Optimisation of Piston Bowl Geometry and Double Injection Strategy in Diesel Engines Using a Bespoke Hybrid Regression-Based Technique

Abstract

This research work is driven by the requirement for more efficient and cleaner internal combustion engines that meet fuel saving demands, stringent emissions regulations, and compatibility with hybrid powertrain technology. Despite the improvements made in current engines as useful industrial power units, the environmental and health effects of their emissions as well as relatively low fuel efficiency are still a cause for further research and development. In this work, a 3D Computational Fluid Dynamics (CFD) engine model is validated against experimental data obtained from a single cylinder Ricardo Hydra diesel engine. The validated engine model is used to simulate the in-cylinder combustion processes and emissions characteristics of the engine using AVL Fire software. The engine model is used to investigate the effects of piston bowl geometry, and double injection strategy on diesel combustion and emissions at low and high engine loads operating with double fuel injection strategy. During the investigation, the synergies between piston bowl geometry, injection timing and injection ratio were analysed. This analysis yielded interesting findings on the influence of these parameters on engine performance and emissions at low and high loads. The results showed that exhaust emissions are more sensitive to piston bowl geometry while injection timing and split injection ratio can significantly influence both performance and emissions. In addition, the re-entrance curvature of the piston bowl geometry showed to influence air-fuel mixing by modifying the surface area of the mixing front closer to the surface of bowl side and base walls. In this investigation, larger radius piston bowls denoted by ORB1 and ORB2 resulted in a larger mixing front area. The investigation was extended further towards the multi-objective optimisation of piston bowl geometry and double injection strategy which was performed using a two-stage optimisation approach. The first stage involved a non-pareto constraining optimisation of piston bowl geometry that considered the results from the initial investigation at low and high loads. The goal was to obtain a piston bowl geometry that was suitable for the second stage of the optimisation process which focused on the double injection strategy. The results from the first stage showed that ORB2 was most suitable at low and high loads due to enhanced combustion phasing stability.

The second stage involved the optimisation of double fuel injection strategy at low and high engine loads. A unique optimisation methodology called the Hybrid Regression-based Technique (HRT) was proposed and implemented for this stage to achieve optimal solutions at significantly reduced computational time and cost. The initial aspect of the HRT involved the implementation of a unique hybrid dataset of designs generated using a coupled DoE-Optimiser algorithm approach which proved efficient in enabling an early indication of the pareto region of the multi-objective domain. The HRT also involved an ensemble approach for regression modelling over the hybrid dataset in a bespoke manner. The bespoke predictive learner/model was a hybrid of multiple models obtained from the consideration of a library of regression methods via a comparative analysis. This hybrid model was coupled with the MOGA-II to obtain optimal solutions for the double injection strategies with respect to each performance and emissions parameter considered. Validation of the results showed that the HRT provided acceptable optimal solutions for double injection strategy. At low load, optimal double injection strategies were characterised by relatively early first injection events occurring at about 15°bTDC which contained 80% to 85% of the total injected fuel and about a 10CA dwell angle. At high load, optimal double injection strategies were also characterised by relatively early first injection events occurring at about 11°bTDC which contained 90% to 95% of the total injected fuel and a dwell angle between 8CA and 12CA. More also, the Hybrid Regression-based Technique was able to achieve these improvements with a significantly reduced computational time of over 80% compared to the conventional optimisation approaches in similar studies. The objectives of this research work were to develop a numerical model capable of predicting diesel combustion and emission under different operating conditions, study the effects that arise from the interactions between piston bowl geometry and injection strategy and to identify an optimisation methodology that tackles the complexity involved in engine CFD multi-objective optimisation. The investigations carried out in this work thereby contribute to the knowledge in this field by offering a new perspective for engine modelling and optimisation involving advanced injection strategy and piston bowl geometry which may be useful in providing insight to other researchers in the quest for further engine development and optimisation.