Virtual Multi-Cylinder Engine Experimental and simulation data "in loop" towards a "virtual multi-cylinder engine" for pre-mixed hydrogen combustion Master’s thesis in Mechanics and Maritime Science Mohammad Rusail Karattuthodi Theodore Pierce Repinz DEPARTMENT OF MECHANICS AND MARITIME SCIENCE CHALMERS UNIVERSITY OF TECHNOLOGY Gothenburg, Sweden 2026 www.chalmers.se www.chalmers.se Master’s thesis 2026 Virtual Multi-Cylinder Experimental and simulation data "in loop" towards a "virtual multi-cylinder engine" for pre-mixed hydrogen combustion Mohammad Rusail Karattuthodi Theodore Pierce Repinz Chalmers University of Technology Department of Mechanics and Maritime Sciences Volvo Group Department of Powertrain Engineering Gothenburg, Sweden 2026 Virtual Multi-Cylinder Engine Mohammad Rusail Karattuthodi Theodore Pierce Repinz © Mohammad Rusail Karattuthodi, Theodore Pierce Repinz 2026. Supervisor: Alaleh Safari, Volvo, Powertrain Strategic Development Supervisor: Djordje Purkovic, Volvo, Powertrain Strategic Development Supervisor: Elvin Zanjani, Chalmers University, Mechanics and Maritime Sciences Supervisor: Erik Svensson, Volvo, Engine Performance and Simulation Supervisor: Jan Eismark, Volvo, Powertrain Strategic Development Supervisor: Jian Zhu, Volvo, Combustion System and Simulation Examiner: Petter Dehlander, Department of Mechanics and Maritime Sciences Master’s Thesis 2026 Department of Mechanics and Maritime Sciences Chalmers University of Technology SE-412 96 Gothenburg Telephone +46 31 772 1000 Cover: Virtual 0D/1D multi-cylinder engine "loop". Typeset in LATEX, template by Kyriaki Antoniadou-Plytaria Printed by Chalmers Reproservice Gothenburg, Sweden 2026 iv Abstract This project introduces an integrated methodology and tool set that couples exper- imental data from a single-cylinder engine (SCE) with a 0D/1D Three-Pressure- Analysis (TPA) model and a multi-cylinder engine (MCE) simulation in GT-Suite. The system developed utilizes measured data from a SCE and uses this to com- pute an apparent burn rate in a TPA model. This apparent burn rate, along with other operational data is extracted from the SCE and imposed onto a 0D/1D MCE simulation environment where the feedback is a set of boundary conditions corresponding to the inlet and exhaust manifold of the MCE. These boundary con- ditions are then used to adjust actuators on the SCE - enabling the continuation of a closed loop system until boundary condition convergence criteria have been met. This concept addressed the two main shortcomings of SCE testing and 0D/1D simulation, respectively. In SCE testing, the entire engine system is decoupled, and each subsystem can be controlled independently for a specific load point. This implies that only steady state operation is possible, while the data that is collected is high fidelity measurement directly from the engine. A 0D/1D MCE simulation can model the system-level engine response, however in the context of premixed hydrogen engines, the combustion models are not mature enough to accurately represent in-cylinder combustion events. The integration of a TPA-derived burn rate into the 0D/1D MCE model allowed the gap to be bridged between simplified simulation and experimental reality. A symbiotic looped workflow was established, and this paper shows that there exists potential to optimize the design quality and time of high-performance MCE con- cepts for future powertrain development. A data sampling and averaging study was conducted to understand how the out- put of the simulation components reacted to cyclic variations. The study showed that key performance parameters are not significantly affected by the sample size when compared to a 200-cycle average; however are more sensitive to the sampling window - suggesting a sensitivity to cyclic variations. A minimum of 50 averaged cycles is required for accurate system results. Furthermore, the output of an initial loop iteration of the ’virtual MCE’ simu- lation tool is compared to a similar experimental MCE as a means to understand the system’s initial conditions and validate the concept. This investigation was in- conclusive, and hardware discrepancies were found to be evident in the test setups. v The continuation of loop iterations was halted by the lack of access to a dedicated SCE test bed. The initial iteration found absolute inlet manifold boundary condi- tion errors of 7.14% for pressure, and 82.4% for temperature. While the absolute exhaust manifold back pressure error was found to be 3.98%. The overarching research question was found to be inconclusive. However, this report sets the foundation for future work and discusses how further research can be conducted to aid in realizing its successful outcome. This thesis offers a set of recommendations for continued progression in realizing this as an industry tool. The thesis was therefore written in a manner to aid in future work and can be used as reference for configuring simulation models, addressing errors in future studies and is useful reference for relevant theory. Keywords: Internal Combustion Engines, Hydrogen Combustion, Three Pressure Analysis, 0D/1D Simulation. vi Acknowledgments Our deepest thanks go to our families and friends, whose constant support and sacrifices have allowed us to be a part of this work at Chalmers University of Tech- nology and Volvo Group. We are eternally grateful for the foundation that made this journey possible. To our supervisors, thank you for sharing your expertise and for your tireless guidance as we navigated the challenges of powertrain engineering and develop- ment. We are honored by the time and patience you invested in our growth, and we carry the lessons learned here with immense gratitude. viii x xi List of Acronyms Below is the list of acronyms that have been used throughout this thesis listed in alphabetical order: AFR Air Fuel Ratio BMEP Brake Mean Effective Pressure CA50 50% Mass Fraction Burned CO2 Carbon Dioxide CO Carbon Monoxide CoV Coefficient of Variance CoVIMEP Coefficient of Variation of Indicated Mean Effective Pressure CFD Computational Fluid Dynamics DI Direct Injection DSL Data Science Lab EGR Exhaust Gas Recirculation EOE End of Energizing FRM Fast Running Models GT Gamma Technologies H2ICE Hydrogen Internal Combustion Engines HC Hydrocarbon HCCI Homogeneous Charge Compression Ignition ICE Internal Combustion Engine IVC Intake Valve Closing IMEP Indicated Mean Effective Pressure ISO International Organization for Standardization KIT Karlsruhe Institute of Technology LPDI Low Pressure Direct Injection MCE Multi Cylinder Engine NOx Nitrous Oxides OEM Original Equipment Manufacturers PFI Port Fuel Injection PID Proportional Integral Derivative (controller) SCE Single Cylinder Engine SI Spark Ignition SOE Start of Energizing TPA Three Pressure Analysis VGT Variable Geometry Turbo xii xiv Nomenclature Below is the nomenclature of indices, sets, parameters, and variables that have been used throughout this thesis. Parameters Pin Inlet boundary pressure Pexh Exhaust boundary pressure Tin Inlet boundary temperature Variables mu Unburned zone mass mf,b Fuel mass transferred to burned zone ma,b Air mass transferred to burned zone mf,i Injected fuel mass eu Unburned zone energy p Cylinder pressure Vu Unburned zone volume hf Enthalpy of fuel mass ha Enthalpy of air mass hf,i Enthalpy of injected fuel mass Qu Unburned zone heat transfer rate xv mb Burned zone mass mf,b Fuel mass transferred to burned zone ma,b Air mass transferred to burned zone eb Burned zone energy p Cylinder pressure vb Burned zone volume hf Bnthalpy of fuel mass ha Bnthalpy of air mass Qb burned zone heat transfer rate xvi xviii Contents List of Acronyms x Nomenclature xiv List of Figures xxiii List of Tables xxvii 1 Introduction 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2 Theory 7 2.1 Pre-mixed Hydrogen Combustion . . . . . . . . . . . . . . . . . . . 7 2.2 Cyclic Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.3 Single Cylinder Engine Testing . . . . . . . . . . . . . . . . . . . . 9 2.3.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3.2 Data Typically Collected . . . . . . . . . . . . . . . . . . . . 10 2.3.3 Testing methodologies Overview . . . . . . . . . . . . . . . . 10 3 0D/1D Simulation and GT-Suite Setup 13 3.1 Overview with a Hydrogen Combustion Perspective . . . . . . . . . 13 3.2 Two-Zone Combustion Methodology . . . . . . . . . . . . . . . . . . 14 3.2.1 Core Mechanism . . . . . . . . . . . . . . . . . . . . . . . . 14 3.2.2 Governing Equations . . . . . . . . . . . . . . . . . . . . . . 14 3.3 Three-Pressure-Analysis Model . . . . . . . . . . . . . . . . . . . . 15 3.3.1 Reverse Run Calculation . . . . . . . . . . . . . . . . . . . . 16 3.3.2 Required Data Input . . . . . . . . . . . . . . . . . . . . . . 16 xix Contents 3.3.3 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 3.3.4 Simulation Methodology . . . . . . . . . . . . . . . . . . . . 18 3.3.5 Consistency Checks . . . . . . . . . . . . . . . . . . . . . . . 19 3.3.6 Result Evaluation and Export . . . . . . . . . . . . . . . . . 20 3.4 MCE Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 4 Methodology and Experimental Setup 23 4.1 Virtual MCE Data Exchange and Loop Development . . . . . . . . 23 4.1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4.1.2 Multi-Cylinder Engine . . . . . . . . . . . . . . . . . . . . . 25 4.1.3 KIT Single Cylinder Engine . . . . . . . . . . . . . . . . . . 27 4.1.4 KIT SCE and MCE comparison . . . . . . . . . . . . . . . . 27 4.1.5 Three-Pressure Analysis Model . . . . . . . . . . . . . . . . 29 4.1.5.1 Gas Exchange . . . . . . . . . . . . . . . . . . . . . 29 4.1.5.2 Combustion System . . . . . . . . . . . . . . . . . 30 4.1.5.3 Injectors . . . . . . . . . . . . . . . . . . . . . . . . 30 4.1.6 Multi-Cylinder Engine Simulation . . . . . . . . . . . . . . . 30 4.1.7 Data Exchange . . . . . . . . . . . . . . . . . . . . . . . . . 32 4.1.7.1 Simulation Setup . . . . . . . . . . . . . . . . . . . 32 4.1.7.2 Burn Rate Profile . . . . . . . . . . . . . . . . . . . 33 4.1.7.3 Boundary Conditions . . . . . . . . . . . . . . . . . 33 4.1.8 Loop Convergence . . . . . . . . . . . . . . . . . . . . . . . 34 4.1.9 Multi-Cylinder Variation and Reference Selection . . . . . . 34 4.2 Research Question 1.1: How sensitive is the loop to Cyclic Variations 36 4.3 Research Question 1.2: How well does the simulated MCE compare to experimental MCE data . . . . . . . . . . . . . . . . . . . . . . . 37 5 Results 39 5.1 Loop Output and Performance . . . . . . . . . . . . . . . . . . . . . 39 5.1.1 Inlet Manifold Convergence . . . . . . . . . . . . . . . . . . 39 5.1.2 Exhaust Manifold Convergence . . . . . . . . . . . . . . . . 41 5.1.3 Operational Efficiency . . . . . . . . . . . . . . . . . . . . . 43 5.2 Research Question 1.1: How sensitive is The Loop to Cyclic Variations 44 5.3 Research Question 1.2: How well does the simulation MCE compare to experimental MCE data . . . . . . . . . . . . . . . . . . . . . . . 50 6 Conclusion 55 7 Discussion 57 7.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 7.1.1 Engine Access . . . . . . . . . . . . . . . . . . . . . . . . . . 57 xx Contents 7.1.2 Data Availability . . . . . . . . . . . . . . . . . . . . . . . . 58 7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Bibliography 63 References 63 A Appendix 1: Supplementary Plots for Research Question 1.1 I B Appendix 2: Effects of the Connecting Rod Length IX C Appendix 3: MCE knock filtering and error checks XI xxi Contents xxii List of Figures 3.1 Flow chart of TPA simulation methodology . . . . . . . . . . . . . . 19 4.1 Virtual MCE data exchange loop required for the realization of the "virtual MCE" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 4.2 Render of the HYCET engine . . . . . . . . . . . . . . . . . . . . . 25 4.3 Render of the KIT single cylinder research engine. . . . . . . . . . . 27 4.4 Comparison of the injector and spark plug angles between the KIT test bench and the KST/HYCET test bench. . . . . . . . . . . . . . 28 4.5 Three Pressure Analysis model and corresponding data sources . . . 29 4.6 0D/1D GT-Suite model of the HYCET multi-cylinder engine . . . . 31 4.7 Excel file used to setup the simulation environments of the TPA and MCE models. The use of parameters are color coordinated for the simulation model. . . . . . . . . . . . . . . . . . . . . . . . . . . 33 4.8 Data exchange between the SCE and virtual MCE which governs steady state closed loop convergence . . . . . . . . . . . . . . . . . 34 4.9 Average indicated work done by each cylinder . . . . . . . . . . . . 35 5.1 Inlet manifold pressure . . . . . . . . . . . . . . . . . . . . . . . . . 40 5.2 Inlet manifold temperature . . . . . . . . . . . . . . . . . . . . . . . 40 5.3 Exhaust manifold pressure . . . . . . . . . . . . . . . . . . . . . . . 42 5.4 Exhaust manifold temperature . . . . . . . . . . . . . . . . . . . . . 42 5.5 Processing time evaluation. . . . . . . . . . . . . . . . . . . . . . . 43 5.6 IMEP output from the loop . . . . . . . . . . . . . . . . . . . . . . 44 5.7 IMEP Relative Comparison with the Standard Input Pressure Trace 44 5.8 Indicated Efficiency output from the loop . . . . . . . . . . . . . . 45 5.9 Indicated Efficiency Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 5.10 Intake Pressure output from the loop . . . . . . . . . . . . . . . . . 45 5.11 Exhaust Pressure output from the loop . . . . . . . . . . . . . . . 46 xxiii List of Figures 5.12 Intake Pressure Relative Comparison with the Standard Input Pres- sure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 5.13 Exhaust Pressure Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 5.14 Intake Temperature Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 5.15 Exhaust Temperature Relative Comparison with the Standard In- put Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 5.16 Cumulative Energy Loss output from the loop . . . . . . . . . . . . 47 5.17 Cumulative Energy Loss Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . 48 5.18 CA50 output from the loop . . . . . . . . . . . . . . . . . . . . . . 48 5.19 CA50 Relative Comparison with the Standard Input Pressure Trace 48 5.20 Burn Duration 0-10% Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 5.21 Log-P log-V diagram of the SCE, simulated TPA and MCE models and the experimental MCE data . . . . . . . . . . . . . . . . . . . . 51 5.22 LogP-logV curve for the simulated and the experimental MCE systems 52 5.23 Pressure traces and inlet manifold pressures compared for the vir- tual and experimental MCE. . . . . . . . . . . . . . . . . . . . . . . 53 5.24 Volumetric Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . 54 5.25 Indicated Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 7.1 SCE and MCE datasets that showed the best agreement used for comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 A.1 IMEP Relative Comparison with the Standard Input Pressure Trace I A.2 IMEP Relative Comparison with the Standard Input Pressure Trace I A.3 Indicated Efficiency Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II A.4 Indicated Efficiency Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II A.5 Intake Pressure Relative Comparison with the Standard Input Pres- sure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II A.6 Intake Pressure Relative Comparison with the Standard Input Pres- sure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III A.7 Intake Temperature Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III A.8 Intake Temperature Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III xxiv List of Figures A.9 Energy Loss Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV A.10 Energy Loss Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV A.11 Exhaust Temperature Relative Comparison with the Standard In- put Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . IV A.12 Exhaust Temperature Relative Comparison with the Standard In- put Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . IV A.13 Exhaust Pressure Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V A.14 Exhaust Pressure Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V A.15 CA50 output from the loop . . . . . . . . . . . . . . . . . . . . . . V A.16 CA50 output from the loop . . . . . . . . . . . . . . . . . . . . . . VI A.17 CA50 Relative Comparison with the Standard Input Pressure Trace VI A.18 CA50 Relative Comparison with the Standard Input Pressure Trace VI A.19 Burn Duration 0-10% Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII A.20 Burn Duration 0-10% Relative Comparison with the Standard Input Pressure Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII B.1 Comparison of the conrod length of the MCE (259 mm) to the conrod length of the KIT SCE (277 mm) on the combustion cycle pressure curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IX B.2 Comparison of the conrod length of the MCE (259 mm) to the conrod length of the KIT SCE (277 mm) on a LogP-LogV curve. . . X C.1 Maximum knock amplitude with a 1.5 bar knock filter applied to the experimental MCE data. . . . . . . . . . . . . . . . . . . . . . . XI C.2 Knocking pressure traces filtered from each cylinder which were out- side of the knock threshold . . . . . . . . . . . . . . . . . . . . . . . XII C.3 Average in-cylinder pressure of each cylinder plotted on a LogP- LogV diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XIII xxv List of Figures xxvi List of Tables 4.1 HYCET/KST MCE geometric properties . . . . . . . . . . . . . . . 26 4.2 KIT SCE geometric properties . . . . . . . . . . . . . . . . . . . . . 27 4.3 MCE geometric properties . . . . . . . . . . . . . . . . . . . . . . . 32 4.4 Convergence tolerance of boundary condition parameters on the MCE to the SCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4.5 IFile sample data operating points . . . . . . . . . . . . . . . . . . . 36 4.6 Combined list of MCE combustion, performance, and gas-exchange parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 xxvii List of Tables xxviii 1 Introduction 1.1 Background Internal Combustion Engines (ICE) play a crucial role in the transportation in- dustry. The sector remains a major contributor to greenhouse gas emissions, with most of the fuel currently in use being fossil fuels. In response, the European Union has committed to achieving carbon neutrality by 2050, which demands a significant reduction in emissions from fossil fuel combustion. Using alternative fuels with low or no carbon content is one solution to reduce the CO2 emissions from ICEs. Hydrogen is one such promising alternative fuel that has zero carbon content. During hydrogen combustion, the primary emissions are NOx along with water in liquid and vapor forms [13]. Some key attributes of hydrogen fuel include a low ignition energy, a high diffusion coefficient, a high laminar flame speed, and a wide flammability range. The wide flammability range makes it suitable for lean operation, potentially resulting in increased engine efficiency due to the reduction in pumping losses and wall heat transfer [4]. However, due to the low ignition energy of hydrogen, abnormal combustion events like backfire and knock can be- come more prominent, potentially inhibiting the engine from reaching high power outputs. These abnormal combustion events are particularly hard to manage in pre-mixed hydrogen combustion systems, where the fuel is injected early on in the compression stroke at a low pressure and is ignited with the use of a spark plug. The NOx emissions can be reduced by adopting measures such as lean-combustion and exhaust gas recirculation [13]. Hydrogen combustion engines require care- fully balanced thermodynamic parameters such as the injection technique, air-fuel mixture, valve timing, and ignition timing to realize stable and efficient engine operation [4]. Despite these challenges, hydrogen combustion offers cost benefits compared to other carbon-neutral means of propulsion - by using the already widespread pro- cess chain of conventional combustion engines [13], as research on hydrogen in- 1 1. Introduction ternal combustion engines is significantly shaped by traditional internal combus- tion engines given that the combustion model is essential for both. Retrofitting conventional four-stroke diesel and petrol engines for hydrogen fuel requires mini- mal modifications to manufacturing processes when compared to completely new power-train technologies. This adaptability allows OEMs to re-purpose existing engine designs while meeting emissions regulations through relatively simple mod- ifications. When designing an engine of this kind, engineers utilize and iterate through tools such as 3D Computational Fluid Dynamics (CFD), 0D/1D modelling, and single- cylinder engine (SCE) testing to develop the appropriate combustion systems. This thesis focuses on the use of 0D/1D simulation tools and SCE testing and how they can be used to realize a full-scale multi-cylinder engine (MCE) concept. 0D/1D simulation methods are a valuable early-stage combustion development tool and are used to provide fast and efficient modeling of complex systems by simplifying their physical representation. In engine development, 0D/1D simula- tions are used to analyze overall system behavior—such as pressure, temperature, and flow rates—across components like cylinders, manifolds, and pipes, without resolving detailed spatial variations. This method is valuable for system-level anal- ysis and early-stage design. It enables engineers to quickly evaluate performance, optimize control strategies, guide designs, and allows for more informed 3D simu- lations of combustion systems and flow components. However, standalone models require the fuel burn to be imposed onto the model - which is a challenging task for pre-mixed hydrogen combustion. The single-cylinder engine is used as a tool to physically prototype and test engine systems. It includes a decoupled testbed configuration that can help to accelerate our understanding and optimize a combustion system before moving to a full-scale engine. A proposed combustion system, once designed with CFD and 0D/1D sim- ulation methods, can be validated in an SCE test bench. Isolating one cylinder enables conducting detailed studies on fuel injection, ignition timing, combustion efficiency, and new technologies without the complexity and cost of a full multi- cylinder engine. This thesis work relies on the integration of experimental data obtained from a pre-mixed hydrogen SCE into a Hardware-in-Loop (HIL) test environment. A particular focus is placed on the acquisition of in-cylinder pressure traces, which are analyzed using a special simulation tool called a three-pressure analysis (TPA) model. This model computes the apparent burn rate of the fuel combusted in the 2 1. Introduction SCE. These burn rates are imposed on a 0D/1D MCE simulation model, enabling a more realistic representation of the combustion behavior of a pre-mixed hydro- gen MCE at system level. The output of the MCE provides a set of boundary conditions that can be re- lated to the physical SCE and used to adjust actuator settings of the test bench at the inlet and exhaust manifolds - achieving steady state convergence between the physical SCE and the MCE model in these boundaries is of interest, allowing us to represent the combustion of an MCE system within a controlled SCE environment. In addition to building up an experimental modeling framework, the thesis will investigate methods for efficient data exchange between the SCE test bench and simulation platforms. It will assess the sensitivity of the system to cycle-to-cycle variations and updates to the boundary conditions which are a challenge in hydro- gen combustion. A core ambition of this project is to demonstrate how a data-driven, near real-time modeling approach can significantly reduce development time and cost in the field of hydrogen engine research. 1.2 Aim This project work evaluates whether data from single-cylinder experimental testing can enhance the accuracy of 0D/1D multi-cylinder concept models. By bridging the gap between physical testing and virtual simulation, the study seeks to ac- celerate development timelines and improve the design quality of next-generation engine concepts. 3 1. Introduction 1.3 Research Questions The central focus of this study is to investigate whether calculating the burn rate from the pressure trace of a physical SCE provides a significant advantage for engine concept development. This section breaksdown the research questions in- vestigated. Specifically, the research explores whether using experimental results to calculate boundary conditions and burn rates for a simulated MCE yields better results than relying exclusively on a stand-alone MCE simulation with a predictive combustion model imposed on the system. • In the context of premixed hydrogen combustion, does calculating the burn rate from the pressure trace of a physical SCE, then using it to calculate boundary conditions for a simulated MCE model, offer an advantage for engine concept development compared to relying on a standalone MCE simulation The overarching research question is a complex and demanding research task. Resource and time limitations during the course of the project hindered the full solution to the proposed problem. Instead, the foundation of this project was laid, which aims to aid in the continuation of its work. This was done by breaking down the research question above into research questions 1.1 and 1.2, which are then further evaluated in the results section of this paper. 1.1. Sensitivity to Cyclic Variations: • How sensitive is the burn rate and boundary condition in the simu- lation due to cyclic variations of premixed hydrogen spark-ignited data input? This analysis examines how sensitive the burn rate and simulated boundary condi- tions are to the cyclic variations inherent in experimental input data from premixed hydrogen spark-ignited tests. The objectives of the study were to assess the sensi- tivity of the simulation’s boundary conditions to the input data and to determine an optimal input data size. The end goal of the study was to determine the min- imum input data size while maintaining sufficient resolution and accuracy in the simulation output. 1.2. Correlation and Accuracy: • How accurately does a system-level 1D MCE model, simulated using burn rates from experimental SCE tests and three pressure 4 1. Introduction analysis, correlate with experimental MCE combustion data? This objective assesses the accuracy of a system-level 0D/1D MCE model. It eval- uates how closely the simulation (driven by burn rates calculated from a TPA model and experimental SCE test data) correlates with actual experimental MCE combustion data from an engine with similar geometry and operating conditions. 5 1. Introduction 6 2 Theory 2.1 Pre-mixed Hydrogen Combustion Hydrogen can serve as a primary energy carrier and can be produced from re- newable sources like wind, solar, and biomass. Unlike carbon-based fossil fuels, which are a major contributor to CO2 emissions, burning pure hydrogen ideally produces only water as a byproduct [10]. Hydrogen combustion offers a lower cost of transition because hydrogen internal combustion engines (H2ICEs) can utilize existing infrastructure developed for gasoline and diesel based engines [14]. Combustion processes are classified into diffusion and pre-mix combustion, based on whether the fuel and oxidizer are mixed before or during the combustion reac- tion. Diffusion combustion occurs when the reactants are not mixed beforehand and must mix in the same region where the combustion reaction takes place. In premixed combustion, the fuel and oxidizer are well mixed before spark onset [6].For internal combustion engines, this is usually done with Port Fuel Injection (PFI) or Direct Injection (DI). Premixed hydrogen combustion offers a viable pathway for high-efficiency, carbon- free energy generation. Hydrogen’s unique physical properties set its premixed combustion apart from hydrocarbons. It has a wide flammability range in air, from 4% to 75% by volume, which enables ultra-lean combustion, where engines can use much more air than needed for a stoichiometric mixture [14]. This allows for unthrottled operations at low to medium loads, eliminating pumping losses associated with the pressure drop across the throttle plates [18]. Ultra-lean com- bustion can also reduce peak temperatures and significantly lower nitrogen oxide (NOx) emissions [10]. Hydrogen has a high laminar flame speed, about 1.85 m/s in a stoichiometric mixture, which is enabled through the fuel’s quick reaction rate and large diffusion coefficient [10]. This enables rapid heat release and high thermal efficiency. When compared to hydrocarbon-air mixtures, hydrogen com- bustion has a significantly lower minimum ignition energy [14]. Hydrogen needs less than one-tenth the energy required to ignite petrol at the same air-fuel ra- 7 2. Theory tio (AFR). Because of the short ignition delay and high autoignition temperature (8̃58K), pre-mixed hydrogen combustion is mainly used in spark ignition (SI) en- gines. Due to these characteristics, hydrogen combustion is highly sensitive to operational parameters. The unique physicochemical properties of hydrogen ne- cessitate advanced fuel handling systems and precise electronic control strategies. While it enables high power density, insufficient control can lead to combustion instabilities, which pose significant risks to hardware integrity, efficiency, and op- erational safety[9] [17] [19]. 2.2 Cyclic Variations Cyclic variations are a phenomenon in SI engines when the combustion process varies from cycle to cycle. The main causes of cyclic variations are in the early combustion phase, especially in the formation of the flame kernel after the spark discharge. These variations are caused by: fluctuation in gas movement at igni- tion, variations in the air-fuel ratio, non-homogeneous air-fuel mixing and internal EGR[6]. These changes are observed as fluctuations in the cylinder pressure, which can lead to incomplete combustion and, in the worst case, misfire. The consequence of this is reduced fuel efficiency and engine performance. In hydrogen engines, it is especially important to control these variations[1]. Depending on the air-fuel ratio, the combustion stability of premixed hydrogen is relatively high, with the Coeffi- cient of Variation (CoV) of Indicated Mean Effective Pressure (IMEP) (CoVIMEP) remaining less than 1% at medium and high loads in certain cases. However, as the air-fuel mixture becomes more lean, it becomes more prone to cyclic varia- tions and misfires, limiting the practical extent of lean operation [4]. Abnormal combustion events are still one of the toughest problems for premixed hydrogen engines. Studies show that reducing cycle variability can increase power output by up to 10% without increasing fuel consumption [3]. Cyclic variations have a direct impact on how a vehicle drives and feels, especially when the engine runs in lean-burn conditions [14]. 8 2. Theory 2.3 Single Cylinder Engine Testing 2.3.1 Purpose The SCE test bed is a build configuration that typically includes an isolated com- bustion system from a MCE. It is widely used in the development phase of an engine or combustion system. It offers a number of benefits which range from direct validation of a combustion system component to fuel testing and research. Practically, the benefits of using a SCE test bed include: • Reduced Complexity: The SCE build simplifies the testing environment. Instrumentation is focused on a stand alone combustion environment, reduc- ing the density of sensors and supporting hardware otherwise required on an MCE testbed. Furthermore, the control logic is reduced; whereas an MCE requires complex coordination of multiple injectors and thermal management systems, an SCE scales these requirements down to a single unit. • Accelerated Development Cycle: Managing design iterations is more efficient with an SCE. The simplified assembly allows engineers to iterate between designs quickly, reducing the downtime associated with engine re- builds in MCE configurations. • Economic Advantage: SCE testing is significantly more cost effective due to a lower component count. In MCE testing, a single design change may necessitate a complete engine overhaul. The modular nature of an SCE allows for more frequent and economical hardware iterations. Considering these practicalities, SCE testing has an important place in powertrain Research and Development (R&D). The need for testing in an SCE is driven by several factors listed as follows [6] [11]. • Design Evaluation and Validation: Allowing for the evaluation of a prototype combustion system. This allows engineers to fine tune the oper- ation settings which define its use case and ultimately strives to validate a combustion systems design. • Safety and Product Liability: Engineers are able to expose the combus- tion system to abnormal combustion, designing to prevent the phenomena and mitigate the inherent negative consequences. • Measuring Improvements: Allows for a means to track improvements and evolve a prototype combustion system. • Statistical Data Collection: Statistical data is collected for fine tuning, reliability analysis, and performance rating. The goal of the engineering team is to disassemble each of these components into their foundational development criteria. In context of the above factors the engi- neering team uses the SCE to collect data for optimizing performance, efficiency, 9 2. Theory emissions, structural integrity, individual component selection/design, and relia- bility and durability evaluation 2.3.2 Data Typically Collected In this project the SCE is a critical component in ensuring the successful devel- opment of the "Virtual MCE" design tool. The data acquisition system (DAQ) embedded into the SCE system is used to recorded crank angle resolved and time averaged data. In general crank angle resolved data includes values sampled from the following locations: • Raw pressure signal from within the combustion chamber. • Inlet manifold pressure • Exhaust manifold pressure • Spark plug ignition current • Injector rail pressure and injector needle valve control current • Calculated parameters calculated in real time from the collected data: – Apparent heat release (Also includes a normalized value) – Apparent heat release rate – Knocking parameter calculation Time-averaged parameters are monitored to characterize the engine’s performance. These measurements typically, among other parameters, encompass Exhaust After- Treatment System (EATS) chemical species, regional temperature sampling, tur- bocharger settings and feedback, and EGR ratios. Collecting data at this temporal resolution allows for a macro-scale evaluation of the system’s thermodynamic, me- chanical and chemical equilibrium. 2.3.3 Testing methodologies Overview Experimental evaluation using a SCE is a targeted process primarily focused on the characterization and optimization of a combustion system’s operational pa- rameters. For a low pressure direct injection (LPDI) system (such as that in this research), the SCE provides a controlled environment to establish the performance across a predefined range of engine speeds and loads. To map the engine’s performance, a Design of Experiments (DoE) approach is employed. This methodology is executed in phases aimed at decoupling complex variables and identifying the sensitivities of the combustion. Testing is performed at discrete steady-state setpoints, defined by engine speed and IMEP according to ISO standard 15550:2016 [6][8]. 10 2. Theory The experimental process generally consists of parametric sweeps where individ- ual variables are iterated while maintaining constant boundary conditions. Key control parameters include: • Energizing (Fuel Injection): Start of Energizing (SOE), End of Energiz- ing (EOE), and rail pressure. • Ignition and Valve Timing: Spark timing based on a desired CA50, which is often set at 8 CAD after top dead center. • Gas Exchange: EGR rates and AFR - provides data for the developmen- t/selection of effective boosting systems and their control. Following parametric sweeps, a comprehensive optimization of the engine’s oper- ation is conducted. This phase aims to synthesize the data collected from the parametric sweep into a coherent calibration strategy. Some primary outputs of the optimization include the following parameters. • Efficiency Metrics: Indicated Brake Thermal Efficiency and Brake Specific Fuel Consumption (BSFC). • Emission Profiles: Specifically NOx concentrations. • Combustion Stability: Usually quantified by CoVIMEP and knock param- eters. 11 2. Theory 12 3 0D/1D Simulation and GT-Suite Setup 3.1 Overview with a Hydrogen Combustion Per- spective Modeling and simulation play an increasingly important role in developing pre- mixed hydrogen combustion engines. Researchers use advanced 3D-CFD to ana- lyze mixture formation and flame propagation, which supports improvements in combustion chamber design. Because detailed chemistry in 3D-CFD simulations is resource intensive, initial studies are often conducted in a 0D/1D environment using tools such as GT-Suite [4]. 0D/1D engine simulations are physics based tools that model how fluids move, heat is transferred, and chemical reactions occur in an ICE. These tools use basic conservation equations for mass, momentum, and energy. The engine’s flow sys- tems are represented as a network of 1D pipes and ducts, while the cylinders are treated as 0D spaces where pressure and temperature can change over time but remain even at any moment [5]. This method allows engineers study key engine features like power, torque, volumetric efficiency, and fuel use without spending much time or money on physical prototypes [12]. For hydrogen engines, 0D/1D modeling is especially helpful for early research on high laminar flame speed and wide flammability range, which require careful system-level control. Modeling hydrogen combustion in a 0D/1D environment is challenging because these approaches assume the cylinder contents are evenly mixed, which ignores the spatial differences and shapes that lead to cyclic variations. This is important because hydrogen combustion reacts strongly to hot spots and how well the gases mix. Since these simple models cannot capture the real physics behind these vari- ations, 0D/1D modeling needs a lot of experimental data and careful calibration to make up for this [15]. 13 3. 0D/1D Simulation and GT-Suite Setup 3.2 Two-Zone Combustion Methodology GT-Power uses a two-zone combustion methodology to simulate the combustion process for nearly all engine models except for Homogeneous Charge Compression Ignition (HCCI) models and certain predictive diesel models. This is the standard approach for predicting in-cylinder pressure and temperature, as it accounts for distinct thermodynamic states of the gases in zones. This is the method used during the course of this research. The two zone models detailed as according to the Gamma Technologies Engine Performance Manual. [5]. 3.2.1 Core Mechanism The two-zone methodology operates in the following manner: • At the start of combustion, the cylinder volume is divided into two homoge- nized thermodynamic zones: an unburned zone and a burned zone. Initially, all cylinder contents, including the fresh air-fuel mixture, residual gases from the previous cycle, and any EGR gases, are placed in the unburned zone. • As the simulation progresses, a portion of the air-fuel mixture is transferred from the unburned zone to the burned zone at each timestep. This trans- fer rate is defined by the burn rate, which represents the instantaneous fuel consumption. The burn rate can either be imposed (non-predictive) or cal- culated (predictive) by the combustion model. • Once the unburned mass is transferred to the burned zone in a given timestep, GT-Power performs a chemical equilibrium calculation for the entire burned zone. The calculation identifies all the atoms of each species present in the burned zone, and from this, it takes into account the equilibrium concentra- tion of the standard combustion products. The resulting species concentra- tion is strongly dependent on the instantaneous temperature and pressure in the burned zone. • After the new composition is determined in the burned zone, the internal energy of each species is summed to find the total energy of the burned zone. GT then applies the principle of energy conservation to solve for the new temperatures in both zones and the resulting cylinder pressure. 3.2.2 Governing Equations The two-zone combustion model solves two primary energy equations in each time step. For the unburned zone, the equation is: d (mueu) dt = − p dVu dt − Qu − ( dmf,b dt hf + dma,b dt ha ) + dmf,i dt hf,i (3.1) 14 3. 0D/1D Simulation and GT-Suite Setup where: mu : unburned zone mass mf,b : fuel mass transferred to burned zone ma,b : air mass transferred to burned zone mf,i : injected fuel mass eu : unburned zone energy p : cylinder pressure Vu : unburned zone volume hf : enthalpy of fuel mass ha : enthalpy of air mass hf,i : enthalpy of injected fuel mass Qu : unburned zone heat transfer rate And for the burned zone, the equation is: d (mueb) dt = − p dVb dt − Qb + ( dmf,b dt hf + dma,b dt ha ) (3.2) where: mb : burned zone mass mf,b : fuel mass transferred to burned zone ma,b : air mass transferred to burned zone eb : burned zone energy p : cylinder pressure Vb : burned zone volume hf : enthalpy of fuel mass ha : enthalpy of air mass Qb : burned zone heat transfer rate 3.3 Three-Pressure-Analysis Model Three Pressure Analysis (TPA) is a modeling method in GT-Suite that calculates the apparent burn rate of an engine cylinder using measured pressure data. The model uses a reverse run calculation, further described in Section 3.3.1, in a two- zone combustion setup. TPA predicts all trapped quantities in the cylinder by 15 3. 0D/1D Simulation and GT-Suite Setup simulating flow through the valves. The method uses three inputs: intake, cylinder, and exhaust pressures. By using the intake and exhaust pressures as boundary conditions, the model predicts the gas exchange process. This allows the model calculate the residual gas fraction and trapping ratio of the cylinder and is seen as a more accurate way to determine the burn rate from measured pressure data, especially at part-load conditions [5]. Two types of analysis can be done using the TPA: • TPA steady: The simulation analyzes measurement data for deriving a single combustion burn rate. The input pressure data is either averaged or limited to a single cycle. • TPA multicycle: This method requires instantaneous pressure data over multiple cycles as input. Since a goal of this thesis is to obtain system-level boundary conditions and system- level stability using a derived burn rate, this study focused on the TPA steady method. 3.3.1 Reverse Run Calculation GT-Power uses a special logic called a reverse run calculation. It uses the mea- sured cylinder pressure from a test environment as input to calculate the apparent burn rate. In the standard forward run calculation, the burn rate is the input, and pressure is the calculated output. The reverse run uses the same governing equations, thermodynamic principles, and two-zone combustion methodology as in the forward run. This ensures that the calculation follows the full chemical and thermodynamic laws applied in a forward run without any simplified assumptions. In a reverse run calculation, the amount of mass transferred from the unburned zone to the burned zone is iterated at each timestep until the calculated pressure matches the measured cylinder pressure input at that timestep. This allows us to reproduce the cylinder burn rate from the test environment in the simulation. GT- Power has two methods for calculating the apparent burn rate using the reverse run methodology [5]. • Closed Volume Analysis • Three Pressure Analysis This thesis focused on the TPA methodology. 3.3.2 Required Data Input The TPA model requires primarily three pressure measurements from the cylin- der of interest: intake port pressure, exhaust port pressure, and cylinder pressure. 16 3. 0D/1D Simulation and GT-Suite Setup These measurements must be crank angle resolved. The data required for success- ful simulation includes: • Dynamic Pressures: Crank angle resolved instantaneous intake, exhaust, and cylinder pressures. • Time-Averaged Temperatures: Measured average temperature at the same locations as the pressure sensors for intake and exhaust. • EGR data: Average air mass flow rate and EGR flow rate. • Fuel injection data: Injected fuel mass, timing and timing. • Spark Timing: For Spark Ignition(SI) engines • Drive Shaft Load: Standard engine dynamometer data for model setup • Geometric Data: For the engine cylinder, valves, and ports are required to build the model. 3.3.3 Model Setup A TPA normally has a single-cylinder engine layout. The model comprises ports, runners (if present), valves, the cylinder, and the Engine crank train component built following standard modeling procedures. Modelling components in this sec- tion are referred to according to the GT Engine Performance Manual. A significant feature of the TPA setup is the implementation of port pressures as boundary conditions using the BoundaryPressureTPA template. This tem- plate uses a special logic to vary the temperature around the measured average in response to the pressure fluctuations and reverse flow. It also manages the ther- modynamic mass composition during both forward and reverse flow to ensure that the fluid properties entering and leaving the cylinder are accurately represented. For engines with external EGR this template allows specifying the EGR fraction directly, which is then used to mix fresh charge with the exhaust products from the cylinder. In the cylinder template, an EngBurnRate reference object must be used for TPA steady, and the Cylinder Pressure Analysis Mode should be set to "Analysis, Full Cycle (TPA)". To reduce numerical noise and prevent pressure signal fluctuations, the model requires specific pipe discretization and time-stepping settings. The discretization length for the intake and exhaust ports should be lowered to around 20% of the cylinder bore for intake ports and 25% for exhaust ports. The maximum time step in the flow control of the run setup must be limited to 0.25 degrees. These changes are required because the BoundaryPressureTPA template attempts to match the input pressure data in the first sub-volume of the attached pipe, and too large 17 3. 0D/1D Simulation and GT-Suite Setup sub-volumes or time steps might generate unnatural feedback fluctuations. Static pressure boundaries can cause pressure fluctuations within the ports when the valves are closed. To ensure that the predicted pressure behavior is consistent with the measured data, a control system is required to damp these reflections. This is done by using the friction multiplier in the pipe sub-volume directly adja- cent to the BoundaryPressureTPA template. When the valve is closed, the mul- tiplier is increased to a large value (around 100) to damp the reflections. When valves are open, the value goes back to being 0 for ports or 1 for runners. If the TPA is adapted from a pre-existing full engine model, the injector tem- plates need to be replaced with an InjPulseConn template, which maintains the injection profile. An inaccurate port injection profile can alter the resulting pres- sure waves. Another factor to consider while setting up the TPA model is the timing of the simulation cycle. The start of cycle attribute in EngineCrankTrain must be set to a value that occurs after IVC but before the start of the combustion event. 3.3.4 Simulation Methodology The TPA model follows the same governing equations and two-zone combustion methodology used in the standard forward run simulations. During the simulation, GT-Power follows a special iterative sequence: • For the first cycle, a dummy burn rate is used, and no pressure analysis is performed. • For the second cycle, the forward run simulation will pause at the start of the cycle and will calculate the apparent burn rate by iterating the mass trans- ferred between the combustion zones until the predicted pressure matches the input measured pressure at each timestep. The injection profile and heat transfer data will be carried over from the previous cycle. Once the entire burn rate profile is calculated, the forward simulation run continues. • This process is iterated from cycle two and beyond until steady state con- vergence is reached. 18 3. 0D/1D Simulation and GT-Suite Setup Figure 3.1: Flow chart of TPA simulation methodology 3.3.5 Consistency Checks GT-Power runs consistency checks to validate the integrity of the model setup and measured data [5]. Deriving the burn rate from measured cylinder pressure involves some errors caused by pressure transducers, sensors, modeling assumptions, and simplified thermody- namics. These combine to a cumulative error, which will almost never be zero. As a result, the total available fuel mass in the cylinder will not match the pre- dicted fuel burned during the simulation. GT-Suite then adjusts the fuel energy to achieve a combustion efficiency or burned fuel fraction derived from input test data using an LHV multiplier. If the required LHV adjustment goes above 5%, it is flagged as an error. The integrated IMEP based on the cylinder pressure must be compared to the measured BMEP. If the resulting IMEP is not greater than the BMEP by a value proportional to the engine’s FMEP, it implies errors in the measured data. For steady-state analysis, the measured pressure profile input is shifted to en- sure the best match at the start of the cycle. A shift of more than 0.5 bar will be flagged as an error. When the cylinder is being compressed before the start of combustion, the integrated fuel burned should be close to zero. If the cumulative burn during this period goes above 2% of the entire fuel, it is flagged. In direct injection models, GT checks if there is sufficient fuel in the cylinder to meet the expected burn rate. It is flagged as an error if the missing fuel goes above 2%. For steady state runs using input measure data, GT-Power compares the simulated air mass at IVC, fuel mass, and air-fuel ratio against the measured 19 3. 0D/1D Simulation and GT-Suite Setup data. If the deviations are above 5%, it is flagged as an error. Even a 1 degree crank angle phasing error will have an impact on the integrated IMEP. It can be identified by examining a motoring pressure trace, which should peak around 0.5-1.5 degree before TDC (set the encoder error at 0.7 CAD BTDC as a rule of thumb). Errors in the absolute pressure might shift the entire profile. It is visible as curvature after IVC and inconsistencies during the pumping loop. Deviations in the designed compression ratio can show up in the LogP-LogV at the end of the compression stroke, as a low compression ratio in an upward curvature and a high compression ratio in a downward curvature. 3.3.6 Result Evaluation and Export Once converged, the results are stored in a .glx file. Plot groups are generated for comparisons between the measured and simulated pressures, apparent burn rate, etc. For TPA steady runs, an EngCylCombProfile object is created, which can be imposed in subsequent forward run models to reproduce the experimental test rig combustion behavior. This allows the TPA modeling methodology to function as a bridge between physical testing and predictive simulation calibration. 3.4 MCE Modeling A MCE model is built from three main parts: EngCylinder objects for each com- bustion chamber, an EngCrankTrain that sets up the mechanical link and firing order, and a set of valves, pipes, and manifold connections for gas exchange. The EngCylinder and EngCrankTrain form the main structure for referencing key phys- ical processes like combustion, heat transfer, and friction. Each cylinder connects directly to the EngineCrankTrain, and GT gives each one a global cylinder num- ber based on its port connection. This number is important because it sets the engine’s firing order. Cylinders connect to the intake or exhaust manifolds us- ing valve connection templates, which define how the cylinder valves flow. These valves link the cylinders to the ports, which are modeled with pipes and flow splits to represent the size and heat transfer features of the cylinder head [5]. The fluid dynamics of the intake and exhaust manifolds have a significant impact on the performance of a MCE. GT models these systems using one-dimensional fluid dynamics, solving the Navier-Stokes equations to represent flow and heat transfer within the piping network. This allows for representing the pulsating flows and wave dynamics that are necessary for estimating volumetric efficiency, power, and torque with different operating speeds. 20 3. 0D/1D Simulation and GT-Suite Setup For MCE models with three or more valves per cylinder, the model must have additional valves and ports connected by flowsplits. To accurately represent the pressure losses due to contraction or expansion, the expansion diameter of the flowsplit connecting to the manifold must match the physical diameter in the cylin- der head. For accurate gas exchange predictions, the model needs well-resolved valve discharge coefficients. This is important in the MCE models with EGR because the backflow of exhaust gases into the intake manifold during the valve overlap period can affect the trapped residual fraction and subsequent combustion simulation. MCE models use two modes in the EngCrankTrain: speed mode and load mode. Speed mode is for steady state simulations where we set the engine speed, and the solver finds the matching brake torque. This mode is efficient because it skips the time needed for the crankshaft to reach steady speed under load. Load mode is for transient studies where a load is applied, and the solver calculates engine speed using brake torque, load torque, and system inertia. During transient simulations in load mode, the rotating inertia of the engine and accessories is important for predicting speed changes accurately. In these simulations, we can set operating conditions like turbocharger waste gate positions and AFR with lookup maps or predictive controllers. This lets the model react to changing loads, which is im- portant for drivability and fuel economy analysis [5]. A full MCE model needs to include sub-models such as turbochargers, charge air coolers, and EGR circuits. Turbochargers act as feedback systems, and their performance can be represented with turbine and compressor maps. Some cases might require the need for mass as well as efficiency multipliers to match experi- mental results from engines running under pulsating conditions[5]. An EngCylCombProfile template may be used to directly impose a burn rate on the model. This is an ideal approach to implement a non-predictive model (TPA). To increase calculation efficiency, GT has a Cylinder Copying feature, which allows the solver to execute detailed calculations on a single cylinder and reproduce the results in the other cylinders. This reduces the CPU time required while main- taining a reasonable accuracy for the steady state. In real-time or long transient simulations, a detailed engine model can be converted into a Fast Running Model (FRM) or by using Mean Valve Engine Modeling. These methods can considerably reduce the simulation time required to run an MCE system [5]. 21 3. 0D/1D Simulation and GT-Suite Setup 22 4 Methodology and Experimental Setup 4.1 Virtual MCE Data Exchange and Loop De- velopment This chapter characterizes the interface developed to bridge the gap between exper- imental engine testing and simulated environments. By decomposing the system into its constituent hardware and software elements, the text explains the op- erational workflow and the optimization of component parameters. The section concludes with an analysis of the control logic and the programming architecture that facilitates real-time data exchange. 4.1.1 Overview A Python-based communication protocol was developed to orchestrate data ex- change between the experimental SCE, the TPA and MCE models, and the test rig data acquisition (DAQ) systems and control systems. This integrated envi- ronment, referred to as ’The Loop’, manages the overall feedback to the system controller and the internal feed-forward data streams essential for system oper- ation. Figure 4.1 provides a schematic overview of the backend data exchange architecture in its intended closed loop implementation. 23 4. Methodology and Experimental Setup Figure 4.1: Virtual MCE data exchange loop required for the realization of the "virtual MCE" Figure 4.1 illustrates the synchronization of time-averaged state data and crank- angle resolved pressure data acquired from the SCE (1), described in section 4.1.3, and shows its integration into the TPA model (2), described in section 4.1.5. The TPA model calculates the apparent burn rate profile - referred to simply as the ’burn rate’ - and is then imposed as a combustion condition across all cylinders of the MCE model (3), which is further detailed in section 4.1.6. To achieve loop convergence, the MCE simulation yields critical thermodynamic outputs, specifi- cally the intake manifold temperature (Tin), manifold pressure (Pin), and exhaust back pressure (Pexh). These parameters are transmitted to the test rig operator, who adjusts the physical SCE environment to reflect the simulated boundary con- ditions. This iterative process continues in closed loop until a set of steady state convergence criteria, detailed in Section 4.1.8, are satisfied. The converged Loop yields output performance metrics, and complex gas dynam- ics and thermodynamics for a simplified 0D/1D environment at a burn rate which is true for its operational conditions with system level fidelity. The output of this MCE model was then compared with experimental results extracted from a MCE (further detailed in Section 4.1.2), which served as the catalyst for the development of the MCE simulation model. 24 4. Methodology and Experimental Setup 4.1.2 Multi-Cylinder Engine An experimental engine was used for validation of the simulated MCE. The refer- ence project was a developed 13-liter, six-cylinder inline spark-ignition (SI) engine specifically engineered for hydrogen combustion. To manage boost pressure and transient response, the engine utilized a Variable Geometry Turbocharger (VGT) regulated via a PID control loop. Furthermore, the system incorporated an Ex- haust Gas Recirculation (EGR) circuit to optimize combustion stability and emis- sions. A rendering of the engine is presented in Figure 4.2. The engine was part of a project titled "HYCET" at Volvo, and was tested at the KST engine test labs in Germany. For this reason the name HYCET and KST MCE are used interchangeably in the text and figures of this thesis. Figure 4.2: Render of the HYCET engine The geometry of some key combustion system components are detail below. 25 4. Methodology and Experimental Setup Variable Description Value V Total volume 13 L rc Compression ratio 9.7 B Bore 131 mm L Stroke 157 mm l Conn. rod length 259 mm Table 4.1: HYCET/KST MCE geometric properties 26 4. Methodology and Experimental Setup 4.1.3 KIT Single Cylinder Engine The experimental phase of this research was conducted using data from a SCE at Karlsruhe Institute of Technology (KIT). Utilized by Volvo for advanced power- train development, the KIT SCE was selected for its architectural commonality with the HYCET MCE project. A render of the KIT SCE is provided in Figure 4.3, with its specific geometric parameters detailed in Table 4.2. Figure 4.3: Render of the KIT single cylinder research engine. Variable Description Value V Total volume 2.12 dm3 rc Compression ratio 9.7 B Bore 131 mm L Stroke 157 mm l Conn. rod length 277 mm Table 4.2: KIT SCE geometric properties 4.1.4 KIT SCE and MCE comparison A standard bowl-shaped piston served as the basis for development between the engines. KIT utilized a Bosch Injector and designed the cylinder head to accom- modate a lateral injector placement and a central spark plug. The lateral injector 27 4. Methodology and Experimental Setup enabled fuel-tumble mixing to be realized - aligning with the mixing methodology used in the HYCET engine. The fundamental arrangement of the injector and spark plug resembled one another sufficiently to be deemed interchangeable for this study. The MCE utilized an injector from Hoerbiger and possessed minor discrepancies in the injector and spark plug angles. Figure 4.4 shows the specific details of this difference. Figure 4.4: Comparison of the injector and spark plug angles between the KIT test bench and the KST/HYCET test bench. Furthermore, in Table 4.1 and Table 4.2 it should be noted that the connecting rod lengths do not match one another. A study was conducted to investigate this effect and referring to Figures B.1 and B.2 in the appendix it was found that no significant change existed. It is important to note that the HYCET specifications are the baseline from which the KIT configuration was built - allowing the KIT SCE to serve as a detailed 28 4. Methodology and Experimental Setup follow-up platform for results obtained from the HYCET MCE. However, since the development of these test setups, experimentation and development was conducted independently. Subsequently, this independence introduced a series of limitations on the quality of available data, which is further detailed in section 7.1.2. 4.1.5 Three-Pressure Analysis Model A TPA model was configured based on the geometry of the KIT SCE. Figure 4.5 below shows a screenshot of the TPA model developed for this project. From a cross section of the SCE the general location of the piezoelectric pressure trans- ducers are shown with arrows denoting the simulation components where these CA resolved pressure traces are imposed. Figure 4.5: Three Pressure Analysis model and corresponding data sources 4.1.5.1 Gas Exchange The gas exchange system within the TPA model comprises the following functional elements: 1. Intake runner (inlet pipe) 2. Intake port bifurcations (intake pipe split) 3. Dual intake and exhaust valve assemblies 4. Combustion system control volume (Acting as a pumping device) 5. Exhaust port junction (merging flow from exhaust valves) 6. Exhaust runner (outlet pipe) High fidelity was maintained by ensuring that the geometric dimensions, specifi- cally diameters, lengths, and volumes, precisely replicate the physical architecture of the KIT SCE. In alignment with Gamma Technologies (GT) modeling conven- tions, the geometry is abstracted into a one-dimensional (1D) framework. Within 29 4. Methodology and Experimental Setup this environment, discrete frictional and pressure losses in the piping are not mod- eled explicitly; instead, these losses are consolidated into discharge coefficients (Cd) specific to the cylinder head, piston, and valve geometry, as per the GT engine performance manual [5]. Furthermore, the cam arrays for the inlet and exhaust valves were explicitly defined in accordance with those that were installed on the KIT SCE. 4.1.5.2 Combustion System The combustion system is parameterized within the crankshaft and cylinder mod- ules, as illustrated in Figure 4.5. These computational objects facilitate the defini- tion of the engine’s reciprocating assembly and combustion chamber architecture, including the piston crown (bowl) geometry, cylinder liner dimensions, and con- necting rod length. Furthermore, the firing order and the thermophysical proper- ties of the constituent materials are specified to replicate the KIT SCE configura- tion. Detailed parameter values are tabulated in Table 4.2. 4.1.5.3 Injectors Fuel injection parameters were implemented using a simplified r square wave step function approach. Although complex needle-lift dynamics and spray-wall interac- tions are present and their affects can be modeled using rail pressure and injector current profiles, this study utilizes a constant mass flow rate to represent the in- jection event. The duration is governed by the start of energizing (SOE) and end of energizing (EOE) setpoints, resulting in a step-function injection profile. This abstraction ensures that the total fuel mass delivered per cycle remains consistent with experimental data while simplifying the boundary conditions for the TPA and MCE models. It should be noted that this assumption does not affect the net energy content of the model but does affect the heat transfer modeling, which considers the effects of turbulence and squish in the combustion gases. The effect of this assumption was considered to be minor in the scope of the research; however is a topic put forward as a recommendation for further development. 4.1.6 Multi-Cylinder Engine Simulation The simulation platform for this research is a 0D/1D GT-Suite model based on the MCE developed during the HYCET project. The initial architecture, comprising of the gas exchange system, turbocharger, EGR loop, and core engine geometry, was constructed by Volvo engineers prior to this study for the HYCET project’s 30 4. Methodology and Experimental Setup development. Consequently, the model is assumed to be a high-fidelity represen- tation of the engine’s physical geometry. To adapt this MCE framework for The Loop’s application, specific geometric refinements were implemented. These mod- ifications included minor adjustments to the inlet and exhaust valve diameters, updating the connecting rod length, and redefining the cam profiles to match the SCE hardware. The final model layout is illustrated in Figure 4.6. Instead this geometry was maintained from the original model supplied by Volvo. Figure 4.6: 0D/1D GT-Suite model of the HYCET multi-cylinder engine The combustion system geometry of the MCE is define in table 4.3 below: 31 4. Methodology and Experimental Setup Variable Description SCE Value MCE Value V Total volume 2.12 L 13 L Ncyl Number of cylinders 1 6 rc Compression ratio 9.7 9.7 B Bore 131mm 131 mm L Stroke 157 mm 157 mm l Conn. rod length 277 mm 277 mm Dinvalve Inlet valve diameter 42 mm 42 mm Dexhvalve Exhaust Valve Diameter 40 mm 40 mm Table 4.3: MCE geometric properties 4.1.7 Data Exchange 4.1.7.1 Simulation Setup To facilitate near real-time synchronization between the experimental test rig and the simulation environments, an automated data-polling function was developed. Upon the deposition of a new experimental dataset into a designated directory on the Volvo servers, the integration program automatically detects the update and initiates the processing workflow. The acquired data is categorized into two distinct streams based on sampling frequency. The first, termed ’slow data,’ comprises time-averaged parameters sam- pled at regular intervals. These are initially captured in raw .mdf and .mdp formats before being processed by the Volvo Data Science Lab (DSL) into high-performance .parquet files. Slow data typically includes thermodynamic state variables, mass flow rates, and fluid temperatures. Conversely, high-frequency data, referred to as fast data, consists of crank-angle resolved measurements. This data includes current measurements, pressure data and several parameters relating to the heat release calculation. Of interest was the intake manifold pressure (Pin), in-cylinder pressure (Pcyl), and exhaust manifold pressure (Pexh), with sampling locations indicated in the SCE, shown in Figure 4.1. Initially recorded as raw .ifile files, these raw files are directly utilized by the TPA upon upload to the Volvo servers. These files are also subsequently converted to .parquet format via the DSL pipeline, providing further options for post-processing. The developed software extracts simulation-critical parameters from both high- frequency (fast) and low-frequency (slow) raw datasets to ensure the TPA and MCE models accurately reflect the physical experimental environment. A Mi- 32 4. Methodology and Experimental Setup crosoft Excel-based interface serves as the centralized data exchange hub between the processed experimental outputs and the simulation environments. By utilizing the GT-Suite Data Link functionality, the simulation objects are automatically updated based on the values populated in this Excel interface. A simplified repre- sentation of this configuration template is illustrated in Figure 4.7. This standard- ized format not only streamlined the parameterization process but also provided a transparent framework for benchmarking SCE experimental results against sim- ulated outputs. Figure 4.7: Excel file used to setup the simulation environments of the TPA and MCE models. The use of parameters are color coordinated for the simulation model. 4.1.7.2 Burn Rate Profile The apparent burn rate profile is the output of the TPA model and is stored as a 2D numerical array which includes the crank angle degree (CAD) sampled every 0.1 CAD and the corresponding fuel mass per CAD during the engine cycle. This burn rate profile is stored as a .trn file. The GT-Suite MCE model is then configured to read this file from its directory and imposed it into each cylinder in the MCE model. 4.1.7.3 Boundary Conditions After imposing the burn rates from the TPA model the MCE will then solve for the flow, heat transfer, momentum and energy balance. Once these have reached steady state convergence the output will be the boundary conditions of the MCE that are required to realize the imposed SCE combustion. This can be visualized with figure 4.8. As mentioned previously the parameters of interest are Pin, Pexh and Tin. A reporting function is used within the developed software to facilitate the extraction of these boundary conditions and the creation of a file to store these values. These values are then passed by the loop application to the test rig operator (or control system) according to figure 4.1. 33 4. Methodology and Experimental Setup Figure 4.8: Data exchange between the SCE and virtual MCE which governs steady state closed loop convergence 4.1.8 Loop Convergence Convergence is defined as the state in which the variation in boundary conditions between successive loop iterations remains within the prescribed tolerance window specified in Table 4.4. This steady-state equilibrium is contingent upon the op- erating point being physically and computationally attainable by both the SCE and MCE platforms. To ensure empirical validity, the convergence thresholds are derived from the measurement uncertainty and resolution limits of the integrated sensor hardware, ensuring that the loop terminates only when further iterations fall below the level of acceptable change. Parameter Accuracy Tolerance Pin ±0.05 kPa ±2 kPa Tin ±1C ±5C Pexh ±0.05 kPa ±2 Table 4.4: Convergence tolerance of boundary condition parameters on the MCE to the SCE 4.1.9 Multi-Cylinder Variation and Reference Selection The MCE simulation treats the engine as a holistic system; consequently, the gas exchange dynamics are explicitly modeled, resulting in non-uniform performance across individual cylinders. This variation is anticipated in both the experimental 34 4. Methodology and Experimental Setup and simulated MCE environments due to the inherent complexities of manifold wave dynamics and mass flow distribution. The variance in performance is illustrated in Figure 4.9, which displays the average work performed by each cylinder in the experimental engine. For the purposes of this study, results from Cylinder 5 (both virtual and experimental) are utilized for comparison to ensure consistency. Cylinder 5 was selected because it represents the median work output recorded during experimental testing, serving as a statis- tically representative baseline for the comparative analysis of the wider iterative loop. This load point was also checked for knock and sensor errors - cylinder 5 passed both knock and error checks and more can be read about this in Appendix C Figure 4.9: Average indicated work done by each cylinder 35 4. Methodology and Experimental Setup 4.2 Research Question 1.1: How sensitive is the loop to Cyclic Variations This study examined how the loop’s boundary conditions respond to the pressure data entered. The pressure data covers the crank angle resolved instantaneous in- take, exhaust, and cylinder pressure, which are used to calculate the burn rate in the TPA model. The research looked at how using different sizes of cycle average data, as well as cycles with median and max IMEP, affects the results. Data from 10 raw data files (.ifiles) were used for this study. Samples 9 and 10 are also used to address research question 1.2. Sample No. IMEP (bar) CoV (%) Lambda Speed (rpm) Sample 1 11,15 4,3 2,801 1100,1 Sample 2 14,01 3,9 2,547 1100,1 Sample 3 11,15 2,8 2,813 1100,1 Sample 4 11,01 4,0 2,786 1100,1 Sample 5 11,09 3,8 3,071 1100,1 Sample 6 9,47 3,2 3,111 1100,1 Sample 7 7,85 4,3 3,073 1100,1 Sample 8 7,82 3,9 3,060 1100,1 Sample 9 14,40 4,1 2,402 1300,0 Sample 10 14,39 4,1 2,427 1300,0 Table 4.5: IFile sample data operating points In the TPA steady method, GT-Power averages all input pressure data into a single average cycle for analysis. For this study, custom cycle data were extracted from the .ifiles and averaged using a Python script with the CATool package. Cycle data of various sizes was extracted and used as input to the loop, ranging from 0-5 cycles to 0-200 cycles, along with cycles with the median and max IMEP. A 200-cycle average was taken as the standard input for each sample. Results were extracted from GT-POST, compiled, and subsequently plotted using Excel. The analysis of the TPA and MCE model simulation results focused on the following parameters: 36 4. Methodology and Experimental Setup Category Parameter Indicated Performance IMEPnet (IMEP720 – Net Indicated Mean Effective Pressure) IMEPgross (IMEP360 – Gross Indicated Mean Effective Pressure) Indicated Efficiency Combustion Efficiency Volumetric Efficiency (Air) Combustion Phasing & Duration Crank Angle at 10% Burned (CA10) Crank Angle at 50% Burned (CA50) Burn Duration 0–10% Burn Duration 10–90% Energy & Losses Cumulative Energy Loss Pressures Maximum Cylinder Pressure Intake Cylinder Pressure Exhaust Cylinder Pressure Temperatures Intake Temperature Exhaust Temperature Table 4.6: Combined list of MCE combustion, performance, and gas-exchange parameters 4.3 Research Question 1.2: How well does the simulated MCE compare to experimental MCE data As mentioned in Section 4.1.2 and 4.1.3, the KIT SCE was derived from the HYCET project’s multi-cylinder architecture to facilitate a direct comparative analysis. By adapting a standard bowl-shaped cylinder head to include a central spark plug and lateral injector, the KIT SCE replicates the fuel-tumble mixing strategy employed by the HYCET/KST Multi-Cylinder Engine (MCE). While minor geometric variations in injector-plug orientation exist, the shared piston geometry and injector specifications render the SCE a high-fidelity platform for following up on MCE results. However, the their independent development of these 37 4. Methodology and Experimental Setup platforms introduces specific data limitations, which are addressed in Section 7.1. Section 5.3 aimed to show how the results from the simulated MCE represented the experimental results from the HYCET engine. The research question was an- swered by analyzing the full engine cycle with a LogP-LogV diagram and discusses the difference that were observed with particular focus on the cylinder pressure, pumping loop and inlet manifold pressures. Analysis was also conducted into dif- ferences observed in the volumetric efficiencies and the indicated efficiencies. The limitation of the study are discussed in section 7.1. 38 5 Results 5.1 Loop Output and Performance 5.1.1 Inlet Manifold Convergence Inlet temperature and pressure are governing parameters for convergence, directly impacting the calculated burn rate. Consequently, these values dictate the bound- ary conditions and overall performance output of the MCE simulation. The iter- ative process aims to constrain these variables within the predefined limits estab- lished in Table 4.4. As illustrated in Figure 5.1, the manifold pressure exceeds the permissible thresh- old during the initial iteration, necessitating further computational cycles. At this specific load point, the KIT and MCE exhibit a relative pressure deviation of 6.7% after the first iteration. According to the convergence criteria, the allowable error is defined as a function of boost pressure, where increased boost levels dictate a more stringent error margin. Consequently, a maximum relative absolute error of 0.8% — equivalent to a differential of ±2 kPa between the MCE and SCE — is required to satisfy the convergence targets. It should be noted that this level of convergence accuracy should be tested and validated. In the absence of a means to validate these convergence limits on a dedicated research engine, it is unknown if these limit are obtainable or if greater tolerances are possible. As a result, it is put forward as a recommendation to test these convergence parameters for future development on dedicated hardware. In Figure 5.2, a thermal discrepancy exists between the measured SCE and simu- lated MCE. To satisfy the imposed burn rate profile, a target inlet temperature of 49.2 C was identified for the MCE. Consequently, further iterative loops are nec- essary to achieve numerical convergence. Consistency between the experimental data and the TPA results further validate the TPA model, exhibiting a relative error of only 0.7%. 39 5. Results Figure 5.1: Inlet manifold pressure Figure 5.2: Inlet manifold temperature 40 5. Results 5.1.2 Exhaust Manifold Convergence Regarding the analysis of the exhaust data extracted from the virtual MCE, it should be noted that data was collected at two distinct locations within the ex- haust system. The first location, indicated by the orange bar in Figure 5.4 and 5.3, corresponds to the exhaust port exit immediately downstream of the cylinder. The second location, indicated by the green bar, is situated at one of the two 3-to-1 collectors (serving cylinders 1-3 and 4-6, respectively). Defining these spatial differences is of critical importance as gas wave dynam- ics vary significantly between these two sections; however, both are relevant to this study. Port-level measurements capture the discrete, high-amplitude pulse dynamics of a single cylinder once per four-stroke cycle, allowing for the charac- terization of localized heat transfer effects. In contrast, the manifold measurement is subjected to the combined flow of all three cylinders connected to the single collector. In this case, the individual gas pulses are superimposed out of phase according to the engine’s firing order and valve timings. This interference results in a time-averaged mean gas temperature and pressure recorded across three cylinders, which increased the effective tem- perature measurement and decreased the measured pressure. The impact of this phenomenon was found to be more significant for temperature, and was compar- atively less pronounced for the pressure measurements. With regards to exhaust boundary pressure, the exhaust readings for the SCE and MCE had smaller relative errors than what was observed in the inlet mani- fold. A relative error of 3.8% was measured. The explicit values fell outside of the convergence requirements defined in table 4.4. 41 5. Results Figure 5.3: Exhaust manifold pressure Figure 5.4: Exhaust manifold temperature 42 5. Results 5.1.3 Operational Efficiency Figure 5.5 shows the time taken by each process independently. The simulations were the largest bottle neck with each submission to a simulation cluster taking approximately 3 minutes regardless of the whether the simulation was submitted to a high performance cluster or if it was simulated on a local computer. Amounting to a total of 6 minutes dedicated to simulation. In addition to this, approximately 2 minutes is required to adjust the boundary condition with a further minimum 2 minutes dedicated to data collection from the SCE and operations to impose this data. Figure 5.5: Processing time evaluation. The expected time to conduct a single Loop iteration is estimated at 10 min- utes. It should be further investigated if the benefits of developing this virtual MCE outweigh this temporal disadvantage. This should consider the fact that this method has the potential to greatly increase the maturity of an MCE concept before a physical prototype is required - which is a timely and costly process in itself. The engineer should also bear in mind that this system would be required to go through the full range of engine tests to realize a high fidelity output of the final MCE product (assuming that R&D is conducted on the MCE continuously). More about this can be read in Section 2.3. 43 5. Results 5.2 Research Question 1.1: How sensitive is The Loop to Cyclic Variations The following section discusses the findings related to research question 1.1. When observing how the loop responds to changes in the pressure trace input, using different sample sizes to average the pressure traces of the same sample file in the loop gave very similar boundary results in the MCE model. Figure 5.6: IMEP output from the loop When the IMEP output using a standard input of 200 pressure trace cycles was compared, the largest change in output IMEP from the loop was less than 3.5%. Plots with all the sample size data have been added to the appendix for reference. Figure 5.7: IMEP Relative Comparison with the Standard Input Pressure Trace The results were consistent for other observed indicated output performance pa- rameters from the loop. The biggest change in the indicated performance param- eters came from using a single pressure cycle with either the median or maximum IMEP. In contrast, using the averaged cycle as input resulted in a relative change of less than 1.5% compared to the standard input. 44 5. Results Figure 5.8: Indicated Efficiency output from the loop Figure 5.9: Indicated Efficiency Relative Comparison with the Standard Input Pressure Trace An analysis of the sensitivity of boundary pressure conditions to the input pressure trace showed minimal changes in boundary pressure outputs compared to the standard input, consistent with the indicated performance results. Figure 5.10: Intake Pressure output from the loop 45 5. Results Figure 5.11: Exhaust Pressure output from the loop The largest relative change was observed when using a single pressure trace with median or maximum IMEP. In contrast, averaging the cycle input over 10 or more cycles reduced the relative change to less than 2%. Figure 5.12: Intake Pressure Relative Comparison with the Standard Input Pres- sure Trace Figure 5.13: Exhaust Pressure Relative Comparison with the Standard Input Pressure Trace An analysis of the sensitivity of boundary temperature conditions to the input pressure trace also showed minimal changes in boundary temperature outputs compared to the standard input. With the single pressure trace with median or maximum IMEP input producing a relative change of less than 1.5%, and by 46 5. Results averaging the cycle input over 10 or more cycles, the relative change is less than 0.5%. Figure 5.14: Intake Temperature Relative Comparison with the Standard Input Pressure Trace Figure 5.15: Exhaust Temperature Relative Comparison with the Standard In- put Pressure Trace Analysis of cumulative energy loss showed that averaging 10 or more cycles resulted in a relative change of less than 4%, while a single pressure trace with median or maximum IMEP input produced a relative change greater than 10%. Figure 5.16: Cumulative Energy Loss output from the loop 47 5. Results Figure 5.17: Cumulative Energy Loss Relative Comparison with the Standard Input Pressure Trace Analysis of combustion phasing indicated that using a single pressure trace with median or maximum IMEP input resulted in a relative change greater than 60%. In contrast, averaging 40 or more cycles reduced the relative change to less than 20%. The relative change in combustion phasing showed a strong sensitivity to cycle-to-cycle variability, and it is probable that the relative change produced by the input ranges will vary depending on the coefficient of variation (CoV) and the operating conditions of the test run. Figure 5.18: CA50 output from the loop Figure 5.19: CA50 Relative Comparison with the Standard Input Pressure Trace 48 5. Results Analysis of burn duration indicated that using a single pressure trace with median or maximum IMEP input resulted in a relative change greater than 30%. In contrast, averaging 40 or more cycles reduced the relative change to less than 5%. Figure 5.20: Burn Duration 0-10% Relative Comparison with the Standard Input Pressure Trace 49 5. Results 5.3 Research Question 1.2: How well does the simulation MCE compare to experimental MCE data Please refer to section 4.1.3; 4.1.6 and 4.3 for descriptions on the specific hardware differences between the models and engines. This section discusses the findings relative to Research Question 1.2, specifically regarding the hardware discrepancies identified between the simulated environ- ment and the experimental configurations. Rather than evaluating the correla- tions between the two domains, the results and discussion focuses on the points of divergence. This shift in focus is necessary to characterize the impact of geometric and structural variations (such as manifold architecture, cam shape, etc) on the resultant gas dynamics. By isolating these hardware-induced variances, the sensi- tivity of the 0D/1D model to physical deviations can be more accurately quantified. The results demonstrate that the iterative loop - spanning from the physical SCE through the TPA to the virtual MCE - exhibits a high degree of accuracy. This correlation indicates that the simulation models are robustly tuned for the KIT SCE experimental setup. However, discrepancies are observed in the simulated MCE trace, particularly during the compression stroke, where we observe the sim- ulated MCE overfitting the SCE and the TPA. Observing the red curve, denoting the HYCET engines’ experimental data, dis- crepancies are localized near Top Dead Center (TDC) and in the regions proximate to Inlet Valve Closing (IVC), Exhaust Valve Opening (EVO), and Bottom Dead Center (BDC). As illustrated in Figure 5.21, these deviations correspond to the commencement of the compression stroke and the concluding phases of the expan- sion stroke. The divergence in these specific regions suggests that the experimental engine has been built with an alternative cam profile - removing the validity of the comparison. It was further observed that the experimental engine exhibits characteristic "bow- shaped" curve in figure 5.21 - during both the compression and expansion strokes. This phenomenon is likely attributable to transient heat transfer effects during the compression stroke, and a synergistic combination of heat loss and evolving gas composition during expansion. These factors dynamically influence the effec- tive polytropic coefficient (n) of the charge and exhaust gases, resulting in the non-linear behavior observed in the figure. Consequently, the curvature reflects 50 5. Results the continuous variation of the polytropic index as a function of the instantaneous cylinder temperature and thermophysical gas properties. Furthermore, the base- Figure 5.21: Log-P log-V diagram of the SCE, simulated TPA and MCE models and the experimental MCE data line inlet manifold pressure and exhaust back-pressure configurations were found to be inconsistently calibrated between the two environments. These findings further underscore the operational and hardware-level inconsis- tencies previously identified within the dataset. The resulting impact on the gas exchange process is clearly illustrated in the simplified pumping loop of the log(p) − log(V ) diagram (Figure 5.22). The divergence in the pumping loop area emphasizes the variation in the work required for induction and scavenging, high- lighting the sensitivity of the iterative loop to valve timing and boundary pressure accuracy. Figure 5.23 illustrates that the inlet manifold pressures are distinctly different. The results indicate a higher boost pressure inherent to the experimen- tal tests compared to the parameters configured in the simulation models. This discrepancy was, however, anticipated due to the independent nature of the ex- perimental and numerical studies. The higher intake pressure in the SCE suggests 51 5. Results Figure 5.22: LogP-logV curve for the simulated and the experimental MCE systems a greater charge density, which must be reconciled in the virtual MCE to ensure high-fidelity synchronization of the boundary conditions within the iterative loop. 52 5. Results Figure 5.23: Pressure traces and inlet manifold pressures compared for the vir- tual and experimental MCE. It was observed, however, that the increased boost pressure did not reflect in the volumetric efficiency metric presented in Figure 5.24. This suggests that the gas exchange systems in the simulation models possess a higher flow potential or rep- resent a more optimized configuration than the experimental hardware. This discrepancy is also reflected in the indicated efficiency metric described in Figure 5.25, where the experimental MCE is approximately 4 percentage points lower than the values recorded in any of the other Loop entities. Given the sub- stantial divergence between the two setups, the systems are currently too dissimilar to draw definitive conclusions. Recommendations on the methodology required to normalize these parameters and proceed with the study are offered in the following section. 53 5. Results Figure 5.24: Volumetric Efficiency Figure 5.25: Indicated Efficiency 54 6 Conclusion This research demonstrates significant potential as a novel development frame- work for MCE prototypes, which integrates empirical premixed hydrogen combus- tion data from a SCE test rig into a 0D/1D TPA model, which calculates a high resolution burn rate and then uses this burn rate as input into a detailed MCE simulation in GT-Suite. This methodology enables observation of a MCE within its system context while maintaining the resolution of a physics informed local combustion event. However, due to the absence of a second iteration and the time scale of which a sin- gle iteration occupied in the closed loop system, this study was unable to generate a sufficient dataset to characterize the progression of near real time co-simulation, verify the precision of the resulting MCE output or justify the potential financial and productivity benefits possible. Furthermore, a direct validation of the 0D/1D- MCE model against existing multi-cylinder experimental data was precluded by disparities in the respective operational conditions and inherent differences in the cam profiles. To resolve these limitations, future work must incorporate synchro- nized experimental configurations and loading conditions, and full access to a dedicated SCE to realize complete loop convergence and system validation. Regarding the processing of input data, this study highlights the risks of uti- lizing a single combustion cycle, even one representing the median or maximum IMEPn, due to the stochastic nature of cyclic variations. Quantitative analy- sis revealed that a 5-cycle average resulted in absolute deviations of up to 3% in boundary temperatures, and 1.5% in both boundary pressures and indicated ef- ficiency, when compared to a 200-cycle baseline. Given the profound impact of cyclic variations on combustion phasing, it is concluded that a minimum sample size of 50 averaged cycles is required as a simulation input to ensure statistical sig- nificance and model stability - an important observation in the pursuit of realizing a near real time HiL system. 55 6. Conclusion 56 7 Discussion 7.1 Limitations During the course of the project strict limitation hindered the full realization out- lined by the project aim (see Section 1.2). This section aims to describe these limitation and highlight how they affect the project. 7.1.1 Engine Access Initially, the experimental phase of this project was centered on a single-cylinder research engine at Chalmers University of Technology. This platform was derived from a 13L heavy-duty truck engine specifically configured for pre-mixed hydro- gen combustion research. However, during the implementation phase, a critical malfunction rendered the engine inoperable. The unavailability of the Chalmers engine presented a significant challenge for the operational validation of the data exchange interface. Without a functional test rig, the closed-loop synchronization between the physical hardware and the numerical models could not be fully realized in a real-time experimental context. To mitigate these limitations and ensure a robust validation of the integrated framework, the project’s scope was redirected to utilize the KIT SCE platform. This transition allowed for the utilization of high-fidelity experimental data to validate the communication and logic of the loop, as described in the preceding sections. Making this change only allowed for an initial loop simulation without any further iteration. 57 7. Discussion 7.1.2 Data Availability A significant advantage of the KIT SCE was the availability of a comprehensive experimental database that closely corresponded to its MCE counterpart—a capa- bility that was lacking in the Chalmers engine platform. However, the integration of these two distinct datasets required a rigorous filtering process to ensure phys- ical and thermodynamic consistency. The scope of this comparative analysis was constrained by discrepancies in the operational envelopes; specifically, the KIT SCE and the HYCET MCE were orig- inally tested at inconsistent engine speeds and load points. This misalignment necessitated the selection of a restricted subset of data. Ultimately, three KIT SCE datasets from "run-in" program tests were identified as the most viable can- didates for comparison. Further analysis revealed only two MCE datasets that provided a sufficient corre- lation to these SCE benchmarks, highlighting the inherent inconsistencies between the independent testing campaigns. Figure 7.1 illustrates the datasets exhibiting the highest degree of correlation between the KIT SCE and the HYCET MCE, which serve as the baseline for the comparative analysis presented in the following results section. These data sets were simulated in the Loop following the process outlined by Fig- ure 4.1. Considering the limitation described above, only an initial loop iteration was possible; the results are therefore not converged. This process forms the basis for further development, and the results should be considered an initial proof of concept. 58 7. Discussion Figure 7.1: SCE and MCE datasets that showed the best agreement used for comparison 59 7. Discussion 7.2 Recommendations To enhance the fidelity or the operation fidelity of the developed system, several important refinements are recommended. Firstly, it is imperative to achieve full closed loop numerical convergence using a load point and operating condition al- ready established within the KST engine dataset corresponding to a build with the same cam profile installed; this will serve as the definitive validation for the virtual MCE concept model and also the functional value of The Loop as an engineering tool. Secondly, the computational, data handling and system operations framework should migrate toward a more robust architecture. Transitioning to a Linux-based environment and adopting more precise programming languages, such as C, in place of the current Python-based scripts would significantly improve processing time, computation speeds and adherence to well developed software engineering standards. This transition should be coupled with the development of a graphical user interface (GUI) to improve usability for the engineer and the test rig operator. Furthermore, regarding the physical accuracy of the model, the current square wave injector approximation should be replaced with a more comprehensive and realistic modeling strategy for better characterization of the induction process. Additionally, the 0D/1D models require a comprehensive calibration phase specif- ically targeting heat transfer coefficients and mechanical friction sub-models to reconcile the efficiency discrepancies noted in this study. A significant area for future research involves the investigation of Fast Running Models (FRM) and xRT solver technologies, available through GT software pack- ages. Technical literature suggests that xRT models facilitate rapid simulation times, potentially enabling real-time "Hardware-in-the-Loop" (HiL) applications where the virtual MCE converges with marginal latency relative