Published Papers

All published research from IESBJ. Open access — no subscription required.

4 papers

EKF-Based SOH Estimation and Adaptive Output Control for Early Thermal Runaway Prediction in LiPo Batteries

Lee, Mu-Won

In high-maneuverability FPV drone operations, instantaneous discharge currents reaching tens to hundreds of amperes significantly elevate the risk of LiPo battery thermal runaway. This study proposes a software-based system that combines an Equivalent Circuit Model (ECM) with an Extended Kalman Filter (EKF) to estimate battery State of Charge (SOC) and State of Health (SOH) in real time, and applies Arrhenius reaction kinetics to predict time to thermal runaway. Using the NASA PCoE Li-ion battery dataset (B0005–B0018, 168 cycles), five comparative experiments were conducted: 1RC vs 2RC ECM, Coulomb counting vs EKF SOC estimation, R0 based vs capacity-based SOH definition, temperature threshold vs Arrhenius alert, and fixed vs adaptive output control. Results show that EKF-based SOC estimation achieved 22% lower RMSE than Coulomb counting, and Arrhenius-based alerts were triggered earlier and more stably within each cycle compared to temperature threshold methods. The adaptive output limiter reduced cumulative risk exposure time by 34%, quantitatively demonstrating thermal runaway prevention effectiveness. Furthermore, a physics-based endurance simulator executing a hardcoded FPV freestyle maneuver sequence was implemented, confirming that an SOH=1.00 battery reaches critical state in approximately 10.5 days and an SOH=0.70 battery in approximately 4.0 days.

EKFSOHBatteryThermal RunawayPredictionLipo+7 more

Jun 25, 202608

Read

Analysis of Accumulated Error in Numerical Integration Methods and Its Application to IMU-Based Position Estimation

Lee, Mu-Won

Numerically integrating acceleration data obtained from an IMU (Inertial Measurement Unit) sensor to estimate velocity and position inevitably leads to error accumulation. In this study, the error terms of the Euler method, trapezoidal method, and Simpson’s method are mathematically derived via Taylor series expansion, and the error convergence rate with respect to sampling period ∆t is analyzed for each method. Furthermore, a Gaussian white noise and bias drift noise model based on actual MPU6050 datasheet specifications is implemented in C, and the cumulative error of each integration method is quantitatively compared using Python. Simulation results show that in single integration, the Euler method exhibits O(∆t), the trapezoidal method O(∆t2 ), and Simpson’s method O(∆t4 ) error characteristics, and a structure in which error is amplified quadratically in double integration (acceleration → position) is confirmed. The trapezoidal method is selected as the integration method with the best error eﬀiciency relative to computational cost for real-time embedded systems such as drones, and its rationale is presented mathematically and experimentally.

Ccalculusimunumerical-methodsquadrotorsensor-fusion+7 more

Jun 6, 202607

Read

Autorotating Samara-Inspired Micro Aerial Vehicle with Deflection Flap Control for Safe-Zone Guided Emergency Descent

Lee, Mu-Won

As drones and eVTOL aircraft become increasingly prevalent, crash accidents caused by battery detachment, propeller failure, and multirotor-to-fixed-wing transition failures have emerged as serious urban safety concerns. In particular, mode transitions in eVTOL aircraft often occur at low altitudes of only tens of meters, making conventional parachute deployment practically infeasible. This study designs and fabricates an emergency descent device inspired by the autorotation of maple seeds (samaras), capable of initiating deceleration immediately even at low altitudes. Asymmetric blade pitch angles(5°/35°) are applied to a two-blade propeller to replicate the differential lift of a samara, and servo-driven deflection flaps are added to the blade trailing edges to enable lateral control based on helicopter cyclic pitch principles. In passive mode (power off), dynamic pressure pins the flaps to the blade surface, providing deceleration through autorotation alone. In active mode (power on), an MLX90640 thermal camera detects people on the ground and generates an avoidance trajectory. A simulator was built by linking a C physics engine and a Python visualization server via TCP socket. Validation results showed mean impact kinetic energy of 1.20 J and mean nearest-person distance of 9.15 m, confirming significant deceleration and avoidance performance. A STM32-based control PCB circuit design was also completed, demonstrating feasibility for integration into real aircraft.

dronecrashsamaramaple seedsautorotation

Jun 5, 202601

Read

Rigid-Body Simulation-Based Probabilistic Analysis of Drone Crash Impact Points and Its Embedded System Application

Lee, Mu-Won

This study utilized the C#-based BepuPhysics2 rigid-body physics engine and applied isotropic drag to the drag model to construct a simulation environment for creating a safety device that predicts the impact point in the event of a drone crash. Through this, the probability distribution was analyzed and developed into a C library applicable to embedded systems. To ensure the versatility of the data, altitude, horizontal speed, and vertical speed were set as the main independent variables. An exhaustive search was conducted using a Grid Sweep method to generate foundational data for constructing a LUT (Lookup Table). Simultaneously, to reflect the uncertainty of the actual atmospheric environment, wind speed, wind direction, and turbulence were assigned as Monte Carlo-based random values. By performing repeated simulations, the probability distribution of the impact point under specific flight conditions was obtained. Through this, a fall prediction C library capable of accommodating various aircraft specifications was established.

DroneCrashPredictSimulationRigit-Body

Jun 1, 202606

Read