Model based Calibration of Engines using DOE 

SANJAI KANNAN

MASTEROF ENGINEERING-AUTOMOTIVE ENGINEERING

PSG COLLEGE OF TECHNOLOGY

In this project, Heavy duty commercial vehicle diesel engine is selected to create a methodology for Engine calibration. Heavy commercial vehicle Engine and the test facilities are studied to understand the technology of engine and the capabilities of test facilities to perform the required tests at Engine level. This particular engine is BS III emission compliant engine and has Common rail and Turbocharger implemented. From ECU study, identified the Operating range of Engine performance and also the technology implemented to control the emissions. The parameters available to vary on Fuel system are: Main injection timing, Fuel pressure, Pilot 1 Injection quantity, Pilot 1 Injection timing, Pilot 2 injection quantity and Pilot 2 injection timing. The parameters available to vary on Air system is Boost pressure (Intake Manifold pressure). However, the implemented maps in the Engine calibration doesn’t include Pilot 2 injection quantity and pilot 2 injection timing.

Divided the Engine parameters into global design and local design where in global design includes Engine speed and Engine Torque and Local design includes the variation of all the parameters to vary at each Engine speed and Torque – Main injection timing Main injection timing, Fuel pressure, Boost pressure, Pilot 1 Injection quantity and Pilot 1 Injection timing. The global design involves 2 parameters – Engine speed and Torque while the local design involves 5 parameters – Main injection timing, Fuel pressure, Boost pressure, Pilot 1 Injection quantity and Pilot 1 Injection timing.

Applied Design of Experiments – Lattice hypercube sampling (LHS) methodology to create reduced test matrix of Engine test data collection points for the entire operating zone. When LHS is applied, created 3024 test points for Engine operation. When Design of experiments is not applied and taking a full factorial test at Entire operating range in a random way takes 343750 test points. Ideally LHS design application reduced the Engine testing time from 2 years to 6 days.

Due to COVID19 situation, Engine was not available to run and collect Engine data for creating calibration solution. However, this was solved by providing Engine simulation data from a similar Engine. However, the technology in the Engine simulated has the variable parameters – Main injection timing, Fuel pressure, Boost pressure and Mass air flow.

Using the Simulated Engine data and Gannet calG (Accelerated calibration platform), the critical emissions for this Engine – NOx and Soot are optimized at local points and for the Emission cycles. From this approach, multiple solutions are created for achieving Emission cycle NOx and soot results. From these calibration solutions, Input parameter maps for each solution (Main injection timing, Fuel pressure, Boost pressure and Mass air flow) are prepared to flash into ECU and test it on engine.

Engine Model selection for Calibration 

ARJUN SIVASANKAR  

B.Tech – MECHANICAL & MANUFACTURING ENGINEERING

MANIPAL INSTITUTE OF TECHNOLOGY

Engine models are widely used in all stages of Engine development from Conceptual design to validation. Having one kind of model for all development stages is not suitable. Now with complexity of Engine development process to meet Advanced emission regulations, the requirements of model accuracies demand much closer

The most followed Engine modelling approaches are Physics based modelling and Numerical based modelling. Physics based modelling approaches are broadly used in conceptual development stage and controls development stage. Numerical based modelling is broadly used for complex combustion modelling and use in Emissions modelling and performance modelling. Physics based modelling involves many approximations and need extensive study on Engine and need detailed inputs of geometry of each component present in the Engine while Numerical based modelling involves creating right set of equations between outputs and inputs including the combination of inputs.

For calibration of Engine performance, the accuracies levels needed from Engine models for Emissions, Fuel consumption and drivability are very close and needs less time to model the responses.

In this project, Different Numerical methods of Engine modelling are applied to Engine simulation data to select the right model for each response of Engine. These selected models can be used in Engine calibration to achieve good results and very much close to real world results. There is no single model, which can help for all responses. So, selecting right models for all responses is very much needed.

For this project, we had selected Passenger car Diesel engine. Engine Test data from Engine test cell at steady state conditions and transient state conditions are used to model, optimize, validate and select right model for all Engine responses. Engine responses considered in this project are NOx and Soot. Inputs for these responses are Engine speed, Engine torque, Fuel pressure, Boost pressure, Main injection timing, Fresh mass air flow and Air fuel ratio.

The mathematical modeling methods used are Linear polynomial regression and Gaussian process – Matern kernel and RBF kernel. Using the steady state Part throttle performance data of entire engine operating zone, steady state emission cycle data and Transient cycle data are considered.

Initially, the linear polynomial modelling and Gaussian process modeling are applied for NOx response for PTP data, Steady state emission cycle data and Transient data.

The results achieved are:

• For PTP data, Linear polynomial degree 3, 4, 5, 6, 7,8 9, 10 and Gaussian process – Matern and RBF kernel are providing good results.
• For steady state data, Linear polynomial degree 2, 3, 4, 8 9, 10 and Gaussian process – Matern and RBF kernel are providing good results.
• For transient state data, Linear polynomial degree 2, 3, 4, 8 9, 10 and Gaussian process – Matern and RBF kernel are providing good results.

Summary:

• For PTP data, the results are within +/- 0.01%
• For Steady state emission cycle data, the results are within +/-3%
• For Transient state data, more than 80% points are within +/-10%
• Overall for the 3 sets of data, Linear polynomial degree 3,4, 8,9,10 and Gaussian process – Matern and RBF kernel are providing good results for NOx response for this engine.

Caution: This results for NOx is good for this Engine only. Other engines to be performed with the similar procedure to select the right Model

When applied the same modelling approaches to Soot response, the results are:

• For PTP data, Linear polynomial degree 3, 4, 5, 6, 7,8 9, 10 and Gaussian process – Matern and RBF kernel are providing good results.
• For steady state data, only Gaussian process – Matern and RBF kernel are providing good results.
• For transient state data, only Gaussian process – Matern and RBF kernel are providing good results.

Summary:

• For PTP data, the results are within +/- 0.01%
• For Steady state emission cycle data, the results are within +/-3%
• For Transient state data, more than 80% points are within +/-10%
• Overall for the 3 sets of data, only Gaussian process – Matern and RBF kernel are providing good results for Soot response for this engine.

Caution: This results for Soot is good for this Engine only. Other engines to be performed with the similar procedure to select the right Model

We had tested for another Heavy duty commercial vehicle diesel engine and we found linear polynomial modelling and Gaussian process modelling are providing very good results.

By using this methodology, the right models for each responses can be selected for a particular engine and can be used for Engine calibration at different test conditions. Using calG, created multiple calibration solutions for the engine using linear polynomial model.

Optimisation of Engine out emissions using Box Behnken and response surface approach

DELINA RAJKUMARI R D

Masters in Control Systems

PSG COLLEGE OF TECHNOLOGY

Engine calibration is human centric process involving lot of trade-offs to optimize at different conditions. With advanced emission regulations, the number of trade-offs and test conditions had increased drastically. This forces to use advance methodologies to use computational algorithms to generate calibration solutions which can satisfy different criteria.

In this project, Engine testing time and optimization time reduction is the main focus.

• For Engine testing, studied on various Design of experiments designs and applied Box Behnken design and Plackett Burman design for global and local design of Engine.
• For Engine optimization, performed Manual calibration using Gannet calG and Engine test data for creating multiple solutions to validate on Engine. Creating each manual solution in calG is very quick and cost effective compared to Manual solution development in Engine test cell.

Real Drive emissions, OBD thresholds, CO2 & NTE norms demand:

• Real drive emissions, OBD thresholds, CO2 & NTE norms optimization involves calibration trade-offs of all emission pollutants at different test conditions and various scenarios.
• Manual calibration to achieve these targets take lot of time, human resources, test facilities and proto vehicles.
• Efficiency of Manual calibration is not high at many cases
• To achieve all the targets providing good fuel economy and drivability, we need to generate calibration solutions using computational algorithms rather than manual calibration.

Engine testing time reduction:

In this project, we had applied Plackett Burman design for Engine global design and achieved 98 Engine speed and Torque points to test at the Entire operating zone. For Engine local design, Box Behnken design is applied and achieved 41 Local points at each global point. Total test matrix points for initial data collection using Plackett Burman design and Box Behnken design are 4018 test points. Full factorial points come around 343750 points. The total tests across the entire operating zone with Design of experiments applied can eb completed in 20 days of Engine test cell facility usage while the full factorial tests take 2 years.

Engine optimization time reduction:

Once the right set of Engine models are selected for optimization, used calG to perform engine calibration and achieved multiple solutions to achieve Steady state emission cycle and Transient state cycle NOx and Soot values. This methodology already saves Engine test cell time drastically. But it involves a lot of human effort to create multiple solutions.

When the number of input parameters increase, test conditions increase and trade-offs needed to optimize increase, this becomes the most complex task to achieve. In Calibration, there is no one solution. Many solutions achieved are compromised for Fuel economy or drivability while maintaining the emission values for different conditions. When advanced regulations are needed and OEMs struggle to develop a product in less time, the automated calibration solutions generation is surely out of the box thinking approach which is the real need for the industry now.

The responses optimized in this project are NOx and Soot. We had selected 3^rd degree Linear polynomial models of NOx and Soot for this purpose. We had applied gradient based optimization methods for standard optimization and cyclic optimization.

Applied standard optimization gradient method to achieve the best possible solution for few steady state points. Once we are able to achieve one solution at each point. We could able to generate multiple solutions for each point. With the results achieved, we had used calG to validate and find out the feasible solutions out of all the generated solutions.

Once standard optimization was successful and able to generate calibration solutions at each point, cyclic optimization is focused to achieve multiple calibration solutions for whole steady state emission cycle.

The results achieved are validated in calG and found the feasible calibration solutions. For those calibration solutions, Engine parameter maps are prepared using calG and created Multiple Engine calibration maps for flashing into ECU.