Model Predictive Diesel Engine Control - LIRMM

51
EECI19 M22: MPC for Diesel Engines Model Predictive Diesel Engine Control 1

Transcript of Model Predictive Diesel Engine Control - LIRMM

EECI19 M22: MPC for Diesel Engines

Model Predictive Diesel

Engine Control

1

EECI19 M22: MPC for Diesel Engines

Components:

1. Compressor

2. Variable Geometry Turbine (VGT)

3. EGR Throttle

4. EGR Valve

5. Intake Manifold

6. Exhaust Manifold

7. Pre-Throttle Volume

8. Pre-Intercooler Volume

9. Combustion

10. EGR cooler

11. Intercooler

Diesel engine schematics

2

High speed diesel engine

EECI19 M22: MPC for Diesel Engines

Diesel air path control

Tracking objectives:

โ€ข Control intake manifold pressure and EGR rate (= EGR valve flow

/ Cylinder Flow) to specified set-points

โ€ข Engine speed and fuel flow are measured disturbances

โ€ข Fast response desired to reduce turbo-lag

Constraints:

โ€ข Input constraints: EGR valve, EGR throttle, and VGT lift have limited

ranges

โ€ข Output constraints are imposed on intake manifold pressure

overshoot, exhaust manifold pressure overshoot, turbocharger

speed, EGR rate, etc.

3

EECI19 M22: MPC for Diesel Engines

Model Predictive Control

4

EECI19 M22: MPC for Diesel Engines

Low complexity explicit LQ-MPC

5

โ€ข Nonlinear inversion to make plant to which MPC is applied โ€œmore linearโ€

โ€ข Rate-based model at a single linearization point

โ€ข Intermittent tightened constraint enforcement

โ€ข Constraint re-modeling (e.g., constraint on turbocharger speed is remodeled as a constraint on intake manifold pressure)

โ€ข Special MPC formulation:

โ€ข Control horizon = 1

โ€ข Prediction horizon = 1

โ€ข LQ-based terminal cost

โ€ข Longer constraint horizon, with intermittent enforcement

โ€ข Low complexity gain scheduling

EECI19 M22: MPC for Diesel Engines

Low complexity explicit LQ-MPC

6

EECI19 M22: MPC for Diesel Engines

Diesel air path controller structure

7

EECI19 M22: MPC for Diesel Engines

Combined EGR valve โ€“ EGR throttle actuator

โ€ข Two actuators are โ€œgangedโ€ together to deliver EGR flow

8

EECI19 M22: MPC for Diesel Engines

โ€ข Define state and control increments and an augmented state:

โ€ข The rate-based model has the following form,

ฮ”๐‘ข๐‘˜ = ๐‘ข๐‘˜ โˆ’ ๐‘ข๐‘˜โˆ’1ฮ”๐‘ฅ๐‘˜ = ๐‘ฅ๐‘˜ โˆ’ ๐‘ฅ๐‘˜โˆ’1๐‘’๐‘˜ = ๐‘ฆ๐‘˜โˆ’1 โˆ’ ๐‘Ÿ๐‘˜โˆ’1๐œ‰๐‘˜ = ฮ”๐‘ฅ๐‘˜ ๐‘’๐‘˜

๐‘‡ .

๐œ‰๐‘˜+1 = าง๐ด๐œ‰๐‘˜ + เดค๐ตฮ”๐‘ข๐‘˜๐‘’๐‘˜ = าง๐ถ๐œ‰๐‘˜

าง๐ด =๐ด 0๐ถ ๐ผ

, เดค๐ต =๐ต๐ท

, าง๐ถ = 0 ๐ผ .

Rate-based model

(assumes ๐‘Ÿ๐‘˜ = ๐‘๐‘œ๐‘›๐‘ ๐‘ก)

EECI19 M22: MPC for Diesel Engines

The MPC optimization problem then becomes,

min ๐œ‰๐‘๐‘‡๐‘ƒ๐œ‰๐‘ +

๐‘˜=1

๐‘โˆ’1

๐‘’๐‘˜๐‘‡๐‘„๐‘’๐‘˜ + ฮ”๐‘ข๐‘˜

๐‘‡๐‘…ฮ”๐‘ข๐‘˜

subject to:๐œ‰๐‘˜+1 = าง๐ด๐œ‰๐‘˜ + เดค๐ตฮ”๐‘ข๐‘˜

๐‘’๐‘˜ = าง๐ถ๐œ‰๐‘˜๐‘ข๐‘˜ = ๐‘ข๐‘˜โˆ’1 + ฮ”๐‘ข๐‘˜

๐‘’๐‘˜ + ๐‘Ÿ โˆˆ ๐‘Œ๐‘˜ ,๐‘ข๐‘˜ โˆˆ ๐‘ˆ๐‘˜ .

Rate-based MPC (contโ€™d)

EECI19 M22: MPC for Diesel Engines

Linear model through system identification

โ€ข A local 7th order linear model was obtained through system identification

โ€ข Nominal operating condition chosen as 1600 rpm engine speed and 30 mm3/stroke fuel rate. Engine operates between 800 and 4000 rpm and 0 and 60 mm3/stroke

โ€ข Model order reduction applied to obtain a 4th order linear model

๐‘ฅ๐‘˜+1 = ๐ด๐‘ฅ๐‘˜ + ๐ต๐‘ข๐‘˜,

๐‘ฆ๐‘˜ = ๐ถ๐‘ฅ๐‘˜ + ๐ท๐‘ข๐‘˜ .

11

EECI19 M22: MPC for Diesel Engines

Computational complexity

Table: Comparison of MPC strategies. ๐‘๐‘ง denotes the number of zonespartitioning the engine operating range, ๐‘๐‘ฆ denotes the number of outputconstraints considered, ๐‘๐‘ denotes the number of constraints in the MPCoptimization problem (inputs and outputs with intermittent constraintenforcement), ฯƒ๐‘๐‘Ÿ is the total number of regions in the explicit solution, ROMand Time are estimated memory and computation time usage for explicitMPC on a ~150 MHz processor. The sampling/update period is 32 ms.

12

EECI19 M22: MPC for Diesel Engines

Low complexity gain scheduling1

EECI19 M22: MPC for Diesel Engines

Simulation results โ€“ nonlinear model

14

EECI19 M22: MPC for Diesel Engines

Engine experiments

1200 rpm 1600 rpm

15

EECI19 M22: MPC for Diesel Engines

2400 rpm

Engine experiments

16

EECI19 M22: MPC for Diesel Engines

Experimental results over NEDC and WLTC

drive cycles

17

EECI19 M22: MPC for Diesel Engines

Robust MPC for air path control

EECI19 M22: MPC for Diesel Engines

Robust MPC for air path control: Experimental results

EECI19 M22: MPC for Diesel Engines

From LQ MPC to Nonlinear MPC

EECI19 M22: MPC for Diesel Engines

From LQ MPC to Nonlinear MPC

EECI19 M22: MPC for Diesel Engines

Nonlinear MPC

EECI19 M22: MPC for Diesel Engines

Basic NMPC computational strategy

EECI19 M22: MPC for Diesel Engines

Basic NMPC computational strategy

EECI19 M22: MPC for Diesel Engines

Basic NMPC computational strategy

EECI19 M22: MPC for Diesel Engines

Basic NMPC computational strategy

EECI19 M22: MPC for Diesel Engines

Basic NMPC computational strategy

EECI19 M22: MPC for Diesel Engines

Towards more advanced NMPC

EECI19 M22: MPC for Diesel Engines

Modeling for Nonlinear MPC

EECI19 M22: MPC for Diesel Engines

NMPC based on move blocked rate-based model

EECI19 M22: MPC for Diesel Engines

MPC for feedforward and feedback

EECI19 M22: MPC for Diesel Engines

Experimental implementation of NMPC

EECI19 M22: MPC for Diesel Engines

Experimental implementation of NMPC

Supervisory MPC

EECI19 M22: MPC for Diesel Engines

minz

๐ฝ ๐‘ง, ๐œŒ =

๐‘˜=0

๐‘

๐‘™(๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐‘ข๐‘˜โˆ’1, ๐‘ , ๐œŒ)

๐‘ . ๐‘ก ๐‘ฅ๐‘˜+1 = ๐‘“ ๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐œŒ, ฮ”๐œ๐‘˜ , ๐‘˜ = 0, โ€ฆ ๐‘ โˆ’ 1,๐œ™ ๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐œŒ โˆ’ ๐œ™๐‘™(๐œŒ, ๐‘ค๐‘) โ‰ค ๐‘ , ๐‘˜ = 0,โ€ฆ๐‘ โˆ’ 1 ,

0 โ‰ค ๐œ’๐‘˜๐‘ก๐‘Ÿ๐‘”

โ‰ค าง๐œ’(๐œŒ), ๐‘˜ = 0โ€ฆ๐‘ โˆ’ 1,

0 โ‰ค ๐‘ž๐‘˜ โ‰ค ๐‘ž๐‘ก๐‘Ÿ๐‘” ๐œŒ , ๐‘˜ = 0โ€ฆ๐‘ โˆ’ 1,๐‘  โ‰ฅ 0

where

๐‘™ = ๐›พ ๐œ’๐‘˜๐‘ก๐‘Ÿ๐‘”

โˆ’ าง๐œ’ ๐œŒ2+ ๐‘ข๐‘˜ โˆ’ ๐‘ข๐‘˜โˆ’1 ๐‘…

2 + ๐›ผ ๐‘ž๐‘ก๐‘Ÿ๐‘” โˆ’ ๐‘ž๐‘˜ + ๐›ฝ๐‘ 

Fuel trackingDampingEGR tracking

๐‘ง = ๐‘ข0๐‘‡ ๐‘ฅ1

๐‘‡ โ€ฆ ๐‘ข๐‘โˆ’1๐‘‡ ๐‘ฅ๐‘

๐‘‡ ๐‘‡

FAR slack penalty

Dynamics

Softened FAR constraint

EGR rate constraint

Fuel bounds

๐‘๐‘–๐‘š,๐‘˜๐‘ก๐‘Ÿ๐‘”

= าง๐‘๐‘–๐‘š(๐œŒ๐‘˜)

๐‘ข =๐œ’๐‘ก๐‘Ÿ๐‘”

๐‘ž

๐‘ฅ = ๐‘๐‘–๐‘š ๐‘๐‘’๐‘ฅ ๐‘ค๐‘ ๐น1 ๐น2๐‘‡

MPC formulation

Experimental results (WLTC)

EECI19 M22: MPC for Diesel Engines

Experimental results

FBRS method for convex quadratic programming

EECI19 M22: MPC for Diesel Engines

Results of computational experiments

EECI19 M22: MPC for Diesel Engines

Symbolic code optimization

Towards economic MPC (eMPC)

eMPC as an outer loop supervisor

Performance outputs:

๐‘ฆ๐‘ ๐‘ก =๐œ™๐‘๐‘‚๐‘ฅ๐œ‚

Measured outputs:

๐‘ฆ๐‘š ๐‘ก =๐‘๐‘–๐‘š๐‘ค๐‘

Targets for inner loop:

๐‘Ÿ ๐‘ก =๐‘๐‘–๐‘š๐‘ก๐‘Ÿ๐‘”

๐œ’๐‘ก๐‘Ÿ๐‘”

Actuators:

๐‘ค ๐‘ก =

๐‘ข๐‘กโ„Ž๐‘Ÿ๐‘ข๐‘ฃ๐‘Ž๐‘™๐‘ข๐‘ฃ๐‘”๐‘ก

๐œ’ = ๐ธ๐บ๐‘… ๐‘…๐‘Ž๐‘ก๐‘’ =๐ธ๐บ๐‘… ๐‘“๐‘™๐‘œ๐‘ค

๐ถ๐‘ฆ๐‘™๐‘–๐‘›๐‘‘๐‘’๐‘Ÿ ๐‘“๐‘™๐‘œ๐‘ค=๐‘ค๐‘’๐‘”๐‘Ÿ

๐‘ค๐‘๐‘ฆ๐‘™

๐œ‚ =๐‘ก๐‘œ๐‘Ÿ๐‘ž๐‘ข๐‘’ โˆ— ๐‘’๐‘›๐‘”๐‘–๐‘›๐‘’ ๐‘ ๐‘๐‘’๐‘’๐‘‘

๐‘“๐‘ข๐‘’๐‘™ ๐‘“๐‘™๐‘œ๐‘ค โˆ— ๐‘™๐‘œ๐‘ค๐‘’๐‘Ÿ โ„Ž๐‘’๐‘Ž๐‘ก๐‘–๐‘›๐‘” ๐‘ฃ๐‘Ž๐‘™๐‘ข๐‘’=

๐œ๐‘๐‘’๐‘ค๐‘“๐‘„๐ฟ๐ป๐‘‰

๐œ™ =๐‘“๐‘ข๐‘’๐‘™ ๐‘“๐‘™๐‘œ๐‘ค

๐‘Ž๐‘–๐‘Ÿ ๐‘“๐‘™๐‘œ๐‘คโˆ—

๐ด

๐น๐‘ ๐‘ก๐‘œ๐‘–๐‘โ„Ž

=๐‘ค๐‘“

๐‘ค๐‘Ž

๐ด

๐น๐‘ ๐‘ก๐‘œ๐‘–๐‘โ„Ž

๐œ’๐‘ = ๐‘๐‘‚๐‘ฅ ๐‘๐‘œ๐‘›๐‘๐‘’๐‘›๐‘ก๐‘Ÿ๐‘Ž๐‘ก๐‘–๐‘œ๐‘›๐‘๐‘–๐‘š = ๐‘–๐‘›๐‘ก๐‘Ž๐‘˜๐‘’ ๐‘š๐‘Ž๐‘›๐‘–๐‘“๐‘œ๐‘™๐‘‘ ๐‘๐‘Ÿ๐‘’๐‘ ๐‘ ๐‘ข๐‘Ÿ๐‘’

Exogenous inputs:

๐œŒ ๐‘ก =๐‘ž๐‘ก๐‘Ÿ๐‘”

๐‘๐‘’Fuel per stroke:

๐‘ž(๐‘ก)eMPC control:

๐‘ข ๐‘ก =๐‘Ÿ๐‘ž

EECI19 M22: MPC for Diesel Engines

Efficiency and NOx modeling

EECI19 M22: MPC for Diesel Engines

eMPC Formulation

minz

๐ฝ ๐‘ง, ๐œŒ =

๐‘˜=0

๐‘

๐‘™(๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐‘ข๐‘˜โˆ’1, ๐œŒ)

๐‘ . ๐‘ก ๐‘ฅ๐‘˜+1 = ๐‘“ ๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐œŒ, ฮ”๐œ๐‘˜ , ๐‘˜ = 0, โ€ฆ ๐‘ โˆ’ 1,

๐œ™ ๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐œŒ โ‰ค ๐œ™lim ๐œŒ , ๐‘˜ = 0,โ€ฆ๐‘ โˆ’ 1 ,

ln(๐œ’๐‘(๐‘ฅ๐‘˜ , ๐‘ข๐‘˜ , ๐œŒ)) โ‰ค ln(๐œ’๐‘lim(๐œŒ)), ๐‘˜ = 0,โ€ฆ๐‘ โˆ’ 1,

๐‘Ÿโˆ’ ๐œŒ โ‰ค ๐‘Ÿ๐‘˜ โ‰ค ๐‘Ÿ+ ๐œŒ , ๐‘˜ = 0โ€ฆ๐‘ โˆ’ 1,๐‘žโˆ’ ๐œŒ โ‰ค ๐‘ž๐‘˜ โ‰ค ๐‘ž๐‘ก๐‘Ÿ๐‘” ๐œŒ , ๐‘˜ = 0โ€ฆ๐‘ โˆ’ 1,

where

๐‘™ = โˆ’๐›ฝ๐œ‚ ๐‘ฅ๐‘˜ , ๐œŒ +ฮ”๐‘ข๐‘˜ฮ”๐œ๐‘˜ ๐‘…

2

+ ๐›ผ ๐‘ž๐‘ก๐‘Ÿ๐‘” โˆ’ ๐‘ž๐‘˜

Dynamics

Emissions constraints

Target constraints

Fuel bounds

Fuel trackingDamping

Efficiency

๐‘ง = ๐‘ข0๐‘‡ ๐‘ฅ1

๐‘‡ โ€ฆ ๐‘ข๐‘โˆ’1๐‘‡ ๐‘ฅ๐‘

๐‘‡ ๐‘‡๐‘ข =

๐‘๐‘–๐‘š๐‘ก๐‘Ÿ๐‘”

๐œ’๐‘ก๐‘Ÿ๐‘”

๐‘ž

๐‘Ÿ =๐‘๐‘–๐‘š๐‘ก๐‘Ÿ๐‘”

๐œ’๐‘ก๐‘Ÿ๐‘”

EECI19 M22: MPC for Diesel Engines

๐‘˜ = 0 ๐‘˜ = 1 ๐‘˜ = 2 ๐‘˜ = 3 ๐‘˜ = 4 ๐‘˜ = 5

Non uniform horizon discretization

EECI19 M22: MPC for Diesel Engines

Numerical solution

โžŠ Linearize dynamics and non-convex

constraints about the current measured state

and form the single-shooting OCP

โž‹ Form quadratic programming sub-problem

about the solution at the previous time step

โžŒ Solve QP and apply the resulting control to

the system

minฮ”๐‘ง

1

2ฮ”๐‘ง๐‘‡๐ป ฮ”๐‘ง + ๐‘“๐‘‡ฮ”๐‘ง

๐‘ . ๐‘ก. ๐ดฮ”๐‘ง โ‰ค ๐‘

๐ป = ๐›ป๐‘ง2๐น ๐‘ง๐‘˜โˆ’1, ๐œŒ๐‘˜ , ๐‘“ = ๐›ป๐‘ง๐น ๐‘ง๐‘˜โˆ’1, ๐œŒ๐‘˜ ,

๐ด = ๐›ป๐‘ง๐‘ ๐‘ง๐‘˜โˆ’1, ๐œŒ๐‘˜ , ๐‘ = โˆ’๐‘(๐‘ง๐‘˜โˆ’1, ๐œŒ๐‘˜)

min๐‘ง

๐น(๐‘ง, ๐œŒ๐‘˜)

๐‘ . ๐‘ก. ๐‘ ๐‘ง, ๐œŒ๐‘˜ โ‰ค 0

๐‘ง๐‘˜ = ฮ”๐‘ง + ๐‘ง๐‘˜โˆ’1๐‘ข๐‘˜ = ๐‘ง 1: 3

1

2

3

EECI19 M22: MPC for Diesel Engines

โžŠ

โž‹

โžŒ 3

2

1

Experimental results

EECI19 M22: MPC for Diesel Engines

Simulation results

โ€ข eMPC is stable over NEDC and WLTC cycles

โ€ข Emissions are reduced compared to the benchmark during hard transients โžŠ

WLTC

1

EECI19 M22: MPC for Diesel Engines

Simulation results (contโ€™d)โ€ข NOx constraint violations are penalized, leading to soft constraint

enforcement and reducing NOx over the cycles

โ€ข Fuel-air-ratio constraint is not always satisfied due to conflict with the

drivability constraint โžŠ

1

EECI19 M22: MPC for Diesel Engines

ReferencesHuang, M., Zaseck, K., Butts, K., and Kolmanovsky, I.V., โ€œRate-based Model Predictive Controller for

diesel engine air path: Design and experimental evaluation,โ€ IEEE Transactions on Control Systems

Technology, vol. 24, no 4, pp. 1922-1955, 2016.

Liao-McPherson, D., Huang, M., and Kolmanovsky, I.V., โ€œA regularized and smoothed Fischer-Burmeister

method for quadratic programming with applications to Model Predictive Control,โ€ IEEE Transactions on

Automatic Control, 2019.

Huang, M., Liao-McPherson, D., Kim, S., Butts, K., and Kolmanovsky, I.V., โ€œTowards real-time automotive

model predictive control: A perspective from a diesel air path control development,โ€ Proceedings of 2018

Annual American Control Conference (ACC), June 27โ€“29, 2018. Wisconsin Center, Milwaukee, USA, pp.

2425-2430.

Walker, K., Samadi, B., Huang, M., Gerhard, J., Butts, K., and Kolmanovsky, I.V., "Design environment for

Nonlinear Model Predictive Control," SAE Technical Paper 2016-01-0627, April, 2016, SAE World

Congress, doi:10.4271/2016-01-0627

Huang., M., Kolmanovsky, I.V., and Butts, K., ``A low complexity gain scheduling strategy for explicit

model predictive control of a diesel air path,โ€™โ€™ Proceedings of 2015 Dynamic Systems and Control

Conference, Columbus, Ohio, 2015, Paper DSCC2015-9654.

Huang, M., Nakada, H., Butts, K., and Kolmanovsky, I.V., โ€œNonlinear model predictive control of a diesel

engine air path: A comparison of constraint handling and computational strategies,โ€ Proceedings of the 5th

IFAC Symposium on Nonlinear Model Predictive Control, Seville, Spain, September 2015, IFAC-

PapersOnLine, vol. 48, no. 23, 2015, pp. 372-379.

EECI19 M22: MPC for Diesel Engines

References (contโ€™d)

Huang, M., Nakada, H., Butts, K., Kolmanovsky, I.V., โ€œRobust rate-based Model Predictive Control of

diesel engine air path,โ€ Proceedings of 2014 American Control Conference, Portland, Oregon, pp. 1505-

1510, 2014.

Huang, M., Nakada, H., Polavarapu, S., Butts, K.R., and Kolmanovsky, I.V., โ€œRate-based Model Predictive

Control of diesel engines,โ€ Proceedings of the 7th IFAC Symposium on Advances in Automotive Control,

September 4-7, 2013, Tokyo, Japan, pp. 177-182.

Huang, M., Nakada, H., Polavarapu, S., Choroszucha, R., Butts, K.R., and Kolmanovsky, I.V., โ€œTowards

combining nonlinear and predictive control of diesel engines,โ€ Proceedings of 2013 American Control

Conference, Washington DC, pp. 2852-2859.