Skip to main content

Adaptive Optimal Estimation and Control for Automotive Engine Systems with Approximate Dynamic Programming

Final Report Summary - AECE (Adaptive Optimal Estimation and Control for Automotive Engine Systems with Approximate Dynamic Programming)

Global production and the use of individual passenger vehicles is to increase by 630m from 2000 until 2030. This requires the development of automotive engine management systems (EMS) for increased fuel economy and low toxic gas emissions, using advanced estimation and control approaches. In particular, future proven powertrain structures, e.g. engines for hybrid powertrains, have been of interest for this project. The results of this project have provided new means of vehicle engine control technologies to both academic and industrial communities. Hence, this project was dedicated to investigate a novel online optimal adaptive estimation and control framework and to study its applications to vehicle engines (e.g. indicated torque estimation and air-fuel-ratio (AFR) control). For this purpose, a recently developed bio-inspired approach, approximate dynamic programming (ADP), and other novel adaptive techniques have been exploited to solve online optimal estimation or the AFR control problem by using adaptation, optimization and learning. Practical implementation for internal combustion (IC) engines has been also carried out for an engine test-rig. Hence, the overall objective of this project was to develop novel optimal adaptive control and estimation methods and to practically achieve improved fuel economy and performance, reduced emissions, for current engines, while reducing the development costs by minimizing the number of the used hardware transducers.

1) Specific technical work that has been conducted to support the overall project is:
i) Engine modeling and validation: Engine modeling techniques have been extensively reviewed, and the mean-value engine model (MVEM) has been further extended by considering exhaust gas recirculation (EGR) and fuel injection dynamics. Collaborating with the University of Bath, we have built and calibrated several engine models using a commercial engine simulation software (GT Power);
ii) Novel adaptive parameter estimation to determine engine model parameters: We have developed a novel, systematic theoretical adaptation framework to achieve fast, robust parameter estimation. These proposed algorithms can achieve finite-time convergence, and better transient response than classical gradient and least squares methods, allowing also online verification of a critical excitation condition for parameter estimation;
iii) ADP-based adaptive observer design and engine torque estimation: We have developed a simple, robust unknown input observer to address the engine torque estimation problem using the measured engine velocity and load torque. A combined dynamic simulator based on IPG Carmaker and Matlab was established. Comprehensive comparisons to other established input observers have been conducted and validated using simulations and experiments;
A novel adaptive observer technique has been developed which will be the basis for further work for optimal observer-based output feedback ADP-control.
iv) ADP-based optimal adaptive control design with application to engine ARF regulation: Adaptive optimal tracking control problem for generic nonlinear systems has been solved by incorporating our adaptive algorithm into the ADP method. The idea of an unknown input observer was further extended to regulate the Air-Fuel-Ratio (AFR) of the engine injection system at its desired stoichiometry value to improve the fuel efficiency and emission performance;
v) Simulation and experimental results: We have built several dynamic engine simulators in commercial software (e.g. Matlab/Simulink, GT-Power), and conducted extensive simulations to validate the theoretical findings (e.g. torque estimation). Moreover, we have conducted AFR control experiments based on a new industry-provided two cylinder engine (under realistic driving cycle) to validate the two proposed AFR control methods with unknown input observer compensation. The practical results show very promising results in terms of AFR regulation response and CO, NOX and THC emission. This new control strategy can reduce offline calibration efforts and costs, i.e. the proposed controllers easily integrate into and improve existing methods and do not require information from lookup tables.

2) Management and Result summary: The project has progressed smoothly and successfully. It achieved and exceeded the planned practical objectives. For the practical work, we opted for a simple, yet highly effective approach using an unknown input observer idea. Major parts of the theoretical objectives have been finalized. In fact, we are still working on further developing the proposed optimal control strategies, for output feedback, but also to better solve the optimal AFR control problem. Moreover, the proposed estimation algorithm is further being studied to address the engine indicated torque estimation problem with less measurements (e.g. engine velocity and wheel rotation velocity) by taking the powertrain and tyre dynamics into consideration.
The research of this project has generated a number of novel estimation and control approaches that have been validated based to newly build engine simulators to address the engine modeling, torque estimation and AFR control. So far, the research results of this project have created 3 journal articles, 1 book chapter and 7 conference papers that have been published or accepted, and another 5 journal articles that are under review (See more details in the next Section 2 Dissemination).

3) Impact: The developed algorithms are potentially suitable for next generation engine production and may be employed by one of our industrial collaborators in the near future. The two industrial collaborators were TATA Motors Plc and Jaguar Land Rover. During the project period, we have launched strong links and collaborations with several academic universities and industrial partners, which have led to several research grants, studentships and potential funding opportunities.
In particular, this project has strong links with two collaborators: the Powertrain & Vehicle Research Centre (PVRC) at the University of Bath (Professor Chris Brace and and Dr Richard Burke), and the associated UK-industrial partner; both of them have supported this project in terms of engine simulation software (GT Power), experimental data and test-rig. Following this collaboration, we have prepared and signed a three-party non-disclosure agreement (NDA). Moreover, this collaboration also created the basis for a new joint PhD-student (Professor Chris Brace will serve as the co-supervisor for Anthony Chen, who will start pursuing his PhD at Bristol on modelling and control for engine and powertrain systems). From September 2016 to January 2017, Mr Yingbo Huang, a PhD candidate at the Kunming University of Science and Technology, was also invited to visit University of Bristol as a Researcher, who has participated in and contributed to the engine experiments in PVRC of Bath.
Moreover, we have liaised with Dr Bohong Xiao (who was a Diesel Application Software Planning Engineer at the Powertrain Calibration Controls, Ford Motor Company, UK), and submitted a project application for Ford’s University Research Programs (URP), as a possible new industrial connection.

4) Further outcomes: The Researcher has also launched a new collaboration with Dr Guang Li from the Queen Mary University of London working on ADP based optimal control for a renewable energy plant associated with the problem of output power maximization. Based on this collaboration, the Researcher has been successfully awarded a Newton Mobility Grant jointly supported by the Royal Society, UK, and the National Natural Science Foundation of China (No. IE150833/61611130213).

5) Knowledge transfer and further dissemination: the knowledge created by this project has been well transferred to the University of Bristol through regular research meetings and discussions with students and colleagues, as well as public seminars. The knowledge has also been shared with the University of Bath through numerous discussions and publications with collaborators. In addition, knowledge transfer within Europe and to China has been achieved through academic visits, student exchanges, talks, and collaborations with Queen Mary University of London, Beijing Institution of Technology, and Kunming University of Science & Technology.

6) Training and teaching: the researcher has also participated in many training courses or workshops offered by the host university's academic staff career development department to acquire complementary skills, and to stimulate professional maturity (See more details in Section 2). He has also significantly contributed to several research grant applications. Moreover, the researcher has been heavily involved in the teaching and student supervision activities, i.e. he has supported the host, Dr Guido Herrmann, for two MSc students and one final year group project (four undergraduate students). Hence, these supervised students advertently also contributed to outcomes, i.e. publications, of this project.