Thesis. 1978. Ocean E.--Massachusetts Institute of Technology. Dept. of Ocean Engineering. / MICROFICHE COPY AVAILABLE IN ARCHIVES AND ENGINEERING. / Includes bibliographical references. / Ocean E.

Digitized by Kansas Correctional Industries

Time-delay feed-forward Artificial Neural Networks are examined for gasoline engine cylinder pressure reconstruction using both measured crank kinematics obtained from a shaft encoder, and measured engine cylinder block vibrations obtained from a production knock sensor. Initially, the study focuses on the information content associated with measured data, which is considered to be of equal importance to the particular network architecture and the training methodology. Several hypotheses are constructed, which when tested, reveal the influence of the data information content on the reconstruction potential and limitations. These hypotheses are tested on real data from a 3-cylinder (DISI) engine. Three distinct ideas emerge through this testing process, which are combined to produce a single pressure reconstruction methodology. Reconstruction results obtained via this methodology, applied to crank kinematics associated with steady-state engine operation, show a marked improvement over previously published reconstruction accuracy. Moreover, in steady-state engine operation, the application of this methodology to acceleration measurements of cylinder block vibration, obtained from a knock sensor, show very significant improvements over previous attempts. But the direct application of this same reconstruction methodology to transient engine operation, proves to be problematic. However, a novel generalisation of the approach in the form of a time-dependent feed-forward neural network is proposed and the required adaptation made to the use of the Levenberg-Marquardt training algorithm. This time-dependent approach has been tested under limited transient conditions and shown in the thesis to give good results, therefore offering considerable potential for use with real engine operation. Overall, the thesis shows that by careful processing of measured engine data, standard neural network architectures and standard training algorithms can be used to reconstruct engine cylinder pressure.

The main driving force behind this research was the need for cleaner and more efficient engines to meet the ever-increasing demands on the modern automobile's emissions. In recent years different studies have been carried out to analyze the combined effects of high-pressure injection, boost pressure, multiple injections, included spray angle and combustion chamber geometry. Though considerable research has shown these technologies can meet the low emission regulations, the careful optimization of the engine operating conditions is still required in order to get the full benefit of the different strategies. With these issues as motivation, the first important objective of this study was to gain a detailed understanding of the mechanisms through which fuel injection interacts with other engine parameters and influences diesel combustion and emissions, and hence to attempt to generalize the adoption of multiple injection strategies with regards to improving diesel engine performance. For this purpose, a modified parameter called “Homogeneity Factor of in-cylinder charge” (HF) was introduced and proposed as a new measure in combustion theory to analyze the combustion characteristics and air-fuel mixing process of diesel engines in more detail. The second part of this research builds upon a detail investigation on the included spray cone angle concept and explores further their use in conjunction with multiple-injection strategies in diesel engines. In addition, an investigation was performed in third phase of this research to analyze the effects of piston geometry on combustion, performance and exhaust emission characteristics. The results showed that employing a post-injection combined with a pilot injection results in reduced soot formation from diffusion combustion and enhances the soot oxidation process during the expansion stroke, resulting in decreased soot emissions, while the NOx concentration is maintained in low levels. It was also found that spray targeting is very effective for controlling the in-cylinder mixture distributions especially when it accompanied with advanced injection strategies. Moreover, the results confirmed that a narrower width of piston bowl has a higher unburned fuel air mixture region and hence results in higher soot emissions but with slightly larger piston surface area the optimum operating point could be obtained.

Due to the inherent nonlinearity of the diesel engine, real-time control of the variable geometry turbocharger (VGT) and exhaust gas recirculation (EGR) valve still remains a challenging task. A controller has to be capable of coping with the transient operating condition of the engine, the interactions between the VGT and EGR, and also the trade-off effect in this control problem. In this work, novel real-time fuzzy logic controllers (RFLC) were developed and tested. Firstly, the proposed controllers were calibrated and validated in a transient diesel engine model which was developed and validated against the Caterpillar 3126B engine test bed located at the University of Sussex. The controllers were then further tested on the engine test bed. Compared to conventional controllers, the proposed controllers can effectively reduce engine emissions as well as fuel consumption. Experimental results show that compared to the baseline engine running on the Nonroad Transient Cycle (NRTC), mean values of the exhaust gas opacity and the nitrogen oxides (NOx) emission production were reduced by 36.8% and 33%, respectively. Instant specific fuel consumption of the RFLC engine was also reduced by up to 50% compared to the baseline engine during the test. Moreover, the proposed fuzzy logic controllers can also reduce development time and cost by avoiding extensive engine mapping of inlet air pressure and flow. When on-line emission measurements were not available, on-board emission predictors were developed and tested to supply the proposed fuzzy logic controller with predictions of soot and NOx production. Alternatively, adaptive neuro fuzzy inference system (ANFIS) controllers, which can learn from fuzzy logic controllers, were developed and tested. In the end, the proposed fuzzy logic controllers were compared with PI controllers using the transient engine model.

The temporal and spatial distribution of fuel in cylinders is a key factor affecting the combustion characteristics and emission generation of a DI diesel engine. The airfuel mixing quality is critical for controlling ignition timing and combustion duration. Avoiding fuel-rich areas within the cylinder can significantly reduce soot formation as well as high local temperatures resulting in low NOx formation. The present investigation is focused on the effects of advanced fuel injections and air path strategies as well as the effects of piston geometry and fuel spray angle on air-fuel homogeneity, combustion process and their impacts on the performance and emission of the engine. A Ricardo Hydra single-cylinder engine in combination with AVL Fire CFD software was used in this investigation. An experimental analysis was conducted to assess the combustion characteristics and emissions formation of the engine under various injection strategies such as different injection timing, quantity, ratio, dwell angles between injections with various exhaust valve opening times and exhaust back pressures. A quan- titative factor named Homogeneity Factor (HF) was employed in the CFD code in order to quantify the air-fuel mixing and understand how the air-fuel homogeneity within the cylinder can influence the combustion and emissions of the engine. The investigation concludes that multiple injection strategies have the potential to reduce diesel emissions while maintaining meaningful fuel economy. Split injection can be used to improve the air-fuel mixture locally and control temperature generation during the start of combustion. Increased air-fuel homogeneity results in fewer fuel-rich areas within the cylinder and contributes to the reduction of soot emission. Extending the pre-mixed combustion phase has a direct effect on the reduction of soot formation while NOx generation is highly dependent on the scale of the primary fuel injection event.

In this study, the Partially Premixed Compression Ignition (PPCI) combustion strategy in the high-speed, direct-injection diesel engine was investigated numerically by KIVA-3V code coupled with detailed chemistry, aiming to find the solution to meet the increasingly stringent emission regulations. Using split-injection, the parameters including injection timing, split-proportion, spray angle and injection pressures have been studied for their effects on combustion performance and emissions. The effects of swirl ratio, EGR rate and boost pressure are evaluated for improving the mixing and combustion of PPCI. The Homogeneity Factor (HF) was proposed for evaluating the quality of mixing and for quantitatively investigating the effects of injection parameters and in-cylinder air motion on mixture formation. Relationships between the quality of mixing and combustion performance and emissions were discussed using this factor. The results showed that HF had well revealed overall quality of mixture and the effects of operating parameters explicitly. Different EGR compositions with varied fractions of CO2 or H2O were applied in PPCI combustion in order to evaluate the effects of EGR constituents on the combustion performance and emissions. Moreover, the parametric study was conducted under a sweep of the 2nd injection timing and EGR rate, for the understanding of the effects of CO2 and water vapour in EGR at different operating modes. The speed range and load range for the PPCI diesel combustion using split injection was investigated. The results showed that the high level of EGR rate limited the implementation of PPCI combustion at high engine load, while the engine speed was limited by increased CO emissions. The application of high level cooled EGR had the potential for extending operating limits. The proposed Premixed Rate (PR) has revealed the correlations between the degree of premixed combustion and ignition delay, together with overall equivalence ratio. Good responses in fuel consumption have been shown with increase PR. And the significant reduce in PR indicated low degree of premixed at high engine load.

Knowledge of the pressure inside the combustion chamber of a gasoline engine would provide very useful information regarding the quality and consistency of combustion and allow significant improvements in its control, leading to improved efficiency and refinement. While measurement using incylinder pressure transducers is common in laboratory tests, their use in production engines is very limited due to cost and durability constraints. This thesis seeks to exploit the time series prediction capabilities of recurrent neural networks in order to build an inverse model accepting crankshaft kinematics or cylinder block vibrations as inputs for the reconstruction of in-cylinder pressures. Success in this endeavour would provide information to drive a real time combustion control strategy using only sensors already commonly installed on production engines. A reference data set was acquired from a prototype Ford in-line 3 cylinder direct injected, spark ignited gasoline engine of 1.125 litre swept volume. Data acquired concentrated on low speed (1000-2000 rev/min), low load (10-30 Nm brake torque) test conditions. The experimental work undertaken is described in detail, along with the signal processing requirements to treat the data prior to presentation to a neural network. The primary problem then addressed is the reliable, efficient training of a recurrent neural network to result in an inverse model capable of predicting cylinder pressures from data not seen during the training phase, this unseen data includes examples from speed and load ranges other than those in the training case. The specific recurrent network architecture investigated is the non-linear autoregressive with exogenous inputs (NARX) structure. Teacher forced training is investigated using the reference engine data set before a state of the art recurrent training method (Robust Adaptive Gradient Descent – RAGD) is implemented and the influence of the various parameters surrounding input vectors, network structure and training algorithm are investigated. Optimum parameters for data, structure and training algorithm are identified.

Wong, Sheung Man. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2010. / Includes bibliographical references (leaves 63-64). / Abstracts in English and Chinese. / Abstract --- p.i / 摘要 --- p.iii / Acknowledgements --- p.iv / List of figures --- p.V / Chapter Chapter 1 --- Introduction --- p.1 / Chapter 1.1. --- Segway Personal Transporter (PT) --- p.1 / Chapter 1.2. --- Existing research using Segway Robotic Mobility Platform´ёØ (RMP) --- p.3 / Chapter 1.3. --- The ICSL Segway Rider --- p.9 / Chapter 1.4. --- Thesis outlines --- p.10 / Chapter Chapter 2 --- ICSL Segway Rider --- p.11 / Chapter 2.1. --- Design concept --- p.11 / Chapter 2.2. --- Design overview --- p.12 / Chapter 2.3. --- Actuating components --- p.14 / Chapter 2.4. --- Electronic and sensing components --- p.24 / Chapter 2.5. --- Software development of Segway Rider --- p.28 / Chapter 2.6. --- Chapter summary --- p.31 / Chapter Chapter 3 --- The grand challenge --- p.32 / Chapter 3.1. --- Objective --- p.32 / Chapter 3.2. --- Experiment --- p.33 / Chapter 3.3. --- Running lane tracking by computer vision --- p.34 / Chapter 3.3.1. --- Color space conversion --- p.36 / Chapter 3.3.2. --- Apply binary threshold --- p.37 / Chapter 3.3.3. --- Edge detection --- p.41 / Chapter 3.3.4. --- Hough transform --- p.46 / Chapter 3.3.5. --- Line analysis --- p.49 / Chapter 3.4. --- Chapter summary --- p.51 / Chapter Chapter 4 --- Stand and stay --- p.52 / Chapter 4.1. --- Introduction --- p.52 / Chapter 4.2. --- Box matching method --- p.53 / Chapter 4.3. --- Image processing steps --- p.55 / Chapter 4.4. --- Experiment --- p.58 / Chapter 4.5. --- Chapter summary --- p.60 / Chapter Chapter 5 --- Conclusion and future works --- p.61 / Chapter 5.1. --- Contributions --- p.61 / Chapter 5.2. --- Future works --- p.62 / Bibliography --- p.63

Digitized by Kansas Correctional Industries