Date of Award


Document Type

Open Access Master's Thesis

Degree Name

Master of Science in Mechanical Engineering (MS)

Administrative Home Department

Department of Mechanical Engineering-Engineering Mechanics

Advisor 1

Mahdi Shahbakhti

Advisor 2

Jeffrey Naber

Committee Member 1

Hoseinali Borhan


Low Temperature Combustion (LTC) regimes have gained attention in internal combustion engines since they deliver low nitrogen oxides (NOx) and soot emissions with higher thermal efficiency and better combustion efficiency, compared to conventional combustion regimes. However, the operating region of these high-efficiency combustion regimes is limited as it is prone to knocking and high in-cylinder pressure rise rate outside the engine safe zone. By allowing multi-regime operation, high-efficiency region of the engine is extended. To control these complex engines, understanding and identification of heat release rate shapes is essential. Experimental data collected from a 2 liter 4 cylinder LTC engine with in-cylinder pressure measurements, is used in this study to calculate Heat Release Rate (HRR). Fractions of early and late heat release are calculated from HRR as a ratio of cumulative heat release in the early or late window to the total energy of the fuel injected into the cylinder. Three specific HRR patterns and two transition zones are identified. A rule based algorithm is developed to classify these patterns as a function of fraction of early and late heat release percentages. Combustion parameters evaluated also showed evidence on characteristics of classification. Supervised and unsupervised machine learning approaches are also evaluated to classify the HRR shapes. Supervised learning method ( Decision Tree)is studied to develop an automatic classifier based on the control inputs to the engine. In addition, supervised learning method (Convolutional Neural Network (CNN)) and unsupervised learning method (k-means clustering) are studied to develop an automatic classifier based on HRR trace obtained from the engine. The unsupervised learning approach wasn't successful in classification as the arrived k-means centroids didn't clearly represent a particular combustion regime. Supervised learning techniques, CNN method is found with a classifier accuracy of 70% for identifying heat release shapes and Decision Tree with the accuracy of 74.5% as a function of control inputs.

On rule based classified traces with the use of principle component analysis (PCA) and linear regression, heat release rate classifiers are built as a function of engine input parameters including, Engine speed, Start of injection (SOI), Fuel quantity (FQ) and Premixed ratio (PR). The results are then used to build a linear parameter varying (LPV) model as a function of the modelled combustion classifiers by using the least square support vector machine (LS-SVM) approach. LPV model could predict CA50(Combustion phasing), IMEP (indicated mean effective pressure) and MPRR (maximum pressure rise rate) with a RMSE of 0.4 CAD, 16.6 kPa and 0.4 bar/CAD respectively. The designed LPV model is then incorporated in a model predictive control (MPC) platform to adjust CA50, IMEP and MPRR. The results show the designed LTC engine controller could track CA50 and IMEP with average error of 1.2 CAD and 6.2 kPa while limiting MPRR to 6 bar/CAD. The controller uses three engine inputs including, SOI, PR and FQ as manipulated variables, that are optimally changed to control the LTC engine.