Browsing by Author "Çelik, Mert"

Now showing 1 - 3 of 3

Metadata only
Electric vehicle model parameter estimation with combined least squares and gradient descent method
(IEEE, 2019) Gözüküçük, Mehmet Ali; Uğurdağ, Hasan Fatih; Dedeköy, Mert; Çelik, Mert; Akdoğan, Taylan; Natural and Mathematical Sciences; Electrical & Electronics Engineering; UĞURDAĞ, Hasan Fatih; AKDOĞAN, Taylan; Gözüküçük, Mehmet Ali; Çelik, Mert; Dedeköy, Mert
Energy management algorithms have a crucial role in electric vehicles due to their limited driving range. For an energy management algorithm to be effective, we should model the vehicle as accurately as possible. That is, not only the structure of the model should be accurate, but also the parameters of the model should be accurate. In this work, we take the model of an electric vehicle and tune three parameters in it based on trip data, namely, vehicle mass, air drag coefficient, and rolling resistance coefficient. We do this by using Least Squares method to set the initial guess and then by optimizing the parameters using Gradient Descent. To the best of our knowledge, this is the first work that simultaneously estimates these three parameters. Our work is also unique in the sense that it combines Least Squares and Gradient Descent.
Metadata only
SOC estimation for li-Ion batteries using extended kalman filter with PID controlled process noise according to the voltage error
(IEEE, 2019) Çelik, Mert; Gözüküçük, Mehmet Ali; Akdoğan, Taylan; Uğurdağ, Hasan Fatih; Natural and Mathematical Sciences; Electrical & Electronics Engineering; AKDOĞAN, Taylan; UĞURDAĞ, Hasan Fatih; Çelik, Mert; Gözüküçük, Mehmet Ali
State of Charge (SOC) estimation is critical for battery powered devices in order to find out the remaining charge level. This process is relatively straightforward when the battery is in the resting state. However, it can be challenging while the device is operating, due to the process disturbances and model uncertainties. Various kinds of approaches have already been proposed in the literature like Neural Networks, Kalman Filtering, and Nonlinear Observers. Nevertheless, proposed methods in the literature do not have fast response for initial condition errors. This paper proposes a new implementation of Extended Kalman Filter, which improves the convergence characteristics of states for SOC estimation. The importance of initial condition errors is articulated in this paper, especially from an automotive perspective.
Metadata only
State of charge estimation for lithium-based batteries
(2020-01) Çelik, Mert; Uğurdağ, Hasan Fatih; Akdoğan, Taylan; Uğurdağ, Hasan Fatih; Akdoğan, Taylan; Demiroğlu, Cenk; Poyrazoğlu, Göktürk; Serif, T.; Department of Electrical and Electronics Engineering; Çelik, Mert
This thesis proposes a new State of Charge (SOC) estimation method for lithiumbased batteries, which o ers a good trade-o between convergence and computation times. Lithium-based battery packages are quite common in the automotive industry and beyond because of their high-power density and dynamic response capabilities. Per a given volume, lithium-based battery cells have much more capacity, higher Crates, and lower internal resistance than other cell chemistries. However, this comes at a cost because of lithium's reactive nature. It is hard to preserve, monitor, cool, and control lithium in a pack within a safe state. For these reasons, battery control, or in other words, Battery Management Systems (BMS) is a major topic in the literature, and estimation of SOC, State of Health (SOH), and State of Power (SOP) are considered as core subfunctions of BMS. This thesis focuses on improving SOC estimation for lithium-based batteries. SOC estimation determines the remaining charge level on the battery and is very critical for battery-powered devices. This process is relatively straightforward when the battery is in the resting state. However, it can be di cult while the battery-powered device is operating, due to process disturbances and model uncertainties. The best performing SOC estimation methods in the literature are based on Kalman Filtering, and they are specifically Extended Kalman Filter (EKF) and Adaptive Dual Extended Kalman Filter (ADEKF). While EKF offers the shortest computation time, it results in a long convergence time. On the other hand, ADEKF o ers short convergence time and long computation time. We propose PID-controlled EKF, which o ers a mid-point in terms of convergence and computation times. The importance of convergence characteristics are also articulated in this thesis, especially from an automotive perspective.