, relevancy analysis, informationcontent, feature extraction

– PID-Controller and PLC controllers and characteristic curves analysis

– Online-validation of existing measurements

– PLC-Programming for set-point correction acceptance (3rd party)

– Writing connection to PLC

5. Common definition of PiT Navigator surfaces and reports6. Software training, fine tuning and monitoring7. Common analysis of suggested changes8. Activation – analysis of results (may be fine-tuning)9. Performance test – result evaluation10. PiT Navigator in continuous optimising mode


Powitec


Solution

References

Discussion


Some ReferencesSome References

Power plant Burner installation

burners/mills

MWel Results with PiT Navigator

KomipoSeocheon TPP

Slag tap fired boilers

202 x 215

Boiler 1 and 2Unburned Carbon in Ash -0,8% abs. Unburned Carbon in Ash -1,2% abs.

EvonikKW Fenne

Shifted boxer 8 / 4 195

Boiler 1η steam generator: + 0,4% abs

λ: 1,25 to 1,18

Auxiliary cons.: -2000MWh/yUnburned C in Ash -0,2% abs.+ 180.000 €/y

VattenfallKW Tiefstack

Front wall 6 / 32 x 252

Boiler 1 and 2η steam generator: +0,3% absλ: 1,22 to 1,15CO -12%NOx -29mg/Nm³Unburned C in Ash -0,5% abs.+ 230.000 €/y each boiler

E.OnKW Scholven

Shifted front wall

16 / 44 x 400

Boiler CResidual C up to -1% abs.+ 590.000 €/y

20 power boilers equipped

with Powitec

12 power boilers in closed

loop optimization

46 closed loop

applications in total

approx. 210+ optical

sensors in the field

approx. 250+ acoustical

sensors in the field

80+ international patents


Powitec


Solution

References

Discussion

PiT Navigator at Evonik power plant FennePiT Navigator at Evonik power plant Fenne

MKV Fenne, Völklingen,

(Evonik New Energies)210 MWel

Shifted boxer firing with 8 burners on 2 sides

in 2 levels each. PiT Navigator controls the

secondary air distribution per burner and

the amount of sludge, since May 2005.

A Performance Contracting was agreed

between Saarenergie (Evonik) and Powitec!

- All savings of 7 years to be shared!

- Powitec operates the PiT Navigator

(installation, maintenance etc.)

8 PiT Multisensors

8 x secondary air, controlled

4 x mill, controlled


Powitec


Solution

References

Discussion

Burners-Air System

• Fuels: ballast coal, heavy fuel oil, mine gas, coke oven gas

• 8 low-NOx burners in a staggered, opposed arrangementat 4 levels

• air staging is conducted by secondary air 1 and 2, shellair at the boiler walls and over fire air as burn out air

• to achieve a reduced NOx level, a lambda value of 0.8 isset at the burners and overall air index is increased with

shell air and over fire air to 1.25 (full load)


Powitec


Solution

References

Discussion

High speed serverHigh speed server


Powitec


Solution

References

Discussion

Air/Fuel Ratio

• It was found, that the coal dust distribution in the

2 coal ducts to the burners of one level ist

uneven.

• The navigator corrects this inbalance by

increasing the combustion air to the burner with

the greater fuel flow and reducing the air to the

other one


Powitec


Solution

References

Discussion

Actuations on the

secondary air at

the 8 burners.

Parallel, the

amount of sludge

(feed into coal

mills) is controlled.

Control of the burner secondary air during 24 hControl of the burner secondary air during 24 h


Powitec


Solution

References

Discussion

Control of the burner secondary air during 1hControl of the burner secondary air during 1h

Time scale changed to one

hour


Powitec


Solution

References

Discussion

Fenne power plant:Fenne power plant:

PiT Navigator in normal operationPiT Navigator in normal operation

boiler operation = 8760 h

Bo

ile

r lo

ad

[%

]


Powitec


Solution

References

Discussion

O2O2--set point depending from boiler loadset point depending from boiler load

‚Old‘ set

point

Set point with

PiT Navigator


Powitec


Solution

References

Discussion

O2O2--set point depending from boiler loadset point depending from boiler load

with/without PiT Navigatorwith/without PiT Navigator

Steam


Powitec


Solution

References

Discussion

O2 reduction = wall corrosion ?

• A great danger in changing the air/fuel-ratio is the oxygencontent reduction at the boiler walls.

• The reduction can lead to chlorine included high-temperature corrosion.

• The MKV combustion chamber is equipped with 120 measuring nozzles.

• O2 and CO are measured in regular intervals, all 3 month.

• Dirk Kiehn, Production Manager MKV Fenne:

„With the PiT Navigator, the boiler wall atmosphereimproved significantly.“


Powitec


Solution

References

Discussion

O2O2--measurement results on boiler walls measurement results on boiler walls without PiT Navigator (without PiT Navigator (λλ=1,25=1,25))

Linke Seitenwand Vorderwand Rechte Seitenwand Rückwand

Measurement of reducing atmosphere20.01.2004


Powitec


Solution

References

Discussion

O2O2--measurement results on boiler walls measurement results on boiler walls with PiT Navigator (parallel to a reduction of O2 from with PiT Navigator (parallel to a reduction of O2 from λλ=1,25 to =1,25 to λλ=1,18=1,18))

Measurement of reducing atmosphere20.01.2004


Powitec


Solution

References

Discussion

Results 2010 at MKV Fenne:

• Improved boiler efficiency through air reduction 1,022 t Coal/year = 2,794 t CO2 /year

• Reduced auxiliary consumption at FD and ID fans2,268 MWh/year = 775 t Coal/year = 2,220 t CO2/year

• Reduced Unburned Carbon in Ash: 372 t Coal/year = 1.018 t CO2/year

• Reduced limestone consumption: 99 t/year

• Savings 2010 in total: 2.169 t Coal and 5.936 t CO2 ≈ 300.000 €

• Increased plant net efficiency of 0,18 %


Powitec


Solution

References

Discussion

Further results in 2006 to 2010 at MKV FenneFurther results in 2006 to 2010 at MKV Fenne

• Boiler efficiency + 0,4%

• O2 amount - 24%

(from 4,2 to 3,2 % respectively λ 1,25 to 1,18)

• Unburned Carbon in ash - 0,15% (from 4,0 % to 3,85%)

• No increase of CO and NOx

• “Significantly improved boiler

wall atmosphere”

• “No slagging at the burners”

• “Increased availability”

Savings of < 180.000 € per year + additional benefits


Powitec


Solution

References

Discussion

ReferencesReferences

2010: German Innovation

Award

Climate and Environment

for Outstanding and

Sustainable Technology

Extensive experience in

Power Plants

20 Power Plants

equipped with Powitec

12 Power Plants in

Powitec’s closed loop

operation

Sizes range from

195 MWel to 700 MWel

46 closed loop

optimisers world wide

Powitec Power en BHEL 2011-11-22

Documents

environment climate

german environment minister

loop boiler control

boiler analysisflame

boiler analysisonline

adaptive boiler optimisation

burnerboiler system

technical director powitec