Demonstrating Optimum HCCI Combustion with Advanced Control

Technology Nick Killingsworth Salvador Aceves Dan Flowers Francisco Espinosa-Loza Tim Ross Lawrence Livermore National Laboratory Livermore, California

Miroslav Krstic Department of Mechanical and Aerospace Engineering University of California, San Diego

HCCI combustion occurs when premixed mixture autoignites, no direct trigger to initiate combustion

Premixed fuel and air Mixture compressedinducted into cylinder until it autoignites Mixture expands

In-C

ylin

der

Pre

ssur

e

Crank Angle

There is a window of allowable HCCI engine combustion timings due to constraints

Peak Pressure/ Peak Pressure Rise Rate

NOx Limit

Misfire Limit/ CO & UHC Limit

Load

Combustion Timing

A Proportional-Integral-Derivative (PID) controller has three terms

error e(t)+

-

+ Controlled System

measured output

Up

Ui

Ud

+

+

desired output U

- Proportional to error: uP (t) = tKe( )

- Integral of error : uI (t) = IT

K ∫0

te(s)ds

de(t)- Derivative of error: uD (t) = KTD dt

PID controller takes control action based on past, present, and future control errors

t t+Td

Erro

r

Time

Present

FuturePast

error+

-

+ Controlled System

measured output

Up

Ui

Ud

+

+

desired output

Gain scheduling of PID parameters required due to

nonlinear nature of HCCI engines Gain Scheduled PID Parameters (K,TI ,TD)

IMEP RPM

0-1 [bar]

1-2 [bar]

2-3 [bar]

3-4 [bar]

0-1000 PID1,1 PID1,2 PID1,3 PID1,4

1001-2000 PID2,1 PID2,2 PID2,3 PID2,4

2001-3000 PID3,1 PID3,2 PID3,3 PID3,4

3001-4000 PID4,1 PID4,2 PID4,3 PID4,4

4001-5000 PID5,1 PID5,2 PID5,3 PID5,4

Online non-model based

controller tuning methods are desirable

• Many methods have been developed – Model based methods

• Model must be derived or estimated • Controller parameters calculated offline

– Non-model based methods • Require offline calculations

– Extremum Seeking • Non-model based • Online tuning procedure

We have applied extremum seeking to tune HCCI combustion timing controller

Experimental HCCI engine setup

)( ky

hz z

+ − 1×+

( )θJ

)cos( kωα )cos( kωα

)(kθ

1−− z

γ

Extremum seeking controller tuning scheme

0 100 200 300 400 500 600 700 50

60

70

80

C o l

d V a

lve

P o si

tio n

(% c

lo s e

d

0 100 200 300 400 500 600 700

6

8

10

12

cycle number

CA

5 0 (C

AD

a T D

C )

setpoint Cylinder 1

Experimental controller tuning

Caterpillar 3406 operated at LLNL the platform for control experiments

• Converted 3406 spark ignited engine

• 15L 6-cylinder engine

Intake temperature effective means to control combustion timing

Heated Intake Mixture

Cooled Intake Mixture

Mixing “Tees”(x 6)

Each cylinder’s intake temperature controlled individually

“Cold” Manifold

“Hot” Manifold

Mixing Tees

Intake 1: Cold

Intake 2: Hot

Carb

Generator

Thr

SC

Exhaust Manifold

T

Intake 1: Cold

Intake 2: Hot

Carb

Generator

Preheater

Intercooler

Thr

SC

Exhaust Manifold

T

Schematic of the HCCI engine and thermal management system used in experiments

Air

Fuel

Load

Carb = CarburetorT = TurbineC = CompressorThr = ThrottleSC = Supercharger

C

ExhaustTo preheater

Exhaust to Stack

Intake 1: Cold

Intake 2: Hot

Air

Fuel Carb

Air

Fuel

LoadGenerator Load

Carb = Carburetor T = Turbine C = Compressor Thr = Throttle SC = Supercharger

Preheater

Intercooler

CC Thr

Exhaust Manifold

TExhaustTo preheater

Exhaust to Stack

Exhaust To preheater

Exhaust to Stack

SC

Real-time control system utilizes cylinder pressure measurements as feedback signal

Real-Time Controller PC running Labview RT OS

Cylinder Pressure

Crank Angle Position

Valve Position

User interface

PIDController

Extremum seeking minimizes cost function by tuning the controller parameters

Square Wave CA50 Reference Signal Intake

HCCI Engine

Calc. CostCA50TemperaturePID Function

Extremum Seeking

Algorithm

Controller Parameters

Controller

Cost

Label things animate

Cost function calculated as combustion timing responds to square wave setpoint change

0 100 200 300 400 500 6005

6

7

8

9

10

11

12

CA

50 (d

eg a

tdc)

setpoint measured

Cost Cost New PID Parameters

New PID Parameters

cycle number

Combustion timing varied from 6 to 10 CAD ATDC, cost function only calculated at later timing (10 CAD ATDC)

Extremum seeking: non-model-based optimization method

+

θ (k)

×

( ) y(k)J θ

Function we are trying to optimize

−γ z −1 z + hz −1

High Pass Filter removes DC component of output y(k)

α cos(ωk) αcos(ωk) Demodulate by multiplication with αcos(ωk), this picks out portion ofPerturb system in order output due to perturbation, αcos(ωk)to generate gradient

Estimate of gradient is then integrated to yield a new θ closer to θ*

0 500 1000 1500 2000 2500 30000

2

4

J(θ)

0 500 1000 1500 2000 2500 30000

1

2

3

time (sec)

PID

Par

met

ers K

10Ti

Extremum seeking used to tune a PI controller for improved setpoint response of combustion timing

14

CA

50 (d

eg a

tdc) 12

10

8

6

setpoint measured

0 500 1000 1500 2000 2500 3000

4

1414

1212

8

CA

50 (C

AD

ATD

C)

1010

CA

50 (C

AD

ATD

C)

6 6

4 4

500 505 510 515 520 525 530 535 540 545 550 3000 3005 3010 3015 3020 3025 3030 3035 3040 3045 3050

8

Cylinder 1

PI controller tuned for fast setpoint response

results in high controller gains

50

60

70

80

Col

d Va

lve

Posi

tion

(% c

lose

d

setpoint 0 100 200 300 400 500 600 700

CA

50 (C

AD

aTD

C) 12

10

8

6

cycle number 0 100 200 300 400 500 600 700

A feedforward term was added to the PI controller

error+

-

+ Controlled System

measured output

Uff

Up

Ui

+

+

desired output, r(t)

- Constant feedforward term: u (t) = K ff r(t) ff

CA

50 (C

AD

aTD

C)

Col

d Va

lve

Posi

tion

(% c

lose

d PI + feedforward controller tuned using ES

feedforward results in less high frequency gain

60

70

80

50 0 100 200 300 400 500 600 700

12

10

8

6

cycle number

setpoint Cylinder 1

0 100 200 300 400 500 600 700

ES optimized PI and PI+FF controllers able to hold

constant combustion timing during load disturbance

1500

1600

1700

RPM

PI PI+FF

0

5

10

CA

50 (C

AD

aTD

C) 20 40 60 80 100 120 140 160 180 200

Varia

nce

of C

A50

0 20 40 60 80 100 120 140 160 180 200

2

1

0 0 20 40 60 80 100 120 140 160 180 200

Time (seconds)

ES scheme used to minimize fuel consumption of experimental HCCI engine

by determining optimal combustion timing setpoint

HCCI Engine

e CA50

CA50SP

-Tintake

CA50 + Σ PI

Extremum Seeking

mfuel .

ES delays the combustion timing 6 CAD reducing fuel consumption by more than 10%

Extremum Seeking is a useful tool

for rapidly improving HCCI controller performance

• We use ES to tune various forms of PID controllers while the engine is operating

• ES can also be used to optimize performance parameters (such as minimize fuel consumption)

• ES greatly aids in automating and optimizing the development of controllers for nonlinear systems like HCCI engines