Adaptive Cylinder Pressure Reconstruction for Production Engines

Lead Research Organisation: University of Sussex
Department Name: Sch of Engineering and Informatics

Abstract

Improving the fuel efficiency of the IC engine is important to meet the growing demand for non-renewable energy, and to reduce the emission of carbon dioxide - a major contributor to global warming. Advanced feed-back control strategies offer an important way of improving engine efficiency for existing designs. But to fully exploit these control strategies, a cost effective, durable, and real-time method of measuring engine cylinder pressure is needed since existing sensors are far too expensive and are seriously undermined by long-term durability issues. The search for alternative means of cylinder pressure reconstruction for production engines has continued for two decades. This search is now of critical importance for both conventional and future HCCI engines. Two indirect pressure reconstruction methodologies have been proposed using either measured crank-shaft motion or measured engine-casing vibration. But although numerous methods have been suggested to exploit these two approaches not one single method has yet been fitted to a production engine. There are two reasons for this: i) the most promising method (recurrent neural network model) is still in need of a suitably tuned training methodology, and ii) fixed-parameter reconstruction models will not in general produce accurate pressure reconstruction on a different engine (even of the same type) owing to the effect of variability arising from normal differences in materials, manufacture, operating conditions, and component wear. A fully adaptive reconstruction technique is needed. This proposal aims to create a robust adaptive cylinder-pressure reconstruction methodology for production engines, and to test this methodology on real engine data. This is timely because preliminary studies point very favourably to the most suitable architecture for multi-cylinder pressure reconstruction, but as yet, it is not known how to train these models, even for application to single test engines. A detailed understanding of the stochastic parameter fitting problem is needed. Only then is it likely a suitable training strategy can be designed. More importantly, to address the needs of an adaptive system, a way has to be found to allow fixed-parameter systems become variable-parameter models. Three novel variable-parameter schemes are proposed and these will be appropriately tested. The big question however is how should such variable-parameter schemes be trained for adaptive reconstruction? This question will be addressed in the project.
 
Description The research has focused on finding a way to create the sensing needed for IC engine combustion control at a fraction of the cost of using cylinder pressure sensors. This is to improve engine efficiency and fuel economy, reduce CO2 and noise emissions, and to enable fault-detection, particularly of poor combustion and engine-misfires, for light-duty vehicles. We now have two accurate methods with Gasoline engine cylinder pressure reconstruction capability. These methods will now allow the production engine adaptive requirement issues to be addressed.

The first method uses a recurrent neural network, trained using a fully recurrent methodology. The allows in-cylinder pressures in IC engines to be reconstructed using measured crank kinematics as input to a suitably configured recurrent (NARX) neural network. This is tested on real engine data to assess reconstruction capability. It avoids the need for extremely expensive pressure sensors used in closed-loop engine combustion control. The challenge addressed is to accurately predict cylinder pressure traces within the cycle under generalisation conditions: i.e. using data not previously seen by the network during training. The NARX neural network is first trained using a tuned fully-recurrent training methodology - the so called 'Robust Adaptive Gradient Descent' (RAGD) algorithm). This involves direct construction and calibration of a suitable inverse crank dynamic model which owing to singular behaviour at top-dead-centre (TDC), has proved very difficult in the past via physical model construction, calibration, and inversion. Training using the RAGD algorithm offers a practical option for assembly line training of NARX networks for engine cylinder pressure reconstruction owing to its high efficiency and reasonable accuracy, not previously possible using alternative recurrent training methods. This method is generally fast but much complicated than the second method.

The second method uses much simpler Time-Delay feed-forward Artificial Neural Networks trained with the standard Levenberg-Marquardt algorithm. The same approach can be applied to reconstruction via either measured crank kinematics obtained from a shaft encoder, or measured engine cylinder block vibrations obtained from a production knock sensor. The basis of the procedure is initially justified by examination of the information content within measured data, which is considered to be equally important as the network architecture and training methodology. Key hypotheses are constructed and tested using data taken from a 3-cylinder (DISI) engine to reveal the influence of the data information content on reconstruction potential. The findings of these hypotheses tests are then used to develop the methodology. The approach is tested by reconstructing cylinder pressure across a wide range of steady-state engine operation using both measured crank kinematics and block accelerations. The results obtained show a very marked improvement over previously published reconstruction accuracy for both crank kinematics and cylinder block vibration based reconstruction using measurements obtained from a multi-cylinder engine. The paper shows that by careful processing of measured engine data, a standard neural network architecture and a standard training algorithm can be used to very accurately reconstruct engine cylinder pressure with high levels of robustness and efficiency.
Exploitation Route More efficient internal combustion engines for light-duty vehicles leading to significant reductions in CO2 emissions, i.e. cost-effective low carbon vehicles with better fuel economy and low harmful emissions. Jaguar Land Rover has continued to be directly involved in steering and supporting this research via the JLR Engineering Centre, Whitley, Coventry, UK and more recently JLR Research, Warwick UK.
Sectors Transport
 
Description The findings of this project have been used as part of a case study in a University (of Sussex) wide seminar entitled: 'Collaboration with Industry for Impact'. This seminar involved both Jaguar Land Rover staff and researchers in drug development. The seminar contrasted how research impact occurs in the automotive and pharmaceutical industries. In January 2017, Jaguar Land Rover (i.e. Combustion Specialist Mr Dave Richardson, dricha69@jaguarlandrover.com) supplied an Impact statement indicating the potential Impact of the research as follows: The UK is a centre for engine development and manufacture with a strong presence in the international engine research arena. The research at the University of Sussex in developing Adaptive cylinder pressure reconstruction for production engines represents a significant advantage as it potentially offers a true alternative to in-cylinder pressure sensing, which is seen as increasingly necessary as legislation requires continued reductions in CO2 and exhaust emissions from vehicles. Current state-of-the-art application of this technology requires expensive hardware, which may not prove to be adequately robust. The research proposes a potentially more cost effective method which, if successful, would give UK engine design and manufacture a competitive edge in this research-intensive field. It would also have clear environmental and societal impact by promoting wider availability and uptake of clean, low-cost mobility solutions. The research projects investigating pressure trace reconstruction by adaptive neural networks has potential as a technology for closed-loop combustion control (CLCC). This type of technology will enable advanced internal combustion (IC) engine combustion systems and the implementation of new technologies. These will improve the efficiency of the IC engine so reducing CO2 emission from the transport sector. Advanced control strategies allowing new combustion and engine technologies to be realised will also contribute to further reductions in exhaust emissions, both gaseous and particulate, that are required to improve air quality. CLCC may also enable engines to adapt to variations in fuels and fuel qualities in world markets and allow the introduction of sustainable alternative fuels without the need to calibrate the engine to all possible fuel blends. The more fundamental aspects of the research carried out at the University of Sussex into the structure and training of adaptive neural networks as part of these projects contributes to all aspects of neural network control systems, not just combustion monitoring and control, so has many potential applications with impact that cannot necessarily be quantified at the moment. The application of the research cannot be directly linked to a current product but is technology that is being considered for future products.
Sector Transport
Impact Types Societal
 
Description Research student top-up funding
Amount £39,000 (GBP)
Organisation Jaguar Land Rover Automotive PLC 
Department Jaguar Land Rover
Sector Private
Country United Kingdom
Start 03/2011 
End 10/2012
 
Description Research student top-up funding
Amount £19,000 (GBP)
Organisation Jaguar Land Rover Automotive PLC 
Department Jaguar Land Rover
Sector Private
Country United Kingdom
Start 09/2012 
End 09/2014
 
Description Research student top-up funding
Amount £19,000 (GBP)
Organisation University of Sussex 
Sector Academic/University
Country United Kingdom
Start 09/2012 
End 09/2014
 
Title Accurate gasoline engine cylinder pressure reconstruction capability 
Description We now have two accurate methods with Gasoline engine cylinder pressure reconstruction capability. These methods will now allow the production engine adaptive requirement issues to be addressed. The first method uses a recurrent neural network, trained using a fully recurrent methodology. The allows in-cylinder pressures in IC engines to be reconstructed using measured crank kinematics as input to a suitably configured recurrent (NARX) neural network. This is tested on real engine data to assess reconstruction capability. It avoids the need for extremely expensive pressure sensors used in closed-loop engine combustion control. The challenge addressed is to accurately predict cylinder pressure traces within the cycle under generalisation conditions: i.e. using data not previously seen by the network during training. The NARX neural network is first trained using a tuned fully-recurrent training methodology - the so called 'Robust Adaptive Gradient Descent' (RAGD) algorithm). This involves direct construction and calibration of a suitable inverse crank dynamic model which owing to singular behaviour at top-dead-centre (TDC), has proved very difficult in the past via physical model construction, calibration, and inversion. Training using the RAGD algorithm offers a practical option for assembly line training of NARX networks for engine cylinder pressure reconstruction owing to its high efficiency and reasonable accuracy, not previously possible using alternative recurrent training methods. This method is generally fast but much complicated than the second method. The second method uses much simpler Time-Delay feed-forward Artificial Neural Networks trained with the standard Levenberg-Marquardt algorithm. The same approach can be applied to reconstruction via either measured crank kinematics obtained from a shaft encoder, or measured engine cylinder block vibrations obtained from a production knock sensor. The basis of the procedure is initially justified by examination of the information content within measured data, which is considered to be equally important as the network architecture and training methodology. Key hypotheses are constructed and tested using data taken from a 3-cylinder (DISI) engine to reveal the influence of the data information content on reconstruction potential. The findings of these hypotheses tests are then used to develop the methodology. The approach is tested by reconstructing cylinder pressure across a wide range of steady-state engine operation using both measured crank kinematics and block accelerations. The results obtained show a very marked improvement over previously published reconstruction accuracy for both crank kinematics and cylinder block vibration based reconstruction using measurements obtained from a multi-cylinder engine. The paper shows that by careful processing of measured engine data, a standard neural network architecture and a standard training algorithm can be used to very accurately reconstruct engine cylinder pressure with high levels of robustness and efficiency. 
Type Of Material Technology assay or reagent 
Provided To Others? No  
Impact The research projects investigating pressure trace reconstruction by adaptive neural networks has potential as a technology for closed-loop combustion control (CLCC). This type of technology will enable advanced internal combustion (IC) engine combustion systems and the implementation of new technologies. These will improve the efficiency of the IC engine so reducing CO2 emission from the transport sector. Advanced control strategies allowing new combustion and engine technologies to be realised will also contribute to further reductions in exhaust emissions, both gaseous and particulate, that are required to improve air quality. CLCC may also enable engines to adapt to variations in fuels and fuel qualities in world markets and allow the introduction of sustainable alternative fuels without the need to calibrate the engine to all possible fuel blends. The more fundamental aspects of the research carried out at the University of Sussex into the structure and training of adaptive neural networks as part of these projects contributes to all aspects of neural network control systems, not just combustion monitoring and control, so has many potential applications with impact that cannot necessarily be quantified at the moment. The application of the research cannot be directly linked to a current product but is technology that is being considered for future products. 
 
Title A power supply system including a resonant mass elastic system and cylinder IC 
Description This is revolutionary rotary generator concept designed as a potential Range Extender engine for hybrid-electric vehicles. The concept exploits a highly efficient, single cylinder resonance cycle based engine that is lighter, more compact, and delivers a much higher power density than equivalent generators currently on the market. It is designed to suffer from less friction than its current counterparts, and is totally balanced, meaning it will create very little noise and vibration. 
IP Reference PCT/GB2454360 
Protection Patent granted
Year Protection Granted 2014
Licensed No
Impact The concept is currently at TRL-2 and has yet to be developed.
 
Title POWER SUPPLY SYSTEMS 
Description This is revolutionary rotary generator concept designed as a potential Range Extender engine for hybrid-electric vehicles. The concept exploits a highly efficient, single cylinder resonance cycle based engine that is lighter, more compact, and delivers a much higher power density than equivalent generators currently on the market. It is designed to suffer from less friction than its current counterparts, and is totally balanced, meaning it will create very little noise and vibration. 
IP Reference EP2215339 
Protection Patent granted
Year Protection Granted 2010
Licensed No
Impact This is at TRL-2 and needs to be developed.
 
Description Presentation to the UK Universities Internal Combustion Engines Group (UnICEG) one-day meeting at Oxfod University September 2015 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact This was a presentation at a full-day UnICEG meeting at Oxford University essentially summarising the main findings of our project. The presentation was very well received from the UK community of IC Engine experts.
Year(s) Of Engagement Activity 2015
 
Description University-Wide Seminar 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Local
Primary Audience Professional Practitioners
Results and Impact The findings of this project have been used as part of a case study in a University (of Sussex) wide seminar entitled: 'Collaboration with Industry for Impact'. This seminar involved both Jaguar Land Rover staff and researchers in drug development. The seminar contrasted how research impact occurs in the automotive and pharmaceutical industries.
Year(s) Of Engagement Activity 2015
 
Description Visit to India with Secretary of State for Business Innovation and Skills and Lecture to Automotive Research Association of India 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact In October 2014 Julian Dunne joined the UK Secretary of State for Business Innovation and Skills (Dr Vince Cable) on a 4-day British High Commission mission to India, where he lectured at the Automotive Research Association of India. This involved both small-scale and much larger social engagements with Indian business leaders in Mumbai. Then a visit to the Tata vehicle manufacturing plant in the city of Pune, followed by a half-day visit and tour of the facilities at the Automotive Research Association of India, also in Pune.
Year(s) Of Engagement Activity 2014