Remaining Useful Life Prediction of Li-Po Batteries in UAVs Using Extended Kalman Filter

Darielson A. Souza, João L. O. Torres, João P. P. Gomes, Leonardo Ramos Rodrigues, Rômulo Nunes de Carvalho Almeida, Vandilberto Pereira PintoJournal of Mechanics Engineering and Automation 2015 · DOI

TL;DR

This paper presents a simple yet effective method for predicting the remaining useful life (RUL) of Li-Po batteries in Unmanned Aerial Vehicles (UAVs) using an Extended Kalman Filter (EKF) and an exponential degradation model. This is crucial for UAV mission planning and operational safety, enabling better decision-making regarding flight duration and recharge needs.

What Problem Does This Paper Address?

Unmanned Aerial Vehicles (UAVs) are increasingly relying on Li-Po batteries for their power, offering advantages like reduced weight and better power control compared to traditional combustion engines. However, a critical challenge is accurately predicting the remaining useful life (RUL) of these batteries. This information is vital for mission planning, allowing operators to determine how long a UAV can operate before needing a recharge, thus preventing unexpected power loss during critical operations. Existing methods may be too complex or not well-suited for the dynamic environment of UAVs.

Methods

The authors propose a methodology that combines an Extended Kalman Filter (EKF) with an exponential model to predict battery RUL. The core idea is to model the battery's degradation over time.

  • Degradation Model: The paper uses a model where the battery's degradation state ($d[k]$) evolves over time ($k$). This evolution is influenced by a rate parameter ($r$) and a white noise process ($w_1$). The measured battery terminal voltage ($m[k]$) is then related to the degradation state plus another white noise process ($v$).
  • Exponential Behavior: The model assumes that degradation has an exponential behavior, driven by the parameter $r$.
  • Parameter Estimation: Crucially, the parameter $r$ is not assumed to be constant. It is estimated iteratively using a random walk model and a Kalman filter approach. This allows the model to adapt to varying degradation rates.
  • EKF Application: The EKF is used to simultaneously estimate the degradation state ($d$) and the rate parameter ($r$) from the measured battery voltages.
  • RUL Prediction: Once the degradation state and rate are estimated, the model simulates the future evolution of the degradation. The RUL is then predicted as the time until the battery voltage (or a related degradation metric) crosses a predefined failure threshold (set at 12.5 V in this study).

Datasets

The authors used data collected from discharge tests on three Li-Po batteries of the type used in the Gyro-200 ED-X8 UAV.

  • Test Rig: A custom test rig was built to discharge the batteries under controlled, constant current conditions.
  • Discharge Currents: Tests were performed using discharge currents of 1.0 A, 1.5 A, and 2.0 A.
  • Data Collection: Battery terminal voltages were recorded at a sampling rate of one sample per minute until the voltage reached 12.5 V.

Key Insights and Ideas

  • Simplicity and Effectiveness: The paper demonstrates that a relatively simple exponential degradation model combined with an EKF can yield good RUL prediction results for Li-Po batteries in UAVs.
  • Adaptive Degradation Rate: The ability to estimate the degradation rate ($r$) dynamically is key. This allows the model to adapt to different discharge conditions and even variations within a single discharge cycle.
  • Impact of Discharge Current: The quality of predictions degrades as the discharge current increases. This is attributed to faster battery decay at higher currents, resulting in fewer data points for the EKF to work with.
  • Confidence Intervals: The predictions are presented with confidence intervals (99%), providing a measure of uncertainty in the RUL estimates.

Key Results

The proposed methodology provided good RUL estimation results, with failure times predicted within five minutes in advance for most datasets. The predictions were visualized for discharge currents of 1.0 A, 1.5 A, and 2.0 A, showing the estimated RUL against the ideal RUL. The confidence intervals generally encompassed the ideal prediction line, indicating the model's reliability.

Limitations and Future Work

  • Sampling Rate: The authors acknowledge that a higher sampling rate for voltage measurements would likely improve the quality of the RUL estimates, as it would provide more data points for the EKF.
  • Discharge Current Variations: The current study focused on constant current discharge. Future work should explore load forecasting when the battery discharge current changes dynamically during operation, which is more representative of real-world UAV scenarios.
  • Alternative Degradation Models: Investigating other degradation progression models beyond the exponential one could potentially lead to more accurate predictions.

Summary updated

BibTeX
@article{souza2015remaining,
  author = {Darielson A. Souza and João L. O. Torres and João P. P. Gomes and Leonardo Ramos Rodrigues and Rômulo Nunes de Carvalho Almeida and Vandilberto Pereira Pinto},
  title = {{Remaining Useful Life Prediction of Li-Po Batteries in UAVs Using Extended Kalman Filter}},
  journal = {Journal of Mechanics Engineering and Automation},
  year = {2015},
  doi = {10.17265/2159-5275/2015.12.006},
}
Figure