Engine modelling and optimisation for RDE Prof. Chris Brace
Overview The need to consider system influences on engine performance for RDE How can we achieve this? Component selection Simulation requirements Experimental requirements Issues around workflow integration
The need The engine is the source of greatest non-linearity (apart from the driver) and so receives the greatest attention in simulation, testing & calibration But, powertrains continue to grow in complexity Legislation is becoming more real world Drivers expectations continue to rise Effect of boundary conditions imposed by the powertrain, and system interactions are critical to engine operation The need for effective system level calibration becomes greater
Factor 2 Legislative drive is just one factor WLTP cycle and RDE give added topicality to the issue BUT, this is the same need that has existed for decades Most production calibrations (and architectures) are found wanting in at least some respects when exposed to real life operation Robustness is already an issue even over NEDC 4 Factor 1 Noise Factor
Real world NOx and CO 2 6 Source REAL-WORLD EXHAUST EMISSIONS FROM MODERN DIESEL CARS A META-ANALYSIS OF PEMS EMISSIONS DATA FROM EU (EURO 6) AND US (TIER 2 BIN 5/ULEV II) DIESEL PASSENGER CARS. PART 1: AGGREGATED RESULTS Vicente Franco, Francisco Posada Sanchez, John German, and Peter Mock The International Council on Clean transportation 214
Real world operation is unpredictable Real world operating envelope several orders of magnitude larger than even WLTP operation How can we calibrate for any possibility?
Whole map useage, dominated by lower speeds
Effect of driver behaviour is pronounced
WLTP has broader Coverage than NEDC, < RDE Possible Solutions: More steady state calibration points Lots of effort, doesn t address transients Drive cycle based optimisation Duty Cycle Specific How can we access the benefits of Design of Experiments with full operating region coverage with dynamic events?
CO 2 remains the long term focus RDE emissions compliance will be achieved through hardware design and robust calibration Currently the RDE procedures will allow measurement of real world CO 2 but is not subject to mandatory limits This will surely change in time Real world CO 2 will only become more important from here on 11
A global approach is difficult, but necessary WLTP, RDE are a target setting and validation exercise, NOT a development process For development we need better insight into system performance and a global optimisation approach Full map compliance at steady state Consideration of driver, powertrain and boundary condition interactions Competent dynamic control at all times 12
Optimisation Hierarchy Hardware selection Steady state optimisation Dynamic optimisation Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 2 4 6 8 1 12 Time (s) predicted 4 2 5-5 5 1 5 1 measured measured Validation 5-5 2 4 6 8 1 12 samples error error
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Short term need for SI Will need to run at stoichiometry over full operating envelope to allow 3 way catalyst to work exhaust temperature over 1C Probably need larger catalyst to cope with high mass flows Likely need for Gasoline Particulate Filter (GPF)
Short term need for Diesel DPF, DOC, with selective catalytic reduction (SCR) to reduce NOx Balance in cylinder NOx reduction (EGR, SOI) with SCR to give compliance and acceptable urea consumption
Longer term impact of RDE We have been designing and calibrating around UDC then NEDC since 197 This has profoundly influenced the thinking of several generations of engineers Future powertrains need to comply in all situations Segmented solutions targeted at low load are becoming less favourable Downsizing with driveability enablers (such as eboost, hybridisation) will accelerate Peak power augmentation and waste energy capture becoming more relevant Complex sizing and optimisation task
Component selection and sizing Effective selection of powertrain architecture is critical but largely left to custom and practice guided by expert knowledge Formal optimisation is needed at an early stage Places great emphasis on modelling environment More work needed on architecture optimisation with sizing and through life costing as an integrated activity Cost Performance Emissions CO 2
MOVEM vision
MOVEM vision for architecture optimisation
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Optimising for full map RDE compliance Initial task is to balance engine out emissons with aftertreatment duty cycle Urea refill schedule needs balancing with DPF loading and CO 2 for Diesel Catalyst sizing and placement tradeoff for lightoff v full load on SI Thermal and environment impact on engine out emissions and aftertreatment performance needs greater insight When tough hardware tradeoffs are needed consider likely contribution to overall running based on observed probability Weighted contribution of high speed, high load running is low Medium speed, medium load is dominant Low ambient temperatures much more significant than in NEDC
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Steady state test enhancements Incorporation of prior knowledge to speed up limit search Iterative on-line DoE to minimise data requirement Sweep mapping to yield data more rapidly Bayesean techniques to incorporate prior knowledge into response models
Iterative online DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Start with a simple DoE design Select next points to test based on points of least confidence
Iterative DoE Process Depends upon close integration between cell and DoE tool Recalculates models of mean and variance on the fly Also opportunity to use a Bayesian approach
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Requirements for the chassis dynamometer All of the precision, control of an engine dyno but with the boundary conditions of a chassis dyno Full instrumentation suite Control over engine actuators, speed, load Full integration with the optimisation suite (and accessible by calibration engineers) Implies a relatively mature powertrain and mule vehicle is available 43
Chassis dynamometer at Bath 44
Chassis dynamometer rebuild at Bath Four wheel drive dynamometer Robot driver with real driving characteristics and direct mode for mapping Wheel torque measurement and control Comprehensive raw emissions measurements In cylinder measurements Battery emulation and EV instrumentation Full PCM access -1 to +5C temperature control Full frontal area road speed fan Humidity control, Combustion air conditioning Altitude simulation?
Operating point control options 1. Vehicle model tractive effort, rolls speed feedback 2. Vehicle and tyre model - wheel torque & speed feedback 3. Vehicle, tyre, TX model - engine torque & speed feedback 4. Engine speed and Relativ Luft control - engine RL and speed feedback via PCM
Critical need for precision of CD test Even the NEDC is a complex test procedure Many factors can adversely affect precision, masking observed changes DoE approach used to assess impact of these setup variables, on chassis dyno 47
Main noise factors in chassis dyno NEDC Absolute Fuel consumption change (g) 15% 1% 5% % -5% Battery discharge (V) PAS pump Speed error Tyre type Engine oil level Pedal Busyness Tyre pressure Vehicle alignment Road speed fan Vehicle mass Effect of HTHS change of.6cp Tie down straps Engine start temperature
Use of high bandwidth data on CD Individual cycle comparison on a crank angle basis Allows detailed analysis of system interactions Pressure (bar) Injection Demand 1 8 6 4 2.5 1 1.5 2 2.5 3.3.2.1 Vehicle Speed (km/h) 1 5 195 39 585 78 12 Time (s).5 1 1.5 2 2.5 3 Crank angle (deg) UDC1 UDC2 UDC3 UDC4 x 1 5 x 1 5 Pressure (bar) Inj Dmd 7 6 5 4 3 2 1.2.1 Crank Angle UDC2 UDC3 UDC4 UDC1
e.g. Hot v Cold FMEP Integral of FMEP removes noise from plot and clearly shows difference between hot and cold start tests Engine Speed (RPM) 3 2 1 2 4 6 8 1 12 14 16 18 Cumulative FMEP (bar.cycles) 3.5 3 2.5 2 1.5 1.5 Cold Start Hot Start Difference 2 4 6 8 1 12 14 16 18 Cycle
Vehicle Speed (km/h) Ability to visualise transient events 1 5 195 39 585 78 12 Time (s) Pressure (bar) 6 5 4 3 2 4 3 2 1 Cycle Number 1 Pressure (bar) 5 3 2 Cycle Number 51 1-2 TDC 2 Crank Angle 4 6-3 -2-1 TDC 1 Crank Angle 2 3
Dynamic design of experiments Based on tools developed by IAV Addresses need to cover the entire design space including rate of change of inputs Very demanding control requirements for host system direct to engine actuators Allows hardware to be characterised independently of control strategy and calibration 52
Dynamic schedule design 1. Define static and dynamic operating range 2. Construct Signals NEDC Amplitude Linear Chirp Logarithmic Chirp.2.4.6.8.1.12 Frequency (Hz)
Optimise space coverage of multi-signal test
Example Emissions Modelling Problem Definition Dynamic inputs Temperature input Emissions output Hot engine test plan Hot engine data acquisition Data Pre-processing Cold start test design Hot/Dynamic modelling Dynamic-Hot engine model y f x Cold start data acquisition Temperature scaling factor Data Pre-processing NOx (Cold/Hot) 1.5 1.5 Torque Based Input 2 4 6 8 1 Oil Temperature ( o C) Combine for general dynamic/thermal model Temperature scaling function modelling NOx (Cold/Hot) 1.5 1.5 Pedal Based input s 2 4 6 8 1 Oil Temperature ( o C) f T NOx Scaling 1.5 NOx Scaling Speed (rpm) NOx (ppm) error predicted Hot/dynamic model validation 3 2 1 2 15 1 5 Target Predicted 2 4 6 8 1 12 Time (s) 5 5 5 1 15 measured General temperature dependant model 1.5 NOx Scaling 1.5 NOx Scaling y f x, T 1 error 5 R 2-5 5 1 15 measured -5 2 4 6 8 1 12 samples 1 Model Validation.5.5 3 6 12 6 12 6 12 6 12 Time (s) 6 2 1 12Time (s) 6 12 Time (s) Time (s) Time (s) 1 MeasuredTime (s) Predicted Speed (rpm) NOx (ppm) 5 NOx Scaling.5 NOx Scaling 1 2 4 6 8 1 12 Time (s) predicted 4 2 error 5
Challenges for dynamic DoE Accurate control/measurement of dynamic conditions Sensor and actuator accuracy and response time in complex measurement chains Non-linear dynamic modelling methods Complex mathematical functions with many coefficients Tend towards physics based models Automated model fitting Dynamic optimisation Deterministic Duty cycle specific Stochastic Requires statistical information about duty cycle
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Requirements for modelling Data driven models can be very accurate but give little insight and no predictive power Physics based dynamic models offer insight and reuseability But today s physics based models are not good enough for calibration Must respond appropriately to changing boundary conditions (e.g. thermal) Need to pass boundary conditions to neighbouring models (e.g. heat rejection) Controller and calibration must be included Must respond appropriately to calibration inputs (such as divided injections) 58
Ultraboost exploring the limits of downsizing Complex air path, demanding dynamics
Improving turbocharger modelling 1D and 3D simulation Hot, pulsed, gas stand On engine mapping 6 Dynamic engine test
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles Optimisation of calibration error predicted 5 4 2 5 1 measured error 5-5 5 1 measured -5 2 4 6 8 1 12 samples Validation testing on road with PEMS
Simulation requirements for optimisation 62 Not enough to optimise over one preset velocity based cycle, or even a family Need a stochastic approach with probability of given manoeuvres factored in At least the system level goals and constraints are given to us! Need to decompose to sub-system level
Optimisation workflow Hardware selection Optimise for full map steady state compliance Steady state test Develop models robust to boundary conditions Advanced powertrain test Generate dynamic engine and aftertreatment models Dynamic characterisation on CD Dynamic simulation of real manoeuvres 2 Speed (rpm) NOx (ppm) 3 2 1 1 5 Measured Predicted 4 6 8 1 12 Time (s) Validation testing on CD over preset cycles predicted 4 2 5 1 measured error 5-5 5 1 measured 5 Optimisation of -5 2 4 6 8 1 12 samples calibration error Validation testing on road with PEMS
Workflow and facilities need improvements Perhaps the biggest challenge is the way large companies traditionally work Design, simulation, build, test ops., calibration all need to be joined up Re-use and improve the initial models throughout the process Most powertrain dynos and chassis dynos are not flexible or precise enough today Neither are their operating practices 64
Conclusions RDE will require systematic engine optimisation in vehicle system context Behaviour dominated by steady state capability Boundary conditions and interactions critical Signoff on random CD cycles unlikely to be a robust process on its own Better use of software tools essential Use the CD to generate a rich dataset Validate advanced models Optimise in software Signoff on random cycles with more confidence Significant implications for test design/operation Precision, Access to engine data and actuators
Contact Chris Brace FIMechE Professor of Automotive Propulsion Deputy Director, Powertrain and Vehicle Research Centre University of Bath BA2 7AY +44 1225 386731 C.J.Brace@Bath.ac.uk