Combined Experimental and Numerical Study of Active Thermal Control of Battery Modules. Fan He. Doctor of Philosophy In Mechanical Engineering

Transcription

1 Combined Experimental and Numerical Study of Active Thermal Control of Battery Modules Fan He Dissertation submitted to the faculty of the Virginia Polytechnic Institute and State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy In Mechanical Engineering Lin Ma, Chair Srinath Ekkad Douglas Nelson Michael Philen Heng Xiao March 20, 2015 Blacksburg, VA, USA Keywords: Thermal management, Li-ion batteries, Computational fluid dynamics (CFD), Reduced-order model (ROM), Active temperature control, Reciprocating cooling Copyright 2015

2 Combined Experimental and Numerical Study of Active Thermal Control of Battery Modules Fan He Abstract Lithium ion (Li-ion) batteries have been identified as a promising solution to meet the increasing demands for alternative energy in electric vehicles (EVs) and hybrid electric vehicle (HEVs). This work describes experimental and numerical study of thermal management of battery module consisting of cylindrical Li-ion cells, with an emphasis on the use of active control to achieve optimal cooling performance with minimal parasitic power consumption. The major contribution from this work is the first experimental demonstration (based on our review of archival journal and conference literature) and the corresponding analysis of active thermal control of battery modules. The results suggest that the active control strategy, when combined with reciprocating cooling flow, can reduce the parasitic energy consumption and cooling flow amount substantially. Compared with results using passive control with unidirectional cooling flow, the parasitic energy consumption was reduced by about 80%. This contribution was achieved in three steps, which was detailed in this dissertation in chapters 2, 3, and 4, respectively. In the first step, an experimental facility and a corresponding CFD model were developed to capture the thermal behavior of multiple battery cells. Based on the experimental and CFD results, a reduced-order model (ROM) was then developed for active monitoring and control purposes. In the second step, the ROM was parameterized and an observerbased control strategy was developed to control the core temperature of battery cells. Finally, based on the experimental facility and the ROM model, the active control of a battery module was demonstrated. Each of these steps represents an important facet of the thermal management problem, and it is expected that the results and specifics documented in this dissertation lay the groundwork to facilitate further study.

3 Preface This dissertation is submitted for the degree of Doctor of Philosophy in Mechanical Engineering at Virginia Tech. The research presented in this dissertation was conducted at both the Department of Mechanical Engineering and the Department of Aerospace and Ocean Engineering, Virginia Tech, under the supervision of Dr. Lin Ma from August 2012 to February This work is to the best of my knowledge original. The purpose of the dissertation is to design and demonstrate a novel active thermal control approach for power battery modules used in electric vehicles (EVs) and hybrid electric vehicles (HEVs). Experimental facilities were developed and numerical models were introduced to study the thermal characteristics of battery modules. The facilities and models were then applied in geometric optimization and active control design so that minimum energy consumption and low temperature rise was achieved. I was the one of the lead investigators for the work in Chapter 2, which involves the development of the experimental facilities and the numerical models. I was responsible for both the overall design of both the experiments and the models, and also the implementation. Dr. Xuesong Li, a recent PhD graduate from Dr. Ma s group, was involved in the early stage of this work. I was the lead investigator for the work described in Chapter 3, which focused on the parametrization of plant and controller variables, an important preparatory step toward the active control system design. iii

4 I was the lead investigator for the work described in Chapter 4, which demonstrated geometric optimization and active temperature control of battery modules. To my best knowledge, the work described here represents the first experimental demonstration of active thermal control of battery modules in archived literature. The geometric optimization was performed using the CFD models described in chapter 2 to determine geometry configuration that can achieve optimal cooling effectiveness. Active temperature control system design was performed based on the ROMs to manage the temperature within the optimal range and also to achieve minimal parasitic power energy consumption. The active control system was experimentally demonstrated and the results compared to modeling analysis. Mr. Haoting Wang, a PhD candidate in Dr. Ma s group, has been working together with me on work described in this chapter since August The following is a list of my publications and presentations related to the work described in this dissertation: Journal papers [1] X. Li, F. He, L. Ma, Thermal Management of Cylindrical Batteries Investigated Using Wind Tunnel Testing and CFD Simulation, Journal of Power Sources, 238(0) (2013) [2] F. He, X. Li, L. Ma, Combined Experimental and Numerical Study of Thermal Management of Battery Module Consisting of Multiple Li-ion Cells, International Journal of Heat and Mass Transfer, 72(0) (2014) [3] F. He, L. Ma, Comparative Study of Thermal Management in Hybrid Power Systems Using Cylindrical and Prismatic Battery Cells, Heat Transfer Engineering, in press, iv

5 [4] F. He, L. Ma, Thermal management of batteries employing active temperature control and reciprocating cooling flow, International Journal of Heat and Mass Transfer, 83 (2015) [5] F. He, H. Wang, L. Ma, Numerical and Experimental Study on Thermal Management of Batteries Employing Active Temperature Control Combined with Reciprocating Cooling Flow, International Journal of Heat and Mass Transfer, submitted, Conference papers and poster presentations [1] F. He, D. Ewing, J. Finn, J. Wagner, L. Ma, Thermal Management of Vehicular Payloads Using Nanofluid Augmented Coolant Rail - Modeling and Analysis, SAE International Journal of Alternative Powertrains, 2(1) (2013) [2] F. He, X. Li, L. Ma, Thermal Management of Battery Module Consisting of Multiple Li-ion Cells for Hybrid Power Systems, 52nd AIAA Aerospace Sciences Meeting, American Institute of Aeronautics and Astronautics, National Harbor, Maryland, USA, [3] H. Wang, F. He, L. Ma, Thermal Management of Energy Storage Systems Based on Battery Modules, submitted to NAWEA 2015 Symposium, Blacksburg, VA, [4] F. He, X. Li, L. Ma, Study of Battery Cooling in Hybrid Electric Vehicles using High-Fidelity CFD and Wind Tunnel Testing, 19th ARC Annual Review Meeting, Ann Arbor, MI, [5] F. He, Y. Ding, L. Ma, Combined Experimental and Computational Study of Battery Cooling in Hybrid Electric Vehicles, 20th ARC Annual Review Meeting, Ann Arbor, MI, v

6 Acknowledgements This research work would not have been a reality without the support and interactions from various organizations and individuals. The research presented in this dissertation was conducted at the Department of Mechanical Engineering and Department of Aerospace and Ocean Engineering, Virginia Tech. It was funded by the Automotive Research Center (ARC), a U.S. Army Center of Excellence in Modeling and Simulation of Ground Vehicles. I would like to thank my advisor Dr. Lin Ma. The advice and support provided by him are key factors for my growth in academic ability and personality during my doctoral program. I especially thank him for always being available for discussions and guidance during my research. I would also like to thank Dr. Srinath Ekkad, Dr. Douglas Nelson, Dr. Michael Philen, and Dr. Heng Xiao for serving as my advisory committee members and advising me on my academic progress. I would like to thank my colleagues and friends: Dr. Xuesong Li, Dr. AJ Wicksham, Dr. Weiwei Cai, Minwook Kang, Yue Wu, Qingchun Lei, Haoting Wang, Wenjiang Xu, Shuai Zhang, and many others who have helped me and collaborated with me during the past 3 years. Last but not the least, I dedicate this dissertation to my parents, for their love, support and sacrifice for me. vi

7 Table of contents Contents Abstract... ii Preface... iii Journal papers... iv Conference papers and poster presentations... v Acknowledgements... vi Table of contents... vii List of figures... ix List of tables... xii Chapter 1. Introduction Introduction and motivation Literature review Scope of this work Organization of the dissertation... 6 Chapter 2. Description of experimental setup and models Experimental setup Description of CFD model Description of ROMs Static ROM Dynamic ROM Comparison between experiments and models Summary and conclusions Chapter 3. Parameterization of plant and controller parameters Recursive least squares algorithm in discrete time domain Luenberger observer design for 1 single cell Luenberger observer design for multiple cells Conclusion Chapter 4. Design optimization and active control Geometric optimization of battery modules Benchmark results vii

8 Comparative studies of different configurations Active control simulation using ROMs Experimental demonstration of active control Experimental setup Experimental and CFD results Conclusion Chapter 5. Conclusions and future work Conclusions Future work References viii

9 List of figures Figure 2.1. Illustration of the experimental facilities. Panel (a). schematics of the experimental setup. Panel (b). layout of cells and thermocouples in the battery pack Figure 2.2. Panel(a). schematic of problem definition. Panel (b). sample mesh of and around cell 1 at U = 5 m/s Figure 2.3. Experimentally measured voltage for cells under different current at different SOCs during discharge Figure 2.4. Internal resistances calculated using measured data under different current at different SOCs during discharge Figure 2.5. Heat generation calculated by experimental data for all cells used in CFD modeling.18 Figure 2.6. Comparison of Nu number obtained by CFD model against past experiments and empirically correlation Figure 2.7. Temperature non-uniformity simulated by the CFD model under two representative inlet air velocities: Panel (a). U= 5 m/s. Panel (b). U = 1 m/s case Figure 2.8. Temperature non-uniformity among different cells in the battery pack at steady state. Panel (a). U = 5 m/s case. Panel (b). U = 1 m/s case Figure 2.9. Schematics of the two-zone static reduced-order model Figure Comparison of the maximum cell temperature predicted by the two-zone model and the CFD model under different Re numbers. Panel (a). Tmax predicted by the twozone model, and the maximum and minimum temperatures predicted by the CFD model. Panel (b). discrepancy of the maximum cell temperature obtained from the two-zone and the CFD model Figure Comparison of the flow outlet temperature predicted by the two-zone model and the CFD model under different Re numbers. Panel (a). Tout predicted by the two-zone model and the CFD model. Panel (b). discrepancy of the maximum outlet flow temperature obtained from the two-zone and the CFD model Figure Schematic of the ROM for a battery module and the notations used in its development Figure Comparison between experimental data and simulations from the CFD model under U=0 m/s. Panel (a). temperature of Cell 1. Panel (b). temperature of Cell Figure Comparison between experimental temperature datasets and simulations from the CFD model of Cell 1 under different enclosure wall-cell distance with U=1.0 m/s. 32 ix

10 Figure Comparison between experimental temperature datasets and simulations from the CFD model of Cell 2 under different enclosure wall-cell distances with U=1.0 m/s Figure Comparison between experimental and simulated pressure under U=1.0 m/s Figure Comparison between experimental and simulated pressure under U=2.5 m/s Figure Comparison between experimentally measured temperature and temperature predicted by the ROM Figure 3.1. Scheme of the online parameterization method and structure of the adaptive Luenberger observer Figure 3.2. Rescaled UAC load profile used in this work Figure 3.3. Inlet cooling air flow temperature for simulation of online parameterization Figure 3.4. Comparison of the cell core and surface temperature of the real plant against those estimated by the adaptive Luenberger observer Figure 4.1. Maximum temperature rise under different conditions Figure 4.2. CFD results of maximum cell temperature of MP1 and MC1 under Tin=Tamb Figure 4.3. CFD results of maximum cell temperature of MP1 and MC1 under Tin=25.0 o C Figure 4.4. Volumetric flow rate required to control the maximum cell temperature within the optimal range for MP1 and MC Figure 4.5. Pump power required to control the maximum cell temperature within the optimal range for MP1 and MC Figure 4.6. Maximum cell temperature for MP1, MP2 and MP3 under cooling scheme 1 and Figure 4.7. Maximum cell temperature for MC1, MC2 and MC3 under cooling scheme 1 and 2.60 Figure 4.8. Comparison of max temperature, pressure drop and pump power for all six configurations Figure 4.9. Diagram of a battery thermal management system with active temperature control and reciprocating cooling flow Figure Block diagram of an observer-based control system Figure Comparison of cell core and surface temperature of the real plant versus those estimated from the observer for three cells Figure Instantaneous flow rate and total cooling flow consumption for three different schemes Figure Temperature variation of cell 1, 4 and 8 for three different schemes Figure Summary of cooling performance under six schemes x

11 Figure Experimental setup of the active temperature control system for the battery module using reciprocating flow Figure Hardware and block diagrams of control system Figure Experimentally measured temperature under NC+UND condition Figure Comparison of CFD simulated and experimental results for battery module under NC+UND condition Figure Experimental results of module temperature under NC+RECP condition Figure Comparison of CFD results and experimental data under NC+RECP condition.. 79 Figure Experimental results of battery module temperature under HYS+RECP condition Figure CFD results for the battery module temperature under HYS+RECP condition Figure Summary of Experimental and CFD data of temperature rise, energy consumption and total cooling flow used under four different conditions Figure 4.24 Comparison of fan energy used under different conditions Figure 4.25 Comparison of amount of cooling air used under different conditions xi

12 List of tables Table 2.1 Parameters and thermophysical properties used in the CFD model Table 4.1 Parameters and properties of cells and modules for MC1 and MP Table 4.2 Geometric parameters of battery modules for comparative study in Section xii

13 Chapter 1. Introduction 1.1. Introduction and motivation Electric vehicles (EVs) and hybrid electric vehicles (HEVs) have been identified as part of our solution to the global issue of energy and environmental sustainability [1-3]. For the research and development of EVs and HEVs, energy storage is one of the key enabling technologies. Various types of batteries, such as Lithium-ion (Li-ion) batteries, Lid-acid batteries, and NiMH batteries, have been proposed and tested for EV and HEV applications. Among all these types of batteries, Li-ion batteries have been widely deployed for energy storage in EVs and HEVs due to several unique advantages, including their relatively high energy density, high power density, stability and low selfdischarge rate when not in operation [4]. However, the performance of Li-ion batteries is significantly restricted by their long term stability and safety characteristics [5], and both issues are related to the thermal behavior of Li-ion batteries. Li-ion batteries are thermally vulnerable and their temperature needs to be carefully controlled for optimal and safe operation [6]. To achieve the optimal operation efficiency and cycling life, the temperature of each Li-ion cell in a battery module needs to be managed within a certain range, requiring the temperature non-uniformity among all cells in the module to be minimized [7]. For instance, a temperature range between -10 to +50 [8] has been reported as the tolerable operation range, and a more restricted range of +20 and +40 [9] has been reported for better performance of Li-ion batteries. However under practical conditions, it is extremely challenging to guarantee such optimal operation mainly due to two reasons. 1

14 First, the thermal loads posed on the battery module are large and highly dynamic, resulting in large temperature variations and non-uniformity among the cells. Second, environmental temperatures under which the vehicles operate could vary significantly beyond these desired ranges. As a result, thermal management represents a major challenge for the safe and efficient use of Li-ion battery modules. Overheating or uneven cell temperature due to the factors mentioned above will cause cell degradation and cell failure. Cell degradation leads to considerable loss of voltage and electric capacity, and cell failure leads to thermal runaway or even fire hazard [10, 11]. Therefore in practical applications, it is necessary to employ active or passive thermal management system to maintain the temperature of the batteries within the desired ranges. Therefore, it is the goal of this dissertation to investigate novel thermal management with active control which can reliably manage the battery temperature within the desired range and minimize the parasitic energy consumption at the same time Literature review Due to the importance of thermal control as discussed in the Section 1.1, the investigation of battery thermal management has attracted considerable research and development efforts. Here, past efforts were broadly divided into two categories to facilitate the literature review. Efforts in the first category focus more on the fundamental issues, and the efforts in the second category more on the developmental issues. Efforts in the first category focus on obtaining a fundamental understanding of the thermal behavior of batteries, such as understanding the fundamental electro-thermal processes and the development of the corresponding governing equations. Notable 2

15 examples of the research work in this category include the models based on firstprinciples to simulate the state of charge (SOC) of Li-ion batteries [12-14], a critical parameter that influences the performance and heat generation rate of Li-ion batteries. Advanced experimental techniques such as neutron imaging are being employed to make in situ measurements of lithium concentration to validate these models. In addition, equipped with a fundamental understanding of the governing processes, other models are being developed to simulate other aspects of battery operations, including the heat generation in battery cells under different operation conditions and battery geometries using one-dimensional (1D) [15-17], 2D, and 3D models [18-20]. Based on a fundamental understanding of the thermal behavior of the batteries and the corresponding governing equations, numerical schemes then can be developed to solve these equations and provides more insights. Two types of numerical models, computational fluid dynamics (CFD) models and reduced-order models (ROMs), are of particular interest to this work. CFD models have been extendedly used as an research and design tool in modern automotive industry [21]. CFD models can predict detailed temperature distribution throughout the cells [22-26] and have been demonstrated as a powerful and practical tool to study and optimize the design of battery systems. In addition, CFD models can also be employed to predict pressure distributions in cases which use air or liquid cooling to achieve heat removal [27, 28]. Nevertheless, the high computational cost of CFD models has limited their applicability for onboard and realtime applications [29], for which ROM are needed. In comparison to high-fidelity CFD models, ROMs reduce the target battery system to a number of discrete lumped elements or states, and predict the thermal behavior at a much lower computational cost, enabling 3

16 onboard and real-time applications [24, 29-32]. There are several challenges in the development of ROMs, including validation and model fidelity. Because ROMs typically involve a series of simplifications and empirical parameters, it is critical to validate them either by high-fidelity CFD models or by experiments. Regarding accuracy, existing ROMs that capture only average temperature may be over-simplified for practical purposes due to the temperature non-uniformity among cells and within a cell. For instance, for cylindrical cells, the core temperature can be up to 10 o C higher than the surface temperature under some operation conditions [33]. ROMs that capture both the cell surface and core temperatures can predict the maximum temperature more accurately [34] and hence are more valuable for practical use. Based on such a fundamental understanding and the corresponding models derived, efforts in the second category focus on specific methods to effectively manage the heat dissipated from the batteries during operation, and various thermal management models and techniques have been proposed [35-37]. These techniques attempted to optimize the thermal management of batteries by adjusting one or more of these following variables: cooling flow pattern [7], the cooling medium [34, 38-42], and the control strategy. The goals of these techniques are to maximize the cooling effectiveness and/or to minimize the temperature non-uniformity and fluctuations. In order to effectively manage the heat dissipated from the cells during operation, various battery thermal management systems have been developed in recent years, which can be broadly divided into three categories: passive thermal management systems, active thermal management systems, and active thermal control systems. Passive thermal management systems remove heat generated by battery cells by natural convection using 4

17 air or liquid [43-45] alone or in combination with phase change materials (PCMs) [39, 46-49], heat pipes[50-53], or hydrogel [54] to enhance cooling efficiency. Active thermal management systems remove heat using forced air or liquid convection [38, 43, 55, 56]. Some active systems can also heat the battery systems in low-temperature environment [9] as energy storage system of ground vehicles may need to be operated under some extreme conditions [57]. Active thermal control systems [58, 59] also remove or heat batteries using forced air or liquid convection, but temperature controllers are deployed in the systems to control the temperature of the batteries to a certain setting point using closed-loop control strategy Scope of this work Based on the above understanding of past fundamental and developmental work, it is the goal of this dissertation to design and demonstrate the active thermal control of battery modules, such that the battery temperature can be reliably managed within an allowed range and the parasitic power consumed by the cooling system is minimized. Firstly, to understand the thermal behavior of the battery module, it is necessary to test the battery module experimentally and to model the battery module numerically. In this dissertation, the two categories of modeling techniques, CFD modeling and reducedorder modeling mentioned above have been employed to analyze the thermal characteristics of the battery module. The models are applied with the specific geometries and boundary conditions of the battery modules, and compared with the experimental data generated by the experiment facilities. Secondly, as mentioned above, EV and HEV applications often feature highly dynamic thermal loads during operation. As a result, passive thermal management 5

18 systems are usually not be sufficient or optimal since the maximum temperature can exceed the acceptable range under peak thermal load. A simple, yet non-optimal, approach is then to operate a passive cooling system always on its maximum capacity, resulting in energy waste under low thermal loads. The cooling problem is further complicated by temperature non-uniformity that exists among cells in a multiple-cell module and within a cell (e.g., between a cell s core and surface). Therefore, it is the goal of this work to explore active control strategies to optimize the cooling performance of battery systems under highly dynamic thermal loads. To accomplish this goal, an active thermal control system using reciprocating cooling airflow was developed and presented with experimental demonstration. By following these two steps mentioned above, the ultimate goal of this dissertation is to develop novel thermal management approaches which can achieve higher energy efficiency by reducing power consumption used in the thermal management system, while the temperature of cells can be managed within the optimal range and the temperature non-uniformity can be minimized at the same time Organization of the dissertation The subsequent chapters of this dissertation are organized as follows: Chapter 2 describes the development of experimental setup and the models used in this research. The experimental setup consisted of battery modules, a wind tunnel, and the corresponding diagnostics and control instrumentation. The models developed included a CFD model for high-fidelity simulation and a ROM for active monitoring and control purposes. The comparison of simulated results with experimental data is also presented in this chapter. 6

19 Chapter 3 presents the parametrization of plant and controller variables, which is an important preparatory work for the active control system design and demonstration. In Chapter 3, the parameterization algorithm is introduced. A Luenberger observer design is presented for both a single cell and also for multiple cells, and representative results of parameterization are provided. Based on the preparatory work in the previous chapters, Chapter 4 reports the optimization and active control of battery modules. First, geometric optimization was performed using the CFD model described in Chapter 2 to determine the optimal geometry and cell configuration to achieve low temperature rise, low parasitic power consumption, and low temperature non-uniformity. Then an active temperature control system was designed based on the ROMs to manage the temperature within the optimal range. The active control was demonstrated experimentally, and the experimental data obtained were compared to simulations results. Chapter 5 finally summarizes and concludes the dissertation, with some recommendations for future efforts. 7

20 Chapter 2. Description of experimental setup and models This chapter focuses on the description of the experimental study and models used in this work. Past work has predominately relied on modeling and numerical simulations, and experimental work has been relatively limited. Therefore, a unique aspect of this work is the combined use of wind tunnel testing and numerical simulations, so that simulations and experiments can be compared under well controlled air flow velocities. The experimental facility developed in this work is capable of testing various battery packing geometries, under different charging and discharging currents, with various cooling air flow velocities. The cooling air flow is generated by an open jet wind tunnel, with flow velocity controllable in the range of 0.5 m/s to 30 m/s (corresponding to ~1.1 to 67 miles per hour), encompassing conditions expected in most practical applications. A corresponding CFD model was developed to analyze the experiments. The CFD model analyzed the problem as a 2D conjugate heat transfer problem. Based on the experimental and CFD results, a ROM was developed for active control and monitoring purposes. The ROM reduced the target battery system to a number of discrete lumped elements or states, so that the governing equations were linearized as a set of ordinary different equations (ODEs). As a result, the ROM can predict the temperature dynamics of battery cells with reasonable accuracy at significantly reduced computational cost, enabling the prospect of active monitoring and control. For the rest of this chapter, Section 2.1 describes the experimental arrangement. Section 2.2 describes the CFD models and Section 2.3 describes the ROMs of the battery modules. In Section 2.4, experimental results were compared with results generated by the ROMs and CFD models. Section 2.5 summarizes and concludes this chapter. 8

21 2.1. Experimental setup Figure 2.1. Illustration of the experimental facilities. Panel (a). schematics of the experimental setup. Panel (b). layout of cells and thermocouples in the battery pack. Figure 2.1 illustrates the experimental facility used in this work. Panel (a) shows the overall schematic of the experimental arrangements, and Panel (b) the dimensions and configuration of one example of the battery modules, including the position of thermocouples used in the tests. As shown in Panel (a), the experimental facility consists of an open jet wind tunnel, a customized battery module as elaborated in Panel (b), and a suite of diagnostics. The open-jet wind tunnel has a m test section and operates with a velocity range of 0 to 30 m/s. The air flow in the wind tunnel was generated by an AF-600 GE fan, and the flow velocity was controlled by adjusting the speed of the fan. A Pitot tube at the exit of the settling chamber measures the real-time freestream velocity. The battery module consists of a customized enclosure and multiple Li-ion cells. The customized enclosure houses the Li-ion cells to be tested, as shown in 9

22 Panel (b). The enclosure was fabricated using plexiglass for its good machinability and sufficient mechanical strength. The windward and leeward sides of the enclosure are open and serve as the inlet and outlet of the cooling air flow. The horizontal distance (i.e., distance in the x direction) between the battery cells and the enclosure walls can be adjusted from 3 mm to 17 mm so that the effects of wall-cell distance on cell temperature and pressure drop can be studied during the tests. The cells used in this study are A Li-ion cylindrical cells (2.3 Ah 3.3V), and Panel (b) shows eight of these cells assembled in a 2P 4S configuration (where P stands for parallel connection and S stands for series connection). The cells are 62.5 mm in height and mm in diameter. The gaps between both ends of the cylindrical cells and the enclosure were minimized so that end effect was reduced. The battery module was tested in the wind tunnel under air flow velocities of 0 m/s, 1.0 m/s, 2.5 m/s and 5 m/s. During the tests, the module was rated at 4.6 Ah and 14.8V with a voltage cut-off limit of 8.0 V. Note that this project focused on generating a basic dataset for model validation under well controlled conditions, and therefore the experiments were intentionally designed to avoid many practical complications. These complications include the geometry of batteries (battery cells of different geometry may have different heat transfer and fluid mechanics properties), layout and packaging of cells, and the various charging and discharging cycles. The diagnostics outfitted to the test facility include a Pitot tube (to monitor the inlet air flow velocity as mentioned before), thermocouples, a pressure rake, and a charger and discharger to set and monitor the current and voltage of the cells during tests. As shown in Panel (b), five K type thermocouples labeled as T0-T4 were placed in 5 different locations in the module at a height of 30 mm to monitor the temperature of the 10

23 inlet air and the cells during tests. The thermocouples were calibrated before use and the accuracy of these thermal couples was 0.3 C under a room temperature of 20.0 C. The pressure rake has a total of 28 parallel probes to measure the total pressure at various points simultaneously. In this example, the pressure rake was set at a height of 40 mm and was placed facing the outlet of the battery module enclosure to measure pressure distribution near the outlet. Two data acquisition (DAQ) systems were used in the tests to log temperature and pressure data. The first DAQ system was a NI 9213 temperature measurement module, which recorded the temperature signal from the thermocouples. In all the tests, temperature measurements were recorded every 10 seconds (the maximum temporal response of the thermocouple is 0.83 ms). The second DAQ system was an Esterline Model 98RK Pressure Scanning System, which recorded the signal from the pressure rake probes. This scanner is capable of measuring 48 different pressure channels simultaneously, and 28 were used in this study. The pressure data were acquired by scanning the pressure rake in several different positions in the x direction and averaged in a 10 second sampling period. During the test, the batteries were charged or discharged at different rates and under different air flow velocities ranging from 0 to 5 m/s (0 m/s simply corresponded to the case where the wind tunnel was turned off). The results are discussed in more details in Section Description of CFD model A CFD model was employed in this work to predict detailed temperature distribution throughout the geometry, including the temperature of the cells. 11

24 Figure 2.2. Panel(a). schematic of problem definition. Panel (b). sample mesh of and around cell 1 at U = 5 m/s. Figure 2.2 provides a schematic illustration of the CFD model used to analyze the thermal management problem in this work. As shown in Figure 2.2a, the cooling air enters from the left, flows across the Li-ion cells, and exits from the right. Each Li-ion battery cell is modeled as a cylinder as shown in Figure 2.2a. Each cell has a diameter D = 42.4 mm and a height H = 62.5 mm (not shown in Figure 2.2) to match the physical dimension of the cylindrical cells used in the experiments. An in-line arrangement was considered in this work, though other configurations can be studied with minimal modification of our existing CFD model and experimental facility. The inline arrangement is characterized by the transverse pitch ( S ) and the longitudinal pitch ( T S ), L which were both set to be 53 mm in this work. Other parameters used in the model include the distance from the inlet of the air flow to the front edge of the first column of batteries ( l i ), the overall length of the battery pack ( l b ), and the distance from the rear 12

25 edge of the last column of batteries to the outlet of the air flow ( l o ). This model empirically set l i = l o = mm, and l b was calculated to be mm with 8 columns of battery cells. This work used the ANSYS/FLUENT 14.0 CFD package to model the thermal management problem as a 2D conjugate heat transfer problem, with the rates of heat release dependent on cell internal resistances. The model considered a fluid zone and solid zone to model the cooling air and batteries, respectively. Uniform quadrilateral meshing method was applied to both the liquid and solid zones, as shown in Panel (b) of Figure 2.2 at an air flow velocity of U = 5 m/s. The adapted boundary layer is shown in the enlarged figure shown on the right in Panel (b), illustrating that the grids near the fluid-solid interface is significantly finer to capture the boundary layer. And in this work, the size of the grids was chosen in such a way that the resultant temperatures are independent of the grid size. Inflation option was used to model the flow boundary layers and to keep the wall y + number within an acceptable range. The y + number, defined below in Eqn. (2.1), was kept around 1.0 in this work as required by the enhancement wall treatment: y y ν τ ρ + w (2.1) In Eqn. (2.1), ν is the local kinematic viscosity, y the distance to the nearest wall (i.e., the thickness of the boundary layer grids here), τ the wall shear stress, and w f ρ f the density of the fluid. For U = 5 m/s case, our meshing method resulted in a total number of 442,133 grids for the 32 batteries cells as shown in Panel (a) of Figure 2.2. For such a 13

26 case, the computational cost was about 4-6 hours using a Dell workstation with six processing cores from two quad-core Xeon 3.20 GHz CPUs. The Reynolds stress and renormalization group turbulence model (i.e., the k- epsilon model) [60, 61] was employed to model the turbulence according to previous CFD studies on tube bank with similar geometries and boundary conditions [62-65]. The boundary conditions used include the velocity inlet, pressure outlet, and no-slip condition at walls and battery surfaces. The Unsteady Reynolds-Averaged Navier-Stokes (URANS) equations were solved by FLUENT to obtain the evolution of temperature distribution, and the Reynolds-Averaged Navier-Stokes (RANS) equations solved to obtain the steady-state temperature distribution. The battery and the air properties used in the CFD model are listed in Table 2.1. The SIMPLE algorithm [66] was used to solve the momentum and energy equations in our CFD model. The second order upwind scheme was applied for both momentum and energy equations. The first order implicit scheme was used for the unsteady simulation. Time steps were chosen in a way that the simulated results were independent of the time steps. For an invariant boundary condition, the time step was 1 second. And if the boundary condition changes during simulation, the time step was reduced to 0.05 second to ensure the convergence and the proper simulation of the real physical condition. To develop accurate numerical models for the battery module thermal management system, the internal resistance of each cell should be measured first so that Joule heat generation can be calculated. In this work, a direct current (DC) method was employed to measure the internal resistance of the cells at different SOCs. In this method, the internal resistance of each cell at a certain SOC was determined in 3 steps. First, the 14

27 cell was discharged by the cycler at a constant current I 1 for a very short period and the terminal voltage was measured as U. Second, the cell was then discharged by the cycler 1 at a constant current I 2 different from I 1 for a short period and the terminal voltage was measured as U 2. In the third step, the internal resistance was calculated using the intermediate results I 1, U, 1 I and 2 U by the following equation: 2 R in U = I U I (2.2) where R is the cell internal resistance at the specific SOC. in Using the 3-step approach provided above, internal resistances of all cells were calculated with measured data. Here the data were measured by a Cadex C8000 Battery Testing System. The voltage accuracy of the testing system is 0.1% full scale range, which is 36 V. The current accuracy is 0.25% full scale range, which is 10 A. Therefore, the uncertainties of the voltage and current measurement are V and A, respectively. Figure 2.3, Figure 2.4 and Figure 2.5 show examples of the internal resistances calculation with experimental results of 4 sample cells selected from the battery module. The cells are marked as C1, C2, C3 and C4 in these 3 figures. Figure 2.3 shows the measured intermediate results used in the DC method, i.e., I, 1 U, 1 I and 2 U for each cell under different SOCs during a discharging process from 2 SOC=90% to SOC=10%. To protect the cells, the SOC of each cell was not cycled to SOC=0. For each cell, I 1 and I were set to be constant as 0.5 A and 5 A at any SOC. 2 The corresponding voltage U 1 and U were recorded using the cycler and presented in 2 the figure from SOC=100% to SOC=10%. Take C1 for example, as C1 was discharged 15

28 from SOC=100% to SOC=0, measured terminal voltage U corresponding to 1 I 1 and U 2 corresponding to I both decreased. 2 Figure 2.3. Experimentally measured voltage for cells under different current at different SOCs during discharge. Using data from Figure 2.3 and the method described in Eqn. (2.2), the internal resistances for all 4 cells at different SOCs during a discharging process were determined and presented in Figure 2.4. As can be seen from this figure, each cell was discharged from SOC=100% to SOC=10%, and the internal resistance of the cell was measured at every 10% of SOC. During the tests, the cells were cooled by the air flow so that the temperature of each cell was maintained lower than 22.9 C. The internal resistances of all cells showed similar trends as SOC decreased from 100% to 10%. Take C1 for example, the internal resistance was 55.1 mohm and decreased dramatically after SOC decreased to SOC=90%. During SOC=90% to SOC=20%, the internal resistance showed 16

29 no significant change, and after SOC became lower than 20%, the internal resistance increased again. Figure 2.4. Internal resistances calculated using measured data under different current at different SOCs during discharge. Inside the batteries, heat conduction was considered only in the radial and azimuthal directions, but not in the axial direction in this 2D model. The rate of heat release (Q) from each battery cell is calculated by: Q It RT 2 = () ( ) (2.3) where I represents the charging or discharging current (which can vary with time, t), and R the electrical resistance of each cell (which depends on the cell temperature, T). Using the equation given above and the data given in Figure 2.3 and Figure 2.4, heat generation during an experiment was calculated and shown in Figure 2.5 for illustration purposes. In this figure, the heat generation at a specific time was derived from the measured internal resistance data, and then the heat generation curve as a function of time was fitted to a polynomial to facilitate its implementation in the CFD 17

30 model. As can be seen from the figure, the heat generation first decreased in the beginning, and then increased towards the end for each cell. Such trends were in agreement with the trends of the internal resistances where the time from 0 to 720 s corresponded to the SOC from 90% to 30%. A user defined function (UDF) was created and applied to model the variable heat generation in the CFD model. Figure 2.5. Heat generation calculated by experimental data for all cells used in CFD modeling. Another important variable in the CFD model is the overall heat transfer coefficient (h), and the values of h generated by our CFD model were in overall agreement compared to those archived in the literature [67] as shown in Figure 2.6. For comparison purpose, a region containing 8 cells in a row as shown in Panel (a) of Figure 2.2 was selected. Different free stream velocities U ranging from 0.02 m/s to 20.0 m/s were tested to cover the range of Reynolds number (Re) where past data exist. Here, the Re number is customarily defined based on the maximum velocity ( U max ) as: 18

31 ρ fumax D Re = where Umax m f ST = U S D T (2.4) where µ f is the dynamic viscosity of the fluid. With U ranging from 0.02 m/s to 20.0 m/s, the Re number ranges from 266 to 266,000, covering the laminar flow regime to fully developed turbulent regime. To match the conditions in [67], the upper wall and lower wall were set to be symmetric boundaries. Again, the model adjusted the boundary layer grid thickness so that the wall y + number was kept around 1.0 across all Re numbers. Figure 2.6. Comparison of Nu number obtained by CFD model against past experiments and empirically correlation. Figure 2.6 compares Nusselt number (Nu) obtained by the CFD model against past experiments and empirically correlations published in [67]. Here, the Nu is defined as: 19

32 where hd Nu = (2.5) k k f is the heat conductivity of air, and h the overall heat transfer coefficient. Here, h is defined as shown below using a log-mean temperature difference (LMTD) [68]: f M fcf( Tf, o Tf, i) ( T8 Tf, o) ( T1 Tf, i) h= and TLMTD = nπ DH T T LMTD 8 Tf, o ln( ) T T 1 f, i (2.6) where M is the mass flow rate of air, T, is the average temperature of air at the outlet, f f o T f,i is the average temperature of air at the inlet (25 C in this research), n is the number of cells in a row (n = 8 for the results shown in Figure 2.6), and T 1 and T are the average 8 temperature of the first cell and last cell in the row, respectively [7]. According to [67], the Nu number can be correlated to Re number as defined in Eqn. (2.4), the Prandtl number of the fluid under the inlet conditions (Pr) and near the wall of the batteries ( Pr ), and the relative traverse and longitudinal pitch (defined as a=s T / D and w b=s L / D ). The results shown in Figure 2.6 were obtained with a=b=1.25, and the correlations developed in [67] reduce to Eq. (5) under these conditions: Nu = 0.51 Re Pr ( Pr / Pr ), when 10 < Re < w Nu = 0.27 Re Pr ( Pr / Pr ), when 10 < Re < w (2.7) As shown in Figure 2.6, the Nu numbers obtained by the CFD model agreed reasonably well with the experimental data and empirical correlations. Quantitatively, the maximum discrepancy between our CFD results relative to the correlation was around +30% for Re < ~1000 and -30% for Re > ~1000 (significant transition phenomenon occur near Re=1000). The modeling of heat transfer from cylinder banks is still not completely understood, and the agreement shown in Figure 2.6 was within the scatter of existing data 20

33 [7]. Other parameters and properties used in the calculation of Figure 2.6 are listed in Table 2.1. Table 2.1 Parameters and thermophysical properties used in the CFD model. Battery properties Air properties Density (kg/m 3 ) ρ c = Density (kg/m 3 ) ρ f = Heat capacity(j/kg K) C c = Heat capacity(j/kg K) C f = 1007 Heat conductivity(w/m K) k c = 32.2 Heat conductivity(w/m K) k f = Mass per cell (kg) M=0.3 Dynamic Viscosity(Pa s) µ f = After the model was developed, it was used to simulate temperature distribution of the battery module. Figure 2.7 and Figure 2.8 show a set of benchmark results for an 8P 4S battery module. Figure 2.7 shows the temperature non-uniformity simulated by the CFD model under two representative inlet air velocities U = 5 m/s (Panel (a)) and U = 1 m/s (Panel (b)) in a battery pack consisting of 32 cells, as configured according to Figure 2.2. The batteries are assumed to be 100% charged initially and start discharging at 10 C (25.2 A). Under these configurations, the volumetric air flow rates are, respectively, m /s and m /s. The simulation was performed assuming that the temperature of the ambient air and the cooling air is stable at 25.0 C. The initial temperature of the cells is also assumed to be 25.0 C. Significant temperature non-uniformity can be observed from these results. Batteries near the inlet of cooling air have lower temperature than batteries near the outlet, which is intuitive because the cooling air is being heated gradually as it passes the battery cells. Under U = 5 m/s, the highest battery temperature is ~28.6 C (a 21

34 temperature rise of 3.6 C ). Under U = 1 m/s, the highest battery temperature is ~37.0 C (a temperature rise of 12.0 C ). Figure 2.7. Temperature non-uniformity simulated by the CFD model under two representative inlet air velocities: Panel (a). U= 5 m/s. Panel (b). U = 1 m/s case. Figure 2.8 further illustrates the temperature rise and the temperature nonuniformity. Here, the maximum temperature on each cell for cells 1-16 (as denoted in Figure 2.7) is shown. These results clearly show the trend of increasing cell temperature and temperature non-uniformity as the cooling air flows downstream. For instance, under U = 5 m/s, the eighth cell s temperature is higher than the first one by ~0.5 C ; and under U = 1 m/s, the eighth cell s temperature is higher than the first one by ~3.5 C. Also, the highest cell temperature occurred at the seventh row and/or the eighth row (as shown in these results), which is an observation that valuable for the development of the reduced-order model to be described in Section 2.3. Furthermore, Figure 2.8 indicates that the cells next to the wall (cells 1-8) have higher temperature than cells in the central area (cells 9-16) both because of the insulating wall assumption made here and because of the boundary layer effects near the wall. 22

35 Figure 2.8. Temperature non-uniformity among different cells in the battery pack at steady state. Panel (a). U = 5 m/s case. Panel (b). U = 1 m/s case. To summarize, a CFD model has been described and applied to study the temperature distribution and temperature non-uniformity among cells in a battery pack. In the next section, ROMs are elaborated and applied to study the temperature dynamics during operation Description of ROMs This section describes ROMs used in this dissertation. Two ROMs, a static ROM and a dynamic ROM, are described in this section. The static ROM was developed by the author to quickly estimate the temperature of cells in a battery module. The dynamic ROM, on the other hand, was derived from existing models with some adjustments and improvements for use in this work. The dynamic model was employed to model the thermodynamic characteristics of the battery module and to aid the design of the control system. 23

36 Static ROM In a simplest reduced-order model (named a one-zone model here), all the battery cells can be considered to have the same temperature as in the analysis of overall performance of tube-bank heat exchangers [68]. However, as shown both by experimental and CFD results discussed above, there are significant temperature nonuniformities among cells in a pack and the hot spots in the pack are of particular interests. Therefore, such a one-zone model is not suitable and an improved model is desired to predict the maximum temperature in the pack. An ideal reduced-order model should be able to capture the temperature of each cell in a battery module during operation. However, in this work, we found that a relatively simple static 2-zone model could predict the maximum temperature in a battery module with very low computational cost. The two-zone model is based on the observation made from the benchmark experimental and CFD results. As Figure 2.7 and Figure 2.8 indicate, the highest temperature occurs in the rows that are nearest to the outlet, i.e., the seventh and eighth rows in the cases of Figure 2.7 and Figure 2.8. Therefore, a reduced-order model is developed, as shown in Figure 2.9, to divide the battery pack into two zones. The second zone includes the last two rows of battery cells, and all the cells in this zone are assumed to have the same temperature ( T max ). The rest of the rows of batteries are included in the first zone, whose temperature is assumed to be uniformly T S. The cooling air will first flow through the first zone, and the air exiting the first zone (at a temperature of T mid ) will enter the second zone and provide cooling to the cells in this zone. The air finally exits the second zone at a temperature of T f, o. 24

37 Figure 2.9. Schematics of the two-zone static reduced-order model. Since each zone is assumed to have a uniform temperature, all the cells in each zone can be analyzed in a lumped fashion as detailed in [68]. Take zone one for example, the analysis will be performed in the following steps. First, the log-mean temperature difference (TLMTD) is calculated by: T LMTD ( TS Tmid ) ( TS Tf, i ) T and S Tmid p Dnh exp( ) = = T T ln( ) T T ρ US C T T S mid S f, i f T f S f, i (2.8) Here, h is calculated using Eqn. (2.6), with the Nu number obtained from the correlation in [67] as shown in Eqn. (2.7). There are three unknowns ( T mid, T, and s T ) LMTD in these two equations. As such, one more equation is needed and this equation is provided as the following equation: 2 Q hπ DHTLMTD I R = = (2.9) Eqn. (2.9) is derived by applying the conservation of energy to zone one under steady state, where the heat transferred out of zone one is balanced by the heat generation by the batteries. This current model is being extended to incorporate unsteady charging and discharging currents. Solving the three equations in Eqns. (2.8) and (2.9) yields T, mid 25

38 T, and s T LMTD. Then T is used as the inlet temperature for the second zone, and similar mid calculations as described in Eqns. (2.8) and Eqn. (2.9) are performed to obtain T max and T f, oof the second zone. Figure 2.10 compares the T max obtained from the above two zones against those calculated by the high-fidelity CFD described in Section 2.2 for a case with eight rows of cells. These results were obtained under the same configurations as those used in Figure 2.7 and Figure 2.8. For the CFD results, the highest and lowest cell temperature were plotted to show the range of temperature variation among cells. The comparison was made under various freestream velocities ranging from 0.2 to 10.0 m/s, corresponding to an Re range of 2,660 to 133,000, encompassing the range of practical interests. Panel (a) of Figure 2.10 shows the comparison of T max from the two zone model compared to the CFD results, and Panel (b) shows the difference between T max obtained from the twozone model versus the maximum cell temperature obtained from the CFD model. Satisfactory agreement is observed from Figure 2.10 between the reduced-order model and the CFD model in terms of the maximum temperature. The two-zone model accurately captures the reduction of the maximum temperature as freestream velocity increases, which suggests its usefulness for predicting the maximum temperature in the battery packing. The computation cost involved in the two-zone model is negligible, making it promising for in situ analysis and active control purposes. Panel (b) shows that the two-zone model overestimated the maximum temperature at Re=2,660 (U=0.2 m/s) by ~ 3 C and underestimated the maximum temperature for other Re numbers. The agreement between the reduced-model and the CFD model improves with increasing Re number, though this is mainly caused by the fact that the temperature of the cells in the 26

39 battery pack becomes increasingly more uniform (which approaches the inlet air temperature) as the Re number increases. Our explanation for the underestimation by the two-zone model for at large Re numbers is that the CFD predicts a lower h comparing to the correlation used in the two-zone model, as shown in Figure 2.6. The CFD model predicts lower h, resulting in less efficient heat transfer and therefore higher cell temperature. Figure Comparison of the maximum cell temperature predicted by the two-zone model and the CFD model under different Re numbers. Panel (a). Tmax predicted by the two-zone model, and the maximum and minimum temperatures predicted by the CFD model. Panel (b). discrepancy of the maximum cell temperature obtained from the two-zone and the CFD model. Figure 2.11 shows the temperature of the cooling air at the exit of the battery pack predicted by the two-zone model and the CFD model. Panel (a) compares the exit temperatures, and Panel (b) shows the difference between these exit temperatures. As can be seen, the temperature difference is generally less than ±1 C except at the lowest Re number simulated (Re = 2,660). Our explanation for the large disagreement at Re = 2,660 is that the disagreement between the h predicted by the CFD model and that by the Zukauskas correlation is large near the transition regime (especially in Re range

40 2000) as shown in Figure 2.6. These results further confirm the validity and accuracy of the reduced-order model. Moreover, the exit temperature of the cooling air is an important parameter for the design of the cooling loop. Figure Comparison of the flow outlet temperature predicted by the two-zone model and the CFD model under different Re numbers. Panel (a). Tout predicted by the two-zone model and the CFD model. Panel (b). discrepancy of the maximum outlet flow temperature obtained from the two-zone and the CFD model Dynamic ROM The schematic and notation of the dynamic ROM are shown in Figure The ROM for a cell (indexed as cell i in a multi-cell module) are described by the following systems of ordinary differential equations (ODEs) [29, 34]: T = 2 c c, i IRin CT T T T c, i s, i R ci, si, s si, = Qf, i Rc CT c (2.10) (2.11) Q T T s, i f, i 1 f, i= (2.12) Ru 28

41 where C is the cell core lumped heat capacity, T ci, the cell core temperature, I the c current applied to the cell, R the internal electric resistance of the cell, T, the cell in si surface temperature, R c the thermal resistance between the core and the surface in the cell, C the cell surface lumped heat capacity, Q f, i the rate of heat removal by the s cooling air flow, T f, i 1 the temperature of the incoming airflow to cell i, and R u the thermal resistance between the cell surface and the cooling flow. This model is a twostate model with T ci, and T si, being the two states. To extend the above model to reciprocating flow, where the direction of the cooling flow alternates periodically, the definition of Q f, iin Eqn. (2.12) needs to be modified to the following equations [7]: Q f, i T = T T Ru T R s, i f, i 1 s, i f, i u when cooling flow is from left to right when cooling flow is from right to left (2.13) T T T f, i= RC u f T T Tf, i 1 = RC u f s, i f, i 1 s, i f, i when cooling flow is from left to right when cooling flow is from right to left (2.14) where C f is the lumped heat capacity of the cooling airflow. Here Eqn. (2.14) is used to calculate the downstream fluid temperature of each cell in a battery module. It is important to point out the assumptions and simplifications made in the above ROM. First, the same cell parameters (i.e., C, c C, s R, c R and u R in ) were assumed to all the cells in the module. Second, it was assumed that the thermal behavior of one row of cells is representative of other rows in the module (essentially assuming that the module 29

42 has a large number of rows and/or the heat transfer effects near the wall of the module enclosure are approximately the same as those between two rows of cells) [8]. Third, heat transfer among cells via conduction through tabs and wires was also neglected. Fourth, though both Joule heat and reversible entropic heat loss contribute to the heat generation of each cell, only Joule heat characterized by the applied current load and internal electric resistance was modeled. Lastly, as can be seen from this discussion, the ROM essentially reduced the original problem into a simplified heat transfer problem by lumping many details of the fluid dynamics and heat transfer into a reduced set of parameters such as C f and C. s Figure Schematic of the ROM for a battery module and the notations used in its development Comparison between experiments and models In this section, simulated results from the CFD model and the ROMs are compared with corresponding experimental data under different conditions. For the CFD model, both temperature and pressure distribution results were compared. For the ROMs, 30

43 however, only the temperature of cells was compared as the ROMs do not capture detailed thermal and aerodynamic data as the CFD model does. Figure 2.13, Figure 2.14 and Figure 2.15 show the direct comparison of the experimental temperature against the results obtained from the CFD model described in Section 2.2. Figure 2.13 shows the comparison between two sets of experimental temperature data against CFD results under U=0 m/s (i.e., with the wind tunnel turned off). The experimental data were obtained under 1.5 C discharging and an ambient temperature of 22.1 C. The CFD simulations were conducted under conditions to match those in the experiments. Results in all 3 figures show good agreement between the measured temperature and that simulated using the CFD model. Temperature ( o C) (a) Cell number: Cell 1 U=0 m/s Discharge rate: 1.5 C (b) Cell number: Cell 2 U= 0 m/s Discharge rate: 1.5 C Time (s) Experimental data CFD simulation Experimental data CFD simulation Figure Comparison between experimental data and simulations from the CFD model under U=0 m/s. Panel (a). temperature of Cell 1. Panel (b). temperature of Cell 2. Besides showing the agreements between the CFD model and experiments results under various conditions, Figure 2.14 and Figure 2.15 also illustrate the effects of wallcell distance. Results in both figures show that the temperature rise is lower with a larger 31

44 wall-cell distance of 17 mm compared to that of 5 mm. A larger wall-cell distance allows more air flow to pass through the battery cells to dissipate the heat generated by the cells. Temperature ( o C) Experimental, 5 mm wall-cell distance CFD, 5 mm wall-cell distance Experimental, 17 mm wall-cell distance CFD, 17 mm wall-cell distance Cell number: Cell 1 U=1.0 m/s Discharge rate: 1.5 C Time (s) Figure Comparison between experimental temperature datasets and simulations from the CFD model of Cell 1 under different enclosure wall-cell distance with U=1.0 m/s. Temperature ( o C) Experimental, 5mm wall-cell distance CFD, 5mm wall-cell distance Experimental, 17mm wall-cell sistance CFD, 17mm wall-cell distance Time (s) Cell number: Cell 2 U=1.0 m/s Discharge rate:1.5 C Figure Comparison between experimental temperature datasets and simulations from the CFD model of Cell 2 under different enclosure wall-cell distances with U=1.0 m/s. The discussion in the previous paragraphs focuses on the comparison of temperature rise obtained from CFD models and experimental tests. The next a few 32

45 paragraphs discuss the comparison of pressure drop obtained from the CFD model and experiments, and the implications for pump power. Figure 2.16 shows a set of measured total pressure (relative to the atmospheric pressure) near the exit of the enclosure in comparison to the CFD simulations under U= 2.5 m/s and an enclosure wall-cell distance of 5 mm. The experimental total pressure data were acquired by the pressure rake described in Section 2.1. As shown in Panel (b) of Figure 2.1, the position of x=0 mm corresponds to the location of the symmetry plane. Therefore, the pressure rake was only traversed in the positive x direction. Figure 2.16 also shows the positions of one column of the battery cells (the right column) and the wall. The configuration of these tests is shown in Figure 2.1. The separation between the two columns of battery cells was 10 mm and the diameter of the battery cells was ~26 mm. Therefore, the right column of cells was placed between x = 5 mm to 31 mm as illustrated in Figure 2.1. For the results shown here, the wall-cell distance was 5 mm and the thickness of the wall was 5 mm, so the wall was placed between x = 36 mm and 41 mm. Figure 2.16 shows a region of high total pressure near x=0 mm due to the high velocity air flow going through the gap between the two columns of cells, a region of low total pressure behind the batteries (between x = ~ 5 mm to 31 mm) due to the wakes formed behind the batteries, a region of high total pressure between the gap of the batteries and the wall (between x = ~31 mm to 36 mm), a region of low total pressure behind the wall (between x = ~ 36 mm to 41 mm) due to the wakes behind the wall, and lastly a region where the pressure gradually transitions to the ambient condition (for x > 41 mm). Figure 2.16 also shows the results generated using the CFD model under the 33

46 corresponding experimental conditions. As shown, the CFD and experimental results are in good agreement. 5 4 U=2.5 m/s Wall-cell distance:5 mm Total pressure (Pa) Experimental data CFD simulation 0 Cells Wall Position (mm) Figure Comparison between experimental and simulated pressure under U=1.0 m/s. Figure 2.17 shows another set of comparison between the CFD and experimental results under U=5.0 m/s, again with a wall-cell distance of 5 mm. The results here have the same peaks and valleys along the x direction as those observed in Figure 2.16, and the overall magnitude of the total pressure is generally higher than that shown in Figure 2.16 because of the higher air inlet velocity. The CFD and experimental results are again in reasonable agreement, confirming the validity of the CFD model U=5.0 m/s Wall-cell distance: 5 mm Total pressure (Pa) Experimental data CFD simulation 0 Cells Wall Position (mm) Figure Comparison between experimental and simulated pressure under U=2.5 m/s. 34

47 The rest part of this section compares the simulated results of ROMs with experimental data shown in Figure This work implemented a series of experimental measurements for validation using the setup shown in Figure 2.1. The diagnostics used are the same as those discussed in Section 2.1. Temperature ( o C) (a) T5, U= 0 m/s (b) T5, U= 1.0 m/s (c) T1, U=5.0 m/s Experimental data ROM Simulation Experimental data ROM Simulation Experimental data ROM Simulation Time (s) Figure Comparison between experimentally measured temperature and temperature predicted by the ROM. Measurements have been performed in a range of operation conditions in terms of the charging current, cooling air velocity, inlet air temperature, and geometry and layout of the cells in the module. Then corresponding simulations were performed using the ROM described above for comparison. Results obtained from the ROM using the parameters provided in [29] generated relatively good agreement with the experimental measurements. As mentioned above, Figure 2.18 shows an example set of experimental measurements compared to model simulations. The experimental results were shown in symbols, illustrating the temperature profile measured at various locations under three inlet flow velocities (U=0, 1.0, and 5.0 m/s). These measurements were made by 35

48 discharging the batteries at 1.5 C-rate. As can be seen from these results, there are some small discrepancies caused both by the assumptions and simplifications made in the model as discussed above, and also by experimental uncertainties such as the accuracy of the thermal couple and a small variation of the inlet air temperature during the tests. But overall, the experimental measurements and simulations using the ROM are in excellent agreement under the operation conditions that we have tested Summary and conclusions In summary, this chapter describes the experimental setup and the corresponding numerical models used to predict the temperature of the battery module in this work. The development of the experimental facility and the models lays the ground work for the active control of the battery modules. The experimental facility developed in this work is capable of testing various battery packing geometries, under different charging and discharging currents, with various cooling air flow velocities. The corresponding CFD model was developed to analyze the experiment results with high fidelity. Based on the experimental and CFD results, a ROM was developed for active control and monitoring purposes. The ROM reduced the target battery system to a number of discrete lumped elements or states, thusly significantly reducing the computational cost while being able to predict the temperature dynamics with reasonable accuracy. As a result, the ROM developed here plays an important role in the ultimate active monitoring and control of the batteries. 36

49 Chapter 3. Parameterization of plant and controller parameters 3.1. Recursive least squares algorithm in discrete time domain Chapter 2 introduced both the static and dynamic ROM developed for the purpose of active temperature monitoring and control. This chapter discusses the parameterization of the dynamic ROM and the development of the observer-based control strategy for the actual implementation of the active monitoring and control. The accuracy of the model parameters affects the active temperature control system in two ways. First, the accuracy of the model parameters directly affects the fidelity of the ROM, which is used to both guide the design of the controller and evaluate its performance by simulation. Second, the accuracy of the model parameters also affects the performance of the observer, which is required for the estimation of quantities that are difficulty or infeasible to measure in practice (i.e., the cell core temperature in this dissertation). The parameters of the models are often calculated based on the specific geometries, layout, and physical properties of the cells [34]. Due to the diversity of cell types and their properties, this method often leads to significant uncertainty beyond the acceptable range [8]. Therefore, it is important to perform parameter identification, and this work performed parameter identification for the ROM in an online fashion [8, 29] based on real-time data. We first describe the online parameterization using a discrete algorithm for a single cylindrical cell without control system. A recursive least squares algorithm as described in [69] was applied here. The observer for battery modules used in Section 3.2 was developed based on the single-cell observer described here. Online parameterization scheme can automatically identify model parameters without human intervention and is 37

50 capable of tracking time-variant system parameters. Studies on cylindrical cell parameterization using online identification algorithm has been reported in the continuous time domain [8, 29, 70]. In this work, an identification algorithm in the discrete time domain was developed. Compared with the algorithm in the continuous time domain, the discrete algorithm is relatively simpler for two reasons. First, the continuous algorithm requires a Laplace transformation to derive the parametric model while the discrete algorithm does not. Second, the continuous algorithm also requires a filter to make the observation and the regressors proper, while the discrete algorithm does not require such a filter. The discrete online parameterization of a single cell was performed by discretization of Eqn. (2.10), (2.11) and (2.12) using the Euler difference approach as shown below: ( T T ) 1 1 C = I R + T T cin,, cin,, 1 2 c n in sin,, cin,, t Rc Rc (3.1) ( T T ) C = ( + ) T + T + T sin,, sin,, 1 s sin,, cin,, f, i 1, n t Rc Ru Rc Ru (3.2) where n is the index of the discretized time step. By manipulating Eqn. (3.1) and (3.2) at time step n and n-1 to eliminate T cin,, and Tcin,, 1, the following discrete estimation model is obtained: T = at + at + ai + at + at (3.3) 2 sin,, 1 f, i 1, n 2 f, i 1, n 1 3 n 4 sin,, 1 5 sin,, 2 where a, 1 a, 2 a, 3 a and 4 a are system parameters derived from 5 C, c C, s R, c R and u as defined in the ROM. The other terms on the right hand side of Eqn. (3.3) (i.e., Tf, i 1, n, R in T, f, i 1, n 1 I, Tsin,, 1, and Tsin,, 2) are the model regressors and inputs to the least square 2 n 38

51 algorithm that is used to estimate the system parameters (i.e., a, 1 a, 2 a, 3 a and 4 a ) in 5 real-time. Once these system parameters are obtained, in combination with two given parameters C and c C, the target parameters in the ROM (i.e., s R, c R and u R in ) can be eventually calculated. Note that the discretization scheme is different with different flow directions. It is relatively straightforward to derive the discretization scheme under different flow directions based on the discretization of one cell as discussed here Luenberger observer design for 1 single cell Based on the parameters identified above, an observer can then be designed to monitor the cell core temperature T ci, (which is difficult if not feasible to measure) by using the signals which are more readily available (which are T si,, T f,0, and I), and this work used the Luenberger observer for this purpose. The ROM developed in this work is a linear system, which can be written in the following generalized format in matrix notations: x = Ax + Bu, y= Cx+ Du (3.4) x= = = T 2 T [ Tc Ts], u [ I T ], y T f s 1 1 RC c c RC c c A = ( + ) RC c s Cs Rc R u Rin 0 C c B = 1 0 RC u s C = [0 1] D = [0 0] (3.5) 39

52 where x and y are the actual system states, x and y the estimated states and output respectively, and A, B, C and D the model parameter matrices. The Luenberger observer for such a linear system takes the form of the following equation [71]: x = Ax + Bu + L( y y), y = C x + Du (3.6) where L is the observer gain. This Luenberger observer is a closed loop observer as a feedback term L ( y y) is included in the equation. For an open loop observer, the speed of convergence between the actual state and the estimated state is the same as the transient response, which is too slow to be applied in practical states estimation. Compared with an open loop observer, the closed loop Luenberger observer can accelerate the convergence of the estimated states to those of the real plant under unknown initial conditions [72] and therefore was used here. The model parameters in the Luenberger observer were acquired in a real-time fashion from the online parameterization described by Eqn. (3.1), (3.2) and (3.3) during operation. The Luenberger observer used here was an adaptive observer, i.e., the parameters were updated in each time step. Figure 3.1 illustrates the scheme of the adaptive observer using the real-time estimated parameters. The input current I, the inlet air flow temperature T f, i 1, and the measured surface temperature were inputs for the parameter identification. It should be noted that among all the 5 system parameters in this scheme, only three of them ( R, two ( C and c c R, and u R in ) were identified at each time step, the rest C ) were assumed to be known parameters since they usually change only s within a small range in practical applications. The adaptive Luenberger observer then adopted the identified parameters to estimate the unmeasurable state T ci, in a closed-loop 40

53 fashion, as the estimated T ci, was corrected using the errors from the feedback loop as shown in Figure 3.1. Figure 3.1. Scheme of the online parameterization method and structure of the adaptive Luenberger observer. Simulations were first performed to verify this discrete scheme of online parameterization based on recursive least squares for a single cell. For comparison purposes, simulations were performed both on a single cell system with an adaptive Luenberger observer and a similar system with an adaptive open loop observer. The initial cell core and surface temperatures ( T ci, and T, si ) were both set to be at 26 o C in these simulations, and the initial guess of the two states assigned to the Luenberger observer and the open loop observer were both 30 o C. The current input profile used here is shown in Figure 3.2, obtained by rescaling the Urban City Assault Cycle (UAC) profile for HEVs [73]. The profile of another measurable input, the inlet air flow temperature T f,0, is shown in Figure

54 Current (A) Time (s) Figure 3.2. Rescaled UAC load profile used in this work Inlet flow temperature ( o C) Time (s) Figure 3.3. Inlet cooling air flow temperature for simulation of online parameterization The key results from these simulations are shown in Figure 3.4, which compares the surface and core cell temperatures estimated by the adaptive Luenberger observer to those of the real plant. As can be seen from the results, the core temperature estimated by the adaptive Luenberger observer converged to the value of the real plant rapidly. Once 42

55 convergence was achieved, the observer was able to provide estimations that matched accurately with the real plant values, which confirms the validity of the discrete online parameterization algorithm. It should be noted that the error between the estimated results and the real values of T si, was negligible (so that the lines representing the real and observed values essentially overlap in Figure 3.4 after convergence is achieved) because the cell surface temperature T si, was a measurable state. In contrast, the error between the estimated and real values of the cell core temperature T ci,, which was taken to be immeasurable in these simulations, was slightly larger but still acceptable. Figure 3.4. Comparison of the cell core and surface temperature of the real plant against those estimated by the adaptive Luenberger observer Luenberger observer design for multiple cells Based on the above observer design for a single cell, here we describe the development of an adaptive observer for estimating the cell core temperatures in a battery module consisting of multiple cylindrical cells. There are several different approaches 43

56 that can be used to estimate the core temperature in a multi-cell systems based on the observer for a single cell. A straightforward approach would be to apply the single cell observer directly for each individual cell in the battery module. Such an approach would require two sensors on each cell to measure its surface temperature and the upward airflow temperature, and is therefore infeasible in practical modules, which usually have a large number of densely packed cells. Hence, this work adopted another approach in which an observer was designed for a row of cells in the target module, and it is then assumed that the core temperature estimated for the cells in this row is representative for other rows. This approach essentially invokes the same assumption that as in the ROM described in Section 3.2, i.e., that the module has a large number of rows and/or the heat transfer effects near the wall of the module enclosure are approximately the same as those between two rows of cells. This assumption has been examined and justified from previous measurements and high-fidelity CFD calculations [74]. Practically, this approach minimizes the number of sensors required as long as the observability can be fulfilled [29]. For the sake of simplicity and clarity, here we describe the observer for a row of two cells cooled by unidirectional airflow. Based on this illustrative description, the observer for more cells and reciprocating flows should become self-explanatory. For a row of two cells, the linear model and the Luenberger observer take the same form as Eqn. (3.4) and (3.6), respectively. When one temperature sensor is implemented on the surface of the 2nd cell, the states, input, output and parameter matrices are given as: 44

57 x= [ T T T T ], u= [ I T ], y = T T 2 T c,1 s,1 c,2 s,2 f,0 s, RC c c RC c c ( + ) 0 0 RC c s Cs Rc R u A = RC c c RC c c ( + ) RCRC u s u f RC c s Cs Rc Ru Rin 0 C c 1 0 RC u s B = Rin 0 Cc (1 ) RC u f RC u s C = [ ] D = [0 0] (3.7) Then an observer described by Eqn. (3.5) and (3.7) is capable of estimating the core temperature in each cell. The observability of this model can be examined by its observability matrix Q as given by [71]: C = CA Q... n 1 CA (3.8) where A is the system matrix, C the output matrix, n the number of states for the statespace model of battery module. If the rank of Q (written as Rank(Q)) equals to n, the state-space model is observable. For the 2-cell model, A and C are given by Eqn. (3.7) and n is 4. It can be easily calculated that Rank(Q)=4, and therefore the 2-cell model is observable. 45

58 It is straightforward to extend the above model for a row of two cells to a row of more cells. However, in practice, with increased number of cells in the row, the number of sensors needed also increases and their locations need to be adjusted to ensure the observability. For instance, for a row cell with 8 cells as examined in the next section, at least three temperature sensors are needed to measure the surface temperature of three cells in the row [8]. By applying two observers with each observer monitoring the airflow from each direction, this multi-cell observer is extended to the case of reciprocating flows Conclusion In summary, this chapter investigated the parametrization of plant and controller parameterization of the ROM, which is the preparatory work for active control system design described in the next chapter. A recursive least square parameterization algorithm is introduced. Then a Luenberger observer was designed to estimate cell core temperature, which is difficult or infeasible in practice are presented. The design of the Luenberger observer was reported for both a single-cell and for module with multiple cells. 46

59 Chapter 4. Design optimization and active control Chapter 2 has described the experimental facilities, the corresponding CFD model, and the ROMs. Both the experimental facility and the models are developed to capture and analyze the thermal behavior of battery modules, laying the ground for seeking the optimal design and control strategies. Chapter 3 discussed the parameterization of the plant and control parameters of the ROM, a key step towards the active control and monitoring. Based on the preparation work described in chapters 2 and 3, this chapter reports the design optimization of battery modules and the experimental demonstration of active control. Section 4.1 reports the geometric optimization to achieve better cooling efficiency. Section 4.2 and 4.3 report the experimental demonstration of the active control of battery modules and the analysis of the experimental results Geometric optimization of battery modules Benchmark results In this subsection, 6 different configurations of battery modules for 2 types of cells are presented for study. The diagrams and geometric parameters of these are named as MC1 (where M stands for module and C stands for cylindrical), MC2, and MC3 for modules consisting of cylindrical cells, and MP1 (where M stands for module and P stands for prismatic), MP2 and MP3 for modules consisting of prismatic cells, respectively. Detailed specifications of the cells and modules of MC1 and MP1 are listed in Table

60 Table 4.1 Parameters and properties of cells and modules for MC1 and MP1. Single cell parameters Battery module parameters Cell type Cylindrical Prismatic Configuration index MP1 MC1 Dimension (mm) Ø Geometric configuration 1P 4S 8P 2S 2L Capacity (Ah) Dimension (mm) Voltage (V) Capacity (Ah) Resistance (mω) Voltage (V) Weight (kg) Power at 3 C- rate (W) Power at 10 C- rate (W) Energy density (Wh/m 3 ) Heat generation (W/m 3 ) Energy density (Wh/m 3 ) As can be seen from the table, the cylindrical cells were the same type of cylindrical LiFePO cells as discussed in Chapter 2, i.e. A123 ANR26650 cells. Each 4 cylindrical cell was Ø26 mm 65 mm in dimension. Each cell had a capacity of 2.5 Ah and a nominal voltage of 3.3 V. The energy density of a cylindrical cell was 179,050 Wh/m 3. Prismatic cells were also used in the study and they were LFP-G20 cells manufactured by AA Portable Power Corp. The LFP-G2O prismatic cells were also a type of LiFePO cells. Compared with the cylindrical cell, each prismatic cell had a 4 larger volume and capacity. Each prismatic cell had a mm cuboid shape, a capacity of 20.0 Ah, a voltage of 3.2 V, and an energy density of 141,000 Wh/m 3. These calculations of heat generation were made under the condition that an individual 48

61 prismatic and cylindrical cell output the same amount of power. For each individual cell, the capacity of a prismatic cell was 20 Ah, larger than that of a cylindrical cell (2.5 Ah). Therefore, when MP1 and MC1 were cycled under the same C-rate at the module level, every single prismatic cell had a higher current than every single cylindrical cell, as every 8 cylindrical cells were connected in parallel in each row to share a current value, which was equivalent to the current value applied on one single prismatic cell. Hence higher heat generation was formed in the prismatic cells when the module-level C-rates were the same for MP1 and MC1. Due to the differences in capacity and volume of individual cells, the modules studied in this work were designed with different sizes of gaps between the individual cells to make the modules have the same overall energy density. It should be noted that the internal resistances shown in Table 4.1 were experimentally determined by averaging over a range of the state of charge (SOC) following the approach described in [7]. Prismatic modules were assembled so that they have comparable voltage, current and dimensions as the cylindrical modules. For example, MC1 and MP1 were designed such that they had the same overall dimension, capacity, voltage, power at 3C rate, and energy density. To match these parameters, MP1 was designed to consist of 4 prismatic cells, in a geometrical layout of 1P 4S 1L, meaning 4 cells arranged in one parallel row, 4 series columns, and one layer. Under the notations, the geometrical layout of MC1 shown in Figure 1b was 8P 2S 2L. Note that the geometrical layout notation for MP1 here was the same as its electrical layout, while the geometric layout for MC1 was slightly different as its electrical layout, which was 8P 4S. Detailed parameters of MP1 and MC1 are presented in the later paragraphs discussing comparative study of different 49

62 module configurations. With the above parameters matched, the heat generation for these modules was different, and it was one of the goals of this study to study the implication of such difference in heat generation for thermal management. It should be noted that the modules were not comparable in the thermal aspect, as the prismatic cell chosen in this study had much smaller internal resistances than the cylindrical cells. The results reported with MP1 and MC1 in this subsection are referred to as the benchmark results. Results obtained with other configurations will be compared against these results to illustrate optimization of battery module design. Figure 4.1 shows the maximum cell temperature rise obtained by the CFD model for MP1 and MC1 under different cycling rate and cooling conditions. Panel (a) shows the results for MP1, and Panel (b) the results for MC1. The model simulated the temperature of all the cells in the module, and the maximum temperature among all the cells is shown here. In all the simulations, the ambient temperature was set as o T amb = 30 C. Two cooling schemes were studied. The first scheme assumed that the cooling air has the same temperature as the ambient environment, i.e., T o in = Tamb ( = 30 C). The second scheme assumed that the cooling air was generated from air conditioning system and therefore had a temperature lower than the ambient temperature (set to be T in = o 25 C in this work). The initial temperature of all the battery cells was set to be the same as the temperature of the cooling air. The volume flow rate of the cooling air was set to m /s for both cooling schemes to make the calculations more comparable. The airflow velocity was specified at the module inlet in this work. 50

63 Figure 4.1. Maximum temperature rise under different conditions. Panel (a) and Panel (b) of Figure 4.1 show the temperature rise for MP1 and MC1, respectively, under 4 different conditions, formed by a combination of the two cooling schemes and cycling rates of 3C and 4C. Several observations can be made from these data. First, the temperature increased with time under all different conditions, and the rate of increase gradually decreased as the system approaches steady state. Second, under the same cooling scheme, higher cycling rate (4C) resulted in a larger temperature rise than lower cycling rate (3C). Third, under the same cycling rate, cooling air with a higher inlet temperature (i.e., ambient temperature of 30 o C) resulted in a larger temperature rise than cooling air with a lower inlet temperature. The heat removal rate due to convective heat transfer is proportional to the convective heat transfer coefficient, the surface area for heat exchange, and the difference between the cell temperature and the cooling air temperature. Under a fixed heat removal rate (specified by cell arraignment and cooling air velocity), a higher cooling air temperature resulted in a higher cell temperature. Fourth, when the data in Panel (a) are compared to those in Panel 51

64 (b), it can be seen that higher temperature rise occurred in MP1 than MC1 under the same cooling scheme and cycling rate, which is due to the larger heat generation of MP1 as shown in Table 4.1. Therefore, these results show that though the prismatic cells offer several advantages compared to cylindrical cells such as single cell capacity and power, they may pose more severe thermal management challenges than cylindrical cells due to their heat generation and larger dimensions. Figure 4.2 shows the steady-state maximum cell temperature for MP1 and MC1 cycled under different C-rates and different ambient temperatures. The ambient temperatures vary from 30 o C to 60 o C. This work considered ambient temperature up to 60 o C and 70 o C (as in Figure 4.2 and subsequent calculations), as 1) the standard climate condition A1 defined by the North Atlantic Treaty Organization (NATO) ranges from 32 o C to 49 o C, and 2) solar radiation can cause an additional temperature increase of up to 20 o C [75]. Figure 4.2 assumed cooling scheme 1, i.e. the inlet temperature of the cooling air was the same as the ambient environment. To help illustrate the practicability of operation under these conditions, the optimal and tolerable ranges are also plotted on Figure 4.2, which are taken to be 40 o C and 50 o C, respectively, as mentioned in the introductory paragraph. Several observations can be made from these results shown in Figure 4.2. First, it is apparent that under this cooling scheme the maximum cell temperature in the prismatic module (MP1) exceeded the optimal range of 40 o C when ambient temperature is above about 32 o C, and the cylindrical module exceeded the optimal range when the environmental temperature is above ~35 o C. Second, the results also show that the maximum temperature of the prismatic cells exceeded the tolerable range of 50 o C when 52

65 the environmental temperature is above 44 o C, and the maximum cell temperature for the cylindrical cells exceeded this range when the environmental temperature is above 47 o C. These observations illustrate again the cooling challenges under extreme conditions, and furthermore the increased challenge imposed by prismatic cells compared to cylindrical cells as aforementioned. Lastly, note that Figure 4.2 shows the maximum cell temperature, and significant temperature non-uniformity can exist in a multi-cell module. Therefore, it is possible for some cells in the module to still operate within the optimal or tolerable range when the maximum temperature exceeds these ranges. Figure 4.2. CFD results of maximum cell temperature of MP1 and MC1 under Tin=Tamb. In contrast to Figure 4.2, Figure 4.3 shows the maximum cell temperature for MP1 and MC1 under cooling scheme 2, i.e. the inlet cooling air temperature was taken to be 25 o C. From the results in Figure 4.3, it can be seen that the maximum cell temperature now can be controlled under the upper tolerable limit except for MP1 under a 5C cycling rate. These results suggest that the cold air generate from the AC system can 53

66 dissipate the heat generated by the cells effectively to extend the operation conditions of the cells. However, the tradeoff is that using the cold air from the AC system will have impact on the cabin temperature and the power requirement on the AC system, and a system level study is needed to find the optimal balance in this tradeoff. It should be noted that in each curve, though cells were immersed in an airflow of 25 o C, the maximum temperature increased when ambient temperature increased. These results illustrate the effects of heat exchange between the battery modules and the ambient environment. When the battery modules are not assumed to be in perfect thermal insulation from the ambient environment, the peak cell temperature increases as the ambient environment becomes hotter, even though the temperature of the cooling air remains fixed. Figure 4.3. CFD results of maximum cell temperature of MP1 and MC1 under Tin=25.0 o C. To illustrate one aspect of such tradeoff, Figure 4.4 and Figure 4.5 show the results of the required cooling air flow rate under cooling scheme 1 and the corresponding pump power, respectively, such that the maximum cell temperature can be 54

67 controlled within the optimal temperature range (i.e., not to exceed 40 o C ). To obtain the results in Figure 4.4, different flow rates of the cooling air were tried to find one that can result in a maximum cell temperature between o o 39.5 C Tmax 40.0 C < <. Note that the volume flow rate is in logarithm scale in Figure 4.4. By comparing all 6 sets of results, it can be seen that the required volume flow rate for MC1 cycled at 3C rate was the most sensitive to the ambient temperature. Different configurations have different cell parameters and layout, resulting in different degree of sensitivity with respect to the ambient air as observed here. For example, cylindrical cells have larger surface area per volume compared to prismatic cells, and therefore are relatively cooled more effectively at a given cooling flow rate. After the required flow rate is obtained, the pressure drop societyd under this flow rate was calculated to determine the required pump power, which are shown in Figure 4.5. If scaled up for a typical full-scale battery pack for HEV, the required pump power for MC1 at 5C rate under an ambient temperature of 30 o C would be W using the data from Figure 4.5. From one aspect, the results shown in these two figures illustrate the importance of optimizing the tradeoff involved in the thermal management of battery modules using the cold air generated by the AC system, as these results shown that the flow rate and pump power change by many orders of magnitude depending on the configurations of the modules, their cycling rate, and the environmental temperature. Such pump power consumption shown here will be further compounded in practice because many modules are employed in practical vehicles, and the power consumption of the battery cooling system could represent a significant percentage of the vehicle s power portfolio. 55

68 Figure 4.4. Volumetric flow rate required to control the maximum cell temperature within the optimal range for MP1 and MC1. Figure 4.5. Pump power required to control the maximum cell temperature within the optimal range for MP1 and MC1 In summary, this section reports the results obtained with the benchmark modules MC1 and MP1. Various parameters of the thermal management of these benchmark modules were studied under extreme environmental temperature, including the maximum cell temperature, required flow rate of cooling air, and the associate pump power. These results form the basis of the following comparative study. 56

69 Comparative studies of different configurations In this subsection, different variations of MP1 and MC1 are studied to illustrate the optimization of module design as presented in Table 4.2, with all the parameters (denoted as P to 1 P ) listed to specify the geometric layout of the cells. As illustrated by 6 the tabulations of these parameters and the corresponding diagram in Table 4.2, MP2 and MC2 have the same overall volume as the benchmark cases (MP1 and MC1), but with a different layout to study the effects of geometrical layout. MP3 and MC3 have a smaller overall volume compared to the benchmark cases, and as a result, the cell-to-cell and cellwall gaps were reduced. The parameters in the vertical direction for the 6 configurations were: the overall height was set to 165 mm, the gaps between the prismatic cells and the enclosure was set to 15 mm, and the gaps between the top of cylindrical cells and the enclosure was set to 10 mm. Some of these modules (e.g., MC1) are designed to mimic a practical package, while others are designed for comparative and optimization purposes. Figure 4.6 shows the maximum cell temperature for all three prismatic modules under different ambient conditions. The ambient temperature varied from 30 to 70 o C. The effects of both cooling scheme 1 and 2 were investigated again. In these calculations, the volume flow rate of the cooling air was all set to m /s, which was the same as that used in the generation of the benchmark results as shown in Figure 4.1, Figure 4.2 and Figure 4.3 to make these results more comparable. 57

70 Table 4.2 Geometric parameters of battery modules for comparative study in Section 4.1. Configuration index Geometric parameters P 1 (mm) P 2 (mm) P 3 (mm) P 4 (mm) P 5 (mm) P 6 (mm) Volume 3 ( mm ) MP \ MC MP MC MP \ MC Configuration diagrams Several observations can be made based on these results. First, as discussed before in the section discussing benchmark result, the use of cold air from the AC unit can manage the heat dissipated from the batteries more effectively than the ambient air, thusly extending the operation range. Second, these results also illustrate the value of optimizing the geometrical layout of the module. MP2 had the same overall volume as MP1, and the same amount of cooling air (in terms of volumetric flow rate) is used. However, a higher maximum cell temperature was observed in MP2 than MP1 due to the difference in cell arrangement. In contrast, MP3 had the lowest maximum cell 58

71 temperature even though MP3 had the smallest overall volume. Our explanation is that under the particular geometries of these modules and the same volumetric flow rate, the cooling flow had a higher Reynolds number (defined by the cell-to-cell or cell-to-wall distances) in MP3 than MP1 and MP2, leading to an enhanced heat transfer coefficient in MP3. However, such enhanced thermal management effectiveness is at the cost of additional requirement on the pump power due to the increased Reynolds number, as to be discussed at the end of this section. Figure 4.6. Maximum cell temperature for MP1, MP2 and MP3 under cooling scheme 1 and 2. Figure 4.7 shows the maximum cell temperature for all three cylindrical modules. The conditions and observations are similar to those discussed above. The volumetric flow rate of the cooling air remained to be m 3 /s, and MC3 again had the smallest overall volume and the lowest maximum cell temperature because of the same explanation for MP3. In all simulations discussed above, the maximum temperature occurs at different locations for different configuration and cell types. 59

72 Figure 4.7. Maximum cell temperature for MC1, MC2 and MC3 under cooling scheme 1 and 2. Lastly, Figure 4.8 shows the pressure drop and pump power loss for all 6 module configurations under the same m 3 /s of cooling air, with the corresponding temperature data. The ambient temperature was set to 30 o C and the inlet temperature of the cooling air was set to be 25 o C (i.e., cooling scheme 2). Several observations can be made from these results. First, as mentioned above, MP3 and MC3 exhibited a higher pressure drop and a higher pump power requirement compared to other configurations, again due to the increased Reynolds number in the narrower cell-to-cell and cell-to-wall gaps. Secondly, MP2 had a lower pressure drop and lower pump power requirement compared to MP1, and on the other hand, also a higher maximum cell temperature as shown by the results in Figure 4.6. These results illustrate the tradeoff in the design of modules between heat transfer effectiveness and aerodynamic effectiveness. Third, the comparison between MC1 and MC2 illustrates the benefit and importance of design optimization using the high fidelity CFD models. Figure 4.7 shows that MC2 had lower maximum cell temperature than MC1, and Figure 4.8 shows that the pressure drop and 60

73 pump power requirement for these two configurations are similar. Therefore, with the same overall volume, same amount of cooling air, and approximately the same pump power requirement, MC2 is superior to MC1 in terms of cooling effectiveness. Figure 4.8. Comparison of max temperature, pressure drop and pump power for all six configurations. In summary, this section describes a comparative study of various module configurations. Several aspects of the module design and optimization have been studied and compared, including the maximum cell temperature, pressure drop, and pump power requirement. The results illustrate both the importance and complexity of such design and optimization. The design and optimization involve a complicated tradeoff among many aspects, including the overall module volume, cooling effectiveness, and pump power requirements. Results obtained on MC1 and MC2 show that it is possible to balance these tradeoffs to obtain an improved design in all aspects. 61

74 4.2. Active control simulation using ROMs Before discussing the experimental work on the active control of battery modules, this section first reports the simulation results based on the ROM and the observer developed in chapters 2 and 3, respectively. Figure 4.9 shows a diagram of the target thermal management system. A battery module consisting of 32 cells was considered, and the cells were assembled in an 8P 4S configuration. It was assumed that any 1 of the four 4 is representative of the other rows. Therefore, the ROM and observer developed above were applied to perform active control simulation on one row of 8 cells in the module, and the performance obtained on this row was assumed to be applicable to other rows in the module. The cells in this row were labeled as cell 1 to 8 from left to right as shown in the figure. Three temperature sensors were applied on cell 1, 4, and 8 to provide the inputs for the closed loop observer to estimate the core temperatures of all the cells in the row. Based on the estimated core temperatures from the observer, a control module was employed to regulate the maximum cell core temperatures by controlling 1) a set of flip valves to adjust the direction of the cooling airflow, and 2) the automotive air conditioning system and a fan to adjust the flow rate of the cooling airflow. The primary goals of the control scheme was to actively adjust the direction and flow rate of the cooling airflow so that 1) the core temperature of all the cells is regulated within a preset range under a dynamic load, 2) the temperature non-uniformity among the cells is minimized, and 3) the amount of cooling flow consumed is minimized (therefore minimizing the power consumption from the air condition and fan system). 62

75 Figure 4.9. Diagram of a battery thermal management system with active temperature control and reciprocating cooling flow. Figure 4.10 shows the block diagram of the observer-based control scheme. The estimated core temperatures of cell number 1, 4 and 8 were sent to the control module. The highest of the three temperatures was used as a feedback signal and compared with the preset temperature. The difference was sent to a hysteresis controller to control the flip valves, the fan, and the air conditioning system. When the highest cell core temperature estimated by the observer is higher than the preset limit, the controller activates the fan and the air conditioner to generate cold airflow to cool the cells. When the highest cell core temperature estimated by the observer falls below the preset limit, the controller deactivates the fan and the air conditioner to stop the cold airflow. When the fan and air conditioner are activated, the controller also adjusts the flip valves to switches the direction of the cooling flow according to a periodic timing signal. The inlet temperature of the cold air generated by the air conditioner was set to remain fixed. 63

76 Figure Block diagram of an observer-based control system. A series of simulations were performed to investigate the cooling performance based on the above scheme, and the key results are summarized in Figure 4.11 to Figure To illustrate the usefulness of the active control scheme and reciprocation flows, these simulations were performed under the following six conditions for comparison purposes. The first condition is labeled as HYS+RECP (dt=100), under which both the hysteresis control and the reciprocating cooling flow were applied. The reversion interval (dt) for the reciprocation flow was set to 100 s. The second condition is labeled as HYS+RECP (dt=500) and it is the same as the first condition except the reversion interval was set to 500 s. The third condition is labeled as HYS+UND, under which hysteresis control and unidirectional cooling flow were applied. The fourth condition is labeled as NC+RECP (dt=100), under which no control was applied and the reciprocating flow was applied at a reversion interval of 100 s. The fifth condition is labeled NC+RECP (dt=500), and it is the same as the fourth condition except that the direction reversion was set to 500 s. The sixth condition was labeled NC+UND, under which no control was applied and the flow was unidirectional. The sixth condition represents the simplest case and serves as the benchmark for comparison. For all six 64

77 cases, simulations were made for a same duration of 5000 s. The initial core and surface temperature of all cells were set to 26 o C, and the temperature of the cold air generated by the air conditioner was set to 26 o C. The standard UAC load shown in Figure 4.5 was used, and the control target of the maximum core temperature was preset to be within 39.5 and 40.0 o C. Under the above conditions, Figure 4.11 first demonstrates the observer can accurately estimate the core temperature for multiple cells. Then Figure 4.12, Figure 4.13, and Figure 4.14 compare the performance of these six cases in terms of amount of cooling air consumed, temperature non-uniformity among cells, and temperature fluctuation over time. Figure 4.11 shows a comparison of the cell temperature of the real plant against that estimated by the observer. The comparison was made for cells 1, 4, and 8 under the first conditions, i.e., HYS+RECP (dt=100). Each panel of this figure shows 4 sets of data: the core temperature of the real plant (e.g., labeled as Tc1 in Panel (a) for Cell 1), the core temperature estimated by the observer (e.g., labeled as Tc1_ob in Panel (a) for Cell 1), the surface temperature of the real plant (e.g., labeled as Ts1 in Panel (a) for Cell 1), and the surface temperature estimated by the observer (e.g., labeled as Ts1_ob in panel (a) for Cell 1). The observer estimated both the surface and core temperature so accurately that the estimated temperature overlaps almost completely on top of real plan temperature as shown in Figure 4.11 for all cells. Such accuracy exhibited in these results confirms the validity of the observer for a multi-cell system. Furthermore, since the active control was turned on in this simulation, these results also illustrate that the observer can estimate the core temperatures of multiple cells accurately in an active control system. 65

78 Temperature ( o C) (a) Cell 1 Tc1 and observed Tc1 Ts1 and observed Ts1 (b) Cell 4 Tc4 Observed Tc4 Ts4 Observed Ts4 Tc4 and observed Tc4 Tc1 Ts1 Observed Tc1 Observed Ts1 Ts4 and observed Ts4 45 (c) Cell 8 Tc8 Observed Tc8 40 Ts8 Observed Ts Tc8 and observed Tc8 Ts8 and observed Ts Time (s) Figure Comparison of cell core and surface temperature of the real plant versus those estimated from the observer for three cells. Figure 4.12a shows the instantaneous flow rate of the cold air under three conditions (only the first 3000 s of data are shown here for clarity). Positive values of the flow rate mean that the flow direction is from left to right, and negative values mean that the flow direction is from right to left. These results show that the sixth case (NC+UND), the simplest benchmark case with no control and a unidirectional flow, behaved as expected. Specifically, after a period of operation, the core temperature reached a certain value. Then the cooling flow was turned on and remained at a constant rate. In contrast, with a simple control scheme (case HYS+UND), the controller actively turned the cooling air on and off, resulting in a reduced amount of cooling air consumed. With reciprocating flow introduced (case HYS+RECP), the controller was able to further fine tune the airflow, resulting in further reduction of the cooling air consumption. Figure 4.12b directly compares the amount of cooling flow consumed by integrating the flow 66

79 rate shown in Figure 4.12a over operation time. Please note that the volumetric flow rates mentioned in this section are calculated based on standard temperature. Figure Instantaneous flow rate and total cooling flow consumption for three different schemes. Figure 4.13 shows the variation in the core temperature for Cell 1, 4, and 8 under the same three cases as shown in Figure The labeling of these results is the same as that used in Figure These results illustrate several key advantages of the active scheme and reciprocating flow. First, it is shown that the combined use of active control and reciprocating flow can reduce the temperature non-uniformity among cells significantly. Volume flow rate (m 3 /s) Consumed cooling air (m 3 ) 20 (a) Time (s) 40 (b) 30 HYS+RECP (dt=100) HYS+UND NC+UND HYS+RECP (dt=100) HYS+UND NC+UND Time (s) 67

80 Figure Temperature variation of cell 1, 4 and 8 for three different schemes. As an example, Figure 4.13a shows that active control and reciprocating flow were able to regulate and maintain the core temperature of Cell 1, 4, and 8 to be almost identical under a dynamic load. Second, these results also show that the combined use of active control and reciprocating flow can significantly reduce the temperature fluctuation of the cells over time. And third, when combined with the results shown in Figure 4.11, the first two advantages were obtained while the amount of cooling air was significantly reduced. Observed cell core temperature ( o C) 40 (a) HYS+RECP (dt=100s) Tc1, Tc4, and Tc8 Observed results: 30 Tc1 Tc4 Tc8 20 Optimal range 40 (b) HYS+UND Observed results: Tc1 Tc4 Optimal range Tc8 40 (c) NC+UND 30 Observed results: Optimal range 20 Tc1 Tc4 Tc Time (s) Figure 4.14 summarizes the above discussion by comparing the performance of all six cases. Figure 4.14a shows that the maximum core temperature was controlled within the preset range in all six cases. Figure 4.14b and Figure 4.14c compare the temperature non-uniformity and total consumption of the cooling air over the 5000 s operation. Here the temperature non-uniformity was defined as the maximum temperature difference within the row of cells at each time step. The minimum 68

81 temperature non-uniformity of 1.0 o C was observed in the HYS+RECP (dt=100) scheme with the minimum amount of cooling air ( m 3 /s ). Increasing the reversing interval of the reciprocating flow to dt=500 s resulted in both increased temperature nonuniformity and consumption of the cooling airflow, illustrating the importance of optimizing the interval. Max cell T ( o C) Total cooling flow (m 3 ) T non-uniformity( o C) (a) Goal of control ( o C) (b) Temperature non-uniformity ( o C) (c) Total cooling flow (m 3 ) HYS+RECP (dt=100) HYS+RECP (dt=500) HYS+UND NC+RECP (dt=100) NC+RECP (dt=500) NC+UND Figure Summary of cooling performance under six schemes. It is also noteworthy from these results that the NC+RECP (dt=100) scheme resulted in a significant reduction in the temperature non-uniformity among cells. Although this case consumed significantly more cooling air than the optimal case (the first case), it consumed about the same amount of cooling air compared to the last two cases, again illustrating the usefulness of the reciprocating flow and the importance of optimizing the reversing interval. Lastly, from the results, it can be calculated that compared with the simplest benchmark case (NC+UND), with the combined use of the 69

82 active scheme designed in this work and reciprocating cooling, the temperature nonuniformity can be reduced from 4.2 to 1.0 o C, and the total amount of cooling flow consumed can be reduced from to m (a 38.5% reduction) Experimental demonstration of active control Using simulation tools, Section 4.2 illustrated the usefulness and effectiveness of active temperature control [76] for battery module thermal management. This section reports the experimental demonstration of the active control of battery modules Experimental setup The experimental demonstration was performed using a 4-cell battery module as illustrated in Figure The experimental setup consisted of three major components: a benchtop wind tunnel, a battery module, and diagnostics and control instrumentations. The wind tunnel consists of two electric fans, an enclosure which formed the tested section, and two flow conditioning sections as shown in Figure The fans were 12 V, 0.84 W electric fans marked as Fan 1 and Fan 2, each had a maximum speed of 1,200 RPM and the speed of the fans were fixed at this maximum speed whenever the fans were operating during the tests in this section. The flow conditioning sections were used to guide and shape the flow from the fan to the inlet of the enclosure (i.e., the test section). The distance between the fan and the enclosure was 150 mm. The enclosure had a cross section area of 60 mm x 60 mm, and a length of 178 mm. The battery module to be tested was housed in the enclosure, and the cooling flow can entered the enclosure either from the left or right (labeled as Inlet 1 and Inlet 2 in Figure 4.15). The battery module consisted of a series of four cylindrical cells (A ) as shown in Figure The cells used in this work had a capacity of 2.5 Ah. The diameter 70

83 and height of each cell were 26 mm and 65 mm, respectively. The four cells were placed in a row inside the enclosure, which are labeled as C1, C2, C3 and C4 as shown in Figure The distance between C1 and Inlet 1, and the distance between C4 and Inlet 2, were both 25 mm. The distance between every two cells was 7 mm, and the distance between the cell and the side wall of the enclosure was 15 mm. There was no gap between the cells top/bottom surfaces and the enclosure s top/bottom walls. Figure Experimental setup of the active temperature control system for the battery module using reciprocating flow. The diagnostics and control instrumentations consisted of 7 K-type thermal couples, a cycler (Cadex C8000 Battery Testing System), and a data acquisition (DAQ) system. The K-type thermocouples were used to measure the temperature at different locations in the experiments, which were marked as T0 to T6 as shown in Figure These thermocouples were calibrated before use and their accuracy was determined to be ± 0.3 C under in the temperature range encountered in this work. The diameter of the thermocouples was 0.25 mm to minimize the disturbance to the flow. During the 71

84 experiments, T0 was used to measure the room temperature, and T5 and T6 were used to monitor the ambient air temperature before they entered the fans. For each cell inside the module, the thermocouples were placed at the middle of the cells in the height direction, toward the side of Fan 2. During the tests, the Cadex C8000 battery cycler was used to charge or discharge the cells, and to monitor and record the voltage and current data of each cell. Figure 4.16a and b show the layout of the hardware and the block diagram of the control system, respectively. As shown in Figure 4.16a, the hardware used here consisted of a personal computer (PC), a NI mydaq Module, and a temperature DAQ system (NI 9213 Temperature Measurement Module). The PC served as a digital hysterics controller as well as a signal monitor, which received feedback temperature signal from the data acquisition system, processed the control strategy, and sent control signals to the fans via the NI mydaq module to control the module temperature actively. The mydaq module had 8 digital outputs, which sent digital signals received from the program running in the PC to 2 L293D integrated circuit chips to control the direction and velocity of Fan 1 and Fan 2, respectively. The NI 9213 data acquisition system was a 16-channel temperature measurement module used to obtain the temperature data from the thermocouples. The shortest temporal response time of the thermocouple was 0.83 ms, and the data acquisition system recorded the temperature every 2 seconds during the tests. 72

85 Figure Hardware and block diagrams of control system. Figure 4.16b shows the active control scheme implemented by the hardware described above during the experiments and simulations. The temperatures of all 4 cells (marked as C1 to C4 in the figure) were converted to digital signal by the NI 9213 and sent to the control module. The highest of the 4 temperatures was used as a feedback signal and compared with the preset temperature. The temperature difference was sent to a hysteresis controller to control the direction and velocity of Fan 1 and Fan 2. When the highest temperature measured was higher than the preset limit, the controller activated the fan to generate airflow to cool the cells. When the highest cell temperature fell below the preset limit, the controller deactivated the fan to stop the airflow. The direction of the air flow was determined by the location of the highest temperature to reduce temperature non-uniformity. For instance, if T2 was found to be the highest temperature, Fan 1 was then activated and Fan 2 deactivated, so that C2 was in the upstream direction of the air flow in the battery module and was cooled more effectively. On the contrary, if T4 was 73

86 found to have the highest temperature, Fan 2 was then activated so that C4 was in the upstream direction of the cooling flow and was cooled more effectively. As simulated in Section 4.2, the goal of this control strategy was to manage the temperature within a preset range (and hence reducing the temperature non-uniformity at the same time) with a minimal amount of parasitic power consumption Experimental and CFD results In this subsection, a series of experiments and corresponding CFD simulations were performed to investigate the cooling performance based on the above scheme shown in Figure 4.16b. To illustrate the usefulness of the active control scheme and reciprocation flows, these experiments and simulations were performed under four different conditions for comparison purposes. Similar to the active control simulations presented in Section 4.2, the four conditions in this section were labeled by abbreviation to simplify the illustration and the labels in the text and figures. Specifically, the first condition was labeled NC+UND, under which no control was applied and the flow was unidirectional. The second condition is labeled as NC+RECP, under which no control was applied and the reciprocating flow was applied at a reversion interval of 60 s. The third condition is labeled as HYS+RECP, under which both the hysteresis control and the reciprocating cooling flow were applied. The fourth condition is labeled as NC+NF, under which air flow cooling was not applied and the cells were cooled by natural convection. These four conditions represent the simplest cases and serve as the benchmark for comparison. In the next a few paragraphs, Figure 4.17 and Figure 4.18 show the experimental and CFD results of battery module cooled under the NC+UND condition, Figure

87 and Figure 4.20 present the experimental and CFD results of battery module cooled under the NC+RECP condition, and Figure 4.21 and Figure 4.22 present the experimental and CFD results of battery module cooled under the HYS+RECP condition. Figure Experimentally measured temperature under NC+UND condition. Figure 4.17 shows a sample set of experimental data taken from the battery module when no control was applied and the flow was unidirectional, i.e., the NC+UND condition. During the test, each cell was discharged from SOC=90% to SOC=30% at a 3C rate (corresponding to a discharging current of 7.5 A for each cell) during a 12-minute-long load period, which started at t=300 s and ended at t=1020 s. Heat was generated in the cells when the load was applied on the module. The module was cooled by a continuous unidirectional air flow with an average air flow velocity of 1.38 m/s generated by Fan 1. As explained above, T0 measured the ambient temperature, and T1, T2, T3 and T4 measured the leeward of C1, C2, C3 and C4, respectively. As shown in the figure, the ambient temperature measured by T0 was about 22 o C. When the current load was applied to the cells at t=300 s, and the temperature of all 4 cells 75

88 increased due to heat generated by the cells. After t=1020 s, the discharging current applied on each cell was reduced to 0, and the cell temperature began to decrease as the air flow generated by Fan 1 cooled the cells. From the temperature results of all four cells, it was found that at any time during the test, the temperature of C1 was the lowest and the temperature of C4 the highest, the reason of which was that C1 was located in the upstream of the air flow generated by Fan 1while C4 was located at the downstream of the flow. Figure Comparison of CFD simulated and experimental results for battery module under NC+UND condition. Figure 4.18 showed the comparison of CFD and experimentally measured temperature results for the cells in the battery module under the NC+UND condition. The conditions for the simulation were designed to match those used in the experimental, i.e., the cells were cooled by unidirectional flow from Fan 1, each cell was discharged at 3C rate simultaneously from SOC=90% to SOC=30%, and the average inlet air flow velocity was 1.38 m/s. From the results it can be seen that simulation results exhibit similar trends as observed in the experimental results, in which the temperature of all 4 76

89 cells increased first when the cells were applied with the current load, and then decreased after the load became 0. Several observations can be made by examining the experimental and CFD results reported above. First, the ambient temperature was stable enough to perform the tests and the test system functioned well. Second, the experimentally measured temperature rise of the cells were no higher than 4.4 o C, indicating the large cooling capability of the NC+UND scheme. Third, the experimental and CFD results are in good agreements for all cells, which indicated that the CFD model can capture the physics of the experiments with reasonable accuracy. Figure 4.19 and Figure 4.20 show the experimental and CFD temperature results of each cell when the battery module was cooled by reciprocating cooling with no control, i.e., under the NC+RECP condition. Figure 4.19 shows the experimentally measured temperature of each cell in the battery module cooled by reciprocating air flow without active control. In this figure, Fan 1 and Fan 2 worked alternatively in every 60s, which resulted in the reversion of the cooling air flow direction in every 60 s. The average air flow velocity is 1.38 m/s when the fans were rotating at their maximum speed. The ambient temperature was about 21.9 o C. All cells were discharged at 3C rate simultaneously from t= 360s to t= 1080s, during which the SOC reduced from 90% to 30%. As can be seen from the results, the temperature of the cells was close to the ambient temperature T0 when no electric load was applied to the battery module. At t= 360 s, the temperature increased due to the heat generation inside the cells as the 7.5 A constant current was applied to the cells. After t=1080 s, the temperature of all four cells decreased as the load reduced to 0 while the reciprocating cooling air flow remained 77

90 cooling the cells. Because the heat generated by the cells was removed by the reciprocating cooling flow, the maximum temperature rise of the battery module during the test was only 4.2 o C. The temperature of each cell was fluctuating due to the reciprocating air flow, which reduced the temperature non-uniformity among all 4 cells. It should be noted that the maximum temperature raise of the reciprocating cooling flow case was slightly lower than that of the unidirectional flow case as the temperature nonuniformity was neutralized by the reciprocating flow. Figure Experimental results of module temperature under NC+RECP condition. Figure 4.20 shows the comparison of CFD and experimentally measured temperature results for the cells in the battery module under the NC+RECP condition. The conditions used in the CFD were again designed to match those in the experiments presented in Figure In this simulation, the cooling air flow direction was reversed every 60 s. The average air flow velocity was set to 1.38 m/s and the ambient temperature was set to 21.9 o C. All cells were discharged at 3C rate simultaneously from t= 360 s to 78

91 t= 1080 s, during which the SOC reduced from 90% to 30%. By observing the experimental data, it can be found that the temperature of each cell was fluctuating due to the reciprocating flow, which is similar to the experimental results shown in Figure The temperature variation of the cells showed a wave pattern, especially for the first and the last cell. Every time the flow changed its direction, the cell temperature rose or fell correspondingly. By comparing the experimental and CFD results, it can be seen that when the boundary conditions of the CFD models were similar to the experimental conditions, the trends of the experimental and CFD results are similar and the results are in good agreements. Figure Comparison of CFD results and experimental data under NC+RECP condition. The good agreement observed between the CFD and experimental results gives us confidence to extend the application of the CFD model to analyze experiments involving active control. In the rest of this section, the CFD model was applied for temperature 79

92 prediction for cases with active control by applying the boundary conditions from the active control experiment. Figure 4.21 and Figure 4.22 show the experimental and CFD temperature results of each cell when the battery module was cooled by reciprocating cooling air with a hysteresis control strategy, i.e., under the HYS+RECP condition. Figure 4.21 shows the experimentally measured temperature of each cell in the battery module cooled by reciprocating air flow combined with active temperature control. The ambient temperature was measured to be 22.0 o C. All cells were discharged at 3C rate simultaneously from t= 240 s to t= 960 s, during which the SOC was reduced from 90% to 30%. Hysteresis control strategy was employed here to maintain the maximum cell temperature to be controlled within a certain range, and in the test corresponding to Figure 4.21, the temperature control goal was set to be 26.5 o C to 27.5 o C. Meanwhile, to reduce the temperature non-uniformity, Fan 1 and Fan 2 worked alternatively so that the air flow was reversing in every 60 s whenever the fans were activated by the controller to produce cool flow. The average air flow velocity was 1.38 m/s when the fans were rotating at their maximum speed. From the results of the figure, it can be seen that the trends of the temperature of the cells were as follows. Firstly, the temperature of the cells remained relatively constant as no loads was applied on the cells during the time t=0 s to t=240 s. Secondly, the temperature of the cells was rising dramatically during t= 240 s to t=690 s, when the current load was applied on the cells while the fans were not activated as the temperature was not larger than the control threshold value 27.5 o C. Thirdly, during the time t= 690 s to t=1240 s, the fans were activated by the control module to regulate the temperature of the cells so that the maximum cell temperature was 80

93 maintained in the range 26.5 o C to 27.5 o C. Since the fans worked in a reciprocating manner, the temperature of the cells was fluctuating during this period. After t>1240 s, the fan stopped as the maximum cell temperature was cooled to be lower than 27.5 o C and the temperature of all cells decreased gradually due to natural convection. Figure Experimental results of battery module temperature under HYS+RECP condition. Figure 4.22 shows the simulated results of the battery module operated under the HYS+RECP condition using the CFD model, which means that the temperature of each cell in the battery module cooled by reciprocating air flow combined with active temperature control. Note that in the simulation with active control, the air flow velocities at both inlets were neither constant nor reciprocating, but irregular values determined from the experiment data, i.e., the air flow velocity inputs calculated using the recorded control signal for Fan 1 and Fan 2 during the test presented in Figure From the simulated results, it can be seen that from t=0 to t=460 s, the temperature of all four cells increased dramatically when the constant current load was applied to the battery module 81

94 while the fans were not activated to generate the cooling flow. After t=460 s, the fans were activated alternatively and intermittently so that reciprocating cooling flow was generated to cool the cells when the temperature of cells exceeded the designed range determined by the hysteresis controller. As a result, the temperature of all four cells fluctuated within the determined range during the rest of the simulation. In summary, the trends of the CFD results were similar to the experimental results shown in Figure The only difference was that the constant current load was applied to the battery module during the entire test, and as a result the simulated temperature did not decrease below the determined range. Figure CFD results for the battery module temperature under HYS+RECP condition. To evaluate the effectiveness of the proposed cooling scheme which combines active control with reciprocating cooling, Figure 4.23, Figure 4.24 and Figure 4.25 summarize the comparison of maximum temperature rise, energy consumption and air consumption in the rest paragraphs in this section. 82

95 Figure Summary of Experimental and CFD data of temperature rise, energy consumption and total cooling flow used under four different conditions. Figure 4.23 shows a summary of maximum temperature rise, energy and total cooling flow consumption for four different cases. From the results, among all four cases the NC+RECP case has the lowest maximum temperature rise, which is 4.2 o C. Compared with the NC+UND and NC+RECP cases, HYS+RECP has a higher maximum temperature rise, which is 4.9 o C, but the difference is not significant. For energy consumption, NC+UND has the highest energy consumption, which is J, and HYS+RECP has much lower energy consumption compared with NC+UND and NC+RECP, which is J. Similar trend can be found in cooling air flow consumption. In summary, the HYS+RECP control strategy reduced parasitic energy 83