Korean Danish PEM Fuel Cell Workshop 2013

MODELLING AND CONTROLLING FC SYSTEMS USING AN ADAPTIVE NEURO-FUZZY INFERENCE

SYSTEM STRATEGY

S Ø R E N J . A N D R E A S E NA S S O C I A T E P R O F E S S O R , F U E L C E L L A N D B A T T E R Y R E S E A R C H G R O U P

SØREN J . ANDREASEN, KRISTIAN K. JUSTESEN,

Outline

S J A @ E T . A A U . D K

2

• Introduction• Methanol reformers and HTPEM fuel cells

• Air heat exchange• Liquid heat exchange

• Critical system issues• ANFIS models• Implementation of ANFIS models

• Feedforward control• Stiochiometry monitoring and control• System speed improvements

• Conclusion and future work

Advantages• PBI-based MEAs have a high tolerance to CO• A liquid fuel, such as methanol is easily accessable and storable• Heat can be utililzed in fuel conversion• System energy density increase is ”cheap”

Challenges• System size and complexity increases• Impurities are introduced• System heat-up

Reformed methanol HTPEM fuel cell systems

S J A @ E T . A A U . D K

3

• Waste heat from cathode air cooled HTPEM Fuel Cell Stacksis used for evaporation of reformer fuel.

• Proper heat integration and design is required to avoid highBoP consumption.

• HTPEM system does not needadditional CO clean up.

Reformer system - air heat exchange

S J A @ E T . A A U . D K

4

Reformer system - liquid heat exchange

S J A @ E T . A A U . D K

5

• Liquid heat transfer can minimize system size and BoP power consumption.

• System logistics are more conventional.

• From a control PoV, more variables to manipulate, larger DoF

• Several system topologies are usabledepending on application utility heat and demand.

• Individual system components are characterized ex-situ, such that fuel cell stack and reformer system behaviour can be separated.

• Fuzzy logic / Neural network models of internalsystem states, can be developed.

• Model based control approaches are implemented in system software and system improvements arequantified.

Control development methodology

S J A @ E T . A A U . D K

6

Serenergy H3-350 350W off-grid battery charger

(HTPEM + SR)

Serenergy H3-50005kW power system

(HTPEM + SR)

Critical HTPEM system issues

S J A @ E T . A A U . D K

7

• Hydrogen impurities, CO, unconverted fuel• Fuel cell stack anode starvation• System complexity, non-linearity• System behaviour during load transients,

and ambient changes• System efficiency

What is ANFIS modelling?

S J A @ E T . A A U . D K

8

Adaptive Neuro-Fuzzy Inference Systems (ANFIS)

A combination of Fuzzy Inference Systems (FIS)and Neural Networks (NN). The resulting model canbe used to describe complex non-linear physicalphenomena, such as a fuel cell / reformer system byusing a human-like reasoning scheme which learnsto imitate a physical system based on experimentaldata.

Basically the chosen model is learns how to behavefrom the experimental data it is subjected to.

• Layer 1: Fuzzyfication layer, converts inputs to fuzzy variables (0-1) and Bell-shaped membership functions.

• Layer 2: Non-adaptive nodes, determining firing levels. FuzzyT-norms and fuzzy AND (product method)

• Layer 3: Non-adaptive normalization layer, each firing level is normalized

• Layer 4: Adaptive layer, how changes in the input affects the output, using adaptive consequent parameters.

• Layer 5: Summation layer

ANFIS modelling

S J A @ E T . A A U . D K

9

Gas composition modeling in a reformed Methanol Fuel Cell system using adaptive Neuro-Fuzzy Inference Systems, Int. Journal of Hydrogen Energy, 2013, K.K.Justesen,S.J.Andreasen,H.R.Shaker,M.P.Ehmsen, J. Andersen, In Press

Experimental analysis

S J A @ E T . A A U . D K

1 0

• Example: Extensive ex-situ reformer output gas measurements using gas analyzers create the foundation for ”learning” system behaviour by an Adaptive Neuro-Fuzzy Inference System (ANFIS) model.

• Proper prediction of gas composition, anode stoichiometry, etc. can enable higher efficiency, reliabilityand lifetime.

Modelling precision

S J A @ E T . A A U . D K

1 1

• The model predicts eachspecies flow with a goodprecision. Carbon Dioxideshown here

Modelling precision

S J A @ E T . A A U . D K

1 2

• Choosing the number of membership functions. Consider no-flow output, precision etc. Noticenumerical differences in the extremes.

• Pump flow determines usablehydrogen flow in the FC anode, but fuel evaporation and conversionneed to be considered.

• A model based approach can beused for proper feedforward contoland determination of system setpoints.

• For example ANFIS modelling canprovide high prescision stateprediction based on experimentalresults avoiding undesired system operating conditions.

System control challenges

S J A @ E T . A A U . D K

1 3

Implementation results

S J A @ E T . A A U . D K

1 4

Differences in anode stoichiometry estimationcomparred to ideal fuel conversion.

Original high current ramplimit to avoid transient”problems”.

Implementation results

S J A @ E T . A A U . D K

1 5

Current ramp rate limit choosen based on dynamicanalysis.

Improvements in stacktemperature controller robustness.

System simulation

S J A @ E T . A A U . D K

1 6

Simple system models can be usedto optimize system control in ”extreme” cases.

Normally negative current steps would result in burner overheating, but with detailed control such criticaltemperature excursions can beavoided.

Improvements in load following capabilities

S J A @ E T . A A U . D K

1 7

Fuel cells are great part load performers, butreformer systems often remove thisadvantageous feature.

Proper control can restore som of thesecapabilities if needed.

• Efficient and reliable HTPEM fuel cell systems are at a development stage, wherecontrol becomes increasingly relevant.

• Fuel cell systems are excellent part load performers, but lag, complex systems dynamics and expensive state monitoring can limit load following capabilities and system performance.

• Knowledge from detailed experimental analysis and experience can effectively beused in advanced control strategies (such as ANFIS based approaches) that areimplementable in real systems.

• Advanced system control can be used to improve performance, i.e. system efficiency, transient performance, load following capabilities and eventually lifetime.

Conclusions

S J A @ E T . A A U . D K

1 8

• How can knowledge of system lifetime be used to improvesystem lifetime?

• To what extent can we train adaptive models by usingdegradation information for both reformer and fuel cellstack, and can we change operating setpoints accordinglyto potentially increase lifetime?

Future work

S J A @ E T . A A U . D K

1 9

Degradation

Degradation?

T h e a u t h o u r s w o u l d l i k e t o a c k n o w l e d g e t h e f i n a n c i a l s u p p o r t f r o m t h e D a n i s h E n e r g y Te c h n o l o g y D e v e l o p m e n t a n d D e m o n s t r a t i o n P r o g r a m m e

Acknowledgements

20

Korean Danish PEM Fuel Cell Workshop 2013

MODELLING AND CONTROLLING FC SYSTEMS USING AN ADAPTIVE NEURO-FUZZY INFERENCE

SYSTEM STRATEGY

S Ø R E N J . A N D R E A S E NA S S O C I A T E P R O F E S S O R , F U E L C E L L A N D B A T T E R Y R E S E A R C H G R O U P

SØREN J . ANDREASEN, KRISTIAN K. JUSTESEN,