Research Paper

Context

UAVs such as delivery drones are equipped with Lithium Ion batteries which have a limited life. There comes a point in the life of the batteries when they need to be replaced because of declining performance and rising maintenance. Predicting the Remaining Useful Life (RUL) of Li-Ion batteries can prevent malfunctions, cost, loss of autonomy, etc.

How to do it?

A Machine Learning algorithm is run across data collected from sensors in the UAVs, the model is trained on the historical data to recognize the point where a battery started declining. Using the trained model the current UAVs are monitored live with their battery health and based on some parameters the current age of the batter along with RUL is predicted.

computers-13-00064-g001-550.jpg

What are the model parameters?

  1. No. of Charge/Discharge cycle (k) (First variable to be predicted)
  2. Voltage, Current, and Discharge Capacity corresponding to K
  3. Cycle of the Fault Threshold (Count of cycles where the fault occurs) The battery’s threshold capacitance is defined. From this information, we can determine the cycle count for the fault threshold.
  4. No. of cycles remaining or RUL (Second variable to be predicted)

How to predict

1. Current number of Charge/Discharge Cycle (k)

Predicting the number of cycles elapsed k is a regression problem. The total number of cycles of the batteries can be extracted from the capacity variation curve per the number of cycles analyzed.

2. Number of cycles remaining or RUL

$$ RUL~=~Cycle~of~the~Fault~Threshold~-~k $$

Algorithms

In the research paper, due to the regressive nature of the task, the following algorithms were used:

  1. Regression Machines for Supporting Vectors OR SVRM (SVM for Regression)
  2. Multiple Linear Regression (MLR)